How do you find your real ICP in your closed-won deals?

You find your real ICP in your closed-won deals by exporting them from your CRM with industry, size, deal cycle and revenue, uploading the file to Claude, and asking one question: what patterns do you see in my best customers? It takes about five minutes, and the answer often differs from the ICP the team wrote down.

I first wrote this up as a post on my LinkedIn, and it is something I see every week. This is the longer version.

Why is your current ICP probably wrong?

Your current ICP is probably wrong because it was built on assumptions instead of patterns. Most teams sit down, brainstorm who their ideal customer is, and write down job titles, company sizes and industries. Then they load those filters into Apollo or Sales Navigator and start sending. Nobody checked the list against who actually bought.

A few weeks later the reply rates are low and the pipeline is full of bad-fit deals. Everyone blames the messaging. But the messaging is usually not the problem. The targeting is.

The old way is founder intuition, sales team opinions and a whiteboard. Those are not useless, but they are opinions. Most teams are running outbound to an ICP they invented, not one they discovered, and the data to discover it is already sitting in their CRM.

What data do you need from your CRM?

You need an export of your closed-won deals with five fields: company name, industry, company size, deal cycle and revenue. That is enough for patterns to show. Start with your last fifty closed-won deals; that is enough data for Claude to find the clusters without a long clean-up first.

In HubSpot this is an ordinary deal export, filtered to closed-won. A few things make the answer better.

• Keep it to companies, not people. Strip contact names and emails. You do not need them, and you should not upload what you do not need.
• Fill obvious gaps first. If industry or size is blank on many deals, enrich the companies in Clay before you upload, or the patterns will be about missing data.
• Include revenue and deal cycle. Size and industry say who buys. Revenue and cycle say who is worth chasing.

What do you ask Claude?

Ask Claude one question: "What patterns do you see in my best customers?" Then read what comes back. You do not need formulas, pivot tables or a data team. The point is to let the patterns in your own deals speak before anyone argues for their favourite segment.

Keep it simple. Upload the data, ask the question, read the answer. If something surprises you, ask a follow-up about it: which industries close fastest, which sizes bring the most revenue, whether anything changed in the last six months.

The same works for a customer list. When a new client comes in with existing customers, we enrich that list in Clay, give it to Claude and ask for trends. Five minutes later we can see which industries they are in, which job titles dominate and where they cluster geographically. We used to ask clients who they thought their best customers were. They guessed. Now we show them the data.

What kinds of patterns come back?

The patterns that come back are usually about size, industry and timing, and they are often not the ones the team expected. A company-size band can turn out narrower than the one in the ICP document. An industry the team was not targeting can dominate the wins. And a recent event, such as a particular hire, can line up with the fastest deals.

That last kind is the most useful, because it is not just a filter. It is a signal you can watch for, so the next account that shows it gets contacted first. Signals like that tend to go untracked simply because nobody looked.

Every pattern is a claim about your market that you can test. Treat it as the starting hypothesis for the next list, not as the final word.

What do you do with what you find?

You turn what you find into targeting: update the filters, the tiers and the signals your team watches. Change the headcount band, add or drop industries, and start tracking the events that showed up before the best deals. Then run outbound to the new targeting and compare the results with the old.

Fix the targeting first, then worry about the copy. A good message sent to the wrong companies looks like a bad message.

Write the new ICP down where the whole team can see it, with the date and the deals it came from. The next person who wants to change it should know what it was based on, and should have to bring data of their own to change it.

Mike Hurley, our founder, starts every market map the same way, by going back into the CRM before building a list. The ICP map in our library shows how we turn the patterns into tiers, and the signal-based list post covers watching the signals. This is the first step of what we build in GTM engineering.

What should you try first?

Export your last fifty closed-won deals from HubSpot with company name, industry, size, deal cycle and revenue. Remove anything personal, upload the file to Claude, and ask what patterns it sees in your best customers.

Then put the answer next to your ICP document. Where they disagree, believe the deals, change one filter, and watch what happens to the next list.