AI and Product Management: Becoming More Evidence-Guided

Generative AI is transforming product management, or so we’re told by AI companies and influencers. The vision appears to be all about output: The PM producing roadmaps, backlogs, PRDs, user stories, and prototypes faster than ever before. Some companies are accepting these claims as truth and pressuring their PMs to use AI, but in my opinion this completely misses the point. AI’s biggest promise is not in accelerating the way we work today, but in helping us switch to evidence-guided, discovery-driven work. The companies that realize this earlier will gain major advantages. 

Before we dive into specifics, let’s start with a basic question that still puzzles many executives.

What Do Product Managers Do? 

The job of product management is to discover and help build and launch the right product.

What makes a product right? Mostly these things:

• It delivers high value to the market (customers, users, partners)
• It helps capture value back for the company (revenue, market share, usage …) 

Delivering and capturing value is the job of the entire organization. The role of product managers is to make sure the company has the right product for the job. Good product companies recognize that planning what to build (roadmaps) and then executing on the plans (PRDs, backlogs, user stories, code) just isn’t good enough. To create high-impact products in the face of uncertainty, the product organization has to also conduct research and product discovery. Product managers play an integral role in the entire product development lifecycle, but they’re especially important in the research and discovery phases, which they drive. 

AI As a Research and Discovery Aid

Many companies underdo research and product discovery because they lack capabilities and skills such as data-collection and analysis, user research, market research, and experimentation. Most organizations I meet are aware of this gap and are trying to close it. There are many off-the-shelf solutions to help, and AI is rapidly making these tools more powerful and easier to use. 

But there’s also a human challenge — research and discovery require us to do a lot of deep analytical thinking — what psychologists call Type-2 thinking —, which is slow, deliberate, and hard for our brains. Many PMs I meet struggle doing this work due to lack of experience or time.  I believe AI can help close this gap too. 

I see these benefits when working with product teams. Tasks such as choosing target metrics, making projections, creating outcomes-based goals, and building models, are hard for many PMs. With the right context-setting and prompts, AI can help anyone complete these tasks at a fraction of the time and effort and with very good results (which is why I started combining AI in my workshops). As long as the human critiques the results and validates the logic, the risk of errors is quite low. More broadly, I see AI as a new thinking tool that augments our human intelligence.

Humans are good at: 

• Quick, shallow, intuitive thinking
• Empathizing with other people (users, customers, colleagues) 
• Communicating and collaborating with other people 
• Understanding the broader context — business, technical, operations, politics…
• Knowing how things work in the real world
• Discerning what is good, tasteful, or moral

Artificial Intelligence is good at: 

• Sifting through data (qualitative and quantitative) and creating summaries 
• Identifying themes, patterns, and insights
• Developing estimations, projections, business models 
• Suggesting goals, opportunities, ideas, assumptions, hypotheses
• Calling out reasoning errors made by humans 

Of course humans can do deep thinking too (we’ve done it all along), and with sufficient time and effort do it better than LLMs, but AI tools can offer a useful shortcut and lower cognitive load. 

Don’t Put AI between You and Other People

Some areas you don’t want to delegate to the bots;  communicating with other people is one of those. As humans we’re much better at understanding the  nuances of subtext, body language, and culture, and we often have a past relationship and broader context to guide us. AI has none of that.

I’m mentioning this because some recommendations place AI squarely between us and other humans:

• AI writing PRDs — a PRD (product requirements doc) is a communication method with your team. In my experience it’s much better if the PRD is mostly co-written over time with the team based on what you discover together. 
• Vibe-coding prototypes (to communicate requirements) — a prototype can be a useful communication tool, but it can also be a distraction and a source of friction between PM, UX, and engineering. Again co-development with your team is much better, and it matters less if you use vibe-coding, no-code dev, prototyping tools, or other methods. 
• AI interviewing users and customers — You definitely don’t want to automate this; you’ll likely annoy the participants and miss a ton of nuance. AI notetaking, summaries, and analysis are great. 
• AI training sales and marketing — Again, there are many benefits in you doing this work in-person rather than auto-generating a video with an AI avatar. The learning goes both ways, and the relationship you build and reinforce is invaluable. 

How Will Product Managers Use AI?

Generative AI is still a rapidly evolving technology that has many limitations (hallucinations is a big one), so it’s hard to predict how and when it will change the way we work. If the theory I presented above is true, and we will start using AI to help us work in a more evidence-guided way, the following table may reflect future PM collaboration with AI: 

The high impact (green) areas are mostly about data analysis and pattern extraction. The medium (yellow) areas are ones where we need to combine analytical thinking with human intuition and company/market context. The low-impact areas (orange) are heavy on collaboration with other humans. 

Competing Narratives 

In our productivity-obsessed society it’s only natural to position AI as an output booster. But we PMs are not about producing artifacts, but about creating high-impact products in imperfect companies. AI may be misused to amplify the traditional way of working (output-focus, feature factories), or it may help us drive the change towards evidence-guided development. Sadly the former is the prevailing narrative at the moment and many companies are pressing their PMs to use AI to automate their work. Some are even firing product managers because of these theoretical performance boosts. The goal of this article is to propose an alternative narrative about AI and product management. If you think others should hear it, please consider sharing.