The advice that doesn't apply to you
Search for how to define an ICP and you'll find the same method everywhere: segment your customer base, analyse retention and expansion by segment, identify the cohort with the best economics, target more of them.
That method is correct and it requires a few hundred customers. With 40, every segment has three or four companies in it, one of them is an outlier that will distort whatever you conclude, and the differences you find are noise.
Most early-stage teams respond in one of two ways. They either skip the ICP entirely and sell to whoever will talk to them, or they write an aspirational profile that describes the customer they wish they had. Both leave you guessing.
There is a third option, and it starts from a different question.
Ask "who succeeded?" not "who bought?"
Your purchase data at this stage is mostly a record of who your founder happened to know and which conversations went well. It reflects your sales access, not your product fit.
Your outcome data is far less contaminated. So sort your 40 customers into three groups, using judgement rather than a model:
- Thriving — using it regularly, expanded or would, would be genuinely annoyed if it disappeared
- Fine — renewed, low engagement, no strong feelings
- Struggling — churned, nearly churned, or heavy support burden relative to their value
Even at 40 customers, this split is usually obvious to anyone who works with them. You do not need a scoring model to know which of your customers love you.
Then look for what the thriving ones share
Look across four dimensions, in this order.
1. Situation, not sector. The strongest pattern is almost never "fintech" or "50–200 employees." It's a circumstance: they just hired their first RevOps person, they're migrating off a legacy system, they've grown past the point where a spreadsheet works.
Situation predicts urgency. Firmographics only predict eligibility.
2. The trigger that started the search. Ask ten thriving customers what specifically made them start looking, and expect three to four repeating stories. Those stories are worth more than any demographic cut, because your entire acquisition strategy is really about intercepting them.
3. Who championed it internally. Same role? Same seniority? Someone new in post, or someone long-tenured? The pattern here tells you who to target and, just as usefully, whose absence predicts failure.
4. What they had in place already. Prerequisites you didn't know you had. If every thriving customer already had clean CRM data or a dedicated ops resource, then that's not a nice-to-have — it's a qualification criterion you've been ignoring.
Write it as an exclusion, not a description
The output most teams produce is a paragraph describing an ideal company. It's pleasant to read and impossible to act on, because it never tells anyone to say no.
A useful ICP is a filter. Write it as the conditions under which you would decline:
We work with B2B SaaS companies between Seed and Series A that have an existing customer base of 30+, at least one person owning revenue operations, and a founder actively involved in GTM decisions. We don't take on pre-revenue companies, or teams where no single person owns the pipeline number.
That version changes behaviour in a sales call. The paragraph version doesn't.
Validate it before you commit
Two checks, both cheap.
Retrospective: apply the ICP to your last 20 lost or churned deals. If it would have excluded most of them, it's carrying real information. If it would have accepted them all, it's too loose to be doing anything.
Prospective: score your current open pipeline against it. Then wait a quarter and see whether the high scorers actually closed at a higher rate. This is the only real test, and it means the ICP you write in March is a hypothesis until roughly June.
Treat it as a rolling hypothesis
The mistake I see after a good ICP exercise is engraving the result. With 40 customers you have a well-reasoned guess, not a finding.
Put a date on it, revisit it every two quarters, and keep a note of the deals that violated it and won anyway. Those exceptions are how you find out the profile has shifted — usually a year before the aggregate data would tell you.
The goal at this stage isn't a correct ICP. It's a specific enough one that you can be wrong in a way you'll notice.
Hilal Tasdan
B2B SaaS Growth Marketing Consultant & Fractional CMO. Partner in Growth.