An ideal customer profile describes the type of company that gets the most value from what you sell, closes fastest, and stays longest. It is not a person and not a wish list. Built properly, it is a short set of attributes you can check from outside a company, so a rep can look at an account and decide in under a minute whether it belongs in the pipeline.
To build one from existing customers:
- Start from outcomes, not revenue. Find the customers who actually got the result you promise.
- Look for what they share across firmographics, structure, and behaviour.
- Keep only the attributes you can observe before talking to the company.
- Write the exclusions too — who this is not for is half the value.
- Test it against last year’s losses and narrow until it separates wins from losses.
The rest of this guide covers each step, how an ICP differs from a persona and a TAM, what to do with fewer than ten customers, and how often to revisit it.
What is an ICP?
An ICP is a description of the company you should be selling to. The best working definition is the intersection of two things: where your product creates the most value, and where you can win most cheaply. Either alone gives you a bad profile. Companies that would get enormous value but take eighteen months and a security review to close are not ideal, and companies that buy instantly but churn in two quarters are worse than no customer at all.
A good ICP is short. Five to eight attributes is typical; more than ten means you have written a description of your favourite customer rather than a filter. Every attribute has to earn its place by changing a decision — if reps would qualify the same accounts in or out without it, delete it.
It also has to be checkable from outside. “Values data-driven decision-making” is not an ICP attribute, because you cannot verify it before a conversation, which means it cannot be used to build a list. “Has a dedicated RevOps function” is, because you can confirm it from a headcount page in fifteen seconds.
ICP vs buyer persona vs TAM
Three different questions, routinely collapsed into one.
TAM is how big is the market — the total set of companies that could conceivably buy. It is a number for investors and board decks. It does not help a rep choose what to do on Tuesday.
ICP is which companies to sell to. It is a filter over the TAM, and it operates at the account level: industry, size, structure, stack, stage, signals.
Buyer persona is which person inside those companies to talk to. It operates at the human level: role, responsibilities, incentives, what they read, what they fear.
The order matters. ICP narrows the TAM to the accounts worth pursuing. Persona identifies who to approach inside each one. And the internal champion is narrower still — the specific individual who will advocate for you, who may or may not match your persona. A team that builds personas without a defined ICP ends up with beautifully targeted messaging aimed at companies that were never going to buy.
How to build an ICP from existing customers
If you have customers, the data is already there. The work is being honest about it.
Start with outcomes, not revenue. List the customers who actually got the result your product promises — not the ones who pay the most. Big logos that never adopted are a trap; they teach you to chase companies that sign and do not stay. If you track it, use time-to-value and expansion rather than contract size.
Then look for what those accounts share. Work across three layers:
- Firmographic: industry, employee count, revenue, geography, funding stage. The blunt filters, and still the most useful.
- Structural: does a specific function exist? Is there a named owner for the problem? A company with no RevOps team cannot buy a RevOps tool no matter how well it fits.
- Behavioural: what were they doing before they bought? Running a manual workaround, using a competitor, hiring into the problem area. This layer predicts best and is used least.
Keep only what you can observe. Go through your list and delete every attribute you could not verify from LinkedIn, a careers page, a website, or a funding announcement. What survives is your actual ICP; the rest is post-hoc description.
Write the exclusions. Explicit anti-criteria — company types you will not pursue regardless of how good they look — save more rep time than the inclusion list does, because they end deals early that would otherwise consume a quarter.
What to do when you have no customers yet
With fewer than roughly ten customers you do not have a pattern, you have anecdotes. Treat the ICP as a hypothesis and design it to be disproven fast.
Start from the problem rather than the market. Describe the specific pain precisely, then ask what has to be true about a company for that pain to be expensive enough to pay to fix. Usually it is a threshold: enough volume, enough headcount, enough complexity. Below that line the problem exists but is tolerable, and tolerable problems do not get budget.
Then pick the narrowest starting segment you can defend. The instinct is to stay broad to avoid missing opportunity, and it is wrong — a broad early ICP produces scattered feedback you cannot learn from. Narrow enough that your customers talk to each other, and word of mouth starts doing work no campaign can buy.
