The year was 2004. The glass-walled tech office in Bengaluru was quiet, long past working hours. It was 11.30 at night on a Tuesday. The release was supposed to go out hours ago, but an issue had surfaced. Without the build, the consultants in Europe—lined up by the product leader to test the release—would be sitting idle. Pressure was quietly mounting.

“This should’ve been released by now,” the product leader said over the phone to Srinivasan Subramanian (Srini), who was our engineering manager then. His tone was tight. “There was no indication of a delay. I didn’t see this coming. I expected better communication. I’ve lined up the consultants to test starting tomorrow.”

Srini had the phone on speaker. He muted it and turned to us — three tired developers, each trying to keep one eye on the code and the other on our manager on the line.

“How much time do you guys need to fix this?”

“Two hours,” my lead said. The two of us nodded.

Srini unmuted the phone. “It’ll be released in the next two hours,” he said.

The product leader paused. “Are you sure?”

“Yes,” Srini replied. “I trust my team.”

He hung up, looked at us and said, “Alright guys, I’m heading home. Call me if there’s anything.”

That was it. No big talk, no frustration, no pressure. Just calm belief.

And we did it. We fixed it, built it, and released it.

That night has stayed with me ever since. It taught me what leadership really is — trust and accountability.
No drama. No speeches. Just the confidence that your team will do what they said they would.

I met Srini recently for this conversation. Over the years, I’ve had many chats with him, some about work and some about life, but this time was different. This time, I was interviewing him. Some conversations make you pause and think a little deeper. This was one of them.

From My Conversation with Srinivasan Subramanian, CTO & Co-Founder, Unvired Inc.

Q: When did the idea of starting up first take root?
Srini: Around 2006 I would say. It wasn’t a grand vision or a business plan drawn on a whiteboard. It began as a silent urge — a need to move, to create, to build something of our own. We had a few projects in mind that we thought could sustain us for six months, maybe a year after we start. That was enough. We said, let’s begin, and we’ll figure it out as we go. What mattered most was building something meaningful, not just running a consulting outfit.

Q: What made you walk away from a stable, well-paying role to start something so uncertain?
Srini: I suppose it came from some kind of restlessness. I was managing people, budgets, plans — but I wasn’t building anymore. The work was comfortable, maybe even enviable from the outside, but comfort has a way of dulling you over time. I missed the thrill of solving problems, of writing code, of seeing something come alive because you created it. Just because you’re good at managing people doesn’t mean that’s the only thing you should do in life. My core was still technology — and I wanted to get my hands dirty again.

Q: Was entrepreneurship something you always imagined, or did it come later in life?
Srini: I think it was in the genes somewhere. My father was a divisional manager at United India Insurance — a stable government job, good position, everything society called success. And then one day, he decided to walk away and start his own consulting company. I must have been in school, maybe ten or eleven. He believed in doing things on his own terms. Watching that taught me it’s okay to leave certainty for conviction. Maybe that seed stayed buried in me for years until it was my turn to take the leap.

Q: You once stepped away from corporate life to chase something of your own, only to return later. What did that experience teach you about yourself?
Srini: Yes, it was in 2000. That phase taught me a lot about timing and self-awareness. When I left, I thought I was ready to build something on my own — but maybe I wasn’t. The stint outside didn’t go as planned, and when I was considered for an opportunity at the company again, I said yes. People often see returning as a step back, but to me it was validation. It meant I had left well — without burning bridges. There’s no greater endorsement than a company wanting you back. It taught me that exits matter as much as beginnings. How you leave often defines whether doors will open again later.

Q: So when did those side projects turn into Unvired Inc— a company with a purpose?
Srini: It happened almost quietly. A few of us had some consulting work that could keep the lights on for six months, maybe a year like I earlier said. But deep down, we knew we didn’t want to be just another services outfit. We wanted to build products, create something of our own. There was no investor deck, no funding, just conviction — that we could do things faster, smarter, and better than large companies weighed down by their own processes. That’s how Unvired was born — out of belief, not a business plan.

Q: The early days of any startup are usually a blur of chaos and hope. What were those first months like for you?
Srini: They were intense and strangely grounding. We had a few projects, mostly helping customers untangle stuck SAP mobile implementations. The momentum felt good, we were solving real problems, earning trust, getting noticed. But about a year in, reality hit. One of our major contracts ended abruptly over a difference in expectations. It was a wake-up call. Suddenly, there was no pipeline, no guarantees. We had to think harder, not just about survival, but about direction. That’s when we started shaping Unvired around products and long-term value, not just short-term work. Those months tested us, but they also made us sharper.

Q: What did you decide to focus on after that turning point? How did Unvired find its footing?
Srini: We realized we were often early — too early, in fact. We were building platforms when the market wasn’t ready for them. Customers didn’t want architecture; they wanted outcomes. That changed how we thought. Instead of talking about frameworks and integrations, we started talking about problems and impact. When you sell a platform, you talk to IT. When you build an application, you talk to business. And business is where conviction lives. That shift, from platform to purpose-built apps, gave Unvired its direction. It made us practical without losing ambition.

Q: Most founders chase funding early on, but you chose to stay bootstrapped. Why?
Srini: We did explore funding once, but it quickly became clear that the ecosystem here was tuned more toward consumer tech, not enterprise software. Investors wanted fast scale and flash. We were building depth. So we decided to grow through revenue instead. It wasn’t easy; in the early days after starting up, there were a couple of times we had to dip into our personal savings just to make payroll. Founders got paid last. That discipline built trust — with employees and clients. In hindsight, I’m glad we stayed bootstrapped. It forced us to think like owners every single day.

Q: You’ve spoken about a difficult period where you had to let people go. What did that experience teach you?
Srini: That was one of the hardest phases of my career. We were going through a downturn, and I had to lay off close to forty people — not because they weren’t performing, but because the business had to downsize. It was gut-wrenching. You build relationships with people, see their families, their hopes, and then one day you have to look them in the eye and tell them it’s over. It taught me that leadership isn’t just about targets and plans — it’s about carrying the emotional weight of your decisions. I think that’s when I realised I wanted to build something where we’d handle people differently, where trust wouldn’t be situational.

Q: What kind of culture did you consciously build at Unvired?
Srini: From day one, we wanted decency to be non-negotiable. In smaller companies, you often hear stories of founders holding back salaries or making exits messy. We were determined to be the opposite. People who leave should remember us for how we treated them, not how long they stayed. Many of our early hires joined as freshers and grew into leaders. That, to me, is success — when people build their careers here and still speak fondly of the place after they move on. We run a tight ship, but never a fearful one. Trust and respect are what keep it afloat.

Q: You’ve seen technologies come and go — mobility, cloud, now GenAI. What keeps you curious after all these years?
Srini: The problems. They never stop changing. Every new wave—whether it was mobile back then or AI now, throws fresh puzzles at you. Some days I think, why am I solving this again? But truth is, that’s what keeps me alive. Technology has kept me sharper than I was ten years ago. Every day, someone walks in with a new challenge, a new unknown, and you have to figure it out. It’s frustrating, yes, but it’s also the best kind of fuel. The day I stop being curious is the day I’ll stop being relevant.

Q: How do you see AI and GenAI shaping the future of enterprise software?
Srini: I think AI’s biggest strength is also its biggest weakness — it’s powerful, but unpredictable. In enterprise systems, you can’t afford that. You need determinism, explainability. So for me, AI isn’t a standalone product; it’s a supporting layer. It should make processes smarter, not replace them. The future isn’t about AI taking over—it’s about AI working quietly behind the scenes, making humans better at what they already do. The real shift will come when people stop selling “AI solutions” and start selling better outcomes powered by AI. That’s when it’ll mature.

Q: Can you share a moment when an AI project didn’t go as expected — something that changed how you think about technology?
Srini: One that stands out was a reconciliation tool we built for a customer. It matched bank statements with invoices using AI. Technically, it worked beautifully. But it failed in reality, because the client didn’t want to change their process. They wanted the same manual workflow, just faster. That’s when it hit me: technology alone doesn’t transform anything. People and processes have to evolve with it. If you plug AI into an outdated system, you’re not modernizing; you’re just automating the mess. That lesson still guides how we build today.

Q: You often talk about problem-solving as the core of engineering. How do you mentor your teams to think that way?
Srini: I tell them not to come to me saying, “something’s broken.” Come and tell me what broke, why you think it happened, and what you’ve tried so far. Half the job is defining the problem well. Most people rush to fix things without understanding them. I use a simple analogy—think of your problem like an equation: X + Y = Z. If you have too many unknowns, you can’t solve it. Reduce the variables. Make some assumptions. Fix what you can, then go after what you can’t. That applies to debugging, to business, and to life.

Q: How do you approach decision-making, especially when the data isn’t clear?
Srini: I’ve always believed you make the best decision with the data you have at that moment. It’s like taking a snapshot of a moving system, you act on what’s visible, knowing that things will evolve. If the data changes, you course-correct. I don’t waste time regretting yesterday’s call; it was the right one for the context I had. Too many people freeze, waiting for perfect information—it never comes. Leadership is about moving when others hesitate, and adjusting when you must. Progress comes from momentum, not paralysis.

Q: You’ve seen co-founders and early colleagues move on. How do you handle those moments as a leader?
Srini: With acceptance. People leave for their own reasons - different life stages, financial needs, family priorities. You can’t hold that against them. What I do wish is that people would talk before they decide, not after. Most conversations about leaving happen once the decision’s already made. I’ve learned not to take it personally. If someone’s mind is set, the best thing you can do is help them exit gracefully. Everyone deserves to leave with dignity. And I’ve always told my team this — leave in a way that they’d want to hire you back. That’s the true measure of how you’ve conducted yourself.

Q: You’ve often spoken about integrity and gratitude in how people leave a company. Why does that matter so much to you?
Srini: Because how someone exits says more about them than how they performed. I’ve seen people leave quietly and with grace, and I’ve seen others walk out entitled, thinking the world owes them something. The truth is, no company is perfect — but if you’ve learned, grown, and built relationships, that’s worth acknowledging. Gratitude isn’t weakness; it’s maturity. Even when someone moves on, I want them to look back and say, “I learned something there.” That means we did something right. In the end, integrity is currency — it travels with you long after the paycheck doesn’t.

Q: After all these years, what does success mean to you now?
Srini: It’s not money, not anymore. Money comes and goes. What stays is the satisfaction of having built something that lasts — and the people who grew along the way. Some of our earliest hires joined as freshers and are now directors, heading teams and running operations. That’s what makes me proud. You realise after a point that your true impact isn’t in the code you wrote or the deals you closed, but in the careers you’ve shaped. If people leave this place wiser, more confident, and still willing to call when they need advice — that’s success.

Q: You’ve mentored several generations of engineers. What do you think has changed the most in young professionals today?
Srini: I think the sense of entitlement has grown faster than the hunger to learn. Many young engineers want instant results — bigger pay, faster growth, everything now. Some even walk in with multiple offers, treating jobs like transactions. But the best ones, the ones who truly grow, still value craft and patience. What worries me isn’t skill — it’s attitude. People underestimate the value of consistency, humility, and curiosity. The industry doesn’t reward noise; it rewards reliability. If you stay grounded, there will always be space for you — no matter how fast technology changes.

Q: Looking back, is there anything you would have done differently?
Srini: I’ve learned not to dwell on regrets. Every decision made sense with the data I had then — that’s all any of us can do. But if I had to pick one thing, maybe I’d have started Unvired’s presence in the U.S. earlier. Being based in India was both a blessing and a challenge. It gave us resilience, but it also slowed us down. We survived longer — sometimes because we could, not because we should. If we’d been in the U.S., we might have failed faster, learned quicker, maybe even grown bigger. But then again, maybe we wouldn’t be who we are today. Every detour built some part of the company’s character — and mine.

Q: What would you tell young founders or engineers who dream of building something of their own?
Srini: Don’t wait for the perfect plan. It never comes. Start small, but start. Be ready to get your hands dirty and your ego bruised. Learn to solve problems, not just chase trends. And most importantly, leave every place, every person, a little better than you found them. Because in the end, businesses are built on trust — and trust is built on how you show up every single day. If you can keep that intact, the rest will find its way to you.

Q: And when you look ahead — after all the storms, pivots, and reinventions — what does the future look like for you and Unvired?
Srini: I see it as a steady ship in moving waters. Technology will keep changing — AI, automation, whatever comes next — but our compass stays the same: build honestly, treat people right, and solve problems that matter. We may not always have the loudest marketing or the biggest funding rounds, but we’ll have a story that lasted. For me, that’s enough. The goal now isn’t speed — it’s substance. To keep building, keep learning, and maybe, inspire a few more people to believe that you can stay ethical, stay small if you have to, and still make a dent.

Final Thoughts from Our Conversation

When our chat ended, Srini leaned back in his chair and smiled the same way he did that night years ago — the night he trusted three developers to pull off a release past midnight. Nothing about him had changed. The same steadiness, the same belief in people, the same confidence that good work always finds its way.

As I walked out, I realised Unvired Inc wasn’t built on slogans or pitch decks. It was built on a way of working — show up, do what you say, and let the results speak.

I’ve worked with many leaders in my career, but if I had to pick one who defined what leadership truly feels like, it would be Srini. No contest. He didn’t lead from the front or push from behind — he simply stood beside you, made you believe you could, and somehow, you did.

There’s something about the mirror.
It never lies. It doesn’t flatter. It just stares back and asks, “Are you doing what truly matters?”

For product leaders, that question can be uncomfortable.
We spend our days discussing strategy, velocity, customer feedback, and release plans. But every now and then, it’s worth pausing and turning the lens inward. The toughest questions aren’t the ones your CEO/CPO asks in a review — they’re the ones you ask yourself when the noise settles.

Here are a few that have stayed with me.

Am I solving a real problem or just shipping features?

In my past, feature requests came in thick — from customers, sales, support, executives, and engineers. Everyone had ideas. Every sprint had velocity goals. Every release had expectations.

But the real measure for a product manager is simple: what problem is this feature trying to solve?

Yes, every ask might address a problem, but not every problem deserves solving right now. Prioritization becomes essential in terms of reach and impact. Think RICE. And yes, you’ll have to disappoint a few. That’s part of the role.

It’s not about pleasing everyone. It’s about focusing on the real problems that matter.

What does success look like, and are we measuring what matters?

I’ve seen teams spend hours assembling dashboards filled with beautiful metrics — most of which meant very little. I’ve done it myself.

Sometimes product managers couldn’t even explain why a specific KPI existed. “Leadership wants to see it” became the default answer. But the better question is: what does success mean to you?

For me, it comes down to one word — value.
Whatever you’re building, ask: does it co-create value with the customer? Are we measuring outcomes or just outputs? When success starts with value, the metrics take care of themselves.

Am I empowering my team, or am I the bottleneck?

I’ve been guilty of this one. I’d bring in new initiatives, ideas, and frameworks — each exciting, each promising to take us from X to Y.

The team would nod in quiet agreement, some resisted, but eventually accepted. Soon burnout crept in. We were drowning in initiative overload.

That’s when I learned that sometimes less is more. What should we not be doing to move us forward faster?
Decision-making needs to be decentralized. The role of a leader is not to push ideas down but to create an environment where ideas rise up.

Empowerment isn’t freedom without direction. It’s clarity without control.

How am I adapting to change and preparing for the future?

Change is uncomfortable because it often means rework. And rework feels like lost progress. But the real test of leadership is how you respond when the landscape shifts.

How flexible are you? How rigid are you?
Do you resist change or reframe it with new data?

The best teams operate on two lenses — one focused on today for today (short-term execution) and another on today for tomorrow (long-term direction). You need both.

Markets, technologies, and priorities will change. If you can’t adapt, you’ll end up explaining why instead of leading what’s next.

What am I not seeing or hearing?

I’ve seen plenty of confirmation bias in my career. Senior leaders claiming, “I know exactly how this works.” Domain experts insisting, “We know our users.”

Humility is rare but essential.
It takes courage to admit, “Maybe I’m wrong.”

The truth is, the higher you go, the more filtered your feedback becomes. You stop hearing the truth unless you go looking for it. Create spaces where people can speak without fear. Seek feedback that stings. Look for data that contradicts.

What you refuse to see will eventually find you.

Would I be proud to be a user of my own product?

If you suddenly switched roles and became a customer, would you be excited to use what you built?

Most product managers unconsciously design for others, assuming someone else will deal with the rough edges. But if you were the user logging in every day, relying on it to do your job — how would you feel?

This question changes everything.
When you experience your product like a customer, empathy becomes real. You start noticing the tiny details that make or break trust.

Great products are born from pride of use. If you wouldn’t proudly use it yourself, why should anyone else?

If I were an investor, would I put my money into this product?

Here’s the toughest one.
If you were an investor looking at your roadmap, traction, and clarity of vision — would you put a dollar into it?

If the answer is no, your why needs work.
The product strategy, the story, the differentiation — something’s missing. Investors, like customers, don’t fund uncertainty. They fund conviction.

And conviction only comes when your strategy connects vision, execution, and business value in a straight line.

Where this all leads to..

When the mirror talks back, it’s rarely kind — but it’s almost always right.

These questions aren’t about guilt or doubt. They’re about awareness — the kind that keeps you grounded when everything around you is moving.

Every product leader should pause once in a while and ask, not to impress anyone or to self-criticize, but to realign with purpose.

Because in the end, the best products and the most respected leaders are built the same way: with the courage to look in the mirror, listen to what it says, and do something about it.

It was September 2003.
I was rushed into the interview room at the Oberoi Hotel on MG Road, Bengaluru. SAP Labs hadn’t yet moved into its campus. The interview was for a developer role.

That’s where I first met Pranjal Dubey — a Development Lead at SAP Labs, and my interviewer for the day.

Within minutes of speaking to him, I sensed it. A thorough gentleman. Calm, composed, with a quiet sharpness.

I got the job. And for the next few years, I was lucky to work alongside Pranjal and many other brilliant colleagues.

Four years later, I moved to a different team, and we didn’t interact as much.

Then, one day, in 2009, word spread. Pranjal was leaving SAP Labs.

I couldn’t comprehend why. He had all the makings of a future leader — capability, respect, trust. Why would someone in the role of a Program Director leave?

I met him in office.

"Why?" I asked.

What followed felt like a scene from a movie.

The moment that changed everything

"Coming from a priestly family," Pranjal began, "I used to visit my ancestral village, Sandalpur, near Indore every year to perform traditional religious ceremonies.

During one such visit in 2006, an elderly man and his son approached me, asking if I could help the boy find a job."

Pranjal enquired about his qualifications. The boy had just completed school. No degree.

“I told them to get a proper degree — then something could be done."

He returned to Bangalore and forgot about it.

"Two years later, I was in Sandalpur again. After the annual rituals, the same father-son duo returned. The father handed me a sheet of paper. A college degree certificate.

They had sold land, sacrificed years of toil, and paid ₹4 lakhs — to buy a fake college degree.

The father believed that paper would guarantee his son a future.”

And then Pranjal said something over coffee that day at SAP Labs — a line I still remember:

"I have to go back. I have a social and moral responsibility to my people."

He had decided to leave behind the safety and comfort of a life he’d built working at SAP Labs to pursue something far more uncertain — and far more significant.

We said goodbye.
I knew Pranjal had gone on to establish himself. But I had to know more.
So in April 2025, I caught up with him for a conversation.

Adjusting to the new norm

"You know, Sandeep," Pranjal said, joining our meeting ten minutes late, "you can't really schedule things out here (in Sandalpur). Best you can do is adjust. I tried to change things. Eventually, I just started flowing with them."

It felt like we’d picked up right where we left off way back in 2009.

“All right,” I said. “Let’s begin from the beginning.”

He started with his father.

"In the '90s, my father was a government executive engineer in Indore. He worked for the water supply department. But the system wore him down.

One day, my father decided to resign, return to our village, and take over as priest after my grandfather passed away.

But my mother, my siblings, and I... we resisted. We told him not to give up his job — for our sake. And he stayed."

“And you ended up doing exactly what your father once wanted to — quit your job and follow what your heart was telling you,” I said.

Pranjal smiled.

“Yes,” he said. “And looking back, I understand what he was trying to move towards. But back then… we were selfish, I suppose.”

Two years after persuading him to continue with this job, his father passed away.

Pranjal, just 22 then, was suddenly next in line to become head priest — a role he didn’t want.

"I told them I’d come each year for the rituals. But I had dreams... a career, a different life."

It was during one of these visits that the father-son duo with the fake degree approached him — and left him shaken.

The real epiphany

"Was that fake degree incident the moment?" I asked him. “The epiphany that pushed you to leave SAP back in 2009?"

Pranjal nodded, but added, "It lingered... but the real shift happened during a leadership course at IIM Bangalore. There was an activity — we had to write our own obituary. A summary of our life’s impact."

He paused and said, "I couldn’t list more than three things I’d done that had meaning to others."

That hit him hard and forced clarity.

“I still wasn’t sure what I wanted to do,” Pranjal admitted. “But one thing was very clear — I didn’t want to get caught in the corporate rat race. Even the idea of becoming an SVP didn’t excite me. I had started a software firm during my MCA days and earned nearly ₹50,000 a month — back in the '90s. So I knew I had the entrepreneurial mindset.”

Over the next year, he met with many founders and CEOs — including Narayana Murthy and K. Dinesh of Infosys, and several others across India.

"And that’s when it began to click," he said. "Education. That’s where I wanted to build."

Finally, after consulting with his mentor, Prof. DVR Seshadri at IIM-Bangalore about the path ahead, Pranjal received advice that would alter the course of his life.

