🤔 Dear Lewis, how do I pivot from traditional PM to AI PM?

A principal product manager faces the AI revolution and wonders if her years of experience are becoming obsolete.

Hey, it's Lewis. Most leadership and career advice feels like it was written by someone who's never actually managed people or navigated a layoff. This newsletter is built from the scar tissue of real moments.

Every week, I share real leadership stories, frameworks, and strategies from my executive coaching practice—all anonymized but absolutely authentic.

Here's what paid subscribers get:

• Executive Coaching: DM me your work and career challenges for personalized guidance

• Battle-Tested Frameworks: Actionable techniques from $1000+/hr coaching sessions

• Complete Access: All premium content including scripts, templates, and frameworks

P.S. Get my Convince Your Manager template to expense this subscription.

Here we are again, my friends, back for another installment of Dear Lewis.

Today, I'm sharing insights from a coaching conversation with a seasoned Product Manager who's eyeing the rapidly evolving world of AI Product Management. She's got 20 years of PM experience under her belt but feels like she's standing at the edge of a strange new frontier.

It's not uncommon to find experienced PMs feeling both drawn to and intimidated by the AI revolution. And as always, I'm here to offer some guidance including:

• Breaking down the real differences between traditional and AI product management

• A practical framework for making the transition successfully

• Specific actions my client took to bridge her knowledge gap

Keep evolving,

Lewis C. Lin

---

One of my clients is Arora (not her real name). She works as a Principal Product Manager at a major tech company. She's been in product for two decades, and she's really good at what she does. She's shepherded dozens of successful products from concept to market. Her teams respect her, her stakeholders trust her, and she's got the battle scars to prove she knows her stuff.

But something's gnawing at her. The world is shifting beneath her feet. Every job posting, every conference, every industry newsletter screams "AI" these days. And she's wondering if her hard-earned PM expertise is at risk of becoming... antiquated.

"I don't want to be the Blockbuster of product managers," she tells me during our first session. "But I don't even know what I don't know about AI product management."

I tell her that her instincts are good. She's not facing obsolescence – she's facing evolution. The core of product management remains: understanding user needs, creating value, and shipping solutions. But AI adds layers of complexity that traditional PMs have never had to navigate.

"It's like you've been playing chess for 20 years," I explain, "and suddenly the game adds a third dimension and probabilistic pieces. The fundamentals still apply, but the strategy and execution are different."

So we dig into what makes AI product management distinct, and what becomes immediately apparent is that Arora's mental model of product development keeps hitting walls when AI enters the picture.

For instance, when I ask her how she'd approach a recommendation feature, she starts by talking about user stories and acceptance criteria. Classic PM thinking. But she keeps tripping over phrases like "the algorithm should know what users want" – treating AI like it's just another deterministic feature to be specced out.

"Here's where traditional PMs get stuck," I tell her. "You're used to telling engineers what to build. With AI, you're defining what success looks like, but the 'how' becomes a probabilistic experiment."

We start working together to reshape her mental models. Here's where we focus:

Data is Not Just an Input – It's Your Foundation

Traditional PMs view data as something that flows through their product. For AI PMs, data is the soil in which your product grows. Its quality, quantity, and characteristics determine what's even possible.

I ask Arora about her current approach to data. She talks about analytics dashboards and A/B testing frameworks. Good start, but not enough.

"In AI product management," I explain, "you'll need to think about data from first principles: What data do we have? What data do we need? How will we get it? How will we validate it? How will we maintain it?"

To illustrate, I share a story about an AI project that failed spectacularly because no one asked basic questions about data freshness. The team built a beautiful ML model to predict user behavior, only to discover their training data was refreshed monthly while user behavior changed daily. The predictions were perpetually stale.

"As an AI PM, you're not just a product manager – you're a data strategist," I tell her. "You'll need to partner with data engineers to design your data pipeline before you even think about model performance."

This hits Arora like a thunderbolt. Twenty years of PM experience, and she's never once been involved in designing a data pipeline. She's always consumed data; she's never had to think about manufacturing it.

Success is Two Metrics with One Master

Next, we talk about how success is measured. Arora is comfortable with product metrics – engagement, retention, revenue. But AI products have a dual reality: they must succeed both as products and as models.

"This is where AI PMs earn their keep," I explain. "You'll be tracking both product metrics and model metrics, and understanding how they relate."

Arora looks puzzled. "What are model metrics?"

"Accuracy, precision, recall, F1 scores – these measure how well your model performs its predictions or classifications," I explain. "But here's the trap: you can have great model metrics and a terrible product, or vice versa."

I share an example of a content recommendation engine with 95% accuracy (great model performance) that users hated because it kept recommending the same five articles (terrible product experience). The model was technically successful but commercially worthless.

"Your job as an AI PM is to define what 'good enough' looks like for model performance in the context of user value," I tell her. "Sometimes a 70% accurate model that users love beats a 99% accurate model that solves the wrong problem."

This is a paradigm shift for Arora. In traditional products, technical performance and user satisfaction usually align. With AI, they can diverge dramatically.

Experimentation Gets a New Dimension

We move on to talk about experiments, something Arora is very comfortable with. She's run countless A/B tests over her career.

"With AI, your experimental framework needs an upgrade," I explain. "You're not just testing features anymore – you're testing hypotheses about what signals predict outcomes."

I draw a quick comparison:

Traditional PM A/B Test: "Does button color A or B drive more clicks?" AI PM Experiment: "Do past purchase patterns or browsing behavior better predict future purchases?"

"The biggest difference," I continue, "is that traditional experiments have clear winners. AI experiments often reveal trade-offs you'll need to navigate strategically."

I share a story about an AI team that discovered their model could optimize for either quick recommendations or accurate ones – but not both with their current data. This wasn't a technical problem; it was a strategic choice about what kind of product experience they wanted to deliver.

Arora's eyes widen. "So I'd be making decisions about accuracy thresholds and confidence levels?"

"Exactly. And explaining those trade-offs to stakeholders who might not understand why your product can't be both 100% accurate and lightning-fast."

From Shipping Features to Managing Model Operations

Next, we tackle what happens after launch. Traditional PMs are familiar with the cycle: ship, measure, iterate. But AI products have a more complex lifecycle.

"Model operations is a whole discipline you'll need to master," I tell Arora. "Models degrade over time as the world changes around them. It's called model drift, and managing it becomes part of your product strategy."

I explain how an AI PM needs to work with data scientists to establish monitoring systems, retraining triggers, and governance frameworks. It's not enough to ship a great model; you need a strategy for keeping it great.

This concept of "living products" that can deteriorate without code changes is foreign to Arora. In her world, products don't break unless someone breaks them. In AI, entropy is constantly at work.

"So I'd be planning for model updates the way I plan for security patches?" she asks.

"Similar, but more strategic," I reply. "You're not just fixing vulnerabilities; you're deciding when and how to retrain your model based on performance degradation and new data."

The MODELS Framework for Transitioning to AI PM

As our session wraps up, I can see Arora processing everything. It's a lot. So I offer her a framework to structure her transition from traditional to AI product management. I call it the MODELS approach:

M - Map Your Transferable Skills Identify which of your current PM skills transfer directly to AI (user research, roadmapping, stakeholder management) and which need modification (requirement definition, success metrics, testing).

O - Own Your Data Strategy Start thinking about data as a product in itself. What data do you need? How will you collect it? How will you ensure quality? How will you maintain it over time?

D - Develop Technical Literacy You don't need to become a data scientist, but you do need to understand ML concepts well enough to have meaningful conversations. Learn about model types, training/testing paradigms, and evaluation metrics.