Module 2
Validate in seven days
Twenty real conversations, a pass or fail threshold set in advance, and a decision.
Validation is not a survey and it is not a model telling you the market looks attractive. It is twenty conversations with people who have the problem, followed by an ask that costs them something.
This module takes a week. At the end you will know whether to build, and you will know it on evidence you gathered rather than evidence you assembled to feel better.
Sign in below to save your answers
What AI research is good for, and what it is not
A model will happily produce a market size, five personas and a list of objections. None of that is validation. It is a summary of what has already been written, which means it contains no information about whether a specific person will pay you next month.
Synthetic personas are worse than useless when they are treated as evidence, because they never say the thing that changes your plan. Real buyers do that in the first ten minutes.
Use models for the parts that are genuinely research: finding who holds the budget in a type of company, mapping the vocabulary a market uses, drafting outreach, preparing the questions you have not thought of, and summarising your own notes after the calls. Never use them as a substitute for the calls.
Your turn
What good looks like
- Assumption 1: disputes cost operators more than the admin time, because of goodwill refunds.
- Assumption 2: the ops manager can approve a $500 a month tool without the CFO.
- Assumption 3: they will accept photos taken by their own drivers, not by an inspector.
Build the list of twenty real people
Twenty is the number because five gives you anecdotes and fifty is procrastination. They must be people who have the problem now, not people who find it interesting.
Name them individually. A list that says 'fleet managers in the UAE' is not a list. A list that says 'Rashid at X, Priya at Y' is. Where you have no route in, write the route you will try: a mutual contact, a cold email with a specific claim, a comment on something they published, an event they attend.
Aim for at least twelve who could actually buy, not just people who feel the pain but do not hold budget.
Your turn
| Name | Company / role | Has the problem because | Route in | Status |
|---|---|---|---|---|
What good looks like
- Rashid, ops manager at a 300-car subscription fleet. Has the problem because his team re-types condition reports. Route in: mutual contact from my old team. Status: call booked Thursday.
- Priya, COO at a 90-car rental operator. Has the problem because she signed off 14k dirhams of goodwill refunds last quarter. Route in: cold email quoting that number from her public interview. Status: sent Monday.
The conversation script
The purpose of the call is to collect facts about the past, not opinions about the future. People are unreliable about what they would do and quite reliable about what they did last month.
So ask three things: what they do today, step by step; what it costs them, in time, money or errors; and what they have already tried to fix it. That last question is the one that matters most. If they have never tried to fix it, it is not a real problem for them, whatever they say.
Do not describe your product until the end, and when you do, describe it in one sentence and then stop talking. Write the notes within ten minutes of hanging up, in their words rather than yours.
Your turn
| Person | What they do today | What it costs | What they already tried | Their words worth keeping |
|---|---|---|---|---|
What good looks like
- Question: walk me through the last handover dispute you personally dealt with.
- Cost: 'Two of us spent most of a morning on it and we still refunded him.'
- Already tried: 'We made drivers take photos. Half of them are blurry and nobody looks at them until there is an argument.'
- That last sentence is the product brief.
Ask for commitment, not encouragement
Encouragement is free, so it tells you nothing. Every good conversation should end with an ask that costs the other person something: a deposit, a prepaid pilot, a signed letter of intent with a date, or a block of their team's time in the calendar.
The ask can be small. A $250 deposit against a pilot is more informative than a hundred people saying it sounds great. If they will not commit anything, ask what would have to be true for them to. Their answer is the roadmap.
Record the ask you made and the answer you got, per person. Not the vibe of the call.
Your turn
| Person | Ask made | Answer | Amount or date committed | Reason if no |
|---|---|---|---|---|
What good looks like
- Rashid: asked for a $500 prepaid four-week pilot on 40 cars. Answer: yes, starting the 12th.
- Priya: asked for the same. Answer: no, needs the CFO. Committed instead to a 30 minute call with the CFO next week.
- Third operator: no commitment, said 'come back when you have a case study'. That is a no.
Set the threshold before you see the numbers
Write your pass and fail lines now, before the results are in. Once you have the data you will find a reading of it that lets you carry on, because everybody does.
A workable threshold names a count and a form of commitment: for example, three paid pilots or five signed letters with dates, out of twenty conversations. Anything less is a no, and a no means you go back to candidate two rather than lowering the bar.
Also write the ambiguous case in advance: what you will do if you get two paid pilots instead of three.
Your turn
What good looks like
- Pass: 3 prepaid pilots of $500 or more, from 20 conversations, within 10 days.
- Fail: fewer than 2. Then I stop and reopen the maintenance scheduling candidate.
- In between: with exactly 2, I run them, and treat week five as the real decision point rather than extending validation.
The judgment call
Mixed signal is the normal outcome, and it is a judgment call. Two paid pilots and eight warm nos is neither proof nor refutation, and running five more conversations rarely changes the picture. What you are really deciding is how much of your own year you will spend finding out.
How to think about it
- Weight money over words, and dates over intentions.
- Separate 'they do not want it' from 'they want it and I reached the wrong person'. Those need different responses.
- Look at whether the yeses share something the nos do not. That is your segment, and it may be narrower than you planned.
- Set the next decision point with a date attached before you continue.
Questions worth sitting with
- Which of the nos would become a yes with a different buyer in the same company?
- What do the yeses have in common that the nos do not?
- How much time and money are you prepared to spend before the next go or stop decision?
- If a friend showed you this exact result, what would you tell them?
Common questions
Can I validate a startup idea with a survey?
Not usefully. Surveys collect opinions about hypothetical futures. Validation needs facts about the past and an ask that costs something: a deposit, a prepaid pilot or booked time.
How many customer interviews are enough?
Twenty conversations, of which at least twelve are with people who could sign a purchase. Beyond that you usually stop learning new things and start delaying the decision.
Is a letter of intent good enough validation?
It is weaker than money but stronger than encouragement, provided it carries a date and a named owner. Treat unsigned or undated interest as a no.