"What you're talking about isn't a startup. It's a social transformation. And social change doesn't happen in a quarter. It takes decades.” Prof. DVR sounded caution. “Maybe you’ll scratch the surface in your lifetime — if you're lucky."

Prof. DVR told him to reflect deeply. If he was serious about building an institute that would uplift rural youth — not just educate them but change the trajectory of their lives — then he had to be in it for the long haul.

"And remember," he added, "the great institutions we admire today — they took over 100 years to build. You’re talking about building one for people who’ve been forgotten by the system. You’re not solving for scale. You’re solving for purpose."

"I knew then... this wasn’t going to be a project. It was going to be my life’s work." Pranjal reflected.

When passion gets ahead of patience

“I thought I could change things faster than expected,” Pranjal admitted, looking back.

Despite Professor DVR’s caution, he was convinced his passion would carry him through.

“I had taken a six-month sabbatical from SAP to test the waters. I stayed in Sandalpur through peak summer — and I loved every bit of it. The work felt meaningful. Uplifting youth felt rewarding.”

He believed he could rewire the ecosystem through sheer will. That once students saw the opportunity, they’d flock to it.

But the ground he returned to wasn’t ready to change just yet.

Sant Singaji Institute of Science and Management (SSISM) began in 2010.
There was no ready infrastructure. The mindsets were fixed. Money was tighter than expected.
And the shift from a stable life in Bangalore to daily uncertainty in Sandalpur began to take its toll on the family. The system pushed back too — paperwork dragged, permissions stalled, and support came slower than it should have.

Even within the village, people wondered why he was trying to fix something they didn’t think was broken.

Every step forward required far more time, patience, and sacrifice than he had ever anticipated.

Through it all, his wife became the pillar of strength.
She had left behind the comforts of Bangalore life they had painstakingly built — and relocated to Indore without complaint.

"Whatever I have been able to build here," Pranjal said, his voice softening, "is because of her strength and sacrifice. She let go of so much, quietly, so that I could pursue this work."

Pranjal had two more strong shoulders to lean on.

My cousin Sankalp Dubey, who had a strong hold over government processes and local permissions, stepped in.” Pranjal said. “He took charge of the toughest parts — getting approvals, managing contractors, and navigating the political landscape — things I knew I wouldn't have been able to handle alone.”

“And my younger brother Anjul Dubey, who left a senior role as an All-India Service Head at Idea Cellular in Mumbai joined the mission.

He had spent years hiring youth from small towns across India, and he understood how rural students thought, what motivated them, and what hurdles they faced when trying to enter the mainstream workforce.”

While Pranjal focused on designing what needed to be taught and how it should be taught, Anjul worked on getting the right students into the institute, building the initial connect with the villages, and ensuring that the first batches had the right attitude and potential.

Meanwhile, Sankalp managed everything on the ground — the permits, the construction, and the day-to-day firefighting that came with building an institution from scratch in a rural area.

It became a team effort — each person handling a piece of the puzzle.

What they built came at the cost of everything else

“We donated five acres of our land and all our savings to set up the institute.” Pranjal said. “I even had to mortgage my house.”

Between savings, loans, and help from a few well-wishers, they raised nearly six to seven crores — enough to put up basic infrastructure, classrooms, hostels, and other facilities to get things moving.
They even took a bank loan of five crores, confident that once placements started and the first few batches graduated, the system would become self-sustaining.

Then reality of rural India came calling.

“We thought 60% of students would pay and the rest could be subsidized. But only 20% paid. And even among the rest, many didn’t see the value of what we were offering.”

The very community he wanted to serve didn’t yet believe in the kind of future he was trying to create for them.

And that hurt. Not because they doubted him, but because they doubted what was possible.

"They joined our institute because it was free and not because they were eager to learn. That was the toughest pill to swallow," he said.

He could’ve walked away. Blamed the system. Blamed the students.
But instead, he looked inward.

Something had to shift. So Pranjal began travelling across the country — visiting institutions, meeting educators, observing what worked and what didn’t. Conversations turned into insights. Slowly, through first principles thinking and fresh perspective, he began to see the problem differently.

“What if we stop chasing quantity and focus only on those who care? What if we build something rigorous, something aspirational — not just accessible?” he reasoned.

That’s when the institute moved from being just another college to something bolder — an engine of transformation.

Degrees were no longer the north star. Progress was measured by what students could do and not by what was printed on a certificate.

What others turned a blind eye to, Pranjal chose to see

The institute began to shed its old skin.

No more chasing volume. If students weren’t hungry, the institute wouldn’t feed them.

“We flipped the model,” Pranjal said.

He introduced a new structure.
No fixed semesters. No spoon-fed lectures. No chasing kids to attend class.

Instead, he designed a level-based system — inspired as much by kung fu movies as by the failures of traditional education.

“It didn’t matter whether students were in first year or third,” he explained.
“They couldn’t sit for placements until they cleared each level — practical tasks, real-world projects, group work. They decided when they were ready. Not the calendar.”

It was radical. And it worked.

For the first time, students weren’t being dragged toward a job.
They were climbing toward it — one level at a time.

Still, I had to ask. “How did they keep going? This model isn’t easy. Kids from village schools aren’t exactly conditioned for self-paced, high-accountability learning. What made them stay the course?”

“Motivation doesn’t start with a task. It starts with a dream,” Pranjal said.
Most of these students had never stepped into an office. Never met someone called an ‘engineer’. Never imagined their name printed on a visiting card.

Alongside redesigning the curriculum, Pranjal focused on rewiring their imagination.

With guidance from his mentor Prof. DVR Seshadri, Pranjal built the academic model around four pillars:

“The day you publicize their success,” Pranjal said, “you don't need to explain the value of effort anymore. Their village sees the car they bought. The house they built. The job they landed.”

That lit the spark. The levels? They gave it shape. And something to grow into.

Masterstroke!

And then came the results.

Subscribe now

Startups began to notice the talent and the transformative effects of experiential learning.

“We had students who hadn’t even given their final exams — but they could build, code, present, sell,” Pranjal said. “Some landed jobs with ₹40–50 lakh packages. No IIT tag. No degree. Just proof of work.”

One such company was Physics Wallah — the edtech unicorn.

“They hired over 150 students from our Management Excellence Group,” he said.
“Not just for technical roles, but for sales, ops, and customer success. Because our kids could speak, hustle, and adapt — especially in Hindi-speaking markets.”

Success wasn’t some distant story anymore. It was someone you knew. Someone you grew up with.

The kid who once rode a second-hand bicycle was now driving a car.
And for a village that had long seen government jobs as the pinnacle of ambition, something shifted.
Sons and daughters were now working at Microsoft, Cognizant, SAP — even at unicorn startups they’d once only heard of on TV.

“You want to change a place?” Pranjal smiled. “Don’t give them lectures. Give them a neighbor who made it.”

Powerful.

“Parents who once said, what’s the use of studying all this?” Pranjal recounted, “started asking how can my daughter get in?”

That was all Pranjal needed. Not a newspaper headline, but a quiet shift in what parents were beginning to believe.

Always thinking long-term, Pranjal knew he couldn’t build a future with a syllabus written in the past.

“The official university curriculum still taught outdated topics. Java applets. JSP. No mention of what startups or real-world employers were looking for.” Pranjal said, “So, SSISM created its own model.”

Three streams were introduced:

Each stream moved beyond textbooks. Students learned how to communicate, work in teams, and lead projects — not just clear exams.

No more fixed semesters. No examinations to prove worth. They moved forward when they were ready. Some raced through the levels. Others stayed until things clicked.

And that was okay. That was the design.

It was working. Slowly, but surely and the results were showing.

And now, the next chapter is taking shape. A full-fledged BTech in Artificial Intelligence and Data Science, under Information Technology Excellence Group (ITEG), is set to launch soon — a step closer to Pranjal’s vision of a rural institute offering the kind of education students once had to leave home to find.

But I couldn’t help asking — what kept it going, especially in those early years when belief was high but the bank balance wasn’t?

Pranjal had taken a ₹5-crore loan to build the infrastructure.
Fees weren’t coming in. Costs kept piling up.
And belief — as strong as it was — doesn’t pay EMIs.

“That phase would’ve broken me if not for one man,” he said.
“Vinayak Lohani.”

Vinayak — an IIT and IIM graduate, a disciple of Swami Vivekananda, and the founder of Parivaar, an organisation working with destitute children in Bengal and Madhya Pradesh.

Pranjal had met him at an event where both were speaking. One thing led to another, and soon Vinayak was not only mentoring him — but connecting him to donors, guiding him on the emotional and operational nuances of fundraising, and reminding him that people do show up — but only after you show up for long enough without them.

“He had this quiet power,” Pranjal told me. “Lived near the Narmada in a simple hut, wore kurta-pyjama every day, and stayed away from the spotlight — even when the most celebrated Indian cricketer came to inaugurate a school he had built, he didn’t show up.”

Yes, you read that right.

The school adjacent to SSISM — a feeder institution designed to bridge the gap between rural children and higher education — was donated and inaugurated by none other than Sachin Tendulkar. And the man who made it possible, Vinayak Lohani (now a Padma Shri awardee), stayed away from the fanfare.

“He told me, ‘This isn’t about me. Let the children have their moment.’ And he meant every word,” Pranjal said, pausing for a moment.

That one act told me everything I needed to know about the kind of people who show up in stories like this.
Not to be remembered. But to make sure someone else gets a chance to be seen.

The long game

"I’m not building a college," Pranjal said, leaning back. "I’m building a possibility.”

“The dream is to turn this into an IIT for rural youth,” Pranjal said, “a place where kids walk in unsure if they even belong — and leave knowing they do.

If IITs could be built before India was even independent, why can’t a village in Madhya Pradesh give rise to the next one?”

He knows he may never see the full outcome.
And he’s fine with that.

“Real change,” he said, “isn’t fast. It’s not visible every day. But one day, it becomes the new normal. And when that happens… you step aside quietly, and smile.”

I sat with that for a while after the call.
It landed differently. I realized it’s something you tell yourself on a day when nothing’s moving, but you still show up.

Here was someone who no longer needed the world to clap to know something mattered. Someone who stayed when it would’ve been easier — and safer — to leave.

And maybe that’s the real story here.
Not of a college. Not a model. Not even a movement.
But of a man who didn’t wait for perfect conditions.
Who just did the work — one day at a time.

And in doing that, reminded me that maybe the real change-makers don’t announce themselves. They just show up. And stay. And build.

After speaking to Pranjal, I’m convinced — the IIT for rural students is no longer a dream. It’s already being built.

Scroll down for a student spotlight few glimpses from SSISM’s journey.

🎓 Student Spotlight: Navin Saran

The boy from Sandalpur who lit the way forward

Navin was born in Sandalpur but grew up in the small villages around it. His family originally came from Rajasthan. His father, like many others in the area, worked hard, earned little, and stretched everything to make life possible. Paying for college education wasn’t just tough — it was unthinkable.

That’s when Sankalp — Pranjal’s brother — entered the picture.

Navin’s father met him by chance, and the conversation turned to a new college opening in Sandalpur. Subsidized fees. Local campus. A focus on students from the region. That was all the convincing he needed.

Navin joined the first batch.

“It wasn’t easy,” he told me. “I failed my first C-language exam.”

Confidence dipped. His father had pinned his hopes on Navin becoming the first graduate in the family. And Navin felt the weight of it. Watching his father borrow money just to keep him in class made it even harder.

“There were moments I really doubted if I belonged here,” he said.

That’s where Pranjal stepped in — not just as the founder, but as someone who stayed back after class, helping them code, debug, and try again.

“He’d sit with us till 1 a.m. sometimes,” Navin recalled. “He made you feel like your future mattered.”

But it didn’t stop at academics.

Noticing Navin’s cricketing talent, Pranjal pushed him into sports. He nudged others toward design, entrepreneurship, public speaking — whatever spark he saw in them.

“He had a way of knowing what you were good at, even before you did.”

Navin still remembers the visits to SAP Labs, IIM Bangalore, and several other IT firms — field trips arranged by SSISM that, for most village kids, were beyond imagination.

“That visit changed something,” he said. “It was the first time I saw what was possible. Pranjal sir and Aditya Trivedi sir put in massive effort to make this happen.”

This is what Pranjal calls one of the pillars of SSISM — Provide exposure.
And what Navin experienced that day? Many others at SSISM did too.

He was offered a job at SAP Labs after clearing the written test and campus interviews. Later, he got selected for the SAP Scholar Program, which let him pursue his M.Tech. Today, Navin works as a data engineer at Booking.com in the Netherlands.

But that’s not how he measures success.

He used his earnings to build a three-storey house for his family.
He took care of his mother’s treatment, paid off loans, bought a car, got insurance, and ensured his brother completed his education.

“All of this,” he said, “happened because of SSISM. Without it, we’d have stayed stuck.”

And what he said next spoke volumes about the ripple effect:

“After I got placed, ten others in my extended family became engineers. They started believing it was possible too.”

Navin also spoke about what SSISM has done for girls in the region.

“Earlier, they’d drop out after school. Now they’re going to college. That wouldn’t have happened without a college like this right here.”

This is about what belief, access, and the right nudge can do — not just for one boy, but for everyone watching from the sidelines.

And if Navin’s story is anything to go by, I’m certain that the students of Sandalpur have a thousand more waiting to be told.

Glimpses from SSISM journey

(Pic: Invendis Founder and CEO - Satish Kulkarni)

It was supposed to be a brief meeting—just another exchange of thoughts to explore Industrial IoT.

But what unfolded over coffee with Satish Kulkarni, Founder and CEO of Invendis Technologies, felt like uncovering a story buried beneath the noise of India’s startup frenzy. A story not of hype, but of heart. Not of exits, but of endurance.

Because Satish didn’t just build a company.
He rebuilt it from the rubble—twice.

And he didn’t do it alone.
Rangarajan, co-founder and COO of Invendis, has been there since day one—holding the operational core steady through every peak and every storm.

As Satish puts it, “Ranga is the nuts and bolts guy—the one who hunkers down and makes sure the windmills of Invendis keep running.”

The rise before the fall

In 2007, Invendis was born with a singular mission: to bring real-time monitoring to telecom towers through custom-built hardware. Back then, “IoT” wasn’t yet a buzzword. Satish and his team called it what it was—a telecom tower remote monitoring system. Designed in India. Built for India.

Early success came fast.
The team landed a multi-million dollar deal with a leading tower company in Bangalore. Satish thought, "This is it. The rest of the industry will follow."

And they almost did.

But then came the 2G spectrum scam.

A visit to the deepest valley becomes inevitable

“In our line of business,” Satish told me,
“there are peaks—and then there are valleys.
And when it’s a valley, it’s a deep, deep valley.”

This was one of them.

Satish had hired talent. He had built for demand.
Invendis had just shipped a multi-million-dollar product to one of India’s leading tower companies. Momentum was on their side.

When the 2G spectrum scam broke around 2009–2010, it wasn’t just a regulatory scandal. It was a nuclear winter for telecom infrastructure. The Supreme Court canceled 122 telecom licenses. CapEx froze. Tower companies halted expansion. Mergers tangled up decision-making.

“It was like an elevator drop from 100 to 0,” Satish said.
“We were too young to have reserves. Too hardware-focused to attract SaaS-hungry investors. And too far from Product-Market-Fit to reassure anyone.”

“We had built an organization for peak demand,” he added.
“And overnight—it all went to zero.”

Yes, let that sink in—
From millions… to ZERO.

People had to be let go. Offices were shut.
And worse than all of that—the market stopped believing.

But there was one belief that didn’t fade.

“Our investors told me, Satish said ‘We won’t let the lights go off in your office.’ And they meant it.”

They funded the company—not for control or equity.
But out of belief. And that belief kept Invendis alive.

As Satish said this, it made me pause.

Because with the right investor mindset, a founder doesn’t just get capital—they get oxygen.
A cushion to operate without constant anxiety, to make decisions that aren't driven by desperation.
To retain people. To rebuild. To think long-term.

It reminded me of the best cricket coaches—the ones who back their players through form slumps and dropped catches.
Because when you're not worried about your spot in the team, you play fearlessly.

“I remain indebted to my investors,” Satish said.

Powerful sentiment.

The valley before the payoff - a familiar pattern

Satish’s story reminded me of something I’d once read in Mastery by Robert Greene—about Paul Graham, the founder of Y Combinator.

After selling his first startup for about $50 million, Graham disappeared into years of silence.
He wrote essays. Explored ideas no one wanted to hear. Invested in his thoughts—while the world moved on.

He stayed in his own valley—ploughing hard, quietly, while no one was watching.

Then one day, Y Combinator launched.
And years of obscurity became the foundation for startup history.

“The terrain will rise—if you’re brave enough to stay in the valley long enough,” Satish said, after a long silence.

And that’s what he did.
He didn’t pivot. He stayed. He rebuilt.
And when the market came calling again—he was already ready.

In the middle of the crash, came a boarding pass. Destination: South Africa.

In the midst of this deep valley in his journey, a small thread of hope came from an unlikely place—South Africa.

American Tower Corporation (ATC), a client that had bought systems in India, was expanding into South Africa and Ghana. They nudged Satish: “Why don’t you speak to the teams there?”

And so he did.

What followed was nothing short of a resurrection.

By 2011, Invendis got its first international order to deploy in Ghana.
Soon after, tower companies across Africa began to follow.

“We came from a country where power cuts were routine,” Satish said.
“So we had already engineered for chaos. For theft. For heat. For broken infrastructure.”

American and Israeli competitors entered with polished solutions built for rule-based systems. But they didn’t know how to survive a 10-hour power cut or prevent on-site tampering.

Invendis did.

“Our designs aimed for precision—but were grounded in the realities we knew all too well,” he said.

(Pic: Invendis Facility in Bengaluru)

Fast forward a few years—and uncertainty returned

By 2020, the remote tower monitoring business was losing steam.
The model was solid—but the technology curve had moved on.

Telecom operators were consolidating their infrastructure, and many began embedding monitoring features directly into OEM equipment.
At the same time, energy efficiency and operational visibility had become table stakes—not differentiators.

“I realized we had to pivot before we became obsolete,” Satish said.

They still had steady revenue coming in from tower monitoring, but Satish wasn’t going to wait around for disruption to hit.

“It was a classic Kodak moment,” he reflected—the kind where you either pivot, or become irrelevant.

At that point, Invendis acquired a router and systems design engineering firm founded by former professionals from Wipro, Cypress Semiconductor, and GE.

The shift made sense—connectivity was no longer just about uptime, but also about intelligence, security, and edge computing.
Industries now needed routers that weren’t just conduits of data, but smart, programmable, and secure gateways.

But no acquisition comes easy. There’s technology to align, people to retain, cultures to absorb, and numbers that rarely tell the whole story.

“The products weren’t ready for the market,” he said. “It took us two years to make them commercially viable. But now—we’re doing really well in this space.”

And as he said that, it made me reflect.

Maybe this is what true leadership looks like—not reacting when things break, but sensing the shift before others see it.
To pivot before decline. To invest before certainty.
To prepare your company not for the now, but for what’s next.

Invendis, under Satish’s leadership, did exactly that.

The heart and culture behind the hardware

Today, Invendis operates in over 32 countries across Africa, Europe, and Asia, and monitors over 150,000 remote sites.

When I asked Satish what he’s proudest of, he didn’t talk about scale.
Not revenue. Not reach.

He talked about his people.

“I don’t run the day-to-day,” he said. “I trust my people. If something’s broken, I’ll know. But that rarely happens.”

This wasn’t hands-off leadership. It was trust—in people, and in process.

He sets the frame.
The team paints the canvas.

And when I asked about growth, he didn’t mention strategy decks or funding rounds.
He offered this:

“Sometimes, it takes you years to get to growth.
And sometimes, in a single week, you grow what you hadn’t in years.”

Then he paused. Shifted tone. And shared something personal.

“My former driver—he once dropped my son to school.
Today, he runs our factory operations in Bengaluru,” he said, with quiet pride.

It wasn’t a one-off story.

Satish had built a place where people didn’t just punch in hours—they grew.
With the company. Alongside it. Because of it.

“Anyone can fire experienced workers and cut costs,” he said.
“But what happens to their families? Their dignity?”

Even when spreadsheets told him otherwise, Satish chose loyalty.
Year after year.

“At the start of every financial year, this thought crosses my mind—do I retain, or do I cut?”
“I always retained,” he said.

He didn’t build culture with a playbook.
He built it with decisions—quiet ones, repeated often.

And when I walked through his electronics assembly unit near Silk Board, I saw something you can’t fabricate.

Pride.
Clarity.
Ownership.

“These aren’t workers,” he said. “They’re the soul of what we build.”

(Pic: Invendis staff at the Bengaluru Facility)

The discipline behind the scenes

As we wrapped up our chat, Satish dropped a line that didn’t sound profound—until I processed it later.

“Even when I’m out of the country, I start work at 9 a.m.
I’ve done that every day since we started.”

No excuses. Just rhythm. Ritual. A founder’s way of showing up.

And I remembered something I once read:

“How you do one thing is how you do everything.”

Satish’s calm consistency was more powerful than any motivational reel.

What I carried with me after this conversation

1/ Grit isn’t loud. It’s what’s left when everything else gives up.
Satish didn’t just survive the 2G crash—he rebuilt from zero.
No safety net. No guarantees. Just belief that it was still worth building.

2/ Resilience is a quiet habit.
The world moved on. Competitors pivoted. Capital dried up.
But he showed up. Again and again.
Sometimes that’s the real work—not vision, but staying power.

