understand · brief intake
Founder interview
The nine questions and what they unlock.
In Brief intake
Nine questions the website can't answer, asked once, stored in the founder's own words. They turn guesses into the few facts that only the founder has, and they mark clearly which answers are about real payers and which are beliefs to test.
Verbatim, always.
Store the answer exactly as given. A paraphrase loses the phrases detectors and opening lines need.
Beliefs are labelled beliefs.
Most answers are tier D. Only answers about real payers, or the founder's own channel sizes, carry tier A.
Ask the gap, not the list.
Skip a question the brief already answers with better evidence. Never read the founder a form.
what it is
A structured customer-discovery interview, pointed at the founder
Industry names: founder interview, discovery call, intake. It borrows from customer-discovery practice (ask about past behaviour, not hypotheticals) but the subject is the founder's knowledge of their buyers, not a buyer.
Its job is narrow: fill the brief fields that need someone who has watched the business. Who paid first and what they were doing just before. Who to refuse. Where buyers gather. What they call the problem. What channels already exist. Everything else should come from evidence the founder doesn't have to remember.
Past behaviour — the interview rule · "Who paid you first?" beats "Who would pay?". The first has an answer that can be checked.
Verbatim answer — capture rule · stored as typed, with the question it answers and the date.
Belief vs evidence — evidence level · most answers are tier D: recorded, used as a starting hypothesis, never as a search term until confirmed.
Gap-driven — ordering · the next question is the most valuable empty field, not question 1, 2, 3 in order.
how it works
Asked once, in the founder's words, in gap order
The interview runs after the site read, so it only asks what the site couldn't fill. In a chat surface it is one question per turn, never a list.
- Read the gaps — empty or weak fields after the site read
- Pick the next question — most valuable gap first; skip what's already known
- Ask, one per turn — plain words, past behaviour, no forms
- Store verbatim — exact answer, question id, date, who answered
- Tag the evidence — A if about real payers or own channels; else D
- Map to fields — each answer fills named brief fields
- Ask one follow-up — only when it unlocks a detector or a not-if line
- Mark confirmable — what later evidence should confirm or kill it
- Read the gaps — empty or weak fields after the site read
- Pick the next question — most valuable gap first; skip what's already known
- Ask, one per turn — plain words, past behaviour, no forms
- Store verbatim — exact answer, question id, date, who answered
- Tag the evidence — A if about real payers or own channels; else D
- Map to fields — each answer fills named brief fields
- Ask one follow-up — only when it unlocks a detector or a not-if line
- Mark confirmable — what later evidence should confirm or kill it
the object
An interview answer
Answers are append-only. A later answer to the same question adds a row; the old one stays, so you can see how the founder's view moved.
interview_answer
| field | type | what it holds |
|---|---|---|
question_id | enum | q1 to q9, or a follow-up id |
answer | text | verbatim |
answered_at | timestamp | when it was given |
fills | list | the brief fields it feeds |
evidence | enum | A when about real payers or own channel sizes; otherwise D |
status | enum | founder-confirmed by definition; later evidence may mark it conflicting |
confirm_by | text | what evidence should confirm it: payers (A), signups (B) or the market (C) |
surface | enum | where it was asked: intake flow, chat, settings |
sample
{
"question_id": "q7",
"answer": "they say 'nobody knows it exists' and 'I'm a builder, not a marketer'",
"answered_at": "2026-10-12T21:40:00Z",
"fills": [
"voice_of_customer"
],
"evidence": "D",
"status": "founder-confirmed",
"confirm_by": "C: the same phrases in public posts",
"surface": "chat"
}question catalog
The nine questions, and the follow-ups worth asking
Grouped by what the answer unlocks: the founder's own truth about payers, the fit filter, what to listen for, or what the pipeline can act on.
founder-truth
Q1. Who paid you first, and what were they doing right before?
- fills
- Who pays today · why they bought
- evidence
- A when it's about real payers
- feeds
- the starting trigger
founder-truth
Q2. What did they use before you?
- fills
- What it replaces
- evidence
- A from payers, D otherwise
- feeds
- doing it by hand, unhappy with a tool
filter
Q3. Who would you refuse as a customer, and why?
- fills
- Who you'd turn away
- evidence
- D
- feeds
- the not-if list
filter
Q4. What has to be true for someone to get value on day one?
- fills
- What they must already have · first result
- evidence
- D, confirmed by B
- feeds
- hard filters, activation
filter
Q5. Who uses it, and who pays for it? Same person?
- fills
- Who uses it vs who pays
- evidence
- D
- feeds
- the role in the filter
listen
Q6. Where do the people you want spend time online?
- fills
- Watering holes
- evidence
- D, confirmed by C
- feeds
- detector sources
listen
Q7. What do they call the problem, in their own words?
- fills
- Voice of customer
- evidence
- D, confirmed by C
- feeds
- detector phrases, opening lines
act
Q8. What channels do you already have, and how big are they?
- fills
- Channels you already have
- evidence
- A
- feeds
- the reachability gate
act
Q9. What do buyers expect that the product can't do yet?
- fills
- What you can deliver now
- evidence
- D
- feeds
- the whole-product gate
Follow-ups
Ask at most one, and only when it turns an answer into something a detector or a filter can use.
after 1 — What did they say, in their words, when they decided?
unlocks: the opening line a sequence can echo
after 1 — Did anyone like them leave? Why?
unlocks: the first not-if line, from evidence
after 3 — Have you ever said yes to someone on that list? What happened?
unlocks: whether the exclusion is real or a preference
after 6 — Name one specific thread, subreddit or account.
unlocks: a source a detector can actually poll
after 7 — Paste a sentence someone wrote that made you think 'that's my customer'.
unlocks: a seed phrase with a real example attached
after 8 — Which of those has ever produced a signup?
unlocks: which channels are real, not just large
failure modes
How it goes wrong, and how to catch it
Leading questions — "Founders struggle with distribution, right?" gets a yes and teaches nothing. · Catch it: ask about past behaviour with names and dates; an eval flags yes/no questions.
Paraphrased answers — The founder's phrase "crickets after launch" is stored as "low traction". · Catch it: store verbatim; the summary is a separate field.
Beliefs used as facts — "My buyers are on LinkedIn" becomes a detector source before anyone checks. · Catch it: tier D never writes a search term until confirmed by C evidence or the founder explicitly confirms.
Form fatigue — Nine questions in one message get one-line answers or none. · Catch it: one question per turn, gap order, stop when the gate fields are filled.
Stale answers — An answer from the first week stays true forever. · Catch it: answers are dated; re-ask when evidence conflicts.
Answering for the founder — The model fills an answer from the site to save a turn. · Catch it: an interview answer must have the founder as its source; otherwise it is a site fact.
data to capture
Record it so a model can learn from it later
Every decision on this page is training data if it is recorded in the guide's shape from day one.
1 · log the decision · 2 · version whatever decided · 3 · freeze the inputs · 4 · keep the negatives · 5 · log the odds · 6 · every human touch is a label · 7 · one chain of ids · 8 · keep the raw
record
- every answer verbatim, with question id, date and surface
- questions asked but skipped or declined
- which fields each answer filled, and the evidence level assigned
- later evidence that confirmed or contradicted each answer
label
per answer: confirmed or contradicted by later evidence (payers, signups, market)
trains
a question-ordering policy (which gap to ask next) and a founder-belief calibration score: how often this founder's tier D answers hold up
test
- knob: question order and wording
- mode: randomized A/B by founder on order; measure fields filled per turn and drop-off
- win rule: more gate fields filled in fewer turns, with no drop in completion