Product discovery as a pipeline: the two judgment calls baked in

Last week I went through a run of product-management repos. Most of them cover a lot of ground at once, everything from strategy to sales.

But one did the opposite. It is called product-discovery-skills, by Else van der Berg, a freelance product lead. Instead of covering everything, she focused on a single thing, product discovery, and went as deep as she could on it.

What made me stop was not the coverage. It was the judgment she wrote into the skills themselves, two things people running discovery get wrong all the time. This set does not just walk the steps. It decides on the user's behalf at exactly the points where people usually decide badly.

I will tell it in three parts. First, what it is. Then the two judgment calls baked in. And finally, the lessons you can actually use.

Part 1What product-discovery-skills is

It is six Claude skills chained into a pipeline, each one picking up where the last left off, following Teresa Torres's continuous discovery.

• Screen first. Take the interview transcripts and check whether the people you talked to match your ideal customer (ICP). If they don't, their input doesn't get mixed in.
• Map the phases. Lay out the steps a customer moves through. This is the first level of the opportunity solution tree.
• Extract opportunities. Pull the problems and needs out of each interview, one at a time, guided by a rubric hundreds of lines long that spells out what counts as a usable opportunity.
• Cluster across interviews. Group opportunities that are really the same thing raised by different people, so one problem doesn't get counted as four.
• Size them. Score which opportunities should be worked on first.

What sets it apart from a generic discovery prompt pack is that it leans on named frameworks, both Torres's opportunity solution tree and jobs-to-be-done thinking (looking at the job the customer is trying to get done), not just "try summarizing the interview." There are clear tests for what passes and what doesn't.

Part 2The two judgment calls baked in

Read only the steps and it is a solid discovery pipeline. But what actually made me stop was the two places where she wrote judgment into it, exactly where people running discovery tend to decide wrong.

One: a misfit is signal, not noise. When you extract opportunities and some of them don't fit the phase map you drew, the normal reaction is to throw them out as noise. This set does the opposite. It treats a misfit as a hint that your own map of the problem is wrong, and invites you back to revise it. It is a loop, not a one-way line.

Two: importance is read apart from how many people raise it. When it comes to ranking, most tools let "how many people mentioned it" decide. This set keeps the two apart and refuses to blend them. She writes the rule in plainly.

"Don't chase an opportunity with median importance below 4, however many people raise it. Better to get one person to truly love the product than ten to kind of like it."

This is the part that is pure taste, not a formula. It is a belief that a problem a small group feels sharply is worth more than a problem the crowd feels lukewarm about. And she chose to bake that belief in as a prompt the system raises, not a call the system makes for you. A person still decides.

Part 3Lessons you can actually use

Even if you never touch this set directly, three things carry over to any discovery or analysis work.

• Don't be quick to throw out the misfit. Data that clashes with the frame you drew is often not an error but a signal that the frame needs fixing. Keep it, and ask what it is telling you.
• Split the two axes people love to mash together. How many people, and how sharp the problem is, are different things. Blend them into one score and small problems many people grumble about drown out big problems a few people really feel. Read them apart, then decide.
• Discovery is a continuous discipline, not a one-off. The loop that sends you back to revise the map when something odd shows up is the heart of continuous discovery. You do it again and again, not once and done.

How to start

Try it on two or three customer interviews you already have.

1. Pull out the problems the customer said, one at a time. Don't lump them together yet.
2. Score two axes separately: how many people raise this, and how sharp it is for the people who feel it.
3. When a problem doesn't fit the picture you had in mind, don't drop it. Write it down and ask where you misread the customer.
4. When you pick what to do first, hold the rule: a problem few people feel but feel sharply comes before a problem many people feel lukewarm about.

product-discovery-skills is a good example of the idea that what makes a skill valuable is not stuffing a framework in, but baking in the judgment that skilled people use to decide, right at the points where most people slip. This is one in a series where I go through repos that pack a discipline into skills, more to come, and a wrap-up piece to close it out.