3/ Leadership isn’t always visible. But you feel it in the way a team shows up.
He doesn’t micromanage. He trusts.
And because of that, people don’t just do their jobs—they protect what they build.

4/ Taking care of people when it doesn’t make sense on paper—that’s the real culture.
I’ve worked in places where values were framed on walls.
Satish lives his. In who he keeps. In how he speaks about his team.
It’s not performative. It’s personal.

5/ Domain depth matters. But so does knowing what the domain overlooks.
He didn’t just build tech for towers. He built for theft, for blackouts, for systems that fail.
That’s not just engineering—that’s awareness.
It’s what happens when you’re not building for a pitch deck, but for the real world.

I walked into Invendis thinking about tech.
I walked out thinking about life.

I met a founder who built something very different:

A company.
A culture.
And a story that doesn’t need a spotlight—because it runs on something stronger.

Conviction.

This podcast shares valuable insights for product managers by examining the turnaround of a nearly 250-year-old retail store.... The owner, Raghavendra Jajee, achieved growth not by adding new features, but by observing customer behavior and removing friction in their buying process.... Key lessons include the importance of understanding user urgency, prioritizing speed and simplicity, and recognizing that empathy drives monetization. The podcast emphasizes that sometimes the most impactful innovations are invisible improvements to the existing experience. Hope you enjoy it!

I didn’t expect to find product wisdom on a scorching afternoon in Kalaburgi.
But that’s exactly what happened.

I was catching up with an old friend—Raghavendra Jajee, or Raghu as we call him—who now runs his family’s retail store selling Ayurvedic goods, dry fruits, pooja items, herbal cosmetics, and more in the heart of Kalaburgi’s bustling market.

It’s not just any shop.

M/s Parasram Jajee, founded in 1779, has been quietly serving generations of customers for nearly 250 years. The store stocks over 2,500 unique SKUs, mostly wellness and Ayurvedic products, and sees nearly 1,000 footfalls on a busy day.

You’d think with that kind of traffic, business would have been booming for decades.

But Raghu shared something unexpected.

“It was flat a few years ago,” he said.
“Same revenue. Month after month. Year after year. We weren’t losing customers—but we weren’t growing either.”

And that bothered him.

So he did something most people don’t.

One morning, he just... observed.

He parked himself behind the counter—no dashboards, no reports, no tools—just eyes and ears, wide open.

And what he saw made him deeply uncomfortable.

Customers were walking in steadily—mostly from nearby villages, operating on tight bus schedules and even tighter shopping windows. They’d request an item. The sales assistant would run to the back. By the time they returned, the customer had remembered another product. Off the assistant went again.

Meanwhile, the line grew. Customers waited, grew restless—and then simply walked away.

“Every customer who walked off wasn’t just a lost sale,” Raghu said.
“It was lost respect. We were wasting their time—and it was hurting our reputation.”

So he stepped into their shoes.
He obsessed over their urgency.
He mapped the experience not as a business owner—but as a customer in a hurry.

He asked himself not, “How do I sell more?”
But, “How do I make it faster and easier for them to buy?”

Subscribe now

That shift sparked a quiet revolution.

Here’s what Raghu changed:

• Expanded the sales counter from two to six people—serving customers in parallel and slashing wait times.

• Reorganized the layout—moving top-selling items to the front and less frequent ones to the back.

• Created festival combo packs—bundling high-demand products for Diwali, Holi, Dasara, and other events.

• Incentivized the sales team with per-item commissions—turning them into active sellers, not passive clerks.

• Placed impulse-friendly products near the counter to drive add-ons and increase value per customer.

• Invested in technology and procurement partnerships—bringing in high-quality, fairly priced goods from across the country.

• Launched an online platform to serve city-wide customers—extending the store’s legacy beyond its physical walls and enabling time-strapped buyers to browse, bundle, and buy from anywhere in the world. Check it out —> https://jajee.in

The result?

Word spread.

“Go to Jajee’s,” people started saying.
“They know what you need—even before you say it.”

Sales grew.
Customer experience improved.
People came back—not just for the products, but for how the experience made them feel.

Raghu wasn’t just running a shop anymore.
He was orchestrating a well-oiled, customer-first system—where speed, empathy, and flow mattered more than inventory size.

So, what can product managers learn from this?

Raghu never built something new. He refined what was already there—with observation, empathy, and clarity of intent.

And that’s where real product leadership begins.

Because sometimes the biggest unlocks don’t come from adding features.
They come from removing friction.

Raghu didn’t ask what more he could give his customers.
He asked what was getting in their way.

When we, as PMs, stop chasing the next shiny roadmap item and start removing obstacles in the current experience, magic happens.

He showed us that:

🟢 Observation beats assumption.
The best ideas rarely come from a dashboard—they come from silent patterns and customer behavior hiding in plain sight.

🟢 User urgency matters more than product features.
Most users don’t want more. They want faster. They want simpler. They want to finish and move on.

🟢 Removing friction is more powerful than adding flair.
Real delight isn’t always visible. It’s often felt in the absence of resistance.

🟢 Empathy drives monetization.
When you understand your users’ journey deeply, you don’t have to sell harder—you just make it easier for them to buy.

🟢 The best innovations are often invisible.
A well-placed product, a reduced wait, a faster checkout—these don’t trend on social media, but they create loyalty that lasts.

So the next time your product growth stalls…
The next time your backlog feels bloated…
The next time you’re unsure what to build next…

Pause.

Walk to your metaphorical counter.
Watch your users.
And ask yourself—

Where are they still waiting?
And what can I remove to help them move forward?

Your team has spent months perfecting the product. The go-to-market (GTM) checklist is complete, the messaging fits the brand, and the marketing plan is airtight. Yet, deep down, there’s uncertainty—you can’t quite articulate what makes your product special in a sentence. Worse, you’re not entirely sure who your message will resonate with.

Sound familiar? You’re not alone.

Many great product launches fail, not because of bad products, but due to weak or vague positioning. Positioning isn’t just about catchy taglines or polished decks. It’s the foundation for every decision in your GTM strategy, from targeting and messaging to sales enablement and pricing.

The truth is, positioning isn’t a marketing task. It’s a strategic imperative that drives success across the board.

What Is Product Positioning, Really?

Product positioning is how you want your customers to perceive your product relative to alternatives in the market. It’s not just about what your product does—it’s about how it fits into the customer’s life and why they should care.

At its core, positioning answers three fundamental questions:

• Who is this for? Your ideal buyer or target customer.

• What category are we in? The market or context your product operates in.

• Why should they pick us? The unique differentiators and benefits your product delivers.

Importantly, positioning ≠ messaging. Messaging stems from positioning, but positioning is the upstream strategic work that informs everything else, providing clarity on what to say and to whom.

A Simple Example

Canva positioned itself not as a design tool for professionals but as “design for everyone.” By focusing on non-designers like marketers, teachers, and small business owners, Canva created a new category of accessible design. This clarity of positioning is what made their marketing and product so effective.

Why Positioning Is the First Lever in GTM

Good positioning is like a compass for your GTM strategy. It ensures every decision aligns with a clear north star. Done well, it shapes every key element of GTM execution:

1. Messaging → Clarity

Positioning provides the foundation for your messaging. Without it, your communications risk being generic and uninspired. With clear positioning, your tagline, value props, and sales pitches resonate deeply with the target audience.

2. Targeting → ICP Alignment

Positioning sharpens your understanding of your Ideal Customer Profile (ICP). It ensures that marketing and sales efforts focus on high-potential segments, reducing wasted effort and increasing efficiency.

3. Sales Enablement → Objection Handling

Clear positioning arms your sales team with the context they need to address objections. It gives them a framework to articulate what differentiates your product and why it’s the best choice for the customer.

4. Pricing → Perceived Value

Positioning influences how customers perceive the value of your product. Premium brands like Tesla or Allbirds command higher prices because of their clear, aspirational positioning.

The 5 Pillars of Great Positioning

To craft powerful positioning, focus on these five pillars:

1. Target Customer

Who exactly is your product for? Be specific and avoid the temptation to say, “everyone.” Clear customer targeting ensures your positioning speaks directly to the right people.

2. Problem Context

What specific problem or need does your product solve for this audience? Define this problem through the customer’s eyes to make it relatable.

3. Product Category

What type of solution is your product? Are you part of an established category, or are you redefining it? Clear category definition helps customers immediately understand your product’s purpose.

4. Differentiators

What does your product do uniquely well that competitors can’t match? Highlighting your differentiators sets you apart and drives competitive advantage.

5. Proof

Why should customers believe you? Evidence like case studies, testimonials, or tangible outcomes helps build credibility.

Example in Action

Replit: Positioned as “the collaborative coding platform for developers,” Replit targets beginner to mid-level developers (target customer), solves the problem of complex and isolated coding environments (problem context), and differentiates with real-time collaboration features (differentiators).

Positioning Isn’t Static—It Evolves with GTM

Positioning isn’t a one-time exercise; it’s an evolving strategy that adapts as your product and market grow. Companies often need to refine their positioning at different stages of the GTM lifecycle:

1. Early-Stage (PMF):

Focus positioning on solving a narrow, urgent problem to find product-market fit.

2. Scaling:

Refine positioning to appeal to adjacent markets without losing focus on your core audience.

3. Expansion:

Update positioning to account for competitive signals, market shifts, or new innovations.

Real-World Example:

Notion started as a note-taking app but evolved its positioning into a “connected workspace,” targeting teams and collaboration—a broader, more scalable market.

How GenAI Tools Can Help You Shape & Test Positioning

AI tools can help you uncover insights, refine your messaging, and test positioning ideas quickly. Here’s how:

• Summarize Customer Feedback

Use Claude or ChatGPT to analyze customer survey responses and extract common themes about pain points or desired outcomes.

• Test Messaging Variants

Tools like Jasper and Writer help you experiment with different tones and wording that align with your positioning.

• A/B Testing for Positioning Pages

Platforms like Mutiny and Google Optimize allow you to test landing page options to see how different positioning strategies resonate.

Product Positioning in Action

To refine your own positioning, try using this simple framework inspired by Geoffrey Moore’s Crossing the Chasm:

Example for an AI Compliance Platform

For regulated businesses seeking to streamline compliance workflows, ComplaAI is an AI-powered compliance platform that reduces paperwork by 60%. Unlike legacy tools, we integrate with existing systems and provide real-time monitoring.

From this framework, you can build:

• Landing page messaging

• SDR call scripts

• Analyst briefing decks

Landing page messaging can be crafted using the framework to highlight the specific needs of your target customer, emphasize key differentiators, and clearly present the proof points that set your product apart from competitors. This ensures that visitors immediately understand how your product addresses their pain points.

SDR (Sales Development Representative) call scripts benefit from the structure by providing a clear narrative for engaging prospects. Using the framework, SDRs can focus conversations around the customer's challenges, the product's unique abilities, and why it outperforms alternatives—all while resonating with the target audience.

Analyst briefing decks also benefit greatly from this positioning framework. You can use it to succinctly articulate your product’s value in a way that aligns with industry trends and differentiates you in a competitive landscape. Providing strong proof points further strengthens your credibility among industry experts and stakeholders.

By leveraging this approach, you create a unified message that strengthens your product narrative across multiple key touchpoints.

Mistakes to Avoid

• Trying to Be for Everyone: Diluted positioning doesn’t resonate with anyone.

• Talking About Features, Not Context: Focus on the problem you're solving, not just the product specs.

• Positioning Once and Forgetting: Updates are essential as markets shift and your business grows.

• Thinking It’s Just a Slide Deck: Positioning should inform everything—messaging, targeting, and even product development.

Why Strong Positioning Is Your GTM North Star

Positioning isn’t just marketing—it’s the strategic compass that guides your entire GTM strategy. With clear positioning, every aspect of your business gains cohesion—your messaging sharpens, your targeting improves, and your product resonates more deeply with your audience.

Before your next product launch, take a closer look at your positioning. Does it answer the key questions? Does it differentiate your product effectively? If not, start from the basics.

And remember, successful positioning begins with your product team and your customers—not just with marketing.

CTA: Need help clarifying your positioning? Start by reflecting on your GTM strategy and your target audience. Or, explore tools like Copy.ai and ChatGPT to test ideas on the fly.

Additional Notes: Clarifying the Audience and Purpose of Product Positioning

Product positioning isn't limited to a single document or audience; rather, it’s a framework that serves multiple purposes for various stakeholders, both internal and external. Internally, product positioning provides a foundational guide for teams such as sales, marketing, and product development. For instance, the sales team can use this positioning to craft focused pitches that align with customer pain points and the product's unique benefits. Externally, it helps customers understand how your offering addresses their specific needs, highlighting its differentiation in the market.

Typically, product positioning can be shared through a combination of documents or materials—this might include a positioning statement, personas, value propositions, messaging frameworks, and go-to-market (GTM) playbooks. Each of these can be tailored for its primary user while maintaining consistency in the core positioning. By fostering alignment across teams and ensuring clarity for customers, it bridges the gap between your internal strategy and external communication, ensuring the message resonates universally.

Example of Internal and External Facing Product Positioning

Internal-Facing Positioning

For internal teams, such as sales, marketing, and product development, the product positioning focuses on equipping them with the tools and clarity needed to articulate the product’s value. For instance, an internal-facing messaging framework might emphasize key pain points the product solves, competitive differentiation, and detailed personas. Example elements might include:

• Core Value Proposition: "Our platform reduces manual processes by 70%, allowing enterprise teams to focus on strategic growth."

• Unique Differentiators: Highlight proprietary technology, speed, or enhanced security features for internal understanding.

• Sales Enablement Points: "When speaking with mid-level managers, focus on how this product streamlines daily workflows."

External-Facing Positioning

For customers and external stakeholders, the focus is on tailoring the message for clarity, simplicity, and emotional impact. A go-to-market message might distill complex features into highly relatable benefits and outcomes. Example communication could include:

• Customer-Centric Message: "Transform the way you work—cut down wasted time and achieve more with our intuitive platform."

• Benefit-Driven Highlights: Focus on how the product will make life easier, save money, or align with customer values.

• Brand Voice Consistency: Ensure the tone remains approachable and empathetic, connecting with the target audience on a personal level.

By maintaining strategic consistency while addressing specific contexts, product positioning seamlessly connects both internal team needs and external customer expectations.

TL;DR:

Product positioning isn’t just a marketing exercise—it’s the strategic core of your GTM success. When done right, it clarifies your messaging, sharpens ICP targeting, boosts sales enablement, and elevates perceived value. This blog breaks down what positioning really is, how to craft it using five key pillars, and how tools like GenAI can help refine and test it. Includes real-world examples (like Replit and Notion), a plug-and-play positioning framework, and common mistakes to avoid. If your GTM feels off, your positioning might be the root cause.

A few months ago, a potential customer asked me:

"Your AI pricing makes no sense. Why should I pay for AI when I don’t even know if it’ll work?"

That moment made me realize—pricing AI isn’t just about setting a number. It’s about trust, perceived value, and strategic positioning. Unlike traditional SaaS pricing, AI has variable costs, ambiguous success metrics, and a constantly changing value proposition.

So how do you price AI effectively? The answer lies in understanding what customers actually pay for.

Customers Don’t Buy AI—They Buy Outcomes

The core principle of pricing remains the same:
✔ Customers don’t buy software.
✔ They don’t buy features.
✔ They buy value—the outcomes your product delivers.

AI pricing should be rooted in the economic impact it creates for the customer, which typically falls into three buckets:

• Cost Savings – Reducing workforce or operational overhead.

• Time Savings – Automating processes for efficiency.

• Revenue Growth – Unlocking new business opportunities.

Understanding where your AI product delivers value is the first step to pricing it correctly.

Beyond Setting a Price —> Aligning Incentives

AI pricing isn’t just about what the customer pays—it’s about how incentives are aligned. Think value co-creation. Your pricing model should ensure:

1. Maximized Revenue – Capture the highest price customers are willing to pay while ensuring perceived value justifies the cost.

2. Sustainable Cash Flow – Unlike traditional SaaS (which has fixed costs), AI incurs ongoing variable costs (e.g., API usage, cloud expenses). Your pricing model should prevent cash crunches.

3. Customer & Vendor Alignment – AI pricing should encourage long-term value realization. Misaligned incentives—such as unlimited AI usage for a fixed price—can destroy margins.

The Biggest Mistake —> Competing on Price

One of the worst things you can do is engage in a price war with competitors.

Take the airline industry as an analogy—there’s hardly any differentiation in pricing models, leading to 2.6% profit margins per seat. If you constantly undercut competitors, you turn your AI into a commodity.

🚨 Common PM Mistake:
Many product managers panic and slash prices when new competition enters. But this is a race to the bottom.

What to Do Instead —> Differentiate, Don’t Discount

Rather than lowering prices, make your product so valuable that customers don’t mind paying a premium.

🔹 Offer seamless integrations that competitors lack.
🔹 Provide customization or compliance features that justify premium pricing.
🔹 Focus on a specific niche where your AI delivers unmatched value.

When customers see you as a strategic investment rather than just another tool, price sensitivity disappears.

Success-Based Pricing - Does It Work?

A growing trend in AI pricing is success-based pricing, where customers pay only when AI delivers tangible results. This is similar to fintech models—Stripe only charges when a payment is processed.

✅ Pros: Customers pay only when they see value, reducing adoption friction.
❌ Cons: AI success is often ambiguous—how do you define a "successful" chatbot response or AI-generated insight?

While success-based pricing sounds attractive, it can lead to unpredictable revenue. Instead, a more reliable model is value-based pricing.

Value-Based Pricing: Is this the smartest AI pricing model? May be.

With value-based pricing, you charge based on the economic impact your AI creates.

🔹 Step 1: Define Expected Value – What specific problem does your AI solve?
🔹 Step 2: Quantify Financial Impact – How much time/money does it save or generate?
🔹 Step 3: Charge a Fraction of That Value – Typically 25-50% of the value created.

Example: AI-Powered Sales Prospecting Tool

Imagine selling an AI tool that automates lead qualification for a B2B SaaS sales team.

• The company has 50 sales reps.

• Each rep earns $80K per year, with total costs (office space, tools, benefits) bringing it to $120K per rep.

• Total sales team cost: $6M per year.

• AI tool reduces time spent on prospecting by 30%, saving $18K per rep.

• Across 50 reps, total savings = $900K per year.

🔹 Pricing Strategy: If the company saves $900K, you can confidently charge $300K per year, ensuring a strong ROI.

This model ensures customers see clear value while you capture a fair share of the economic benefit.

Cost-Based Pricing - Why It Fails for AI

Traditional SaaS often uses cost-plus pricing, adding a markup on development and infrastructure costs. But this approach doesn’t work for AI.

🚨 Why Cost-Based Pricing Fails:
❌ Software margins should be 80-90%—cost-based pricing erodes profitability.
❌ Customers don’t care about your costs; they care about outcomes.
❌ AI costs (like API fees) fluctuate—pricing based on costs is unpredictable.

Example: Cost-Based vs. Value-Based Pricing

Let’s say your AI costs per customer (API fees, cloud hosting, engineering support) amount to $200K per year.

• Cost-based pricing with a 50% markup → You charge $300K per year.

• Value-based pricing (considering your AI delivers $2M in savings) → You could charge $700K, ensuring healthy margins.

AI pricing should be based on value—not cost.

Should You Publish Pricing or Use ‘Contact Sales’?

A big debate in SaaS is whether to display pricing publicly or require prospects to contact sales.

🔹 Rule of Thumb:
✔ Self-serve SaaS? → Publish pricing.
✔ Enterprise AI? → Use ‘Contact Sales’ to offer flexible, high-value pricing.

Enterprise customers have wildly different value equations—if you pick one random price, you’ll either overcharge or undercharge.

Instead, use a hybrid approach: Offer a starter plan ($99/month) for small teams, while keeping enterprise pricing flexible.

The Hidden Pitfalls of BYOK (Bring Your Own Key) Pricing

Some AI vendors allow customers to bring their own API key (BYOK) to manage costs—similar to leasing a car but still having to buy gas.

🚨 Why It Fails:
❌ Finance teams prefer a fixed spend over fluctuating API costs.
❌ Customers don’t want to manage API provisioning.
❌ If their key gets revoked, your product’s functionality suffers.

Better Alternative? Bake API costs into your pricing model to avoid friction.

AI Pricing as a Competitive Advantage

AI pricing isn’t just about numbers—it’s a powerful tool that drives revenue, differentiation, and long-term success.

✔ Avoid price wars—they commoditize your product.
✔ Charge based on customer value, not costs.
✔ Use enterprise flexibility—don’t lock AI into fixed pricing models.

💡 Does your AI pricing align with the value you deliver? If not, it might be time to rethink.

PM-Dev Chronicle Part 1: Walking the Slippery Bridge

Part Two

A week later, Deepthi walked into the meeting room again.

Same room. Same sleep-deprived faces. Same aroma of over-brewed coffee. It felt like déjà vu.

But something was different this time.

The last meeting had been a battle of perspectives—developers viewing feasibility, product trying to push vision. It wasn’t hostile, but it wasn’t productive either. This time, she wasn’t here to convince them. She was here to collaborate.

She had spent the past week listening, researching, and gathering facts. No wishful thinking. No vague assumptions. Just a practical path forward.

Deepthi settled into her chair and placed her laptop on the table. The subtle hum of keyboards slowed. Eyes flickered away from screens. Good. They were listening.

"Glad you all are here," she began, her tone lighter than last time. “I’ve been digging into this, and I want to propose something different. Instead of forcing real-time dashboard updates, how about ‘near’ real-time?”

