Ideal customer profile: the definition, and a template that survives contact
An ICP is a filter you run against a list, not a demographic form. The definition, how it differs from a persona, and a template with the signal axis most omit.
An ideal customer profile describes the kind of company that buys fast, stays, and tells other people. It is a filter you run against a list. It is not a demographic form and it is not a persona.
The test for whether yours is real: hand it to somebody who has never met your customers, give them a list of two hundred companies, and see if they can sort it. If they cannot, you have written a description rather than a filter.
An ideal customer profile is a company-level filter that predicts who buys quickly and stays. It should contain five things: the shape of the company, the trigger that makes the problem urgent, the observable signal that the trigger has fired, the disqualifiers, and what success looks like after they buy. Every field must be checkable against a real company in under a minute.
ICP, persona and TAM are three different objects
They get used as synonyms constantly and they do three different jobs.
TAM is a count for a pitch deck. Every company that could conceivably buy. It is calculated once a year and it changes nothing about your Tuesday.
The ICP is the filter. It is the one you actually work with, and it should be narrow enough to exclude most of the TAM. If your ICP includes 60 percent of your addressable market, it is not a filter, it is a mission statement.
The persona is the human who has to say yes. Their job title, what they get measured on, what they are afraid of, what language they use for the problem. One deal typically involves one ICP and three or four personas: the person with the problem, the person with the budget, and whoever can veto it.
| TAM | ICP | Persona | |
|---|---|---|---|
| Describes | A market | A company | A person |
| Used for | A slide | Choosing who to contact | Choosing what to say |
| Reviewed | Yearly | Quarterly | When the role changes |
| Fails when | Nobody believes it | It includes everyone | It is invented rather than interviewed |
Use the ICP to pick companies. Use the persona to write the message. Mixing them produces the classic failure where a team targets "VP of Operations at mid-market manufacturers" and then cannot explain why any specific manufacturer should care this week.
The axis most ICP templates leave out
Nearly every ICP template on the internet is a fit form. Industry, headcount, revenue band, geography, tech stack. All of that is useful and all of it is static.
Static attributes tell you who to talk to. They cannot tell you when. And "when" is most of whether outbound works.
A perfect-fit company with nothing happening is a company that will ignore you politely for a year. The same company three days after they posted a job for the role that owns your problem is a different prospect entirely, and the only thing that changed is timing.
So a usable ICP has two axes: the fit attributes, and the observable signal. Most teams build the entire list from the horizontal axis and then wonder why perfect-fit accounts never respond.
The template
Five fields. Fill each one with something checkable.
1. Shape. The static attributes, kept to the three or four that actually predict success. Not the ten you can imagine.
Example: construction and civil engineering firms, 20 to 200 staff, doing work that requires permit approval, in markets where approval takes more than six weeks.
2. Trigger. What makes the problem urgent enough to spend money on. This is not a pain point. It is the event that turns a background annoyance into a line item.
Example: they have taken on more concurrent projects than their current approvals process was built for, usually visible as a hiring push or a run of new project announcements.
3. Signal. How you detect that the trigger has fired, from outside the company, without asking. Where those signals get posted is its own problem. This is the field that makes the profile usable.
Example: an open role for a project coordinator or permit expediter, three or more projects listed as in-progress on their site, or a public complaint about approval delays in a trade forum.
4. Disqualifiers. Who looks like a fit and is not. Write these down, because they are what stops a list from silently filling with near-misses.
Example: firms doing only residential renovation (no permit complexity), firms over 500 staff (they have already built this internally), anyone whose work is single-jurisdiction.
5. What good looks like after they buy. The outcome that made your best customers stay. This field is what lets you tell whether the ICP is still correct in six months.
Example: they cut average approval turnaround by at least two weeks within a quarter, and a second team inside the firm asks for access without being sold to.
Building it from ten customers
You do not need a research budget. You need about ten real paying customers and two hours.
- List them. Paying, not trialling. Ten is the floor.
- Rank them. By how fast they bought, how long they stayed, and whether you would take ten more like them. Fast and sticky at the top.
- Take the top three and the bottom three. Ignore the middle for now. The middle is where every attribute looks equally common.
- Ask what the top three share that the bottom three do not. Not what they all have. What separates them. Industry is rarely the answer. Usually it is a situation: a specific constraint, a specific team structure, a specific thing they were already trying to fix.
- Ask what happened at each of the top three in the 60 days before they bought. That is your trigger field, and it is usually the most surprising thing in the exercise.
- Write the disqualifiers from the bottom three. They are your best source of what a near-miss looks like.
If the top three share nothing, you do not have an ICP problem. You have a product that is solving several different problems for several different markets, and the ICP exercise has just told you that. That is a useful result even though it is not the one you wanted.
Testing whether it is real
Three checks. A profile that fails any of them will not survive being used.
The sorting test. Give it to somebody who has never met your customers, with a list of 200 companies. Can they sort the list into yes and no? If every judgement call comes back to you, the criteria are in your head rather than in the document.
The minute test. For each field: can you check it against a real company in under a minute, from public information? This is the evidence standard applied to the profile itself. "Values operational efficiency" fails. "Has more than three concurrent permit-stage projects" passes.
The exclusion test. Run it against your last twenty lost deals. It should exclude most of them. If it would have let them all through, it is not filtering anything.
Reviewing it
Quarterly, and immediately after any month where more than one good customer churned.
An ICP is a hypothesis about who succeeds with what you built. Churn is the strongest available evidence that the hypothesis has drifted, and it usually drifts in a specific direction: the profile gets quietly widened over a few quarters as the team chases deals slightly outside it, until it describes everyone and filters nobody.
The signal that this is happening is easy to spot. Your win rate holds steady but your time-to-close and your churn both creep up. That is what selling to the edge of the profile looks like on a dashboard, and the fix is to narrow the document rather than to work the pipeline harder.
One thing to do today
Open your CRM. Take the five customers you would most like ten more of. For each, write one sentence describing what was happening at that company in the two months before they bought.
If four of the five sentences rhyme, you have found your trigger, and that single field will do more for your outbound than any amount of firmographic filtering. It is also the field an AI sales agent needs before it can write anything worth sending.