Validating Demand Before You Build: A Founder's Checklist
How to test whether anyone wants your AI product before building it: the evidence that counts, the signals that fool you, and a two-week sequence.
Published 2026-08-28 · 4 min read
Most AI products are built before anyone checks whether the problem is worth solving, because building is more comfortable than asking. Validation is not a phase that delays the work — it is the cheapest part of the work, and the part that determines whether the rest was worth doing.
This is what counts as evidence, what fools founders into false confidence, and a sequence that takes about two weeks.
Evidence that counts
| Signal | Why it is real | How to get it |
|---|---|---|
| Someone is already paying to solve it | Money is the only unambiguous expression of demand | Ask what they pay for today, and to whom |
| A manual workaround exists | Effort spent is effort they would rather not spend | Ask to see the spreadsheet, the process, the person they hired |
| They describe the problem unprompted | It is salient enough to surface without a leading question | Ask about their workflow, not about your idea |
| They ask when they can use it | Forward commitment, however small | Notice it — do not prompt for it |
| They introduce you to a colleague | Reputation spent on your behalf | End every conversation by asking who else has this problem |
| They give you a deposit or a letter of intent | The strongest signal available pre-product | Ask. Most founders never do, because they fear the answer |
Signals that feel like validation and are not
- "That sounds really useful." Costs nothing to say and is almost always true. Politeness is the most common false positive in founder research.
- Enthusiasm from other founders. They are evaluating the idea as an idea, and they are your peers rather than your buyers. Their approval feels like progress and is neither.
- Waitlist signups from a launch spike. A signup with no follow-up conversation tells you someone clicked a button. It does not tell you they have the problem.
- Friends and former colleagues. They are optimising for your feelings. Useful for feedback on a working product; useless for demand.
- "Would you use something that…?" Hypothetical questions produce hypothetical answers. Ask what they did last time, not what they would do.
- Traffic without activation. Visitors prove a headline worked. They do not prove anyone wants the product.
The two-week sequence
- 1
Days 1-2: write down what you believe
The problem, who has it, what they do today, what they would pay, why now. Be specific. This is the list of hypotheses you are about to test, and writing it first is what stops you hearing only confirmation.
- 2
Days 3-9: twenty conversations
People who have the problem, found by searching for the complaint rather than the category. Ask about their workflow. Do not pitch. Record their exact words, what they currently do, what it costs them in time or money, and what they have already tried.
- 3
Days 10-11: look for the pattern
If ten or more describe the same problem in similar words and all have some workaround, demand is plausible. If the descriptions vary wildly, you have found several different problems and need to pick one.
- 4
Days 12-14: test commitment
Go back to the five most promising people and ask for something that costs them: a deposit, a letter of intent, an hour on a call with the whole team, or an introduction to whoever holds the budget. This is the step that separates interest from demand.
What each outcome means
| Result | Read |
|---|---|
| Everyone has the problem, nobody has a workaround | The problem is real but not urgent. Worth watching, not worth building yet. |
| Workarounds exist, nobody will commit | You are early, or your solution does not match how they think about it. Talk to ten more before changing the idea. |
| A few will commit, most will not | Normal. Narrow to the few — they are your first segment and probably your only one for a year. |
| Nobody describes the problem you expected | The most valuable outcome. You just saved six months, and you now know what the real problem is. |
| Everybody commits immediately | Suspicious. Check whether you are only talking to people who want to be nice to you. |
How much validation is enough
Enough when three things are true and you can state each in a buyer's words rather than your own:
- You can name the specific person who has the problem, including something about their situation.
- You know what they do today and what it costs them.
- At least a few have given you something that cost them, however small.
Beyond that, more research has diminishing returns and the remaining questions can only be answered by putting something in front of people. The failure mode is not validating too little — it is validating forever as a way of avoiding the part where someone might say no.
Validation is not about proving the idea works. It is about finding out what the idea actually is, before you spend six months building the wrong version of it.
More in Cold start
- The 10 Mistakes That Kill an AI Product LaunchA diagnostic list of the ten failures behind a flat AI product launch, ordered by how often they happen and how cheaply each one is fixed.
- What to Do in the 30 Days After Your AI Product LaunchThe month after a launch is when most founders stop and most of the value is created. A week-by-week plan for converting a spike into something durable.
- How to Cold Start an AI Tool in a Crowded CategoryWhen thousands of products claim the same category, generic positioning is unrecoverable. How to narrow until the competition stops being direct.