She glanced at Arjun, the architect—the one who would be most critical.

A pause.

Then, one by one, heads lifted from screens.

"I spoke with some of our integration partners," she continued. "Turns out, some APIs can handle updates every 10 to 15 minutes. Others, just once an hour. Instead of forcing everything to be instant, what if we create a system that adapts dynamically?”

This time, no immediate resistance. No scoffs or dismissals. That was progress.

Arjun, now interested, leaned forward slightly.

“Interesting.”

It wasn’t a yes, but it also wasn’t a no.

Deepthi pressed on.

"For APIs that allow it," she continued, "we use webhooks instead of polling. That way, we get updates only when something changes, instead of hammering them with requests every second.”

Raghav, who had been silently absorbing the discussion, let out a low whistle.

“That actually makes sense,” he muttered.

Encouraged, Deepthi kept going.

“And for APIs that have stricter limits, we stagger the updates—some every few minutes, others less frequently. No unnecessary traffic, no overload. That way, we optimize responsiveness without burdening the system.”

This time, the silence wasn’t filled with tension. It was filled with thought.

Raghav tapped his fingers on the table. “I can work with that.”

Arjun rubbed his chin, his face no longer skeptical but contemplative.

“Not a bad suggestion. That’s actually… smart.”

Deepthi noticed something shift in the room. The discussion had turned from 'why it won’t work' to 'how we can make it work.'

That was the breakthrough.

She leaned back, giving them space to process.

“How about we meet in a couple of days after we discuss what design considerations need to be made?” Arjun suggested.

Deepthi nodded. "Sounds good. I'll schedule a time for us to meet."

The meeting wrapped up. But this time, something had changed.

She had walked in last time trying to convince them.
She walked out today having earned their trust.

Not by pushing harder.
Not by debating louder.
But by speaking the technical language.

She still couldn’t write a single line of code.

But now, she understood how to communicate in a way that mattered.

As Deepthi gathered her things, she overheard Raghav and Arjun chatting quietly.

“Honestly, I didn’t think she’d come back with something useful,” Raghav admitted.

Arjun smirked. “Yeah, I thought this was gonna be another ‘PM asks for the impossible’ meeting.”

“Same.” Raghav nodded. “But she actually listened. She didn’t push unrealistic stuff—she gave us something that works.”

Arjun tapped his pen against the table. “It’s rare. Most PMs just push business needs without thinking through constraints. This was different.”

Deepthi pretended not to hear, but inside, she smiled.

She had cracked the code.

It wasn’t about knowing how to build the solution.
It was about knowing how to have the right conversations about it.

Author’s Note:

A bridge isn’t built by one side alone.
It’s built when both sides move towards each other.

That’s how great products are made.
That’s how great teams work.
That’s how friction turns into flow.

"Product management is really the fusion between technology, what engineers do, and the business side of things."

– Marissa Mayer, Former Google Product Leader & Yahoo! CEO

The roads leading to the office bustled with street vendors enthusiastically engaging with their customers, offering fresh idlis and steaming cups of filter coffee. The aroma filled the air, blending with the early morning rush of Bengaluru.

Inside SynergyTech’s glass-walled office, however, the atmosphere was far from lively. The warmth of the sun streamed in through the windows, but inside the conference room, the mood was pensive. It was time for yet another roadmap review session—one of many that the product and engineering teams held regularly.

Deepti Sharma, the Product Manager, walked in with quiet confidence. She looked sleep-deprived, but that was nothing unusual. Everyone else in the room looked just as exhausted. Despite this, her energy remained high. She had an exciting feature lined up for discussion, something she believed would make their enterprise customers sit up and take notice.

“Real-time insights.” She announced as she took her seat, scanning the room. “That’s what they want. That’s what will set us apart. Imagine a dashboard where every metric updates instantly—live, with no delays, no batch processing. Real-time, as it should be.”

Before the inevitable technical skepticism could set in, Arjun Dubey, the system architect, leaned forward with a measured look.

“Deepti, before we talk about how to do this, have you validated why customers need real-time insights? Are they just asking for it because it sounds good, or is there a concrete business case?”

Deepti was prepared for this. She had spent the past few weeks talking to key customers, analyzing how they used the product, and understanding their pain points.

“Yes, I have,” she responded confidently. “Three of our largest enterprise clients have expressed frustration over delayed decision-making because they rely on batch-processed reports. One of them even built a workaround by exporting data and refreshing it manually every 30 minutes.”

A few heads nodded, showing interest.

“The biggest issue,” she continued, “is that operational teams are flying blind. They make decisions based on outdated data. One customer told me that by the time they catch an issue in supply chain tracking, it’s already too late to act. If they had real-time insights, they could intervene sooner, prevent losses, and improve efficiency.”

There was a pause. This wasn’t just a nice-to-have feature; it was tied to real business impact.

“Fair enough,” Arjun said, exchanging a glance with Raghav Bhat, the senior backend engineer. “Now let’s talk feasibility.”

That was the cue.

“Live updates,” Raghav said, leaning back in his chair. “From all integrated sources?”

“Yup.” Deepti nodded.

“And by ‘live,’ you mean?” Raghav probed further, his tone cautious.

“As soon as the data changes, the dashboard reflects it instantly. No lag.” Deepti’s confidence didn’t waver. She had expected this question.

Raghav exhaled sharply. “You say it like it’s achievable with the wave of a wand.”

“It’s not magic,” Deepti countered, folding her arms. “We have the technology to make this happen.”

Raghav leaned forward, elbows on the table. “Technology comes with limitations. Right now, our system processes data in batches. Event-driven architecture. That means we queue data, process it, and update in intervals. It works because if something fails, we retry in the next cycle without causing system-wide chaos.”

His voice turned firmer. “What you’re asking for is a synchronous system—where every time someone opens the dashboard, we fetch fresh data from multiple third-party sources. What happens when one of those APIs is slow? Or worse, when it goes down?”

Deepti hesitated. She wasn’t ignorant of tech, but she hadn’t fully considered failure scenarios—at least, not like this.

“Can’t we… somehow make the data exchange faster?” she asked.

A soft chuckle rippled through the room.

And then, from the corner, Arjun spoke up again, his voice patient and composed. “Deepti, you like food, don’t you?”

Deepti frowned. “I—what?”

“Restaurants,” Arjun clarified. “Right now, our system works like a fine-dining kitchen. Orders are placed, prepared, and served in sequence. Some dishes take longer, some shorter, but the process remains efficient. What you’re asking for is a live kitchen, where every customer stands at the counter, waiting while the chef prepares their meal instantly. Now imagine what happens when one order takes longer than expected.”

Deepti could picture it now—a growing queue, restless customers, one delayed order cascading into another, the entire kitchen slowing to a painful halt.

“Oh.” The realization hit her.

“Exactly.” Raghav leaned back again. “And that’s without even talking about rate limits. Some APIs won’t let us make more than a certain number of requests per minute. If we hit that limit? The feature dies.”

Silence filled the room. No one wanted to be the first to speak.

Finally, Deepti broke the quiet. “Okay, so what’s the alternative? Because customers still want something close to real-time.”

Arjun nodded approvingly. “Now that’s the right question.”

“We tweak the model,” Arjun suggested. “Instead of actual real-time, we reduce the batch processing window. Maybe every 10 minutes instead of every hour. That way, users get near real-time data without breaking the system.”

“And,” Raghav added, “we should talk to our integration partners. We need to figure out what their APIs can actually handle, so we don’t build something that collapses the moment we scale.”

Deepti exhaled, nodding. “I’ll reach out to them before we meet next.”

As the meeting wrapped up, Raghav and Arjun exchanged knowing glances.

“You think she’ll get it this time?” Raghav asked.

“She’s starting to,” Arjun replied.

The aroma of coffee lingered, a silent witness to the friction, learning, and quiet victories that shaped the team.

Deepti had walked into the meeting expecting a simple yes, but walked out with something far more valuable—a deeper understanding of trade-offs in technology. Being a product manager wasn’t just about pushing for features; it was about balancing customer needs with technical realities, understanding trade-offs, and working with engineers, not against them.

She had won trust by validating the business need, proving that real-time insights weren’t just a nice-to-have but a real problem worth solving. But she had also learned a hard truth—what customers want and what’s technically feasible don’t always align.

The engineers had done their part too—challenging assumptions, ensuring scalability, and guiding the conversation toward a practical solution. They weren’t roadblocks; they were protecting the product from breaking under its own weight.

This wasn’t about winning or losing—it was about finding common ground.

But the real question remains…

Would this chasm between product and engineering ever be bridged?

And more importantly… what would Deepti do next?

Find out in Part 2, publishing next.

“If there's one thing in AI that one should pay attention to, it's Agentic AI.”

-Andrew Ng

The AI Agentic Workflow is a fascinating development. The more I explore how complex it is for machines to autonomously mimic human behavior, the more I realize how much we take our cognitive and reasoning abilities for granted.

Everyday tasks like booking a meeting, driving, or making tea involve a series of coordinated actions that feel effortless because our brain optimizes them over time. Through neuroplasticity, repeated behaviors strengthen neural pathways, making these tasks feel second nature. But for machines, achieving human-like reasoning requires computational models that simulate cognitive processes—blending deterministic logic for structured tasks with probabilistic reasoning for uncertainty.

The rise of agentic AI workflows is making this seemingly impossible feat a reality, and that’s exciting. As I explore this space, I want to map which aspects of AI behavior are deterministic (rule-based and predictable) and which rely on probabilistic decision-making (adaptive and uncertain).

You may have heard of rule-based automation and robotic process automation (RPA)—which are purely deterministic. These systems follow predefined workflows where given the same input, the same output is always produced. However, AI extends beyond deterministic logic into probabilistic reasoning, where uncertainty and adaptive decision-making play a role.

It’s important to note that all AI processes fall into one of these three categories:

• Deterministic AI follows strict rules and produces predictable outputs.

• Probabilistic AI assigns probabilities to decisions based on uncertainty and adaptation.

• Hybrid AI combines deterministic rules with probabilistic decision-making, ensuring structured execution while allowing adaptability.

Now that we’ve set the context, let’s explore the three core characteristics of AI agentic workflows. If you’re interested in learning more about Agentic AI, check out my earlier blog on the topic.

Three core characteristics of AI agentic workflows

1. Continuous Learning – Agentic workflows integrate past interactions, data, and user feedback to continuously improve results. Unlike traditional automation, these systems evolve over time through reinforcement learning or adaptive models.

2. Reflection & Independent Decision-Making – AI agents proactively critique their own outputs, identify flaws, and refine their approaches without requiring explicit user input. This mimics human metacognition, allowing AI to self-correct and optimize behavior.

3. Intelligent Collaboration – AI agents work with other systems, microservices, or humans to divide tasks, exchange information, and reach goals more efficiently through multi-agent systems. (source: scalablepath)

Applying this concept to flight booking

A great example of this in action is Operator.ai, an exciting development from OpenAI that applies AI agentic workflows effectively. Although I haven’t personally tried it (since it’s not available in India yet), the demo showcased how an AI agent can autonomously process complex tasks, making intelligent decisions while consulting with humans when needed.

If you’re a product manager trying to map this concept, here’s how I broke it down for a flight booking use case:

Can this be related to any existing architectures?

You can think of AI agentic workflows like microservices in software development. Just as microservices break down a monolithic system into smaller, independent services, AI agentic workflows decompose a complex process into micro-agents—specialized AI components that execute specific tasks while coordinating with others.

Like microservices communicating via APIs, these AI agents exchange structured data, prompts, or function calls to complete their tasks efficiently. Designing these micro-agents with an API-first mindset—along with effective context management and reasoning mechanisms—is crucial for scalability and seamless automation.

Even in the human brain, completing a complex task—like checking the rearview mirror while reversing a car—requires millions of neurons firing in parallel, with each neuron playing a specialized role in a larger network. These interconnected neurons work collaboratively to handle various sub-tasks, such as visual perception, motor control, and spatial awareness. Similarly, agentic AI systems distribute cognitive load across multiple task-specific agents that interact and collaborate to execute tasks more efficiently, mimicking human-like automation.

Exciting times ahead! :)

— Deep Nishar, former VP of Product at LinkedIn

Making the move from engineering to product management isn’t just about learning new skills—it’s about thinking differently. Engineers focus on how to build things right, but product managers focus on what’s worth building and why.

That shift might sound simple, but in reality, it’s a big mental rewiring. If you’ve ever found yourself asking, "Why are we even building this?" or "Does this actually solve a problem?"—you’re already thinking like a PM.

Here are 14 key shifts that will help you make the leap. And no, this isn’t just theory—I’ll break down why each of these matters in real-world work settings.

1. Transition from “How” to “Why”

Great engineers focus on how to build. Great product managers focus on why it should be built.

from jumping to solutions → defining the problem

We love solving problems. The moment we see one, our instinct is to start coding a fix. But here’s the trap: If you don’t fully understand the problem, even the best solution is useless.

Have you ever built a feature, only to realize later it wasn’t what the user actually needed? Happens all the time.

Before jumping in, ask:
✅ What’s the actual pain point?
✅ Who is facing this problem, and why?
✅ What will change if we solve this?

💡 Takeaway: Great products start by solving the right problem, not just executing the perfect solution

2. You’re Not a Designer—But You Need One

You might understand UI and UX, but that doesn’t make you a designer.

from “looks good to me” → designing for users

Engineering-led design often focuses on functionality—"It works!" But product-led design focuses on intuitive user experiences—"Does this make sense for the user?"

Ever used a clunky, hard-to-navigate app that technically worked fine but was frustrating? That’s what happens when design isn’t prioritized.

Work with designers, test early, and focus on usability.

💡 Takeaway: A well-designed product makes users feel smart—not confused.

3. Prototype Before You Build

Think before you code.

from shipping code → validating ideas

Every developer has had that moment: "We built it, but no one is using it." That’s why PMs focus on prototyping, testing, and validating before committing resources.

Start with mockups, wireframes, or even paper sketches. Talk to real users. The sooner you get feedback, the less painful it is to pivot.

💡 Takeaway: Validate early to avoid wasting time building something nobody needs.

4. Think Beyond Features—Focus on Impact

More features don’t make a product better—solving real problems does.

from “what can we build?” → “what should we build?”

Teams love feature roadmaps. But ask yourself—do users actually need all these features?

Instead of adding feature after feature, focus on:
✅ How does this improve the user’s experience?
✅ Does it solve a core pain point?
✅ Will this drive adoption, retention, or revenue?

Simplicity wins. Great products do a few things really well.

💡 Takeaway: More features don’t make a product better—solving real problems does.

5. Be Data-Driven, Not Just Opinion-Driven

Gut feelings are great—but data tells the real story.

from “i think” → “the data shows”

Have you ever debated what feature should be prioritized, and everyone had a different opinion? PMs cut through the noise by looking at data:

• User behavior analytics (What are users actually doing?)

• A/B testing (What’s working, what’s not?)

• KPIs & business impact (Are we moving the needle?)

If you can’t measure it, you can’t improve it.

💡 Takeaway: Decisions based on data drive better products than those based on assumptions.

Subscribe now

6. Understand the Market & Your Competition

Your product doesn’t live in isolation—users have choices.

from “we’re unique” → “how do we compete?”

It’s easy to assume our product is special. But how do users see it compared to alternatives?

• What pain points do competitors solve well?

• What’s missing in the market?

• How can we differentiate without just copying?

Great PMs understand market shifts, trends, and competitor strategies.

💡 Takeaway: Your product doesn’t live in isolation—know the landscape to stay ahead.

7. PMs Wear Multiple Hats Daily

One moment you’re talking to engineers. The next, you’re with marketing. Then customers.

from “deep focus” → constant context switching

As an engineer, deep work is valued. But PMs need to adapt constantly:
✅ Strategist: Defining the roadmap
✅ Customer Advocate: Understanding pain points
✅ Marketer: Positioning the product
✅ Support Partner: Helping resolve issues

Adaptability isn’t a skill—it’s a superpower.

💡 Takeaway: Success comes from balancing strategy, execution, and communication across teams.

8. Shift from TAD to PRD

Technical Architecture Docs (TADs) explain how to build. Product Requirement Docs (PRDs) explain what to build and why.

from execution → strategic alignment

Engineers are great at breaking down how something should be built—choosing the right architecture, ensuring scalability, and optimizing performance. But without clear direction on what and why, teams can end up building something technically perfect but strategically irrelevant.

A PRD (Product Requirement Document) shifts the focus to:
✅ What problem are we solving?
✅ Who is the user, and what are their needs?
✅ What does success look like?

A well-written PRD ensures everyone—engineering, design, sales, and leadership.

💡 Takeaway: Cross-functional alignment is important even before a single line of code is written.

9. Know your go-to-market (GTM) playbook

Building is just step one. Getting people to use it is step two.

from “we built it” → “how do we get adoption?”

Many teams celebrate a product launch but fail to plan for adoption. A feature doesn’t sell itself—you need a strong go-to-market (GTM) strategy to ensure people:
1️⃣ Know about it
2️⃣ Understand why they should use it
3️⃣ Actually adopt it

A great GTM strategy covers:
✅ Positioning: Who is this for, and why should they care?
✅ Messaging: How do we clearly communicate the value?
✅ Distribution: What channels (email, social, partnerships) will drive adoption?
✅ Onboarding & Activation: How do we ensure users experience value fast?

💡 Takeaway: A product’s success isn’t measured by launch—it’s measured by adoption.

10. Sales & Marketing are your allies

Engineering + Sales/Marketing = 🚀 Product Success

from working in silos → collaborating early

PMs don’t just work with engineering. Sales and marketing are key partners in ensuring product success.

• Sales teams interact with customers daily—they know what features prospects want.

• Marketing positions the product—crafting messaging that resonates with the target audience.

• Customer Success gathers feedback—helping refine what works and what doesn’t.

💡 Takeaway: PMs don’t just build products—they help sell them, too.

11. Talk to customers—but really listen

Users don’t always say what they need. Observe behavior, ask the right questions, and uncover real insights.

from “what do you want?” → “what problem are you facing?”

Customers will often ask for features—but that doesn’t mean those features solve their real pain points.

Instead of just taking requests, dig deeper:
✅ Why do you need this?
✅ What problem are you trying to solve?
✅ How are you currently working around it?

Watch how users behave, not just what they say. Sometimes, insights come not from direct feedback, but from how people actually interact with the product.

💡 Takeaway: Users are great at describing their problems—not always at suggesting the best solution.

12. Prioritization: the hardest but most important job

100 feature requests. 10 engineers. What do you build first?

from “let’s build everything” → “let’s build what matters most”

PMs deal with endless requests—from customers, sales, leadership, and engineering. But resources are always limited, so what gets built matters more than how much gets built.

Prioritization frameworks help decide:
✅ Impact vs. Effort – Is this a high-value feature or just a nice-to-have?
✅ Urgency vs. Long-Term Benefit – Does it solve an immediate pain or align with our strategic vision?
✅ Customer Value vs. Business Value – Does this help users and drive revenue?

💡 Takeaway: Prioritization isn’t just about what to build—it’s also about what NOT to build.

13. Develop an API-led mindset

Modern products aren’t standalone—they connect, integrate, and scale.

from features → ecosystems

APIs aren’t just a technical decision—they’re a strategic enabler. PMs who think API-first design products that integrate seamlessly into users’ workflows.

• APIs extend product reach – Let third-party apps enhance your offering.

• APIs create ecosystem lock-in – Users depend on your platform’s integrations.

• APIs open new revenue streams – API monetization is a growing business model.

💡 Takeaway: Think beyond products—build platforms.

14. Finally, have fun building products!

Product management is exciting. Stay curious, experiment boldly, and enjoy the process.

💡 Takeaway: Great PMs don’t just build things—they create impact.

Wrapping It Up

Moving from engineering to product management isn’t just about picking up new skills—it’s about changing the way you think. Engineers are great at figuring out how to build things. But as a PM, your job is to ask, "Are we even building the right thing?"

This role isn’t about having all the answers—it’s about asking better questions, making tough trade-offs, and thinking beyond execution to impact. You’ll work across teams, talk to customers, analyze data, and sometimes, just trust your gut. You’ll have to say no more than you say yes and learn to be okay with that.

At the end of the day, great PMs don’t just build products—they solve problems that matter. And that’s what makes this role so exciting.

💡 Final thought: You’re not just shipping features—you’re shaping the future of what people use and love.

Which shift has been the hardest for you? Let’s talk—drop a comment or DM me! 🚀

Artificial intelligence has steadily progressed, transforming from simple rule-based systems into sophisticated tools capable of natural conversations, predictions, and even creative tasks. Yet, for all its advancements, traditional AI has remained largely reactive—it responds to inputs, follows predefined rules, and operates within rigid boundaries. Agentic AI is poised to change that narrative entirely. This new wave of AI doesn’t just wait for commands; it takes initiative, makes decisions, and acts with autonomy. It’s not just an upgrade; it’s a transformation.

Agentic AI is the next big step forward. It shifts AI from being a passive assistant to an active collaborator. Imagine the leap from a static recommendation system that suggests washing machines based on your browsing history to an AI that evaluates your specific needs, compares real-time options, analyzes energy ratings, and even places the order—all autonomously. That’s the potential Agentic AI brings, and it’s already beginning to take shape.

At its heart, Agentic AI introduces the concept of “agency” into artificial intelligence. Traditional AI systems excel at solving narrowly defined problems by processing fixed inputs to deliver predictable outputs. For example, an AI chatbot might answer customer queries by drawing on a predefined database. However, it cannot adapt to new data or navigate tasks that fall outside its training.

