I spent an afternoon at Napkin Thinking, a casual talk where Rolf Biernath, founder of Biernath Consulting and a thirty-year veteran of 3M, walked through his Build What Wins™ framework. Rolf is also co-author of Chapter 24 in The Generative Organization: AI Playbook for Exponential Leaders, an Amazon bestseller.

He opened with three numbers from McKinsey’s 2025 State of AI report.

1. 73% of advanced manufacturing firms are not using AI agents in product development.
2. Only 5.5% of organizations capture more than 5% of operating earnings from AI.
3. High performers are three times more likely to have redesigned their workflows end-to-end.

Build What Wins screens every product investment across four dimensions: viability, defensibility, feasibility, and timing. Does the market want it, and will it pay enough? Can we keep winning once others see us do it? Can we actually build, test, and ship it? Is the window open now or in two years? The value of this framework is the urgency to ask these questions in week one, rather than months in.

“Focus on the most likely failure mechanisms—the ones that are easiest to detect early. I’ve built a workbook that helps with that, including questions like: who already holds patents in this space? Are we up against players like Google or Amazon, or should we be thinking about working with them? It’s not that missing a checkbox means your product will fail—it just means it’s a factor you need to account for as you develop.”

The dimension Rolf spent the most time on was feasibility, because feasibility has cheapened. Automotive aerodynamics simulation that used to take 50 hours on 500 processor cores now runs in under an hour on a single graphics processor, with better than 95% accuracy. Crash-safety validation that used to demand months and multiple physical prototypes can be predicted from historical data before the part exists. Precision aerospace-machined parts that took four to sixteen weeks to source and ship in one to three weeks through services like Hadrian. The implication is uncomfortable. If your moat was the ability to absorb long, expensive iteration cycles, your moat is gone.

He mentioned three shifts that rewrite the old prototyping playbook.

1. AI now is a filter you use to kill bad designs fast.
2. The prize is the training data you build by instrumenting your own work before you automate it.
3. Copilots are giving way to agentic loops that reshape the work itself.
   

The advice that was useful in 2023 is, in his words, actively wrong now.

He told a 3M story that made the cost concrete. A team spent two years developing a product. They took it to customers, who said the product was great and they would take it for free if it came bundled with the thing they actually needed. Two years to learn that what they developed was worth “free” for their customer. Rolf argues that these questions today can be asked in week one.

In the end, we had a conversation about leadership in product development environments. Art Fry, who was in the audience, said, “Leaders build businesses, managers operate them”. As an organization ages and the original builders move on, you end up with people placed in roles to run a recipe. They never had to create anything. Innovation costs money and carries the risk of failure, and operators do not take that risk easily. That is the mechanism behind innovation stalling in big companies. I was happy to hear that message resonate with the audience, as one of my passions is fast-tracking lifesaving devices past slow giants and into patients’ hands. As I venture into my product/process innovation journey, it is reassuring to hear some of the foundational minds behind 3M, as we know it today, confirm that speed is largely an operational problem and should be addressed for the sake of innovation.

I left thinking about how often teams polish the wrong prototype because the week one questions felt too uncomfortable to ask. Build What Wins gives you a way to ask them on purpose, in week one. The framework is trademarked, so Rolf is clearly planting a flag.