Ask a founder who their ideal customer is and you will usually get a filter.
Series B to D. Two hundred to two thousand employees. North America and Western Europe. Fintech, healthtech, or logistics. VP of Engineering or above.
That is a prospecting query. It tells a sales development rep which rows to pull out of a database. It does not tell a copywriter what to write, a designer what to show first, or a product lead what to cut. Which means it is not doing the job an ideal customer profile exists to do.
The question a real ICP answers
Not "who could buy this" but "who has a problem so shaped that our product is the obvious answer, today."
Two companies can match every firmographic box and be completely different customers. A five hundred person fintech that just failed an audit and a five hundred person fintech that passed one comfortably last year are not in the same market. The first one has a deadline, an internal champion with something to prove, and a budget that appeared last week. The second one has a nice-to-have and a calendar that is already full.
Same company on paper. Different buyer entirely.
The thing that separates them is situation, and situation is what most ICP documents leave out.
What to write down instead
Start with the trigger. What happened, in the four weeks before they contacted you, that made this a priority? Someone left. A customer complained in a way that reached the board. A tool got deprecated. A number moved in the wrong direction and somebody had to explain it. There is always an event. Find it and write it down in the buyer's language, not yours.
Then the alternative. What are they doing right now instead of using you? This is almost never a competitor. It is a spreadsheet, a contractor, three engineers on a Slack channel, or a policy of ignoring the problem. Your real competition is the status quo with a headcount attached, and your pitch has to beat that, not the other logo in the category.
Then the cost of doing nothing, in their terms and preferably in their numbers. If nobody inside the company gets hurt by waiting, you do not have a deal, you have a conversation that will stay pleasant for four months and end nowhere.
Then who feels it first, who pays, and who can block it. Those are usually three different people, and the pitch that wins over the first often alarms the third.
Notice that none of this is headcount. Headcount tells you where to send the email. The rest tells you what the email says.
The narrowing test
Read your ICP and ask which deal from last quarter it disqualifies.
If the answer is none, you have documented your customer list rather than made a decision. An ICP that includes everyone you have ever sold to is a description of the past. An ICP that rules things out is a strategy, because it commits you to being worse at something in exchange for being obvious at something else.
Founders resist this at Series A and B for a reason that makes sense. Revenue is not so abundant that you want to turn any of it away, and narrowing feels like volunteering for a smaller company.
What actually happens is the opposite, and the mechanism is not mysterious. A broad claim has to be re-explained to every buyer, because it was not written for any of them. A narrow claim is repeated by buyers to each other, because it named their exact situation and that is a thing people quote. Narrow positions travel. Broad positions have to be carried.
Brian Chesky has a version of this for product: design something perfect for ten people before you try to scale it to a million. The ten are not a market size, they are a standard. You cannot build the perfect experience for a demographic. You can build it for a person in a situation.
Where the ICP has to show up
A document nobody reads is not an ICP, it is a file.
The test of whether yours is real is whether it is visible in artefacts. The homepage headline should be unusable by a competitor selling to a different situation. The first question on the sales call should be the trigger question. The demo should open on the screen that matters to this buyer rather than the one that demos best. The roadmap should have at least one thing on it that is there because this buyer needs it, and at least one thing off it for the same reason.
If your ICP never caused you to delete a feature request, remove a section from the site, or say no to a qualified-looking lead, it has not touched the business yet.
The part that takes discipline
The ICP will be wrong the first time. Write it anyway, in specific language, and let the market correct it.
The failure mode is not choosing wrong. It is choosing so vaguely that nothing can be wrong, which also means nothing can be learned. A profile that says "mid-market companies that care about security" survives any evidence, which is another way of saying it produces none.
Write the version that could be embarrassing. Then go find out.