Because it describes an aspiration rather than a decision. Most ideal customer profiles read like a wish: mid market to enterprise, growth minded, values innovation. Nothing in that sentence tells a salesperson which meeting to take or a marketer which campaign to cut, so nobody consults it.
A useful ICP is uncomfortable. It rules things out. If yours has never caused an argument, it is probably not specific enough to be doing any work.
We build websites for B2B software companies, and the ICP is the input that decides almost everything about the site. So we have a fairly opinionated way of running this exercise.
An ideal customer profile describes the type of company you serve best. It is about the organisation, not the individual: size, industry, structure, situation and the problem they have. That is the whole scope.
It is not a persona. A persona describes a person inside that company, their role and what they care about. You need both, and confusing them produces a document that is too abstract to target and too specific to be strategy.
It is also not your total addressable market. The market is who could buy. The ICP is who you should chase, which is deliberately smaller. Teams that treat these as the same thing end up with an ICP that excludes nobody.
Three ways, and we see all three regularly. The first is aspiration. The document describes the enterprise logos the founders want rather than the companies that actually buy, renew and refer. Marketing then builds for a buyer who has never purchased.
The second is breadth. Every stakeholder adds a segment they do not want to lose, and the profile widens until it covers most of the economy. A profile that includes everyone provides no guidance, which makes it feel safe and useless at the same time.
The third is that it is unfalsifiable. Values innovation and cares about efficiency cannot be checked against a real company, so nobody can use them to decide anything. Every attribute in a good ICP should be something you could verify from public information in under a minute.
Three parts. First, firmographics you can check: employee count range, industry, geography, and the technology or business model they run on. These are the filters that make a list buildable.
Second, a trigger. What has to be true right now for this company to want you? A recent funding round, a new hire in a specific role, a migration, a compliance deadline, or a growth stage that breaks their current process. The trigger is what turns a fit into a reason to buy this quarter.
Third, a disqualifier list, which we will come back to because it is the part everyone skips and the part that does the most work.
Look at outcomes, not revenue. Rank your customers by retention, expansion and how quickly they got value, not by contract size. The largest customer is often the worst fit, because they bought despite the product rather than because of it.
Take the top ten and the bottom ten and look for what differs structurally. Not personality, not enthusiasm. Things like whether they had someone whose actual job was the thing you solve, or whether they were already doing the process manually before they bought.
That second pattern is the most reliable signal we see. Companies already doing the work badly convert and retain far better than companies who would need to start doing it. Selling a solution to people without the problem yet is the slowest path there is.
Because it is the only part that changes behaviour this week. A list of who you are for gets nodded at. A list of who you are not for lets a salesperson end a bad call politely and lets a marketer cut a campaign without a meeting.
Write it as specific, checkable rules. Fewer than a stated number of employees. No one in a named function. Operating in a market you cannot legally serve. Already committed to a platform you cannot integrate with. Each rule should be something anyone could apply from a company's website.
Then track how often it is overruled. Every exception is either a mistake you are repeating or a signal that the profile is out of date, and you cannot tell which unless somebody is counting.
It removes the excuse for being vague. Building a list used to be the constraint, so people wrote broad profiles because narrow ones were hard to action. That is no longer the situation. Apollo states on its own site that its database holds 240 million contacts and 30 million companies, and that over 600,000 companies use the platform.
When a list of that specificity can be assembled in an afternoon, a broad ICP is not pragmatism. It is an unmade decision. The scarce resource is attention, not records, and a narrow profile is how you spend attention well.
The same shift applies to enrichment. Tools like Clay and Apollo can check most firmographic criteria automatically, which means your ICP attributes should be chosen partly for whether they can be checked in bulk. An attribute nobody can filter on will never be used.
Directly, and more visibly than most teams expect. The homepage has to name the buyer clearly enough that the right person recognises themselves and the wrong person leaves. That is only possible once the ICP is settled, which is why we run this exercise before writing copy.
It also decides your proof. Case studies should feature companies that look like the ICP, because a prospect scans for someone resembling themselves before reading a word. Our notes on case study page design cover how to make that recognition fast.
And it decides how much you narrow. If you genuinely serve two distinct profiles, the site has to handle that deliberately rather than by averaging them into one vague message. Our piece on homepages with two audiences covers the ways that can work.
Review it quarterly against closed deals, and give it an owner. The test of whether an ICP is real is not whether it exists but whether sales can recite it. If they cannot, it is a document rather than a decision, and you should ask them what they use instead, because that is your actual ICP.
Watch for drift in both directions. New customers arriving from a segment you did not target is useful information, not a nuisance. Some of the best repositioning we have seen started with somebody noticing that a third of new revenue came from an industry nobody had planned for.
Keep the document short. One page, three sections, checkable attributes. Long ICP decks get built once and never opened, which returns us to the original problem.
Rank your customers by retention and expansion rather than size, take the top ten, and write down what is structurally true of all of them. Then write five disqualifiers specific enough that a salesperson could apply them from a prospect's website in a minute.
Share both with sales and ask whether they would actually use them. If the answer is no, keep rewriting until it is yes, because an ICP nobody uses is worse than none at all. If you want help turning that profile into a site that speaks to it, or a look at whether your current pages match the customers you actually want, we are happy to walk through it at phoenix.studio.
Tell us where you want to go. We'll tell you how we'd get you there.