Problem interview
An Interview about the last time an ICP member hit the Problem, what they did and what it cost, closed by an ask for a commitment.
A problem Interview asks an ICP member about the last time the Problem happened: what they did and what it cost. It ends with an ask for a commitment, and its answers become graded Evidence.
Past behavior, not opinions
Everyone is generous about the future and accurate about the past. Ask whether someone would pay, and most say yes. Ask what they did the last time the Problem struck, and they tell the truth.
A problem interview asks only about things that happened. The moment, the workaround, the hours or dollars it took, and who else was involved are facts the person remembers. Praise, predictions, and feature requests are opinions, and they weigh zero.
The interview never pitches. Describing the solution turns a witness into a critic, and critics give opinions.
The script
The opener names a moment, not a topic. The probes follow the person's story, and the close asks for something that costs them.
opener: Tell me about the last time you chased a client for missing receipts.
probes:
- What happened right before that?The script is itself a variant. Openers are tested like any other copy, against the share of interviews that surface a recent, costly instance.
Grading the answers
Each answer is graded by what it cost the person to give it.
| Answer | Grade | Why |
|---|---|---|
| "I lost two days reconciling receipts last month." | behavior | a past act, with a cost |
| "I pay a temp $600 every quarter to do this." | behavior | money already spent on a workaround |
| A $40 refundable deposit | commitment | costs the giver money |
| An introduction to the partner who pays | commitment | spends the giver's reputation |
| "I would pay $40 a month for that." | opinion | free to say |
| "This is a great idea." | opinion | praise weighs zero |
Each graded answer links to the Problem as Evidence, with the quote verbatim and the person it came from. Evidence counts distinct ICP members, so ten answers from one person count once.
The ask
Every interview ends with an ask that costs the person something. It can be a refundable deposit on the reserve page, a pre-order, an introduction to whoever pays, or a second call with their own data.
A yes is commitment-grade Evidence. A no is Evidence too, recorded with its reason. A run of noes for one reason is the most useful result a week of interviews produces.
Who interviews, and how many
In the Idea state, the Founder owns the Problem interviews count, the state's one Metric that matters. The Researcher finds Prospects inside the ICP, books the interviews, and runs them; the Founder reads every transcript.
An Agent that runs interviews introduces itself as an agent. Under a human's Mandate, it can instead invite people as that human, within the Mandate's limits on volume and topics.
Each person has one contact budget across the Studio. Sibling Startups testing the same Problem share one deduplicated list, so no one hears from five variants of one Idea.
When to stop
Interviewing earns its cost only while the next interview could change the next Decision. When five in a row repeat what the record already holds, stop and run the cheapest test that asks for money. Test demand picks it.
A receipt-chasing tool for independent bookkeepers holds eight interviews in its first week. Six describe a costly instance in the last month, and three take the $40 reservation. The Problem holds, so the next week's test is a pre-sale, not more talk.
After the first sale, the Signal interview takes over. It applies the same discipline to paying Customers and asks why they bought.
Next
- Test demand: what to run once interviews stop changing the plan.
- Wants: how interview answers join the record of unmet demand.
- Office hours: where the week's interviews become a Decision.