The ICP problem in SaaS is structural. Most ideal customer profiles are built by looking at the existing customer base and asking: what do our best customers have in common? This is backward-looking analysis that's good at describing who you've successfully sold to, not who your product is actually best suited to serve.
The customer who bought your product is not the same as the customer your product creates the most value for. They might overlap significantly. They often don't overlap completely.
The gap between "who bought" and "who you create most value for" is where your ICP work needs to happen.
How to find the real ICP:
Start with retention data, not sales data. Which customer cohorts have the highest 24-month retention rates, the highest NRR, and the lowest support cost to serve? These are the customers your product genuinely fits. Build your ICP from the retained, expanding cohort, not from the sold cohort.
Look at product usage depth, not just breadth. Customers who use 80% of your product's capabilities are a different ICP than customers who use 20%. The deep-usage customers have a different job to be done, different organizational context, and different buying behavior.
Interview your best customers about the problem they had before they found you. Not what they use your product for now — what problem they were trying to solve before they bought it. This reveals the real purchase trigger, which is more specific and actionable than any demographic ICP attribute.
Find the anti-ICP. The customers you churn most quickly, support most heavily, and win deals with who then expand the least — these customers define the edges of your ICP just as clearly as your best customers define the center.
Your ICP should be a specific, testable hypothesis with retention data to support it. If it's a marketing persona, it's too vague to be useful.