What is an ideal customer profile (ICP)?

An ideal customer profile is a written description of the specific type of customer you are the best possible choice for: their industry, size and stage, business model, the exact job they hire you to do, and the attributes they share. It describes a company or segment, not a single person. The point of an ICP is not to list everyone who could pay you; it is to name a narrow group with similar problems so you can build for them precisely, learn faster from their feedback, and let word spread among buyers who already talk to each other. A good ICP is specific enough that you can look at a prospect and say plainly whether they fit or not.

How do you define your ideal customer profile?

Start from the customers you already serve well and look for the attributes that repeat: industry, company size and stage, business model, technical sophistication, how they buy, and what they value. Then write the shared job to be done, the specific problem this group hires you to solve. Braintrust's Ankur Goyal frames it as picking customers with similar attributes so you can zero in on the problems they care about. Keep it to one page and make it exclusionary on purpose: if your ICP does not rule out most of the market, it is too broad to guide what you build. Revisit it as you learn, but resist widening it just because someone with a checkbook asked.

Why should a startup focus on a small number of customers?

Because a small group with shared attributes lets one insight serve many. When your customers look alike, the feature you build for one is the feature the next ten need, so your learning compounds instead of scattering. Trying to make everyone happy dilutes your effort and produces a product that is adequate for many and essential to no one. Founders building AI-native services companies warn about the early demand trap: capping the number of pilot customers to a small handful so you can actually serve them well, rather than saying yes to every logo and serving all of them badly. Depth with a few of the right customers beats breadth across the wrong ones.

What is the difference between an ideal customer profile and a buyer persona?

An ideal customer profile describes the type of organization you are the best fit for: industry, size, stage, business model, and the problem they share. A buyer persona describes the individual human inside that organization who you actually sell to: their role, what they are measured on, and what would make them champion you internally. You need both, but in order. The ICP tells you which companies to go after; the persona tells you who to reach and what to say once you are inside. Getting the ICP wrong is more expensive, because it sends your product roadmap and your sales effort at the wrong companies entirely.

Ideal Customer Profile: Who to Serve First

Pick your customer before you pick your roadmap. The single decision that shapes what you build, who you sell to, and how fast you find product-market fit is your ideal customer profile, and most founders skip it. They try to be useful to everyone and end up sharp for no one. In a Greylock conversation about why high-taste companies like Stripe, Instacart, and Airtable use Braintrust, founder and CEO Ankur Goyal describes the opposite move: get very specific about which customers to bet on, then go all in on them. Here is how to define your ideal customer profile and why a small set of the right customers compounds faster than a long list of the wrong ones.

The short version

Your ideal customer profile (ICP) is a written description of the specific type of customer you are the best possible choice for: their industry, size and stage, business model, the job they are trying to do, and the attributes they share. It is not a wish list of everyone who could pay you. The point is to name a narrow group with similar problems so you can build for them precisely, learn faster, and let word spread among people who already talk to each other. Goyal is blunt about the payoff: "If you try to make a very small number of people that have similar attributes ... then you can really zero in on what the problems are that they care about." The mistake to avoid is the reflex to keep your ICP wide so you do not turn away revenue. A wide ICP does not win more customers; it just makes your product mediocre for all of them.

What an ideal customer profile actually is

Define your ICP in terms of attributes that repeat, not adjectives that flatter. Three layers do most of the work:

Firmographics: industry, company size and stage, and business model. "Series A to Series B fintechs" is a profile. "Fast-growing companies" is not.
The shared job: the specific problem this group hires you to solve. If two customers use your product for genuinely different jobs, they are probably not in the same ICP.
Behavioral attributes: the traits that make them act alike, such as technical sophistication, how they buy (bottom-up adoption versus a top-down committee), and what they actually value.

Keep the ICP (the type of company) separate from the buyer persona (the individual inside it). The ICP tells you which companies to chase; the persona tells you who to reach once you are in the door. Get the ICP right first, because it points your whole roadmap and go-to-market at a set of companies, and pointing them at the wrong set is the expensive error.

Why a few high-taste customers beat chasing everyone

Braintrust makes evaluation and observability tools for teams shipping AI products, and its customer list reads like a who's who of careful engineering organizations: Stripe, Instacart, Airtable, and others. Ask how a young company earned that roster and Goyal's answer is not "we said yes to everyone." It is the opposite. "The thing that we did well is that we were very specific about which customers we wanted to make a bet on," he says. They chose a small number of discerning companies with shared attributes and worked to make those specific customers genuinely happy.

