How we use AI in product discovery

Published July 28, 2025 | Written by Hannes D’Hulster

Share article on:

In the 1920 we would say: we connect to the electricity, in the 2000 we would say let’s get on the internet in 2025 we say: let’s ask chatGPT, Gemeni, <insert your favorite LLM>

Like the internet and electricity before, the use of this technology will become normal and in that process will disrupt a few sectors. We see VCs and so Start-ups focussing on AI in the last years, these innovators are figuring out if we can automate services like sales, data creation and recommendation (think e-commerce). The energy transition from reliable coal and gas to volatile sun and wind is also a great candidate for being successful with AI.

What with about the product design craft?

Time to experiment

Not only the start-ups are innovating, with close attention of the corporates, but everyone is figuring out if and how we can use these new technologies to our benefit.

It is clear that the technology is not ready to take over the PM or design job at this moment:

Everything looks the same: because trained on what is existing, innovation is by definition very hard for these tools. They’re just statistical decisions.
Getting the last 20%: text are pretty hollow, images still contain hallucinations, videos don’t know what happens if someone passes behind a tree. For now, what we get is never production ready
They do as they’ve been told: their possible autonomy grows, they get better at understanding what we mean but unravelling the problem and engineering the right prompt for it, is still the role of the human.

This means, for product innovators (business, design, tech,…) that it is time to understand how this technology may impact our way of working in the future. Not only waiting for that future but also helping to shape the future. By understanding, by experimenting, by adjusting. Also by calling bullshit an by abandoning certain tools.

Short anecdote form our hiring adventures

Experiments on Discovery

One of the subjects we help start-ups with is the process of Discovery. It is all about figuring out the user’s problem and the possible software solution to this. This includes wireframing and UI design but it also part of Product management and requires a good deal of research. Benjamin wrote an article on tools for UX/UI design, go check it out, we’ll focus a bit more on research and pattern finding.

Understanding context: for desk research on how systems work, what legislation is involved and what competitors are in the market ChatGPT is our friend. By getting in the conversations, asking follow up questions and checking assumptions you can get a pretty good first overview of the problem.
Transcribing interviews: yes they jump in zoom calls without asking: the transcription bots. They do often a good job in creating an overview, sometimes even pointing out highlights in this text.
Thematic analysis: Finding patterns, finding wholes in your qualitative research and drawing conclusions seem not (yet) a good idea. We’ve tried, but the quality is low, Chatty misses a lot of nuance and you get out of touch with your subject.
Brainstorming: Let’s say “maybe”. Yes it can come up with solutions that have worked before but those are not always fully applicable to the situation. Also: for many of us brainstorming is the most joyful experience within the discovery process. We don’t like to outsource that.

While hiring PMs we present them with a list of around a 100 insights from several target audiences about the same product. Their assignment is to find the patterns and come up with a plan for solving for those. We reduced the preparation time from 1 week to 1 hour because of Claude and their friends. We see that candidates ignore AI suggestions after 15 minutes. Really understanding of underlying problems from statements is really not yet a machines job.

We’re very curious on how AI will keep supporting start-ups to disrupt energy and other sector, we’re following closely, experiment and adopt where it make sense.