How narrow should my ICP be at seed stage?

Narrow enough to name the list, say no, and hear the same story twice

A seed-stage ICP is narrow enough when you can write down the companies that fit, describe their problem in the words they use, point to the event that makes it urgent this quarter, and refuse a prospect who does not match. It is too narrow when the constraints are arbitrary and you cannot fill a list.

Most advice on this question stops at “go narrow”. That is right, but it is not an answer. The practical answer is a set of tests, and the rest of this page gives them. The parent question, how to define the profile in the first place, is covered in how to define the right ICP for an early-stage startup. This page assumes you have a draft and asks how tight it should be.

Many doors, one chosen. The others are not closed; they wait for a later cycle.

Why the default error is too wide, not too narrow

Founders widen their profile because narrowing feels like giving up revenue. The evidence from people who have watched many startups points the other way: the profile that is too wide is the common failure, and the one that is too narrow is rare and easy to fix.

Sybill quotes Salesforce in its list of ICP mistakes: “The number one mistake people make when drafting an ICP is casting too wide a net for their ideal customers.” Its own test is blunt: “If your ICP cannot help a rep say no, it is not focused enough.” See 8 common ICP mistakes. Startups.com calls a profile like “small and mid-market businesses across multiple verticals” “a refusal to choose”, and states that “The strongest ICPs are narrow on purpose” in its ICP lexicon entry. Stackmatix puts the sequence plainly: “Most startups win by going narrow first and widening later”, in its startup ICP guide.

Paul Graham frames narrowness as a tactic rather than a limit in Do Things that Don’t Scale: “Sometimes the right unscalable trick is to focus on a deliberately narrow market.” His example is Facebook: “At first it was just for Harvard students.” The image he uses is a fire kept small until it is hot enough to add logs. At seed stage you are building heat, not reach.

Narrow on problem and trigger, not on firmographics

The right axis for narrowing is the situation, not the company size. A profile that says “one problem, one workaround, one trigger, one buyer who can say yes” is narrow in the way that matters. A profile that says “companies with a certain headcount in one region” is narrow in a way that teaches you nothing.

a16z’s framework lists specific, measurable qualities as the mark of a good profile: “Great ICPs consist of specific, measurable qualities.” It also warns that a big deal with a large company is not yet an ICP; check whether you solve a meaningful problem across the organization or a point problem for one team. See a framework for defining and refining your ICP. Valor VC’s five tests for a seed ICP are all about situation: who is eager to refer you to a peer in the same role, who with the same title does not need you, what recent event makes your product urgent, which sector feels the problem most severely, and whether the list of accounts looks like your pipeline for the next two years. Read defining the seed stage startup’s ICP.

Axis	Narrowing that teaches	Narrowing that is arbitrary
Problem	One workflow that breaks, in the buyer’s words	“Needs better productivity”
Trigger	A new hire, a failed audit, a tool that was cancelled	None, or “growing companies”
Buyer	One role who owns the pain and the budget	A list of six titles
Alternative	One workaround they use today	“No current solution”
Firmographics	Only what predicts the problem above	Headcount band, region, funding stage chosen for the database filter
Exclusions	Who you will refuse and why	None written down

Stackmatix calls the exclusions row the most profitable part of the template, because knowing “who not to chase protects your limited capacity.” Write that row first; it is the fastest way to find out whether your profile can say no.

Keep fit and readiness apart. Fit is whether the account has the problem, the workaround and the buyer; readiness is whether the trigger has fired and someone is looking now. Sybill lists “Confusing ICP fit with purchase readiness” among its eight mistakes, and its fix is literal: “Maintain separate fields for ICP fit and purchase readiness.” An account that fits but is not ready stays in the profile and waits for its trigger; an account that is ready but does not fit is the warm lead your exclusions row exists to refuse. A profile that folds the two into one score narrows on the wrong thing, because it drops good accounts that are merely early.

The four tests that tell you it is narrow enough

Your ICP is narrow enough when it passes four tests: the list test, the language test, the trigger test and the refusal test. Fail one and the profile is still a category, not a customer.

1. The list test. Can you name the accounts? Open a spreadsheet and write down the companies that fit, with the role you would contact. If you cannot get past a dozen names without guessing, the profile is not findable. If the list runs into the thousands, it is not a wedge.
2. The language test. Can you describe the problem in words you heard from a buyer, not words you wrote in a deck? Keep a file of phrases from calls. If the profile uses your category name and the buyers use a different one, the profile is wider than the market you actually talk to.
3. The trigger test. Can you name the event that moves someone from tolerating the problem to looking for a fix? Without a trigger you will contact people who agree with you and never buy. Valor VC’s version of the question is which recent event at a company makes the solution urgently needed.
4. The refusal test. Can a founder or a rep tell a warm inbound lead no, with a reason written in the profile? Sybill’s rule applies: a profile that cannot help you say no is not focused enough.

Openskool adds the reason this matters early: “A narrower initial segment can make language, outreach, onboarding and feedback more comparable.” A narrow profile makes every conversation count twice, once as a sale and once as a data point. Read its ICP lesson.

