Every startup playbook tells you the same thing: define your ICP before you start selling. So you sit down, think about who your product is for, and write something like "B2B SaaS companies, Series A to Series B, 50 to 200 employees, decision maker is VP of Sales or Head of Revenue."
That feels right. It's specific. It sounds like a real customer profile.
It's also almost certainly wrong.
The assumption problem
Your first ICP is always a guess. You haven't sold anything yet. You're describing who you think will buy based on who you built the product for. But the people you built it for and the people who actually buy it are often very different.
A founder builds a project management tool for engineering teams. They target CTOs at mid-stage startups. But their first 5 paying customers turn out to be operations managers at agencies. Nobody saw that coming. The product worked the same way, but the buyer was completely different from what they assumed.
This happens more often than not. Your assumption-based ICP is a starting point, not a destination. The problem is that most founders treat it as fixed and never update it.
A filter is not a profile
"Series A SaaS, 50-200 employees, VP of Sales" is not an ICP. It's a filter. It describes demographic attributes. It tells you nothing about whether these people actually need your product right now.
A real ICP answers different questions:
- What problem are they feeling right now?
- What event triggered that problem?
- What are they currently doing about it?
- Why is their current solution not working?
- What would make them switch this week?
The difference between a filter and a profile is context. "VP of Sales at Series A SaaS" is a filter that matches thousands of people. "VP of Sales who was promoted 6 weeks ago and is building an outbound team for the first time at a company that just raised $5M" is a profile that matches a handful of people who probably need you right now.
"We stopped targeting a job title and started targeting a moment. Our reply rate went from 3% to 22%."
How to build a data-driven ICP
Your real ICP emerges from patterns in your actual interactions. Here's how to find it.
Step 1: Track everything
For every outreach attempt, record:
- Who you reached out to (name, title, company)
- What signal triggered the outreach (funding, hiring, pain post, competitor complaint)
- What happened (no reply, replied, booked call, converted)
- What industry they're in
- What company stage they're at
You don't need a fancy CRM for this. A spreadsheet works. The point is that after 50 to 100 outreach attempts, you'll have enough data to see patterns.
Step 2: Look at your wins, not your list
After you've done some outreach, stop looking at who you targeted and start looking at who responded. Specifically:
- Which leads replied? What did they have in common?
- Which leads booked calls? What made them different?
- Which leads converted? What pattern do you see?
You might discover that your best leads all share a signal you weren't tracking. Maybe everyone who converted had recently posted about a pain point on LinkedIn. Maybe they all came from companies with 10 to 30 employees, not 50 to 200 like you assumed. Maybe the title that converts isn't VP of Sales but Head of Operations.
The patterns are in the data. But you have to collect the data first.
Step 3: Build a feedback loop
Your ICP should update every month. As you do more outreach, you learn more about who actually buys. Use that to refine your targeting:
- Month 1: You target your assumed ICP. Reply rate is 5%. You get 2 calls.
- Month 2: You notice both calls were with founders at pre-seed companies, not Series A VPs. You shift targeting.
- Month 3: With the new targeting, reply rate is 15%. You get 6 calls. 2 convert.
- Month 4: Both conversions came from companies that recently raised funding. You add "raised in last 30 days" as a signal filter.
By month 4, your ICP is completely different from what you started with. And it's based on real data, not assumptions.
The review step: training your system
The fastest way to build a data-driven ICP is to review your leads before you reach out. For each lead:
- Thumbs up: This looks like a real prospect. Good company, good signal, right role.
- Thumbs down: Wrong fit. Maybe they're a competitor. Maybe they're too big. Maybe the signal isn't relevant.
- Reclassify: This person isn't a buyer, they're a partner. Or a champion who can introduce you.
Every thumbs up and thumbs down teaches you something about your real ICP. After reviewing 50 leads, you'll know exactly what a "good lead" looks like for your specific product. Not in theory. From experience.
And if your tools are smart enough to learn from those reviews, the next batch of leads gets better automatically. That's the difference between a static ICP written on a whiteboard and a living profile that improves every week.
Your ICP is alive
The founders who find product-market fit fastest are not the ones with the best initial ICP. They're the ones who update it the most.
Your first guess will be wrong. That's fine. The point is to start reaching out, collect data from real interactions, and let the patterns show you who your actual customer is.
Stop treating your ICP as a fixed definition. Start treating it as a hypothesis you're testing every week. The data will tell you who your customer really is. You just have to listen.