Agentic AI changes this dynamic. It empowers systems to plan, adapt, and operate independently in real-time. This shift is made possible by advancements in large language models (LLMs) like GPT-4, Claude, and Google Gemini. These models have moved beyond generating static responses to understanding workflows, making decisions, and collaborating with other AI agents or tools. Paired with frameworks like LangChain and Autogen, Agentic AI orchestrates multiple components to complete complex tasks.

Consider warehouse management. A traditional AI might help optimize inventory based on past sales trends. An Agentic AI, on the other hand, could dynamically analyze inventory levels, adjust supply orders in response to real-time demand, and reroute shipments during disruptions—all without human oversight. This ability to operate iteratively—reflecting, learning, and improving with each step—is what sets Agentic AI apart.

Real-World Applications: Beyond Hypotheticals

Agentic AI might sound futuristic, but its use cases are already making waves in industries like human resources, healthcare, logistics, and marketing. Let’s take a closer look at how it’s transforming these sectors.

In human resources, onboarding new employees often involves repetitive and time-consuming tasks. Preparing personalized training schedules, assigning materials, and tracking progress typically require significant manual effort. An Agentic AI can take over these responsibilities, analyzing the new hire’s role, customizing onboarding paths, scheduling sessions, and even monitoring progress. HR professionals are then free to focus on strategic initiatives rather than mundane administrative tasks.

Healthcare presents even more compelling opportunities. Imagine an AI system monitoring a hospital’s intensive care unit. If a patient’s condition changes suddenly, the system could cross-reference their medical history, suggest potential interventions, and alert the medical team with actionable insights—all in real time. This capability not only enhances operational efficiency but can also save lives by identifying critical issues faster than human staff alone.

In marketing, Agentic AI systems are being used to automate entire campaigns. From creating content tailored to specific demographics to analyzing real-time engagement metrics and refining strategies, these systems handle the operational details, allowing marketers to focus on creativity and innovation. For instance, a marketing agent could identify trends on social media, generate ad content, and optimize campaigns dynamically, adjusting to audience responses instantly.

Product Owners (POs) can think of agents as specialized team members with distinct personas, each responsible for a specific part of a process. Just like in a team, where different roles collaborate to achieve a goal, agents work together to handle tasks efficiently. Each agent can be imagined as a persona with unique strengths and responsibilities, making it easier to relate to their functionality.

For instance, in the HR onboarding example:

• The Strategist (Role Analyzer Agent): The persona who understands the new hire’s role and gathers all the relevant details, like an analyst setting the groundwork for a project.

• The Tailor (Onboarding Customizer Agent): The creative problem-solver who designs personalized onboarding plans, much like a UX designer customizing user journeys.

• The Organizer (Scheduler Agent): The meticulous coordinator who aligns schedules and ensures everything is on track, just like a project manager.

• The Watchdog (Progress Tracker Agent): The diligent observer who monitors progress and flags potential issues, similar to a QA lead keeping an eye on quality.

By imagining agents as personas, POs can better understand how each "team member" contributes to the process, delegates tasks, and works collaboratively to achieve the desired outcome—just like orchestrating roles in a product team. This perspective makes the functionality of Agentic AI intuitive and relatable.

Subscribe now

The Building Blocks of Agentic AI

Creating an Agentic AI system involves several key components, each playing a vital role in ensuring the system’s effectiveness. Let’s break it down:

• Planning: The system must first understand the overall goal and break it into smaller, manageable tasks. For example, optimizing a supply chain might involve analyzing inventory levels, scheduling shipments, and prioritizing key orders.

• Execution: Here, the AI carries out the tasks using tools, databases, or other agents. For instance, a logistics AI might coordinate with suppliers, check traffic conditions, and adjust delivery schedules in real time.

• Refinement: Agentic AI systems don’t just execute; they learn and improve. By incorporating feedback loops and evaluation metrics, they refine their outputs to deliver better results over time.

• Collaboration: Multi-agent architectures allow specialized agents to work together. One agent might handle data retrieval, another might analyze trends, and a third might focus on execution. These agents collaborate to tackle complex challenges efficiently.

  

Image: Components of Agentic Workflow, source: Vellum.ai

Frameworks like LangChain simplify the process of building these systems, providing prebuilt tools and workflows that developers can customize to suit their needs. For example, a developer might create an AI for managing customer support, equipping it with tools for sentiment analysis, knowledge retrieval, and real-time response generation.

Challenges and Considerations

While Agentic AI offers incredible potential, it also comes with challenges. Ensuring reliability, maintaining ethical standards, and balancing autonomy with human oversight are critical considerations. For instance, a healthcare AI managing patient data must adhere to stringent privacy regulations while providing accurate and timely recommendations.

Another challenge lies in error handling. Agentic AI systems must be equipped with safeguards to detect and address issues. A failure in one part of the workflow could have cascading effects, making robust validation mechanisms essential.

Human-in-the-loop systems can mitigate some risks by allowing humans to review and approve critical decisions. For example, an AI managing a marketing campaign might seek approval before launching new content or making major budget adjustments. These systems combine the efficiency of AI with the expertise and judgment of humans.

A Glimpse Into the Future

The future of Agentic AI is expansive, touching virtually every industry. In logistics, autonomous agents could optimize global supply chains, responding to weather disruptions or geopolitical changes in real time. In education, AI tutors might create personalized learning plans, adapting content to suit each student’s pace and interests.

One particularly exciting area is urban management. Imagine cities where AI systems monitor traffic patterns, adjust signals dynamically to reduce congestion, and even manage energy consumption based on real-time demand. These applications go beyond convenience; they address pressing global challenges like sustainability and efficiency.

While some experts view Agentic AI as a stepping stone to Artificial General Intelligence (AGI), its immediate value lies in solving practical, real-world problems. By embedding these systems into workflows, businesses and governments can achieve efficiencies, reduce costs, and enhance decision-making processes.

Building and deploying Agentic AI systems is not without its hurdles, but the benefits far outweigh the challenges. These systems are more than just technological tools; they are collaborators capable of augmenting human intelligence and creativity.

As Agentic AI becomes more embedded in our processes, it will not only change how we work but also how we think about problem-solving and innovation. This isn’t just an evolution of AI—it’s a shift toward a future where humans and machines work together seamlessly to achieve more than either could alone. The journey is just beginning, but the possibilities are endless.

Disclaimer: The views and insights presented in this blog are derived from information sourced from various public domains on the internet and the author's research on the topic. They do not reflect any proprietary information associated with the company where the author is currently employed or has been employed in the past. The content is purely informative and intended for educational purposes, with no connection to confidential or sensitive company data.

Image Courtesy: Star Sports

Artificial Intelligence (AI) is surely making its mark in media broadcasts, bringing a new dimension to how live telecasts are produced and experienced. A recent screen grab from the India vs. Australia Test match illustrates just how far AI has come in creating immersive viewing experiences. The person in the center of the frame, positioned by AI, gives the impression of standing right on the field, blending seamlessly into the action around him. It’s subtle but effective, drawing the viewer’s attention and making the presentation feel more dynamic and connected to the event.

From a viewer's perspective, this AI-driven positioning creates a sense of closeness to the game, almost as if the presenter is part of the field rather than simply standing in front of a static background. It enhances engagement and ties the commentary to the match in a visually compelling way. However, while this feature is undoubtedly impressive, it’s not without its shortcomings.

One glaring issue is the lighting. The open stadium creates a mix of bright and shaded areas, and the AI struggles to balance them. This results in uneven lighting across the frame, with certain parts appearing overly exposed while others remain in shadow. It’s a small detail, but one that subtly detracts from the overall viewing experience. These imperfections highlight the limitations of current AI systems in understanding and adapting to complex environments like outdoor stadiums.

For product managers, this example provides an important lesson: launching a product or feature that’s "good enough" to deliver value is often more impactful than waiting indefinitely for perfection. The telecast, while not perfect, was live and immediately started adding value by engaging viewers. This decision allowed the team to collect real-world feedback—an invaluable tool for improvement. Waiting for everything to be perfect could have delayed the launch and missed the chance to understand how users interact with the feature. This demonstrates how progress comes from action, not hesitation, and how valuable insights can be gained when a product is out in the world.

This approach captures the essence of a Minimum Viable Product (MVP):

Minimum Viable Product (MVP) refers to a product or feature that is functionally complete and addresses a core problem for its target audience. While it is ready for real-world use, it deliberately focuses on the essentials and invites user feedback to refine and polish the experience. An MVP is not the rawest version of a product but one that balances usability and readiness, ensuring it can deliver value while revealing areas for improvement through user insights.

Early releases are not just about speed to market—they’re opportunities to learn, iterate, and improve. In this case, the AI-driven framing and positioning feature is already improving the user experience, even though there’s still room to make it better. By putting the feature out there, broadcasters have been able to see how it performs, identify what works, and understand where adjustments are needed. Without taking this step, progress would have been stalled, and the full potential of the feature might have remained untapped. Each iteration builds on the last, turning feedback into concrete improvements and moving the product closer to its ideal state.

This brings us to an important question: how should product managers think about MVPs?

When you’re building a product or feature, it’s tempting to chase perfection. We want everything polished and seamless before releasing it to the world. But here’s the thing: if you wait too long, you miss the chance to learn. The cricket telecast example is a perfect reminder. Was it flawless? No. But it was live, functional, and valuable. It made the broadcast better, even with its imperfections. That’s the spirit of an MVP—getting out there, solving a problem, and using real-world feedback to improve.

As a product manager, your MVP doesn’t have to be groundbreaking—it just has to start moving the needle. Here are a few ways to approach it:

Find the Heart of the Problem

Think of your MVP as a way to answer one key question: What’s the most important thing your user needs? Everything else can wait. For the cricket telecast, the goal was to make the presenter feel like part of the game. Did they nail the lighting? No, but they got the framing right, and that mattered more in the moment.

💡 Try this: Picture your feature as a tree. The trunk is the core problem your users face, and the branches are all the extra things you could add. Build the trunk first. Your MVP should stand tall even without the branches.

Make It Valuable, Even If It’s Simple

Your MVP shouldn’t feel like a prototype; it should still provide value. Users should be able to say, “This helps me.” It doesn’t need all the bells and whistles—it just needs to work. In the broadcast example, the AI framing created a sense of connection for viewers, even though the lighting wasn’t perfect.

💡 Ask yourself: What’s the one thing your feature must do well to make users feel like it’s worth their time?

Don’t Be Afraid of Feedback

The beauty of an MVP is that it’s a conversation starter. Once it’s out in the world, users will tell you what works and what doesn’t. Feedback is your map to the next version. Think about the AI broadcast team—they now know they need to improve lighting adjustments. Without going live, they wouldn’t have seen this issue in action.

💡 Remember: Feedback isn’t criticism—it’s free advice from the people who matter most: your users. Build a system to collect it, whether it’s through surveys, analytics, or even casual chats with early adopters.

Done Is Better Than Perfect

Perfection is a moving target, and chasing it can paralyze progress. Imagine if the broadcasters had waited until the lighting was flawless—they might have missed the chance to showcase the AI framing entirely. Instead, they launched, learned, and improved. That’s the game plan for MVPs: act, observe, refine, repeat.

💡 Think this way: Your MVP is like planting a seed. It won’t grow into a tree overnight, but it needs to get into the ground to start growing at all.

This is what makes MVPs and AI in broadcasting—and many other industries—so exciting. It’s not just about the technology itself but also about how these small, incremental improvements drive meaningful progress. Each release, even with its imperfections, takes the product one step further. With every cycle of feedback and improvement, the user experience becomes better, and the product gets closer to delivering its full potential. It’s a reminder that sometimes, the most important thing is simply to take that first step, knowing that perfection isn’t required to start creating value and learning for the future.

AI is changing the way products are built and experienced, and Product Managers (PMs) are now stepping into roles where traditional skills meet cutting-edge technology. Moving into an AI Product Manager (AI PM) role is not just about learning new concepts; it’s about understanding how to integrate AI capabilities like sentiment analysis into products that truly solve user needs.

Sentiment analysis is one of the simplest ways to explore the power of AI. It helps decode user emotions and feedback, giving valuable insights to improve customer experience. This blog will cover:

• How to approach product management with an AI mindset.

• A step-by-step guide to importing and using a pre-trained sentiment analysis model with Python.

• Practical tips to run the code and use it for building user-focused features.

The goal is to make AI approachable for PMs and provide a hands-on way to start exploring its potential. Whether you are new to AI or already curious about how to apply it, this guide will help you take the first steps.

Thinking Like an AI Product Manager

Traditional Product Managers focus on features, user journeys, and achieving product-market fit. While these remain important, stepping into the role of an AI Product Manager means evolving your thought process to account for the unique capabilities and challenges of AI.

• Shift to Outcomes: The focus moves beyond delivering features to understanding the outcomes AI enables. These might include automating repetitive tasks, making accurate predictions, or significantly enhancing user experience. Instead of asking, "What feature should we build?" the question becomes, "What value can AI create for the user?"

• Data as the Foundation: AI is only as good as the data it learns from. A strong AI PM understands the lifecycle of data—what data the product generates, how to clean and organize it, and how to use it for training or fine-tuning models. Building a product roadmap now involves considering not just features but also the quality, availability, and privacy of data.

• Experimentation First: AI thrives in environments that allow rapid experimentation and iteration. Testing ideas, validating models, and measuring impact become part of the workflow. Unlike traditional PMing, where features often follow a linear development path, AI capabilities demand iterative cycles to refine and deliver genuine value.

Take sentiment analysis as an example. Integrating it into a CRM application could transform how customer queries are managed. By classifying queries as positive, negative, or neutral, sentiment analysis enables personalized responses, better prioritization of support tickets, and smarter allocation of resources. This is the kind of impact that comes from thinking like an AI PM—focusing on outcomes, leveraging data, and embracing experimentation to drive meaningful results.

Understanding Sentiment Analysis in AI Products

Sentiment analysis is a natural language processing (NLP) technique used to identify the emotional tone behind a piece of text—whether it’s positive, negative, or neutral. This capability is widely applied across various domains to improve user experiences and decision-making processes. Common examples include:

• Customer feedback analysis: Understanding how users feel about your product or service to prioritize improvements.

• Social media monitoring: Tracking sentiment trends to measure brand perception or the success of marketing campaigns.

• Prioritizing customer complaints: Quickly categorizing and addressing urgent or critical issues for better service delivery.

Why Use Pre-Trained Models?

Building and training AI models from scratch requires deep technical expertise, a significant amount of data, and computational resources. For most applications, this isn’t practical. Instead, AI Product Managers can save time and resources by leveraging pre-trained models. These models have already been trained on large datasets and fine-tuned for specific tasks, making them ready to use with minimal additional effort.

One example is Hugging Face’s distilbert-base-uncased-finetuned-sst-2-english, a model fine-tuned for sentiment analysis tasks. By using such models, AI PMs can quickly prototype, test, and deploy features like sentiment analysis without the need to build everything from scratch. This approach enables faster iteration and ensures that your focus remains on solving user problems and delivering value.

Adopting pre-trained models not only lowers the barrier to entry but also helps AI PMs gain hands-on experience with AI technologies, which is essential for making informed product decisions.

Hands-On: Using Sentiment Analysis with Python

Below is an example of using Chainlit and Python to implement a sentiment analysis feature with minimal effort. This code demonstrates how to integrate a pre-trained model, classify user input, and respond dynamically based on sentiment. The workflow is straightforward, leveraging the power of Hugging Face's distilbert-base-uncased-finetuned-sst-2-english model for text classification.

The Code Breakdown

The Python code provided demonstrates how to integrate a pre-trained sentiment analysis model into a conversational AI application using Chainlit and FastAPI. This approach enables quick deployment of a sentiment analysis feature that responds dynamically based on user input. Let’s walk through the key components step by step.

1. Importing the Libraries

import chainlit as cl from transformers import pipeline

• Chainlit: A framework designed for building conversational AI applications. It simplifies the process of creating chat interfaces that handle real-time user interactions.

• Transformers: A versatile library by Hugging Face that provides pre-trained AI models for a wide range of natural language processing (NLP) tasks such as sentiment analysis, text generation, translation, and more.

These libraries form the backbone of the application, enabling you to combine conversational interfaces with advanced AI capabilities.

2. Loading the Pre-Trained Model

@cl.cache def load_model(): 

return pipeline("text-classification", model="distilbert-base-uncased-finetuned-sst-2-english")

• pipeline(): A utility from the Transformers library that simplifies the process of loading and using pre-trained models for specific tasks. In this case, we’re using a text classification pipeline, which is designed for tasks like sentiment analysis.

• model="distilbert-base-uncased-finetuned-sst-2-english": Specifies the pre-trained model to use. This particular model is fine-tuned for sentiment analysis, classifying text as positive or negative with high accuracy.

Why Use Caching?

The @cl.cache decorator ensures the model is loaded into memory only once during the application’s lifecycle. This avoids the overhead of reloading the model with each user interaction, significantly improving performance and reducing latency for subsequent requests.

By caching the model, the application becomes more efficient, especially when handling multiple user queries in real-time.

3. Handling User Input

@cl.on_message async def main(message):     
try:       
user_query = message.content

• This block is where the application interacts with the user in real time. Here's what happens:

  • @cl.on_message: A decorator provided by Chainlit to listen for and handle incoming messages from the chat interface. Each time a user sends a message, this function is triggered.

  • message.content: Extracts the text content of the user’s query. This allows the application to process the user's input dynamically.

  • user_query: Stores the user’s input for further processing, including passing it to the sentiment analysis model for classification.

  By capturing the user’s input seamlessly, this block sets the stage for analyzing the query and generating a meaningful response, making it the cornerstone of the conversational flow.

4. Performing Sentiment Analysis

result = classifier(user_query) 

intent = result[0]['label'] confidence = result[0]['score']

• Here’s what happens in this step:

  • classifier(user_query): The pre-trained model processes the input text provided by the user. Using the text classification pipeline, it analyzes the sentiment of the text and generates a prediction.

  • intent: Extracts the sentiment label from the model’s output. The label indicates whether the sentiment is POSITIVE, NEGATIVE, or, in some cases, NEUTRAL.

  • confidence: Captures the confidence score of the prediction. This is a numerical value (between 0 and 1) that represents how confident the model is about its classification.

  For example, if a user types "I love the service," the model might output:

  [{'label': 'POSITIVE', 'score': 0.98}]

  Here, the intent is POSITIVE, and the confidence score is 0.98, indicating high certainty in the result.

  By extracting these values, the application is equipped to interpret user emotions and tailor responses accordingly. This step bridges the gap between raw AI output and actionable insights.

5. Generating a Response

if intent == "NEGATIVE": 

response = "We're sorry to hear about your experience. Our team will look into this issue promptly." 

elif intent == "POSITIVE": 

response = "Thank you for your positive feedback! We're glad you're satisfied with our service." 

else: response = "Thank you for reaching out. Could you please provide more details about your query?"

• In this step, the application uses the predicted sentiment (intent) to craft a relevant and empathetic response. Here's the breakdown:

  • Negative Sentiment: If the sentiment is classified as "NEGATIVE," the response acknowledges the user's dissatisfaction and assures them that the issue will be addressed. This builds trust and shows proactive support.

  • Positive Sentiment: For a "POSITIVE" sentiment, the application expresses gratitude and reinforces the user’s positive experience, fostering goodwill and customer satisfaction.

  • Neutral or Uncertain Sentiment: When the sentiment is unclear or neutral, the response politely requests more details, ensuring the conversation remains constructive and user-focused.

  By tailoring responses to the user's sentiment, the application creates a more personalized and engaging experience. This dynamic interaction is key to enhancing user satisfaction and delivering value in conversational AI.

6. Sending the Response

await cl.Message( content=f"**Intent:** {intent} (Confidence: {confidence:.2f})\n\n**Response:** {response}" ).send()

• This step delivers the generated response to the user through the Chainlit chat interface. Here's how it works:

  • cl.Message(): Creates a new message object for the chat interface. The content parameter is used to specify the message text.

  • Formatted Output: The message includes the following:

    • Intent: Displays the predicted sentiment (e.g., POSITIVE, NEGATIVE) to provide transparency about the analysis.

    • Confidence: Shows the model’s confidence score (formatted to two decimal places), offering insight into the reliability of the prediction.

    • Response: Provides a context-specific reply based on the sentiment analysis result.

  • .send(): Sends the message back to the user in real-time, ensuring a seamless conversational flow.

    This ensures the user feels acknowledged and understood, closing the loop on the sentiment analysis process with a polished, user-friendly interaction.

Sentiment Analysis Chat Response Using Chainlit

How to Run and Execute This Code

Setting up and running the sentiment analysis chatbot is straightforward. Follow these steps to get started:

1. Install the Dependencies

Before running the code, ensure you have the required libraries installed in your environment. Use the following command:

       pip install chainlit transformers

This installs Chainlit for creating the conversational interface and Transformers for working with the pre-trained model.

2. Save the Code

Save the provided Python script into a file named sentiment.py in your working directory. This will serve as the main application file for the chatbot.

3. Run the Chainlit App

Launch the Chainlit server by running the following command in your terminal:

      chainlit run sentiment.py

Once the server starts, you’ll see a URL in your terminal output indicating where the chatbot is hosted.

4. Interact with the App

Open the URL (e.g., http://localhost:8000) in your web browser. You’ll see a clean chat interface where you can interact with the sentiment analysis application.

