Start with customers who bought, used the product and stayed, then compare suitable losses. Record the trigger, pain owner, previous solution, decision process and prerequisites. Treat the resulting ICP as a working rule that can be revised, not a demographic slogan.
Choose the right comparison group
Do not build an ICP only from every company that signed a contract. Include customers who activated and stayed, plus accounts that looked promising but stalled or churned. Compare the buying moment and operating conditions, not just industry and headcount. The won/lost ICP guide offers a fuller deal-review method.
Find discriminating facts
Look for the previous solution, the problem owner, the event that made the problem urgent, the buying committee and the resources needed to succeed. A useful ICP says who is unlikely to succeed as well as who might buy. A list of broad attributes that applies to half the market will not help a marketer choose targeting or a seller decline a bad-fit request.
Write a testable rule
For example: “Teams with an active multi-location rollout, a central operations owner and enough data to integrate are promising; teams seeking a one-off report without an owner are not.” That is a hypothetical rule, not a claim about any actual client. Test it against recent deals and new enquiries. Record where it misclassifies a good opportunity.
Keep exceptions visible
An unusual customer may become a new segment, or it may be a costly exception. Track it without immediately rewriting the ICP. Revisit the rule after enough deals mature to judge adoption and economics. The objective is better marketing and sales decisions, not a perfect static persona.
An ICP evidence grid
Create rows for recent customers, churned accounts, stalled deals and rejected enquiries. Use columns for trigger, problem owner, previous solution, required capability, budget path and observed outcome. Add a column for evidence quality: direct buyer statement, CRM note or internal inference. Do not treat all three as equally reliable. Look for conditions that distinguish successful customers from merely interested ones.
Write a provisional “yes, maybe, no” rule. A yes account has the trigger and prerequisites seen in strong customers. A maybe account lacks one piece of evidence and needs a question answered. A no account lacks a condition the product currently requires. Test the rule with sales on the next ten enquiries and record where it fails. Marketing can then change targeting and page language, while product can see whether a repeated exclusion signals a future opportunity. Keep the rule small enough for a seller to use on a live call.
Revisit exclusions when the product or onboarding requirements change.
Frequently asked questions
Should churned customers be in an ICP analysis?
Yes. A sale followed by poor activation or churn can show that the company was not a good operating fit despite an initial purchase.
Can the ICP change after a new segment wins?
Yes. Document the evidence and update the rule when repeated behavior supports it, rather than changing it after one attractive account.
Is your ICP broad enough that every company appears to fit?
Bring a mix of wins, losses and wrong-fit enquiries. We can turn those cases into a more useful qualification and messaging rule.
Talk through the problem ↗See how we could work together