Write the hypothesis down with a review date. After the next ten deals, compare what you predicted against what actually closed. The gap is your real ICP.
The attributes that actually matter
Most ICPs over-index on firmographics because they are easy to filter on. The attributes that separate a good ICP from a mediocre one are usually structural or behavioural:
- Does the problem have an owner? A named person whose job is the thing you fix. No owner means no champion, no budget, and no urgency.
- Are they already spending on the problem? In headcount, in a competitor, or in a maintained workaround. Existing spend proves the problem is worth money to them, which is the hardest thing to establish from scratch.
- What is the trigger frequency? Does this type of company regularly generate the buying signals you can act on? A segment that fits perfectly but never changes is a segment you cannot prospect into.
- How many people have to say yes? This drives cycle length more than deal size does, and it is knowable in advance from company size and structure.
- Can they adopt it? Technical prerequisites, data availability, and integration requirements. A perfect-fit company that cannot implement for two quarters is a churn risk, not a win.
How to test and narrow your ICP
An ICP is a claim, so test it against evidence you already have.
Run it backwards over the last twelve months. Score every closed-won and closed-lost deal against the profile. A working ICP produces a visible separation — wins cluster high, losses cluster low. If wins and losses score about the same, the profile is describing your market rather than filtering it, and every attribute is suspect.
Check your churn against it too. Customers who fit the ICP and churned anyway are the most informative accounts you have: either an attribute is wrong, or one is missing.
Then narrow. Most teams’ first honest reaction to the scoring exercise is that their ICP is too broad, and the fix is to cut the segment that produces the most losses even if it also produces some wins. Broad ICPs feel safe and produce diffuse pipeline, slow cycles, and messaging that has to be vague enough to cover everyone.
How to write it down so the team actually uses it
Most ICPs die in a slide. The format that survives contact with a working rep is a one-page checklist, not a narrative, and it has four parts.
Must-haves. Three to five attributes that are non-negotiable, each phrased as something a rep can verify in under a minute. “Has a named owner for revenue operations” works. “Is growing fast” does not, because two people will interpret it differently.
Strong indicators. Attributes that raise confidence but are not disqualifying on their own. These are what separate a good-fit account from a great-fit one, and they are where recent signals belong.
Disqualifiers. Written explicitly, and given equal space. This is the section reps read most, because it gives them permission to walk away, which is the hardest thing to do without cover from a document.
One sentence on why. For each must-have, a short reason drawn from evidence. An attribute nobody can justify is one that will quietly get ignored the first time a rep finds an exciting account that fails it.
Then make it operational. A profile that lives in a strategy doc is a reference; a profile that determines which accounts appear in a rep’s queue on Monday morning is a system. If your ICP does not change what shows up in front of the team, it is not doing any work — and that gap, not the quality of the thinking, is why most ICPs stop mattering within two quarters of being written.
When to revisit it
Twice a year is a reasonable default, plus any time one of these happens: you ship something that changes who gets value, you raise and need to grow into a bigger market, win rates move materially in one segment, or churn concentrates somewhere specific.
Resist rewriting it after a single unusual deal. One great customer outside the profile is not a signal — it is the reason the profile has edges. Three of them in a quarter is a signal.
How Alfa fits
The gap between having an ICP and using one is where most of the value leaks out. A profile written in a strategy doc does not change what a rep does on Monday unless it becomes a list of actual companies, refreshed as the market moves — and building that list by hand is the work nobody has time for, so it gets built once, ages, and quietly stops being used.
Alfa starts from what you sell and the buyers you care about, then turns market movement into a live stream of matching accounts, the likely champions inside them, and the reasons they may care now. The ICP stops being a document and becomes the thing that generates the pipeline.
You keep the judgment. You define who you sell to and refine it as you learn; Alfa keeps the list of real companies in sync with that definition instead of letting it decay.
For a worked example of what that output looks like, see our similar companies lists, where one named company is turned into the ten that most resemble it, with the profile and buyer for each.
With the market defined, the next two steps are watching for buying signals inside those accounts and finding the champion who will advocate for you. Building and running that loop is the job we call the GTM Builder.