From Code to a Deployed AI Feature

Building a working prototype is just the beginning. To transform the sentiment analysis chatbot into a product-ready feature, the following steps can help you bridge the gap between development and deployment:

1. API Deployment

Expose the sentiment analysis model as an API to make it accessible to other applications.

• Use FastAPI: Create a lightweight and efficient REST API that processes text input and returns sentiment labels and confidence scores.

• Consider AWS Lambda: If you're looking for a serverless solution, AWS Lambda can host your model as a scalable function. This is ideal for minimizing infrastructure management.

Example:
The API could accept customer feedback as input and return structured data like:

{ "sentiment": "POSITIVE", "confidence": 0.98 }

2. Integration with CRM

Integrate the API into your Customer Relationship Management (CRM) system to automate processes and enhance workflows.

• Ticket Categorization: Automatically classify incoming tickets as positive, negative, or neutral to prioritize responses based on urgency.

• Response Generation: Leverage the sentiment results to provide AI-powered, pre-drafted responses that customer service representatives can review and send.

Impact: This reduces manual effort, improves response times, and ensures consistency in customer communication.

3. Scalability

To handle real-world usage and growing demand, scalability is key.

• Deploy on AWS Elastic Beanstalk: Package the application and deploy it with Elastic Beanstalk for automatic load balancing and scalability.

• Containerize with Docker: Use Docker to package the application along with its dependencies. Deploy the container on Kubernetes or other orchestration platforms to scale seamlessly across multiple environments.

4. Monitoring and Optimization

Once deployed, monitoring performance is essential:

• Use CloudWatch (AWS) or Prometheus/Grafana to track usage, response times, and errors.

• Regularly retrain the model with updated data to improve its accuracy and relevance.

Example Use Case in Production

Imagine your CRM is receiving thousands of customer feedback entries daily. By deploying this sentiment analysis feature:

• Negative tickets are flagged and prioritized for immediate action.

• Positive feedback is acknowledged automatically with pre-crafted responses.

• Neutral feedback is analyzed for trends, helping the product team identify areas for improvement.

Transitioning from code to a deployed feature involves technical steps but adds immense value to your product. This approach showcases the power of AI in delivering scalable, customer-focused solutions.

Insights for Emerging AI Product Managers

1. Simplify AI with Pre-Trained Models: Using pre-trained models like Hugging Face’s offerings streamlines development, making it easier to prototype and deploy AI features without needing extensive data science expertise.

2. Focus on Real-World Use Cases: Sentiment analysis isn’t just about text classification—it’s a tool to enhance customer experiences. Automate responses, prioritize critical issues, and identify feedback trends to drive actionable improvements.

3. Embrace Hands-On Learning: Practical exercises like this provide a solid understanding of AI workflows—from importing a pre-trained model to integrating it into a functional application. Experience is key to bridging the gap between theory and application.

4. Foster Collaboration with Technical Teams: AI Product Managers act as the bridge between business goals and technical solutions. Understand the technical aspects enough to communicate effectively and align AI capabilities with user needs and product strategy.

These approaches empower AI Product Managers to navigate the complexity of AI and deliver meaningful features that create value for users and businesses alike.

Transitioning into AI Product Management demands curiosity, a willingness to learn, and practical experience. Working on projects like this sentiment analysis app helps you build the confidence and skills needed to integrate AI capabilities seamlessly into your products. At its core, AI is not just about advanced technology—it’s about crafting intelligent solutions that create meaningful value for your users.

Disclaimer: The views and insights presented in this blog are derived from information sourced from various public domains on the internet and the author's research on the topic. They do not reflect any proprietary information associated with the company where the author is currently employed or has been employed in the past. The content is purely informative and intended for educational purposes, with no connection to confidential or sensitive company data.

In the world of artificial intelligence (AI), particularly with generative AI and large language models (LLMs), real-time inference is critical to delivering meaningful and responsive applications. Inference is the phase where AI models, already trained on extensive data, process new inputs to make predictions or generate responses. Inference is what powers real-time interactions in applications from customer service chatbots to medical diagnosis support. However, effective deployment of inference-based AI workloads requires robust infrastructure, highly efficient processing, and flexible, scalable architectures. Enter NVIDIA’s Inference Microservices (NIM), a solution designed to optimize and simplify the deployment of inference models, whether in the cloud or at the edge.

For product managers, understanding the advantages, applications, and deployment options for inference-based AI is essential to harnessing AI’s full potential. NVIDIA NIM offers a modular, efficient, and cost-effective approach, enabling the seamless deployment and scaling of powerful AI models across diverse environments. This article explores how NIM addresses the challenges of inference-based workloads and provides actionable insights for product managers.

Understanding Inference-Based AI Workloads and GPU Utilization

AI models require two main phases to function effectively: training and inference.

1. Training Phase: During training, an AI model learns patterns from a vast dataset, adjusting its parameters to minimize errors. This is a resource-intensive process typically performed offline and requires significant computational power.

2. Inference Phase: Inference is where the trained model applies its knowledge to make predictions on new data in real time. This phase is critical in production environments, where responsiveness can directly impact user experience and overall system performance.

When deploying large language models, the inference process consumes substantial GPU memory and processing power, as each new response is generated token by token. Each token’s generation requires the model’s weights to be loaded onto the GPU, demanding significant resources. Product managers can achieve better performance and reduced latency by optimizing GPU usage, particularly through techniques like tokenization and memory-efficient caching.

The Shift from Monolithic to Microservices Architectures in AI

Traditional software systems were often built using monolithic architectures, where all components were interconnected in a single application. While monolithic systems have a straightforward structure, they pose challenges in terms of scalability, flexibility, and maintainability. As the need for adaptability and responsiveness grew, microservices architectures emerged, breaking applications into smaller, independent services. Each service can be developed, deployed, and scaled separately, bringing multiple benefits:

• Modularity: Isolated services make updates and maintenance easier and safer.

• Scalability: Specific components can scale independently according to demand, optimizing resources and reducing costs.

• Flexibility: Diverse technologies can be integrated, enabling faster adaptation to evolving business requirements.

This shift toward modularity has also influenced AI deployments. Instead of relying on a single, large model, AI systems are now adopting specialized models, each focused on a distinct function. NVIDIA’s Inference Microservices framework brings this microservices architecture to AI deployments, offering modular solutions that can be easily deployed and scaled based on application needs.

NVIDIA NIM: Empowering Product Managers with Scalable, Flexible AI Deployments

NVIDIA NIM enables product managers to simplify AI model deployment through containerized microservices. This allows AI models to operate as independent services that can be efficiently scaled across cloud and edge environments. The modular nature of NIM’s architecture is particularly useful in high-demand applications, as each containerized model can operate independently, enabling optimized resource utilization and greater agility.

NVIDIA NIM API Catalog Page

Key Features of NVIDIA NIM

1. Containerized Models: NIM provides AI models pre-packaged as containers with all necessary dependencies. This allows for deployment flexibility across cloud, on-premises, and edge environments, ensuring faster time-to-market and a streamlined setup process.

2. Dynamic Resource Allocation with Batch Processing: NIM supports dynamic batching, enabling the efficient processing of multiple requests simultaneously. This feature is especially useful in high-traffic applications, such as recommendation engines and real-time chatbots, where optimized resource allocation is essential for maintaining low latency.

3. Memory Optimization with KV Cache: For applications requiring continuous interaction, like customer service bots, maintaining context between exchanges is crucial. NIM’s Key-Value (KV) cache retains processed data, allowing the model to remember context without recalculating it for each new token. This significantly reduces latency and improves the user experience by enabling faster responses.

4. Precision Optimization and Model Quantization: NIM includes advanced precision optimizations, such as FP8 quantization, which reduces the memory footprint of models without sacrificing accuracy. These optimizations make it possible to deploy inference-based AI on resource-limited devices, a significant benefit for real-time applications running on edge devices.

Deployment Options for NIM on Amazon SageMaker and EKS

For organizations already utilizing cloud platforms, NVIDIA NIM can be deployed on services like Amazon SageMaker and Amazon Elastic Kubernetes Service (EKS), both of which provide extensive support for scalable AI workloads.

Amazon SageMaker

Amazon SageMaker is a fully managed machine learning service that supports the building, training, and deployment of ML models. By integrating NIM with SageMaker, product managers gain access to a scalable, managed infrastructure that simplifies deployment, tracking, and resource management. SageMaker handles monitoring, resource scaling, and versioning, enabling teams to focus more on optimizing the user experience rather than managing complex infrastructure.

Amazon EKS

For teams requiring customized configurations, Amazon EKS offers a managed Kubernetes service that integrates seamlessly with NIM. Deploying NIM on EKS allows for precise control over containerized environments, which is beneficial for enterprise applications with complex requirements. Product managers can take advantage of EKS’s orchestration and scaling capabilities to manage high-demand AI workloads efficiently.

Comparison with Competitive Alternatives: AWS Bedrock, Hugging Face, and Others

While NVIDIA NIM offers an optimized framework for deploying containerized AI models with exceptional support for GPU-accelerated workloads, several other platforms provide valuable features that may suit different business needs or deployment preferences. Here’s a brief overview of some competitive alternatives and how they compare:

1. AWS Bedrock: Amazon’s Bedrock platform provides a range of foundation models and is fully integrated with Amazon Web Services. Bedrock excels in offering managed AI services, including popular open-source models and proprietary solutions developed by AWS. For product managers who prioritize a highly integrated, managed environment with broad access to AWS resources (like data lakes, pipelines, and security), Bedrock may be an appealing choice. However, NIM's direct optimization for NVIDIA GPUs offers performance advantages for compute-intensive, real-time applications where speed and latency are critical.

2. Hugging Face: Hugging Face is a widely-used platform for open-source models, including LLMs and multimodal models. With a vast model repository and community support, Hugging Face appeals to product teams looking for flexibility and an extensive range of model choices. While Hugging Face’s platform is excellent for experimentation, training, and deploying smaller or customized models, it may lack the deep GPU integration and performance optimizations that NIM offers, particularly for high-demand applications requiring extensive parallel processing.

3. Google Vertex AI: Google’s Vertex AI offers an end-to-end platform for deploying, managing, and scaling machine learning models, with strong integration into the Google Cloud ecosystem. Vertex AI benefits from Google’s robust infrastructure and ease of use for both training and inference workloads. However, it may not offer the same containerization flexibility and edge deployment capabilities as NVIDIA NIM, which is tailored for performance and efficient resource usage in both cloud and edge environments.

4. IBM Watson Machine Learning: IBM’s Watson Machine Learning platform supports a range of AI workloads and is known for its enterprise-grade capabilities and security features. It’s a strong choice for industries requiring regulatory compliance, such as healthcare and finance. However, for organizations prioritizing GPU-accelerated performance and low-latency inference at scale, NIM’s close alignment with NVIDIA hardware could offer distinct advantages.

Edge Computing and Real-World Applications of NIM

The modularity and efficiency of NVIDIA NIM make it well-suited for edge computing applications, where data is processed closer to the source. Edge AI offers several advantages over cloud-only deployments, including reduced latency, lower bandwidth consumption, and enhanced privacy.

Real-World Use Cases for Edge Deployment

1. Smart Home Security: By deploying NIM on edge devices, smart home systems can perform face recognition and other security tasks in real time without sending data to the cloud. This not only enhances privacy but also reduces latency, making interactions seamless and secure.

2. Industrial IoT and Predictive Maintenance: In manufacturing, edge-based AI can monitor equipment and alert operators to potential issues before they result in downtime. This capability is crucial in predictive maintenance, allowing companies to prevent costly equipment failures.

3. Healthcare Monitoring: Edge AI is revolutionizing healthcare, enabling real-time monitoring of patient data. Medical devices equipped with NIM-enabled AI models can provide instant feedback, alerting healthcare professionals to potential risks in real time, which is critical in high-stakes scenarios.

Anticipated Advancements in NVIDIA’s Edge GPU Technology

NVIDIA's commitment to advancing edge AI is evident in its development of platforms like the Jetson AGX Orin, which delivers unprecedented performance for robotics and autonomous machines. This platform provides 6x the processing power of its predecessor, enabling complex AI computations directly on edge devices. Looking ahead, NVIDIA plans to release new AI chips annually, with the next-generation Rubin platform expected in 2026. This accelerated roadmap aims to enhance AI capabilities significantly, offering more efficient and powerful solutions for edge applications.

Impact on Various Industries

The advancements in edge AI are poised to revolutionize multiple sectors:

• Healthcare: Edge AI enables real-time patient monitoring and diagnostics, allowing for immediate interventions and personalized treatment plans.

• Manufacturing: Predictive maintenance powered by edge AI can foresee equipment failures, reducing downtime and optimizing production efficiency.

• Retail: In-store analytics can enhance customer experiences through personalized recommendations and efficient inventory management.

• Transportation: Autonomous vehicles equipped with advanced edge AI can process data locally, improving safety and decision-making capabilities.

For product managers, staying abreast of these developments is crucial. Integrating cutting-edge edge AI technologies can lead to more responsive, efficient, and innovative products, providing a competitive advantage in the marketplace.

Product Management Strategy: Making Informed Choices for AI Workloads

For product managers overseeing AI deployments, selecting the right infrastructure and deployment strategy is essential. Here are key considerations and strategic steps to help guide decision-making:

1. Cloud vs. Edge Deployment:

   • Cloud Deployment: Ideal for applications requiring extensive computational resources, such as large-scale data processing or models needing frequent updates. Cloud offers virtually unlimited scaling but can introduce latency for applications with real-time needs.

   • Edge Deployment: Best for low-latency, real-time applications like smart devices or IoT. Deploying on the edge allows data processing closer to the source, reducing latency and bandwidth costs. Edge deployment is particularly useful in remote or bandwidth-limited environments where immediate response times are critical.

   • Decision Criteria: Product managers should assess the application’s latency tolerance, data sensitivity, and bandwidth constraints when deciding between cloud and edge. For applications requiring immediate processing or high data privacy, edge is preferable, while cloud is better suited for intensive, centralized data processing.

2. Budgeting for GPU Resources:

   • GPU costs are a significant part of the AI infrastructure budget. Product managers should consider demand forecasting to predict GPU usage accurately, avoiding over-provisioning while ensuring sufficient resources for peak demand.

   • Implementing dynamic resource allocation—a feature available in NIM—can help manage costs by adjusting resources in real time based on workload requirements, maximizing GPU utilization.

   • ROI Calculation: To calculate the return on investment for inference-based AI workloads, product managers can use a combination of performance metrics (such as response time improvements or user engagement metrics) and cost metrics (like GPU hours saved through dynamic batching or edge processing).

3. Case Study: Implementing NIM in a Customer Service Chatbot:

   • Scenario: Imagine a retail company, “ShopSmart,” which wants to enhance its customer service experience by deploying an AI-powered chatbot to handle real-time queries, product recommendations, and personalized support.

   • Strategy: ShopSmart chooses NVIDIA NIM for its deployment, leveraging both cloud and edge options. During peak shopping hours, NIM’s dynamic batching optimizes GPU usage by processing multiple queries simultaneously on Amazon SageMaker. During off-peak hours, the chatbot is run on edge devices in-store, allowing local response to customer queries, saving bandwidth, and ensuring quick responses.

   • Results: By combining cloud and edge deployments, ShopSmart achieves an optimal balance of responsiveness, cost efficiency, and scalability. GPU costs are managed effectively, and the chatbot provides fast, relevant responses without overwhelming infrastructure costs. Product managers at ShopSmart can monitor metrics like time-to-first-response and user satisfaction to continually optimize the service.

Subscribe now

Model Optimization and Performance Monitoring

To maximize the effectiveness of inference-based AI, product managers should consider key performance metrics. NVIDIA NIM supports several optimization techniques, such as model quantization and precision adjustments (like FP8), to reduce memory usage and increase processing speed. These optimizations are particularly advantageous for edge devices, where computational resources are limited.

Metrics for Monitoring Inference Performance

Tracking specific metrics is essential to ensuring optimal performance. Product managers should consider:

• Time-to-First-Token: Measures how quickly the model generates its first response, a key metric for user experience.

• Inter-Token Latency: Evaluates the speed between generating consecutive tokens, crucial for applications where response times are critical.

• Total Generation Time: Assesses the entire response duration, helping teams gauge efficiency for high-frequency applications.

These metrics provide actionable insights that help product managers refine models, monitor performance, and align with user expectations.

Leveraging Fine-Tuning and Retrieval-Augmented Generation (RAG)

For specialized applications, product managers might consider fine-tuning and Retrieval-Augmented Generation (RAG). Fine-tuning involves training a pre-trained model with additional data, enhancing its relevance for specific industry needs. Meanwhile, RAG combines retrieval-based methods with generative AI to enhance contextual accuracy by pulling information from external sources as needed.

These methods are invaluable in applications that require a high level of specificity, such as financial analysis or legal document processing. Fine-tuning and RAG allow models to generate more accurate, contextually relevant outputs, improving user satisfaction and engagement.

NVIDIA’s Model Catalog: Options for Diverse Business Needs

NVIDIA NIM’s model catalog offers a diverse selection of pre-trained models optimized for different domains, from general-purpose language models like nvidia/llama-3.1-nemotron-70b-instruct to industry-specific models for healthcare, retail, and more. Product managers can select from these ready-to-deploy models, reducing time-to-market while ensuring high-quality, contextually relevant outputs. For organizations with specific needs, this catalog provides a solid foundation that can be further customized.

A New Era in AI with NVIDIA NIM

Inference-based AI workloads are transforming real-time applications, and NVIDIA’s Inference Microservices (NIM) framework empowers product managers to deploy AI models flexibly and efficiently across cloud platforms like Amazon SageMaker and EKS, as well as in edge environments.

NIM’s features — including containerized deployment, dynamic batching, KV caching, and precision optimization — make it ideal for managing resource-intensive AI workloads. As edge computing and real-time engagement grow, leveraging NVIDIA’s GPUs and optimized inference techniques will be crucial for maintaining a competitive edge.

For product managers, mastering inference-based AI and understanding NIM’s capabilities can redefine how they deploy and manage AI solutions, reshaping user experiences and unlocking new opportunities. By exploring NIM’s features, product managers can create scalable, high-performing AI applications that meet the dynamic demands of modern consumers in an AI-driven world.

Disclaimer: The views and insights presented in this blog are derived from information sourced from various public domains on the internet and the author's research on the topic. They do not reflect any proprietary information associated with the company where the author is currently employed or has been employed in the past. The content is purely informative and intended for educational purposes, with no connection to confidential or sensitive company data.

Generative AI is rapidly becoming essential in regulated fields like life sciences, finance and insurance, where data-driven insights, compliance, and the ability to deliver accurate results are paramount. For product managers building SaaS solutions in these spaces, selecting the right generative model isn’t about opting for the most advanced technology; it’s about aligning the model with strategic goals, regulatory requirements, and available technical resources. In a field where options from OpenAI, Anthropic, Meta, Google, and others continue to evolve, this choice requires more than just technical insight—it requires a strategic perspective.

This blog dives deep into the considerations product managers should weigh when selecting a transformer model for SaaS applications. We’ll explore model architectures, unpack the true impact of pricing, and discuss long-term strategic factors that can ensure a sustainable, future-proof choice.

Understanding Transformer Architectures: Tailoring Model to Task

Transformer architectures fall into three main categories—encoder-only, decoder-only, and encoder-decoder—each with its unique strengths. By understanding the differences, product managers in life sciences can make better decisions on which models will align with their specific goals, such as improving patient care, enhancing research accuracy, or streamlining regulatory compliance.

1. Encoder-Only Models (e.g., BERT)

What They Do Best:
Encoder-only models are optimized for understanding and analyzing input data without generating extensive output. This makes them ideal for classification, data extraction, and analysis tasks where accuracy and interpretability are essential.

Life Sciences Use Case:
In life sciences, BERT could be employed to sift through large sets of clinical trial data to classify patient symptoms and side effects, helping researchers identify patterns without manually combing through thousands of records. With an encoder-only model, it’s possible to extract insights on patient demographics, treatment outcomes, and correlations across studies, providing a clearer picture of trial efficacy and safety profiles.

2. Decoder-Only Models (e.g., GPT-4)

What They Do Best:
Decoder-only models specialize in generating fluent, coherent text, making them suitable for tasks requiring high-quality text output. These models are widely used in summarization, content creation, and real-time response generation.

Life Sciences Use Case:
GPT-4 could be used to summarize complex genetic research papers, generating concise reports that are easy for cross-functional teams to understand. For example, researchers who need an overview of cutting-edge studies in genomics can use a decoder model to create readable summaries from dense, technical language. This enables faster knowledge transfer, reduces the cognitive load on team members, and ensures that everyone stays aligned with the latest developments in genetic research.

3. Encoder-Decoder Models (e.g., T5)

What They Do Best:
Encoder-decoder models excel at both understanding and generating text, making them suitable for tasks that require data comprehension followed by coherent output. This type of model is effective for translation, summarization, and structured output generation.

Life Sciences Use Case:
A model like T5 could assist in creating structured patient reports by translating data from patient registries into readable summaries. For instance, in rare disease research, encoder-decoder models can analyze patient data, identify notable trends, and generate comprehensive reports to assist medical teams in clinical decision-making. This capability can help aggregate critical patient information into user-friendly formats, improving communication between researchers and healthcare providers.

Strategic Impact: Real-World Scenarios and Model Fit

By focusing on real-world scenarios, product managers in life sciences can better understand the potential impact of each architecture type. From streamlining clinical trial analysis to enhancing collaboration on genomic research, these use cases provide a tangible way to assess how models can support specific tasks within the life sciences field. Choosing a model type isn’t just a technical decision—it’s a strategic one, determining how effectively the model can contribute to the organization’s overarching goals.

