Your ICP is wrong. Here’s how to fix it.

Most B2B sales teams build their Ideal Customer Profile from assumptions — industry, headcount, geography. The evidence that would actually make it accurate is sitting untouched in their CRM.

Cameron Evans · Founder & CEO, Haku

19 June 2026 · 8 min read

Every sales team has an ICP. Most of them are wrong.

Not wildly wrong — directionally, they usually point at the right industry or company size. But directionally right and operationally useful are very different things. A directionally right ICP still sends your reps after thousands of accounts that will never buy. It still produces pipeline that stalls at the same stages every quarter. It still results in your team spending the majority of their working week — research suggests as much as 64% of selling time — pursuing prospects who were never going to convert.

The reason is almost always the same. The ICP was built from the inside out — from assumptions about who should buy — rather than from the evidence of who actually bought, and more importantly, why.

The difference between a demographic filter and an ICP

Ask most sales leaders to describe their ICP and you will hear something like: mid-market SaaS companies, 50 to 200 employees, UK or US, selling B2B, with a sales team of at least five. That is a demographic filter. It is not an ICP.

The distinction matters enormously. A demographic filter describes what a company looks like on paper — its size, its sector, its geography. What it cannot tell you is whether that company is in a position to buy right now, whether they experience the specific pain your product solves, or whether anything has happened recently that makes them more receptive to your outreach than they were six months ago.

A genuine ICP goes further. It describes not just what your best customers look like, but what was happening in their business at the moment they decided to buy. What triggered the initial conversation. What the competitive context was. How long the cycle took and why. What made them renew — or what made them churn. That level of specificity is what separates a targeting model that compounds from one that stays permanently approximate.

“A demographic filter tells you who might buy. An evidence-based ICP tells you who is ready to buy right now — and why.”

Why CRM data is the most underused asset in B2B sales

The evidence required to build a genuinely accurate ICP already exists in most organisations. It lives in the CRM — in the closed-won deals that contain the actual story of how and why a customer bought, and in the churned deals that reveal where the targeting was wrong.

The problem is that most teams do not analyse it. CRM data is used for pipeline reporting, forecast calls, and activity tracking. It is rarely interrogated for what it actually contains: a detailed record of every sales motion your team has run, what worked, what did not, and the underlying characteristics of the accounts where the deal closed versus the ones where it stalled.

When you look at closed-won deals properly — not just the firmographic attributes, but the qualitative context — patterns emerge that no amount of market research or persona workshopping would surface. You find that a disproportionate number of your best customers had recently hired a new VP of Sales before engaging you. Or that the deals that closed fastest were almost always triggered by a specific pain — fragmented data tools, low reply rates, inaccurate contact data — that a rep had identified and led with. Or that the customers with the highest lifetime value shared a characteristic you had never thought to filter for.

The two types of ICP data most teams ignore

Closed-won analysis: what actually drove the purchase

Closed-won analysis is the practice of going back through every deal your team has closed in the last 12 to 24 months and asking a different set of questions to the ones that go into a standard CRM record. Not just what industry were they in and how many employees did they have. But: what triggered the conversation? What was the business context at the time — a funding round, a new hire, a growth push, a failed tool? What did the rep say that resonated? How did the champion describe the problem internally?

The answers reveal the real ICP — not the company type you think should buy, but the situational context that actually predicts purchase. That context is reproducible. Once you know that a new VP of Sales hire at a Series B SaaS company triggers a six-week evaluation cycle that closes at a specific rate, you have a signal worth building a prospecting motion around.

Retention and expansion data: what separates good customers from bad ones

The second data source most teams leave untouched is retention and expansion. Not all closed deals are equal. A customer who churns after six months and leaves a negative review is a fundamentally different data point to a customer who renews, expands, and refers three other companies. Building an ICP from closed-won data alone, without filtering for customer quality, will optimise your targeting towards the wrong accounts.

When you layer retention data on top of closed-won analysis, the picture sharpens further. The customers who stayed and grew tend to share characteristics the churned ones do not. They had a specific internal champion. They had a clear use case that matched the product's core value rather than a peripheral feature. They were at a stage of growth where the problem was acute rather than theoretical. These are the signals that a genuine ICP is built on.

Why subjective data is killing your targeting

There is a subtler problem that sits underneath the CRM analysis issue, and it is the one most sales teams never address: the data used to build most ICPs is subjective by nature.

ICP workshops typically involve a cross-functional group of people — sales, marketing, customer success, sometimes a founder — answering questions about who the ideal customer is based on their individual experience. The sales rep who has closed mostly enterprise deals will anchor on enterprise. The founder who built the product for a specific customer type will anchor on that type. The result is a consensus view that reflects the loudest voices in the room, not the data.

Subjective ICP data also fails to account for survivorship bias. The customers your team remembers and talks about most are the memorable ones — the big logos, the smooth deals, the enthusiastic champions. The deals that quietly closed and quietly churned without ever generating a case study rarely make it into the ICP conversation. But they are often the most instructive signal about where your targeting is drifting off course.

What it looks like when you get it right

An evidence-based ICP is not a static document. It is a live model that updates as new deal data comes in, that reflects the actual behaviour of your best customers rather than an idealised version of them, and that is specific enough — as one framework puts it — to disqualify at least 70% of the prospects your team might otherwise pursue.

That level of specificity feels counterintuitive to most sales leaders. The instinct is always to keep the addressable market as wide as possible. But a narrower, more precise ICP consistently produces better results: shorter sales cycles, higher conversion rates, lower churn, and reps who spend their time on accounts that are genuinely likely to close rather than accounts that look right on paper.

The teams that get this right tend to have one thing in common: they treat their CRM as an intelligence asset, not just a reporting tool. They go back to their best deals and ask why they closed. They look at their worst deals and ask what the early signs were. And they build their prospecting motion around the patterns that emerge — not the demographic filters that feel intuitive but rarely reflect reality.

How Haku builds your ICP from the evidence you already have

This is precisely the problem Haku was built to solve. Before Haku surfaces a single account or builds a single list, it ingests your business data — your CRM history, your closed-won deals, your case studies, the products you sell, the customers who stayed and grew. It uses this to build a proprietary model of what your best-fit accounts actually look like, grounded in real purchase behaviour rather than assumed demographic profiles.

That model then drives every prospecting decision. When a rep asks Haku to find accounts that match their ICP, they are not getting back a list of companies that fit a generic filter. They are getting accounts that share the underlying characteristics — the situational context, the growth stage, the signals — that preceded purchase in your existing customer base. Accounts where the evidence suggests a genuine buying window, not just a demographic resemblance.

The result is a smaller list of accounts that are significantly more likely to convert, reached at the right moment, with outreach informed by what you already know works. That is the difference between an ICP that sits in a slide deck and one that actually drives pipeline.