🧠 Theory Thursday: Product Discovery Without the Guesswork

Art by @basilonmypizza: https://lnkd.in/eF8FkWzN - https://basilhefti.ch/

Herbert Simon called it “satisficing”: we settle for what’s good enough, not ideal. AI systems like GPT change the game: they help us find what we actually want.

Recently, I was looking for a wifi router that could handle two land lines, each entering the building at a different point. Not a mainstream use case. GPT helped me describe it clearly and narrow down the search. In the near future, agents will not stop at the search, they will also buy for us. (I am on the lookout for this, but have not yet encountered it, not even with budget limits).

Now here’s the theory: product search is a revealed preference. A data point. And these are underused. AI systems to interact with customers: yes. But listen to what these interactions tell you: not typically done. Even without interviews, surveys, or workshops, this is called product discovery. From real customer intent.

It’s efficient. Turning chat logs into product insights costs less than running a single focus group. And it’s not trivial. Conversations are personal data: GDPR, HIPAA, and similar frameworks demand we strip identifiers, secure logs, and stay transparent. You also need to separate niche one-offs from genuine patterns with market potential. And of course you need to prioritize: not everything can actually be built.

The approach works best in digital products. Spotify, for example, used audiobook and podcast data to recommend audiobooks, driven by a model called 2T-HGNN (2T for the “two towers”, audiobooks and podcasts; and HGNN for heterogeneous graph neural networks). Similarly in banking, when users keep asking “Where can I see my limits?”: that’s an UI signal. Or in logistics, repeated requests for multi-depot fallback routing can spark new offerings.

We often look at AI from automation viewpoints. But AI also makes demand visible. It shows what people would do, if we gave them the right product. Product discovery is hiding in the logs.

What’s something you were looking for that was surprisingly hard to find?

• Art by @basilonmypizza: https://lnkd.in/eF8FkWzN https://basilhefti.ch/