The mechanics of why this compounds are simple. When your customers share attributes, one insight serves many. The problem you solve for Stripe is close to the problem you solve for the next high-taste engineering team, so every lesson you learn lands more than once. Goyal makes the same point in a separate conversation about evals: the fastest path to product-market fit is to focus on a specific group of customers with similar attributes, because that focus lets you build the right thing instead of a little bit of everything. High-taste customers also compound in a second way. They are demanding, they know what good looks like, and delighting them produces a product other serious buyers trust and reference.

The trap: early demand pulls you wide

The hardest part of holding an ICP is that early demand fights you. When you are hungry for revenue, every interested logo looks like a customer you cannot afford to refuse, and it is tempting to stretch the product to fit each one. Founders building AI-native services companies have a name for this: the early demand trap. Their advice is to cap the number of pilot customers to a small handful, so you can actually serve them well rather than saying yes to everyone and serving all of them badly. Trying to make everyone happy dilutes your effort and, as Braintrust's own experience shows, leads to less effective solutions than a narrow, deliberate bet.

The founders who resist the pull tend to describe their target in almost uncomfortably narrow terms. In a YC Root Access interview, one founder building an AI-native ERP described their segment this way: "We very narrowly focus on a certain segment which was like tech companies that had outgrown QuickBooks." That is a real ICP. It names a business type, a stage, and the exact moment of pain. Another YC founder put the general principle plainly: it is more effective to focus on one specific problem initially than to try to do everything at once. A tight ICP is how you turn that focus into a customer list you can name.

Go all in on the customers you pick

Choosing a narrow ICP only pays off if you then commit to it. Braintrust's edge was not just picking the right customers; it was the depth of the relationship once picked. "We just roll up our sleeves and go all in," Goyal says of how the team works with the customers it bets on. That depth is the return on a narrow ICP: because you are not spread across every possible buyer, you can give the few you chose the kind of attention that turns them into references and design partners.

There is a security dimension here that fits a narrow ICP well. When your ideal customers share attributes, they usually share compliance and data-handling expectations too. A segment of regulated fintechs, for example, will all want the same SOC 2 posture, the same data-residency answers, and the same access controls. Build that once for the segment and it serves every customer in it, instead of improvising a different security story for each unrelated logo. A focused ICP makes trust cheaper to earn, because you are earning it the same way with everyone you serve.

What to do this week

Write your ICP on one page: industry, company size and stage, business model, the shared job to be done, and the two or three behavioral attributes your best customers have in common.
Pressure-test it for narrowness. If your ICP does not rule out most of the market, it is a wish list, not a profile. Cut it until you can look at any prospect and say fit or no fit in one sentence.
List your current customers against it. Mark who fits and who does not. The misfits are usually where your roadmap is getting pulled off course.
Separate the profile from the person. Under the ICP, write the one buyer persona inside those companies you actually sell to, and what they are measured on.
Cap your pilots. Pick a small handful of on-profile customers to go all in on this quarter, and say a polite no, or a "not yet", to the off-profile demand that will try to widen you.

Your ideal customer profile is the decision every other decision inherits, so it is worth making on purpose instead of by default. If you want the operating system around it, from choosing your ICP to running the company with AI, that is what we teach in the AI Operating System for Startups. Once you know who your ICP is, the next choice is how to reach them: see sales motions, top-down vs bottom-up for which motion fits your buyer. For turning customer signal into what you build, see AI for product managers, and for delivering the finished outcome to those customers, see service as software.

Sources

Why High Taste Organizations Use Braintrust (Greylock), the conversation this article distills, where Braintrust's Ankur Goyal explains being specific about which customers to bet on and going all in on them.
Braintrust's Ankur Goyal on Why Evals Are the Core of AI Development (Greylock), Goyal on focusing on a specific group of customers with similar attributes as the path to product-market fit.
How to Build an AI-Native Services Company (Y Combinator), on the early demand trap and capping pilot customers to a small handful you can serve well.
The ERP for the AI Revolution is here (YC Root Access), a founder's narrowly defined segment ("tech companies that had outgrown QuickBooks") as a model ICP.
This Is The Next Industry AI Will Disrupt (YC Root Access), on focusing on one specific problem initially rather than doing everything at once.
Background on Ankur Goyal and Braintrust: Ankur Goyal on LinkedIn, and reporting on Braintrust's customers (Stripe, Instacart, Airtable) in Forbes. Goyal previously founded Impira, acquired by Figma.