Building Resilience with Upstream Data Quality

Product leaders know that high-quality data is the bedrock of AI’s success, especially in regulated industries. Ensuring this quality upstream—before data ever hits the machine learning pipeline—is essential. This is where Data Contracts come into play, a concept that may sound technical but ultimately serves a straightforward purpose: clarity.

A Data Contract is essentially an agreement between data producers (like clinical data systems) and data consumers (AI models and SaaS applications) to set standards for data integrity and structure. For product managers, implementing these contracts helps guarantee that only validated data moves downstream, reducing errors and aligning with both regulatory and operational expectations.

Pricing: Beyond Initial Costs to True ROI

While model capabilities are important, pricing structures can often become the deciding factor. Life sciences applications tend to process large datasets, and the costs associated with model usage can escalate quickly. Here’s a closer look at the hidden and ongoing costs to consider.

1. Usage-Based Pricing

Understanding Usage Costs:
Most providers, like OpenAI and Anthropic, use a token-based pricing model. In life sciences, where data volume can be substantial, this pricing structure can lead to fluctuating costs, especially when usage spikes during high-demand periods, such as during regulatory reviews or clinical trials.

Recommendation:
To mitigate unpredictable costs, product managers should analyze historical data to estimate usage patterns and potential costs. Working with providers that offer custom pricing or cost caps can also help manage spending during periods of increased data processing.

2. Infrastructure and Computational Power

The Hardware Demand:
Advanced transformer models often require powerful GPUs to operate effectively. In-house deployments of models like Google’s Gemini or Meta’s Llama 4 may necessitate GPUs like the NVIDIA H100, which, while powerful, come with a significant financial investment. For life sciences, where models might process high volumes of patient or trial data, this infrastructure can become a core component of the total cost.

Recommendation:
Consider whether the current infrastructure can support these models, and if not, evaluate the long-term benefits of investing in additional GPU resources. Alternatively, cloud providers such as AWS or Google Cloud offer specialized AI hardware that can meet these computational demands without the upfront investment in physical infrastructure.

3. Customization and Fine-Tuning Costs

Fine-Tuning for Specific Needs:
In life sciences, model customization can be critical. Fine-tuning allows for adaptation to specific datasets, such as optimizing a model for interpreting clinical trial outcomes or analyzing patient narratives. Platforms like Hugging Face facilitate fine-tuning, but it comes with added costs in terms of resources and time.

Recommendation:
Assess the ROI of fine-tuning by calculating how much it could improve accuracy or efficiency. For example, if fine-tuning reduces the need for manual data processing in clinical research, this could save on operational costs and accelerate insights, providing measurable returns on investment.

4. Inference Costs

Running Predictions at Scale:
Inference costs refer to the expenses incurred each time a model generates predictions. For high-frequency applications in life sciences—such as real-time patient data monitoring—these costs can accumulate quickly, especially when using fine-tuned models that demand more processing power.

Recommendation:
Implement a usage tracking system to monitor real-time inference costs. This approach helps observe cost fluctuations, allowing to make adjustments to stay within budget and avoid unexpected expenses.

5. Comparative Pricing Across Providers

Evaluating Pricing Models:
While some providers offer competitive pricing structures, it’s crucial to balance these against the technical fit for your application. For instance, Google’s TPUs may be more affordable than NVIDIA GPUs, but they may not be suitable for every model type or workload.

Recommendation:
Conduct pilot tests across multiple providers to evaluate not only cost but also performance under life sciences-specific tasks. For example, consider the latency, accuracy, and model response times, as these metrics are crucial for applications dealing with real-time health data or patient records.

Balancing ROI with Long-Term Costs

Beyond initial expenses, the potential for long-term ROI is key. In life sciences, a carefully chosen model can reduce manual labor, increase compliance accuracy, and improve overall patient outcomes. A cost-benefit analysis that projects these savings will help justify the investment to stakeholders while providing a realistic view of any hidden costs associated with scaling.

Subscribe now

Strategic Considerations: Ensuring a Sustainable Model Choice

Selecting the right generative model is a strategic decision that has lasting implications on product success and adaptability. Beyond immediate capabilities, the model should align with your organization’s long-term vision and technical goals.

1. Scalability and Adaptability

Preparing for Growth:
In life sciences, data volumes and complexity tend to increase over time. Models that offer scalability, like Llama 4 and Claude 3.5, are built for growth, though they may require robust infrastructure. When handling data from diverse sources—such as patient registries, trial results, and genomic information—scalability is essential to keep performance consistent.

Recommendation:
Opt for a model that integrates well with cloud infrastructure, allowing for flexible scaling. Providers who support scalable configurations and modular upgrades are preferable as they minimize disruptions when scaling up data processing capabilities.

2. Compliance and Data Privacy

Meeting Industry Standards:
Data privacy and compliance are crucial in life sciences. Whether processing patient records, clinical trials, or genetic data, any model must meet strict regulatory standards like GDPR or HIPAA. Providers like Anthropic emphasize AI safety, which aligns well with the ethical and legal considerations that life sciences require.

Recommendation:
Request detailed compliance documentation from providers to ensure they support necessary standards. Engaging a third-party compliance expert to verify these claims may provide additional assurance. Some providers even offer custom compliance solutions tailored to regulated industries, which can streamline the integration process.

3. Reliability and Ongoing Support

Ensuring Consistent Performance:
Regular updates to AI models are common, as they often bring enhancements or bug fixes. In life sciences, where accuracy and reliability are critical, choosing a provider with a history of consistent support and maintenance can make a difference.

Recommendation:
Look for providers that offer strong customer support and frequent updates. Ask about their track record with reliability, especially in fields where even minor discrepancies in data output can impact patient care or research outcomes.

4. Provider’s Innovation Track Record

Supporting Future Needs:
The AI landscape evolves rapidly, and it’s valuable to work with a provider that has a history of innovation and improvement. Providers like Google and OpenAI regularly introduce advancements and contribute to the research community, signaling a commitment to staying ahead.

Recommendation:
Research a provider’s history, recent contributions, and standing within the AI community. Providers with a proactive approach to innovation are likely to offer models that can adapt to emerging needs in life sciences, keeping your product competitive and capable of leveraging new AI capabilities.

A Product Manager’s Framework: Practical Steps for Informed Decision-Making

Choosing the right transformer model is a multi-faceted process. Here’s a step-by-step guide to help product managers make a well-rounded, strategic decision:

Step 1: Define Objectives and Key Metrics

Set clear, specific objectives for the model. For example, is the goal to streamline data analysis in clinical trials, improve patient record summarization, or enhance data compliance? Define measurable metrics that align with each goal, such as improved processing time, accuracy rates, or reductions in manual labor.

Step 2: Run Pilot Tests with Real-World Data

Test models with data specific to your use case, such as anonymized patient records or research papers. This will help gauge real-world performance and reveal potential issues like latency or misclassification that could affect practical deployment.

Step 3: Leverage Platforms for Fine-Tuning and Integration

Platforms like Hugging Face can simplify the fine-tuning process and offer compatibility with multiple cloud providers. For example, life sciences applications often benefit from models fine-tuned on specialized datasets, and Hugging Face allows for controlled testing to ensure optimal results.

Step 4: Involve Cross-Functional Teams Early

Involve stakeholders from compliance, engineering, and operations in the decision-making process. Their perspectives can reveal challenges you may not anticipate, ensuring the model fits within both technical and regulatory frameworks.

Step 5: Monitor Regularly and Iterate

Once deployed, regularly monitor the model’s performance and costs, and plan for updates as requirements change. Life sciences applications demand precision, so adjustments based on observed performance trends can prevent potential issues.

Making a Strategic Choice for Sustainable Growth

In life sciences, selecting the right transformer model goes beyond technical specifications. It’s about finding a solution that enhances your product, aligns with regulatory requirements, and is capable of scaling as needs evolve. By evaluating the nuances of model architecture, pricing structures, and strategic considerations, product managers can make informed decisions that support immediate objectives while preparing for future growth.

This thoughtful, strategic approach isn’t just about adopting the latest technology; it’s about choosing a model that becomes a reliable partner in advancing your product’s impact in life sciences, now and in the future.

Disclaimer: The views and insights presented in this blog are derived from information sourced from various public domains on the internet and the author's research on the topic. They do not reflect any proprietary information associated with the company where the author is currently employed or has been employed in the past. The content is purely informative and intended for educational purposes, with no connection to confidential or sensitive company data.

Large Language Models (LLMs) are fundamentally changing the way industries utilize, interpret, and apply data-driven insights. Sectors like agriculture, finance, healthcare, and energy are no longer relying on generic, one-size-fits-all AI solutions. Instead, they are pivoting towards more tailored approaches that directly address the unique challenges and requirements of each industry. By employing techniques like Fine-Tuning and Retrieval-Augmented Generation (RAG), these models are evolving beyond their original broad applications, enabling them to provide detailed, context-specific recommendations and insights that are essential for informed decision-making.

This approach not only enhances the model’s understanding of domain-specific terminology and processes, but also allows it to incorporate real-time, dynamic information from external sources, making the output more relevant and actionable. Furthermore, the development of newer lightweight models, such as Mistral’s Ministral family, marks a pivotal step towards making these AI capabilities accessible on edge devices. These advancements pave the way for more focused, adaptable, and resource-efficient AI solutions that can operate effectively in environments where low latency, privacy, and localized processing are crucial. This guide takes a closer look at these transformative techniques and explores how they can open new opportunities for specialized applications across various industries.

From Generalist to Specialist: Fine-Tuning LLMs for Industry-Specific Expertise

When thinking about how LLMs operate, it's clear they come with a broad foundation of knowledge—a general understanding that can answer many types of questions with reasonable accuracy. However, real-world applications in industries like agriculture, finance, or healthcare often require more than just general insights. They demand a deeper, context-specific understanding of specialized terms, processes, and scenarios that are unique to each field. This is where fine-tuning steps in to bridge that gap.

Fine-tuning takes an existing pre-trained model, which has already been trained on a wide variety of data, and refines it using datasets tailored specifically to a particular industry. By focusing on data relevant to the domain—such as crop management practices, weather patterns, or pest control techniques in the case of agriculture—the model becomes equipped to handle specialized queries. Take, for example, the case demonstrated by Microsoft’s research, where fine-tuning was applied to an agricultural context. This process led to a 6% improvement in accuracy when it came to providing responses related to farming activities. Essentially, the fine-tuned model could now generate precise recommendations based on established agricultural standards and practices.

However, there is a trade-off that product managers and decision-makers need to consider. Fine-tuning is not a simple plug-and-play technique; it demands a substantial investment in both time and resources. It involves curating large and reliable datasets that accurately represent industry knowledge. Moreover, fine-tuning requires significant computational power to retrain the model effectively, which may not always be readily available depending on the scale of the project. Another consideration is the need for continuous updates. Industries are dynamic, and the knowledge base of even the most robust models will need regular updates to stay aligned with current practices, regulations, and new research findings.

Despite these challenges, fine-tuning remains a crucial step in transforming a general-purpose LLM into a specialized advisor capable of addressing intricate domain-specific questions with depth and relevance. By focusing on curating high-quality data and allocating the necessary resources, organizations can fine-tune models to become reliable assets in their sector.

Adding Dynamic Context with Retrieval-Augmented Generation (RAG)

While fine-tuning equips a model with essential domain-specific knowledge, there are plenty of situations where that alone isn’t enough. The reality is that industries are not static; they continuously evolve and require responses that reflect current events or fresh information. This is where Retrieval-Augmented Generation (RAG) becomes invaluable, as it enables the model to dynamically incorporate the most recent, relevant data from external sources.

The way RAG functions is by leveraging the foundational understanding gained from fine-tuning and enhancing it with updated information retrieved from external databases or knowledge repositories. Essentially, it’s a strategy that marries a model’s internal knowledge with real-time insights, allowing it to provide answers that are both accurate and current. Take the example of a farmer seeking guidance on when to plant seeds: while a fine-tuned model might offer general advice based on established agricultural practices, RAG enables that same model to consider the latest rainfall data or the current market prices for seeds. As a result, the recommendation becomes not only more precise but also more practically useful.

The impact of combining RAG with fine-tuning is significant. In trials conducted by Microsoft, applying RAG to a fine-tuned model resulted in an additional 5% increase in accuracy for industry-specific tasks. This layered approach, where fine-tuning builds a robust baseline of knowledge and RAG keeps that knowledge flexible and up-to-date, helps LLMs excel in real-world contexts that are constantly changing.

However, there’s a complexity to implementing RAG effectively. It requires maintaining a well-organized and regularly updated external database, which can be a logistical challenge. This database must be curated meticulously to ensure that the information remains accurate and relevant over time. Despite this, the payoff is substantial. The integration of RAG enables LLMs to transcend the limitations of their pre-trained data, offering the kind of dynamic and responsive intelligence that many industries, like agriculture, truly need.

In the broader scope, the advantage of RAG is its ability to provide a bridge between historical knowledge and present-day realities. This capacity to adjust and contextualize answers based on the most recent and relevant data enhances the value of LLMs in practical applications. For sectors that are influenced by fluctuating variables—be it weather in agriculture, financial markets, or evolving regulations in healthcare—RAG becomes a vital tool in ensuring that the insights generated remain timely and applicable.

Models on the Edge: Mistral’s Ministral Family and Lightweight AI Solutions

As the demand for edge-based AI solutions continues to grow, recent advancements in LLMs are reshaping what’s possible on smaller devices. A standout example is Mistral AI’s Ministral family, which includes the Ministral 3B and Ministral 8B models. These innovations are specifically crafted for edge applications, making them capable of running effectively on devices such as smartphones, laptops, and even some IoT setups. By focusing on delivering high-performance capabilities with a reduced parameter count, Mistral AI has made significant strides in pushing the boundaries of where and how these models can be deployed.

The Ministral 3B model is geared towards mobile scenarios, enabling advanced functionalities on smartphones or lightweight devices without consuming excessive resources. In contrast, the Ministral 8B is built for more robust environments, catering to devices with greater GPU capacity and computational power. This model is ideal for laptops or specialized hardware setups that demand higher performance while still maintaining efficiency.

One of the standout features of the Ministral family is its efficiency in handling contextual information while using a fraction of the parameters that traditional large-scale LLMs might require. Despite their smaller size, Ministral models have shown impressive results in key benchmarks, outperforming similar-sized offerings from major players like Google and Meta. They have demonstrated capabilities in areas such as multilingual support and common-sense reasoning, proving that being lightweight doesn’t mean compromising on depth or accuracy. For instance, in several multilingual and common-sense reasoning tests, Ministral’s models consistently delivered results that exceeded expectations, showcasing their strength in versatile, real-world applications.

Image: Ministral 3B and 8B models compared to Gemma 2 2B, Llama 3.2 3B, Llama 3.1 8B and Mistral 7B on multiple categories (source: mistral.ai)

What’s particularly valuable about these lightweight models is their accessibility and practicality. For developers and organizations looking to deploy AI on consumer-grade hardware, the Ministral family offers an attractive alternative to relying solely on cloud-based solutions. By operating efficiently on edge devices, these models minimize the need for continuous cloud connectivity, which can lead to lower operational costs, reduced latency, and enhanced user privacy.

The ability of the Ministral family to perform effectively on edge devices isn’t just about convenience—it opens up a range of new possibilities for applications that demand real-time processing and autonomous functionality. From smart home systems that need to respond quickly to user commands, to agricultural devices that offer on-the-spot recommendations based on sensor data, lightweight models like Ministral enable AI to be embedded seamlessly into our everyday technology.

These advancements reflect a shift towards making AI more distributed and closer to the point of interaction, whether that’s a phone in a farmer’s hand or a laptop guiding critical decisions in a corporate office. It’s not merely a matter of shrinking the size of models; it’s about intelligently designing them to fit into a broader, decentralized AI ecosystem where they can respond and adapt in real time, directly at the edge.

Optimizing LLMs for the Edge: Techniques in Pruning, Distillation, and Quantization

Deploying large language models on edge devices introduces specific challenges due to the limited memory, power constraints, and the need for efficient processing speeds. This is where optimization techniques like pruning, distillation, and quantization become crucial.

Pruning serves to streamline the model by trimming away less critical or underutilized parameters, effectively reducing the overall size of the model. Think of it as a way to remove unnecessary branches from a tree while keeping its core structure intact. This method helps the model operate more efficiently on devices with constrained resources, ensuring that only the essential aspects remain to deliver robust performance.

Distillation, on the other hand, focuses on creating a more compact version of the original model. The approach involves training a smaller model—often referred to as a "student" model—to replicate the behavior and capabilities of the larger "teacher" model. By transferring the essential knowledge from the larger model, distillation allows the student model to perform nearly as well, but with a significantly smaller footprint. This makes it particularly advantageous for deploying AI on smartphones or IoT devices, where processing power and memory are more limited.

Quantization is all about reducing the precision of the model’s weights and computations. By converting 32-bit weights to lower-bit versions, quantization decreases memory usage and speeds up inference times without drastically affecting the model's accuracy. For applications like providing farmers with real-time advice in remote areas, where connectivity might be unreliable, quantized models can deliver fast, reliable insights without causing battery drain or excessive resource consumption.

Connecting the Dots: Personas in Agriculture and Data-Driven Decision-Making

In agriculture, the impact of LLMs extends beyond individual users, creating a ripple effect that influences an interconnected network of stakeholders. Each of these personas depends on accurate, timely information to make data-driven decisions:

Image: Personas in Agriculture (source: microsoft.com)

Farmers, at the core, rely on precise insights to make key decisions about crop cycles, assess weather-related risks, and identify market opportunities. They’re the primary recipients of AI-driven advice that informs critical day-to-day choices in their fields.

Agronomists and Input Providers are the advisors, guiding farmers on selecting the best seeds, fertilizers, and pest control measures. Their role is to ensure that farmers are equipped with the right resources to maximize yield and maintain crop health.

Consumers, meanwhile, drive the demand for transparency and sustainability in agricultural products. They push producers towards delivering higher quality and accountability, influencing the types of data that LLMs must track and interpret.

Retailers and Consumer Packaged Goods (CPG) Companies aim to streamline supply chain operations while maintaining consistent quality standards. Their focus on efficiency and quality depends on accurate predictions and insights into production and logistics.

Bankers provide the financial backbone of the agricultural sector, assessing risks and extending financial support based on reliable data and projections. Their decisions are informed by a comprehensive understanding of market trends and risks, underpinned by data-driven analysis.

Policy Makers shape the agricultural framework by creating accessible policies, ensuring farmers comply with regulations, subsidies, and safety standards. Their decisions rely on a deep understanding of sector needs and impact.

Researchers / Data Scientists drive innovation in agriculture, developing AI models to automate workflows and deliver predictive insights. Their work informs strategic decisions, optimizing crop management and resource allocation.

The interconnected nature of these personas demonstrates how LLMs can serve as a powerful tool for enhancing decision-making and fostering collaboration throughout the entire agricultural ecosystem. By delivering precise and actionable insights to each stakeholder, LLMs enable a more cohesive and responsive approach to agricultural management.

Tailoring Edge Solutions for Farmers

For farmers working in remote regions, reliable connectivity can be a significant hurdle. They often can’t depend on cloud-based solutions due to inconsistent or non-existent internet access. This reality makes it essential to bring AI directly to their devices. By deploying fine-tuned and RAG-enhanced models on edge devices like smartphones, farmers can receive real-time, location-specific advice right where they need it. The use of distilled and quantized versions of these models ensures that these advanced AI systems run efficiently without overburdening the limited resources of the devices in use.

Integrating these edge models with localized document repositories enables the devices to function independently, allowing farmers to access crucial information even when they are disconnected from the internet. This autonomy is particularly valuable in rural areas where connectivity is a luxury rather than a given. Furthermore, pairing these edge models with on-device sensors creates a system capable of delivering hyper-local insights. The AI can take in real-time data related to soil conditions, climate factors, and crop status, and provide highly tailored recommendations, increasing the model’s relevance and practicality for on-the-ground decision-making.

Embracing Practical AI with Lightweight Models

The development of models like Mistral’s Ministral family signals an important shift toward more practical AI solutions designed specifically for edge computing. These models are compact, but still pack a punch in terms of functionality, making them a viable choice for a range of applications beyond agriculture, including smart home systems and autonomous technologies. Their design addresses the need for AI that can operate effectively without continuous cloud access, which not only saves bandwidth but also reduces latency and enhances data privacy.

For product managers, this shift calls for a strategic approach in selecting which models to deploy. The focus shouldn’t just be on choosing the most advanced model but on balancing the device’s capabilities with the need for up-to-date and context-aware information. Models like Ministral 3B and 8B offer real-world alternatives to traditional cloud-based setups, making them adaptable to various scenarios where reliability and efficiency are critical.

Insights for Domain-Focused Product Managers

Deploying large language models (LLMs) in specialized fields like agriculture requires a careful balance between cloud-hosted solutions and edge deployments. While cloud-based LLMs offer substantial capabilities and computational power, lightweight models designed for edge devices bring efficiency and real-time responsiveness. For product managers focusing on specific domains, this dual strategy introduces unique challenges that need to be managed effectively.

Key Considerations for Model Consistency and Performance

User Experience and Output Coherence: When using different LLMs for cloud and edge—such as a large model hosted on cloud infrastructure and a smaller edge model like Mistral’s Ministral series—it is crucial to ensure both models consistently deliver accurate insights. Differences in architecture or training data can lead to variations in output, which might confuse end-users who depend on coherent information across multiple devices.

