Product discovery
The vision named the buyer. Fourteen interviews named someone else.
One target group. One persona, 95% complete, with zero interviews behind it. Watch a discovery canvas find the user it never mentioned.
The completeness meter counts what somebody typed. Everyone reads it as what the team knows.
Steps 01–02
A complete canvas with nothing behind it
FlowDesk sells shift scheduling to mid-size logistics operators. 400 customers, mostly Benelux.
Their product vision had one target group in it: operations managers. So the discovery workspace had one persona, and it looked healthy — five sections filled, 95% complete, last edited fourteen months ago.
It had zero interviews attached. Everything in it had been written by people who had met that persona in a sales call, which is a real experience of a person, though a different person from the one who opens the product on a Monday.
A completeness meter can only measure how much you have written.
Step 03
Four hundred customers, and no description of the person who uses it
Personas built from target groups inherit the vision's blind spots.
Fourteen churn interviews had been sitting in a folder since February. Read together, eleven of them named somebody the canvas had never mentioned: the dispatcher.
The dispatcher signs nothing and attends no demos. They open the schedule every morning before the depot does, and in FlowDesk's data model they didn't exist, which is why they couldn't log in without somebody buying them a seat.
A vision that names only buyers produces a product that serves only buyers. That's a structural failure, and it becomes visible the moment personas have to cite their evidence.
Step 04
Jobs, pains and gains, drafted from the transcripts
AI does the first pass across fourteen interviews. A human keeps or kills every line.
Speed is the smaller half of it. Every job and every pain keeps a link to the sentence a customer said, so the canvas can be argued with.
One pain came back in eleven of the fourteen transcripts, in eleven different phrasings: the schedule lives in somebody else's account. Nobody had written it down as a product problem, because on its own each one sounded like a complaint about a colleague.
WHY IT MATTERS DOWNSTREAM
A gain phrased as "stops being the last to find out" can be tested. A gain phrased as "improved visibility" can only be agreed with. The wording your personas carry decides whether an experiment can be written at all.
Step 05
Every assumption feels important. Only some are uncertain.
Importance decides whether it matters. Uncertainty decides whether to test it.
Nine assumptions came out of the dispatcher persona. Scored on both axes, two landed in the corner that deserves an experiment, and the company had been arguing about neither of them.
"Setup effort is the main blocker" scored high on importance and low on stated uncertainty, because everyone in the room was sure. That confidence was the problem: the belief had never been checked, so its low uncertainty score was itself an assumption.
The output of discovery is one question worth answering next.
Step 06
One new persona changed the business model
Customer segments sync from personas, so the canvas can't quietly disagree with the research.
FlowDesk's lean canvas had one customer segment. After the dispatcher it had two, one that pays and one that uses, and those are two different businesses to be in.
The value proposition moved with it: from publish a schedule without spreadsheets to the whole depot sees the same schedule, whoever pays for it.
One revenue line became uncomfortable. Seat expansion had been quietly assumed, and if dispatchers should get in for free, that assumption needs a test of its own.
From here discovery hands off. A1 became a two-week experiment, and the dispatcher persona became the owner of an opportunity.
What discovery is for
To find out who you forgot
Vision, personas and business model sit in one place so that adding a single person to your research can't leave the rest of your plan quietly unchanged.
Fair questions
What people ask about product discovery
- We already have a product vision. Do we start over?
No. In the demo the vision statement never changed, because it was right. What changed was the target groups under it, and everything that syncs from them. Bring the vision you have; the gaps show themselves.
- How much of the persona does AI write?
The first draft, from transcripts you supply, with every line linked back to the sentence it came from. You keep, edit or delete each one. A persona with no evidence attached stays marked as unevidenced however complete it looks.
- Do we need the business model canvas if we're not a startup?
Only if your research might contradict how you make money. In the demo, giving dispatchers free access collided with an unexamined seat-expansion assumption, which is worth catching before it becomes a pricing argument.
- What makes an assumption high-risk?
Both scores have to be high. An assumption everyone is genuinely certain of can wait, and an uncertain one that changes nothing can wait longer. Nine assumptions usually produce two or three worth an experiment.
- How long does this take?
It depends on how fast you can connect the interviews you already have. For a ten-person team that has typically taken about five days. In the demo, the reframe took two weeks and used research that had been sitting in a folder since February.
Who opens your product every morning and isn't in your vision?
Bring the interviews you've already done. The first thing discovery usually finds is a person.
FlowDesk is the Shorter Loop demo workspace. The canvases, the evidence links and the way one persona propagates is live in it — open it and check any number here.