Deployment Complexity: Managing a hybrid deployment strategy across cloud and edge devices involves handling different infrastructures, resource constraints, and maintenance protocols. Product managers need to establish a streamlined deployment strategy that reduces disruptions and keeps updates synchronized across platforms, maintaining compatibility and reliability.

Training and Fine-Tuning: Aligning training datasets and fine-tuning processes for each model is vital to ensure consistent interpretation and response to input data. Inconsistencies in training parameters or data sources could lead to diverging results, impacting critical decisions across the domain.

Balancing Cloud and Edge Models for Domain-Specific Needs

As more efficient models enter the market, finding the right balance between fine-tuning, deployment strategies, and real-time information retrieval becomes crucial. This balance ensures that AI delivers relevant and actionable insights to all key stakeholders—from analysts and engineers to advisors and policymakers. The objective is to build an ecosystem where every decision is backed by accurate, up-to-date information, empowering stakeholders to make informed choices and drive meaningful progress.

Ultimately, the challenge and opportunity lie in adopting these technologies thoughtfully. By focusing on deployment specifics and aligning model capabilities with real-world requirements, product managers can create a more resilient and responsive system, translating advancements in AI into improved decision-making and sustainable growth within their respective domains.

Disclaimer: The views and insights presented in this blog are derived from information sourced from various public domains on the internet and the author's research on the topic. They do not reflect any proprietary information associated with the company where the author is currently employed or has been employed in the past. The content is purely informative and intended for educational purposes, with no connection to confidential or sensitive company data.

The VRIO framework is a tool for evaluating a company's resources and capabilities to understand if they can serve as a source of sustained competitive advantage. The image illustrates a step-by-step analysis based on four key factors: Valuable, Rare, Inimitable, and Organized. Each of these criteria builds on the previous one, leading to either a competitive disadvantage, parity, or a sustainable advantage. Here's how you can implement this framework in product management decisions.

1. Valuable (V): Is the feature or capability valuable to your customers?

In practice, this means evaluating whether a new feature or product enhancement provides a solution to an essential customer pain point or creates value for the company in terms of revenue or customer satisfaction. For example, introducing a streamlined dashboard in your SaaS product that saves users time might add significant value.

• How to apply it: Conduct user research, surveys, or gather feedback to understand the key pain points and align your roadmap priorities with those that solve high-impact issues. Use KPIs like user engagement, NPS (Net Promoter Score), or customer churn reduction to assess if a feature is valuable.

• Result: If a feature does not provide value, it’s not worth prioritizing. If it does, move on to the next step in VRIO.

2. Rare (R): Is this feature or capability rare among your competitors?

Once you've identified a valuable feature, the next question is whether it's rare. In other words, do your competitors already offer a similar solution? If your competitors have already saturated the market with this feature, it may offer only competitive parity.

• How to apply it: Perform a competitive analysis by studying the product offerings of your key competitors. Identify what they do not have or what they haven't mastered yet. For instance, if your competitors lack advanced integrations with third-party tools, developing this capability may give your product an edge.

• Result: If the feature is both valuable and rare, it has the potential to give you an advantage. If not, you achieve competitive parity.

3. Inimitable (I): Is the feature difficult to imitate?

Even if your feature is valuable and rare, it can still be replicated by competitors. To gain a sustained advantage, your capability or feature should be difficult to copy. This can stem from technical complexity, proprietary technology, or even a superior user experience that is hard to replicate.

• How to apply it: Assess the feasibility of competitors copying your feature. If you have proprietary algorithms, extensive AI models, or deeply integrated solutions, these are hard to replicate. Features that require significant investment in technology or expertise are less likely to be copied.

  • For example, a unique AI-based recommendation engine may take years for a competitor to develop.

• Result: If your feature is hard to copy, you are moving toward a temporary competitive advantage.

4. Organized (O): Is your organization ready to support and capitalize on this feature?

Finally, the most valuable, rare, and inimitable features are useless if your organization is not structured to deliver them effectively. This includes having the right team, processes, infrastructure, and resources to develop, launch, and support the new feature or product.

• How to apply it: Ensure that your team has the necessary skills, that your development processes can handle the feature’s complexity, and that you can bring it to market efficiently. For example, if your team lacks expertise in AI, implementing an AI-driven feature could stall unless you first organize the resources.

  • Also, assess your post-launch processes, like customer support, monitoring, and scalability, to ensure that the feature's value can be maintained over time.

• Result: If your organization can’t support the feature, the competitive advantage might remain unused or unexploited. However, if your organization is equipped and ready, you have achieved sustained competitive advantage.

Putting VRIO into Action: Example in a SaaS Environment

Imagine you're a product manager for a CRM SaaS product, and you're evaluating a potential new feature: AI-driven customer insights.

1. Valuable: Does it solve a key customer pain point? Yes, it helps businesses identify trends in their customer data more efficiently, reducing the time spent on manual analysis.

2. Rare: Do competitors offer this? No, not many CRMs in your market have built-in AI capabilities for this level of customer insight.

3. Inimitable: Is it hard to replicate? Yes, because the insights rely on proprietary AI algorithms that would be difficult and time-consuming for competitors to develop.

4. Organized: Does your team have the necessary skills, and is the infrastructure in place? Yes, your company has invested in an AI research team, and your system architecture is built to support new AI features.

Now, let’s tie it all together and discussing how we apply the VRIO framework in a structured way for decision-making. The table below provides a visual summary of several features and how they map across the VRIO criteria:

• Support of Obsolete Reporting Tool: Since this feature fails in all four categories, it’s a clear example of a competitive disadvantage.

• Basic Mobile App for Retail: While it provides value, it’s a common feature and easy to replicate, resulting in competitive parity.

• New UI Design Tech: This feature provides value and is rare, but is easy to replicate, leading to a temporary competitive advantage.

• Blockchain-Based Data Security: Though this feature is valuable, rare, and difficult to replicate, the organization lacks the resources to fully implement it, making it an unexploited competitive advantage.

• AI-Powered Personalized Recommendations and Open Platform for Third-Party Retail Extensions both meet all VRIO criteria, providing sustained competitive advantage.

Each feature can be mapped to its corresponding competitive outcome. By using this framework, product managers can prioritize features in their roadmap that not only solve customer problems but also offer long-term differentiation and advantage. This structured approach ensures the right balance between immediate needs and strategic growth.

After working through the VRIO framework in detail, it’s essential to understand how the insights derived from this framework can directly inform and enhance your product strategy. The next step is to connect these competitive insights to the Innovation Value Map (IVM), which is a strategic tool to ensure that your product roadmap has a balanced mix of innovations, from incremental to disruptive.

The diagram below demonstrates how features evaluated through the VRIO framework can be mapped to different quadrants of the Innovation Value Map:

• Routine Innovations (Quadrant 3): Features like the Basic Mobile App for Retail fall here. Though they provide value, they are common across competitors and easy to replicate. They help maintain customer satisfaction but do not differentiate the product in the market.

• Disruptive Innovations (Quadrant 2): Open Platform for Third-Party Retail Extensions is an example of a disruptive innovation. It introduces new business models that extend the capabilities of the SaaS product and provide sustained competitive advantage.

• Architectural Innovations (Quadrant 1): Blockchain-Based Data Security fits into this quadrant. While this feature offers a high competitive advantage, the lack of organizational readiness makes it unexploited. When the organization is prepared to fully support the technology, this will move toward sustained advantage.

• Radical Innovations (Quadrant 4): Features like AI-Powered Personalized Recommendations and New UI Design Tech fall into this category. They introduce significant new technological capabilities, giving the product a temporary or sustained competitive advantage depending on whether they are difficult to replicate and whether the organization can fully support them.

By mapping VRIO-evaluated features to the IVM, product managers can clearly see which areas of their product strategy require focus. Features that offer sustained competitive advantage should be prioritized for long-term market leadership, while features that maintain competitive parity might be necessary to meet immediate customer needs without offering significant differentiation.

If you're interested in learning more about how the Innovation Value Map can guide your roadmap decisions, check out my blog post on Crafting a Balanced SaaS Roadmap. This will provide deeper insights into the balance of incremental and disruptive innovations to ensure your product remains both competitive and innovative over time.

The combined insights from the VRIO framework and the Innovation Value Map (IVM) provide a strategic foundation for making informed product roadmap decisions as shown below:

By evaluating features through the VRIO framework, product managers can determine whether features offer competitive disadvantage, parity, temporary, or sustained competitive advantage. This ensures the prioritization of features that not only address customer needs but also strategically position the product in the marketplace.

Once features have been evaluated, mapping them to the IVM helps to balance innovations across routine, disruptive, architectural, and radical innovations. This strategic alignment ensures that the roadmap is well-rounded, addressing both immediate customer needs and long-term market leadership.

By leveraging both VRIO and IVM, product managers are empowered to make data-driven decisions, balancing short-term gains with long-term strategic advantages, ultimately leading to a roadmap that ensures sustained product success and differentiation in a competitive market.

Disclaimer: The views and insights presented in this blog are derived from information sourced from various public domains on the internet and the author's research on the topic. They do not reflect any proprietary information associated with the company where the author is currently employed or has been employed in the past. The content is purely informative and intended for educational purposes, with no connection to confidential or sensitive company data.

As a SaaS product manager, you’re responsible for delivering a product that not only meets current customer demands but also anticipates future market shifts. However, you face constant pressure to balance immediate customer needs with long-term innovation. How do you avoid the trap of focusing on incremental improvements at the expense of game-changing innovation? The answer lies in balancing your roadmap.

Your roadmap is the blueprint for success, but the challenge often lies in ensuring this balance. Focusing too much on any one type of innovation can lead to a product that’s either stagnant and predictable or overly complex and disconnected from customer needs.

This is where the Innovation Value Map (IVM) comes in as a powerful tool for guiding your product development strategy. The map divides innovation into four quadrants: routine, radical, disruptive, and architectural. Each represents a different type of innovation that impacts your product in unique ways. By ensuring your roadmap contains a mix of innovations from at least two quadrants, you avoid a one-sided approach and create a product that’s both agile and forward-looking.

The diagram below illustrates how innovations can be categorized across the four quadrants of the Innovation Value Map, with examples drawn from various industries.

Here’s how each quadrant contributes to the overall health of your SaaS roadmap and why a balanced approach is essential for long-term success.

The Routine Quadrant: Sustaining the Present

Routine innovations are essential to keeping your SaaS product relevant and improving daily customer experiences. These are small, incremental changes like adding new filtering options, improving patient data entry UI, optimizing page load and checkout speed, or allowing customized notifications. Such updates are often driven by customer feedback and serve to enhance the product's usability without fundamentally changing how it works.

In SaaS, routine innovations increase customer satisfaction by addressing direct requests for slight feature improvements. These may not be game-changing, but they show that you're responsive and committed to constant improvement.

However, too much focus on routine updates can lead to stagnation. A roadmap filled solely with routine work risks becoming one-dimensional, leaving little room for more exciting innovations. Without balancing these smaller updates with radical or disruptive innovations, your product could struggle to stand out from competitors pushing for more transformative features.

Routine work is necessary, but it should always be balanced with innovations from other quadrants to ensure both immediate customer satisfaction and long-term product differentiation.

The Radical Quadrant: Pushing Boundaries

While routine innovation keeps the lights on, radical innovation sets the foundation for long-term success by creating new features or rethinking existing processes to deliver substantial value. These innovations aren't always driven by customer requests but instead come from market research, emerging trends, and internal vision. They often involve cutting-edge technology that redefines how users interact with your product.

For example, AI-driven demand forecasting could replace manual reporting in retail SaaS, predicting inventory needs based on customer behavior and trends. Similarly, machine learning for patient outcome prediction could transform healthcare platforms by offering real-time insights that are far beyond existing capabilities.

Another notable example of radical innovation is Google’s machine learning algorithms in Google Ads. Google transformed how advertisers target users by using machine learning to predict which ads are most likely to perform well based on user behavior. This innovation went beyond routine ad placement updates and enabled more efficient, personalized targeting, giving Google a competitive advantage in the digital advertising space. It added long-term value by enhancing user engagement and driving higher conversion rates for advertisers.

In SaaS, radical innovations can sometimes seem like gambles because they require significant resources to develop and often remain untested in the market. Yet, when successful, radical innovations deliver high rewards by differentiating your product and attracting new customer segments.

Consider a SaaS platform focused on sales automation. While routine innovations might involve UI improvements or adding more customization in reporting, a radical innovation could involve integrating AI to predict customer churn—a feature that competitors may not yet offer. This feature would not only help upsell current customers but could also bring in a new segment interested in advanced analytics.

Radical innovation doesn't mean abandoning your core product. Instead, it requires rethinking how your product can better serve your customers in the future. By placing radical innovations alongside routine updates, your roadmap remains stable yet future-oriented, ensuring both short-term satisfaction and long-term growth.

However, while radical innovations can transform your product, without a foundation of architectural stability, they may fail to scale. Similarly, routine updates keep customers happy, but without disruptive innovation, your product might struggle to maintain market leadership. This balance is key to creating a roadmap that can address both immediate and future needs.

The Disruptive Quadrant: Changing the Game

If radical innovations push boundaries, disruptive innovations break them. Disruptive innovations aim to transform how customers interact with your product or even how they work altogether, often introducing entirely new business models. These innovations are high-risk, high-reward projects that have the potential to reframe your entire value proposition and drastically alter market expectations.

In the SaaS world, disruptive innovations can be especially powerful in saturated markets where differentiation is difficult. For example, consider an open platform for third-party retail extensions, which allows developers to integrate new capabilities directly into a SaaS e-commerce platform. This shifts the product from a single solution to a flexible ecosystem that enables a wide range of customizations and functionalities.

Another example is the introduction of telemedicine solutions for virtual healthcare, which transforms how healthcare services are delivered by enabling remote consultations and care. Similarly, a subscription model for manufacturing software can completely change how businesses access high-end tools, shifting from one-time purchases to recurring revenue streams, which can open up new market segments.

While disruptive innovations can catapult your product into new categories, they come with their own challenges. They require significant investments in time, resources, and risk-taking. Existing customers might be hesitant to adopt such large shifts, and there’s no guarantee of market success. However, when disruptive innovations succeed, they can open doors to new markets and expand your reach significantly.

Atlassian is a great example of how disruptive innovation can pay off. Known for products like Jira and Confluence, Atlassian disrupted the enterprise software market with its freemium model, offering tools for free and monetizing through premium upgrades. As an HBR article on innovation strategy notes:

"Long-term investments in research are risky... But when closely linked to a company’s business strategy, they can lead to substantial, sustained growth."

Atlassian’s approach not only democratized access to powerful tools but also laid the groundwork for long-term success, proving that disruptive innovation, when aligned with a company’s long-term strategy, can result in sustained market leadership.

Balancing disruptive innovations on your roadmap ensures that your product doesn’t become complacent. While continuing to serve your current market with routine and radical innovations, disruptive features prepare you for the next wave of industry transformation, keeping your product ahead of the competition.

The Architectural Quadrant: Building for the Future

Architectural innovations focus less on individual features and more on the broader system infrastructure that supports your product. These innovations might not be immediately visible to customers, but they are critical for ensuring that your product remains scalable, adaptable, and sustainable. Architectural innovations can include overhauling the platform’s technology stack, introducing API ecosystems, or creating integration frameworks that allow other services to interact seamlessly with your SaaS product.

For example, a SaaS company might implement microservices architecture to improve performance and scalability, enabling faster updates and easier maintenance. Similarly, blockchain technology could be introduced to secure patient data across healthcare platforms, ensuring both privacy and scalability as the platform grows. Another example might be IoT-enabled supply chain monitoring, which allows real-time tracking of goods across multiple locations, providing critical data for decision-making in industries like manufacturing and retail.

While these changes might not generate immediate excitement or customer-facing buzz, they are essential for future-proofing your product. Architectural innovations lay the foundation for long-term success by ensuring that your platform can evolve with changing customer demands and advances in technology.

Without these foundational improvements, even the most feature-rich product will eventually hit limitations in terms of scalability and flexibility. However, architectural innovations require significant foresight and investment, as they often don’t offer the immediate returns that routine or radical innovations might provide.

By including architectural improvements in your roadmap, you ensure that your product remains robust and adaptable over time. Although the returns may be less immediate, these innovations set the stage for future radical or disruptive changes, enabling your product to continue growing and evolving.

Share

The Danger of One-Sided Roadmaps

Now that we've explored each quadrant, it’s important to recognize the risk of focusing too much on any one type of innovation. A roadmap that is heavily skewed toward routine improvements will struggle to stand out in the marketplace. One that focuses exclusively on disruptive innovation might alienate existing customers who need stability and reliability. Over-investing in architectural innovation could lead to a product that’s technically brilliant but lacks the features users care about.

The best roadmaps are those that strike a balance between the quadrants. For instance, if you’re planning a major release in the next year, consider mixing routine updates (such as feature enhancements requested by customers) with radical or disruptive features that will create excitement in the market. Complement these with architectural improvements that will enable future development and scalability.

A balanced roadmap ensures you’re both delivering short-term value and building for long-term growth. It allows your product to evolve at a pace that’s sustainable, minimizing the risk of burnout for your development team and confusion among your users.

Just as companies need an overarching innovation strategy to align their efforts, a product roadmap requires a strategic approach to ensure that no single type of innovation dominates. As an HBR article on innovation strategy points out:

“Without an innovation strategy, innovation improvement efforts can easily become a grab bag of much-touted best practices... A company without an innovation strategy won’t be able to make trade-off decisions and choose all the elements of the innovation system.”

The same applies to product management—without a clear innovation roadmap, you risk focusing too much on one area at the expense of others.

Incorporating Stakeholder Inputs Across Quadrants

Roadmap development is not just a product team's responsibility. Stakeholders from across the organization—such as engineering, customer support, sales, and marketing—play a critical role in shaping innovation priorities. To ensure a balanced product roadmap, it's essential to gather insights from stakeholders across the organization—engineering, sales, customer success, and more. As an HBR article on innovation strategy points out:

“Diverse perspectives are critical to successful innovation. But without a strategy to integrate and align those perspectives around common priorities, the power of diversity is blunted or worse becomes self-defeating.”

Similarly, without alignment between cross-functional teams on your product roadmap, you risk conflicting priorities and missed opportunities. Each team brings valuable insights that help ensure a balanced mix of innovations:

• Engineering: Provides technical feasibility and insights into architectural innovations that ensure scalability and sustainability.

• Customer Support: Shares recurring customer pain points, feeding into routine and incremental updates that enhance user satisfaction.

• Sales and Customer Success: Offers real-world feedback on potential market opportunities or customer demands that could inform both radical and disruptive innovations.

• Marketing: Helps identify emerging trends or competitive pressures, pushing for disruptive or radical features that differentiate the product.

By collaborating with these stakeholders, you ensure that your roadmap isn't one-sided or isolated. Inputs from various teams help balance short-term customer needs with long-term strategic goals, aligning innovations across routine, radical, disruptive, and architectural quadrants.

Crafting a Roadmap with Balance in Mind

When planning your product roadmap, use the Innovation Value Map as a guide to ensure balance. Here's a structured approach to achieving this:

1. Assess Your Current Roadmap: Start by categorizing your existing features and initiatives into the four quadrants. Are they mostly routine? Are there any radical or disruptive ideas in the mix? Does your architecture support future scalability?

2. Identify Gaps: If your roadmap leans heavily toward one quadrant, consider introducing innovations from another. For example, if most of your initiatives are routine, explore opportunities for radical innovation that could differentiate your product.

3. Set Priorities: Not all innovations need to be pursued simultaneously. Establish a timeline for when different types of innovations will be introduced, ensuring that you’re delivering both short-term wins and long-term growth opportunities.

4. Collaborate with Stakeholders: Engage with both customers and internal teams to understand their priorities. This will help you balance immediate needs with future-looking innovations that might require more investment.

5. Iterate Regularly: Roadmaps are not set in stone. Revisit the innovation value map periodically to ensure your product development is staying balanced as new technologies emerge and market conditions change.

Balancing Innovation for Long-Term Success

In SaaS, delivering on short-term customer needs while keeping an eye on future growth is a balancing act that requires strategic planning. The Innovation Value Map (IVM) provides a framework to ensure that your roadmap is not one-sided, helping you avoid the pitfalls of over-prioritizing one type of innovation. By integrating routine, radical, disruptive, and architectural innovations into your roadmap, you create a product that’s both reliable and ready for the future.

A well-balanced product roadmap ensures that routine updates, radical breakthroughs, disruptive innovations, and architectural improvements all work together to drive long-term success. As an HBR article on innovation strategy highlights:

“The vast majority of profits are created through routine innovation. But different kinds of innovation can become complements rather than substitutes over time.”

Similarly, balancing innovation types in a product roadmap ensures both short-term customer satisfaction and long-term growth. As a product manager, your goal is not just to maintain the present but to build a product that can thrive in tomorrow's market. A roadmap balanced across the Innovation Value Map ensures you’re ready for whatever challenges and opportunities the future holds.

What are you waiting for? Let’s get those IVMs rolling and start innovating! 🚀

Disclaimer: The views and insights presented in this blog are derived from information sourced from various public domains on the internet and the author's research on the topic. They do not reflect any proprietary information associated with the company where the author is currently employed or has been employed in the past. The content is purely informative and intended for educational purposes, with no connection to confidential or sensitive company data.