How to tell it is too narrow

A profile is too narrow when the constraints are arbitrary, when the list is too short to run a test, or when you exclude buyers who have the identical problem and trigger. Too narrow is rare, and it is usually a symptom of filtering by database fields instead of by situation.

Three signs, each with a fix:

• You cannot fill a list for one cycle. You have added a constraint that does not predict the problem. Remove the constraint that was easiest to select and hardest to defend.
• A prospect outside the profile tells the same story in the same words. Your firmographics are too tight and your problem definition is right. Keep the problem, loosen the firmographic, and log the exception as a hypothesis.
• The buyer, the user and the budget holder are three people. The profile is narrow on company and wide on committee. Narrow the buyer to the role that owns the pain and the budget, and widen the company criteria if needed.

Too narrow is cheap to correct because the list grows when you remove one line. Too wide is expensive because every conversation argues with the last one, and you cannot tell which part of the message failed.

What the one measured example shows

Most writing on ICP width is opinion from practitioners and investors. The clearest measured example of narrowing is Superhuman’s product-market fit work, where segmenting to the most demanding users raised the fit score before any product change.

Rahul Vohra describes using Sean Ellis’s survey, where “the magic number was 40%” of users saying they would be very disappointed without the product. Superhuman scored 22 percent. After splitting the respondents and focusing on the segment that was already very disappointed, he reports: “After segmenting to focus on the very disappointed set of users, we were at 33%.” The score then rose to 58 percent over three quarters of product work aimed at that segment. The profile of the high-expectation customer, defined by Julie Supan as “the most discerning person within your target demographic”, was built from those respondents. Read how Superhuman built an engine to find product-market fit.

Two cautions. These are self-reported numbers from one company, so treat them as an illustration of mechanism, not a benchmark. And the narrowing happened on a group that already used the product; before you have users, you narrow on interviews and on who takes a next step.

How to test the width without a sales team

You test the width of a profile by running two or three adjacent slices against each other with the same message and counting who replies, who recognizes the problem, and who takes a next step that costs them something. Elapsed weeks matter less than contacts per slice. How long such a slice test has to run before you decide is in how long to run a GTM experiment.

Outbound Republic’s playbook for validating a profile proposes testing “two to three adjacent segments simultaneously” and warns that “Below 100, you cannot distinguish genuine ICP misfit from bad luck.” Its sample guideline of 150 to 200 contacted prospects per segment is the author’s own rule and not sourced, so use it as an order of magnitude. See how to validate a new ICP in 30 days. If outbound volume like that is out of reach, the interview version works at seed: ten conversations per slice, the same three questions, and a tally of who described the problem before you did. The method for that is in how to validate an ICP before outreach, and the record you keep is in the startup ICP evidence record.

Scorecards like ImpelHub’s, which grade candidate profiles “on a 1 to 10 scale for all 11 criteria”, help you rank slices before you test; see ICP grading. They do not replace the test. A high score on a scorecard is your opinion with decimals.

Widen in sequence, one variable at a time

You widen when the current list is nearly worked, the conversations have stopped surprising you, and a second slice has shown the same problem and trigger without you steering the interview. You widen one variable at a time, and you keep the old profile as the control.

Openskool’s line is the right frame: “Focus is a sequence decision, not a claim that other customers will never be served.” Stackmatix observes that “Most strong ICPs are the result of two or three refinement cycles, not a single workshop.” Plan the cycles rather than defending the first draft.

Give the cycles a date. Stackmatix’s rhythm for the early stage is to “Revisit yours every 6 to 8 weeks or after any major product change.” Sybill, writing for teams with more history, says to “Review ICP performance quarterly and conduct a deeper annual refresh”, and to pull the review forward “when new alternatives emerge, product capabilities change, a new segment converts, or retention and expansion patterns shift.” At seed, six to eight weeks is the better clock, because one cycle of outreach or ten interviews is about what you can finish in that time, and the review reads that evidence rather than a feeling. Put the review date in the record below and do not widen before it.

Stage	Profile	What you learn	Gate to the next stage
Wedge	One problem, one trigger, one buyer, a named list	Whether the message lands and a next step follows	The list is nearly worked and the stories repeat
Adjacent	Same problem, one variable changed (role, size, sector or alternative)	Whether the motion survives the change	The adjacent slice matches the wedge on recognition and next-step rate
Horizontal	Several slices that share the problem	Which slices need their own message	Each slice has its own list, message and owner

If two slices both pass the gates, choosing between two startup customer segments gives a way to decide without averaging them. Rewrite the profile from call evidence, one criterion at a time, and version each change.

Pitfalls

• Narrowing by the filters your prospecting tool offers instead of by the problem.
• Treating a large deal with a large company as proof that large companies are the profile.
• Blending two segments into one message so the market slide looks bigger.
• Skipping the exclusions row, so nobody can say no to a warm lead.
• Reading one friendly customer as a segment.
• Widening because the current list feels small before the stories have started to repeat.
• Letting the profile live in a deck rather than in the CRM fields where every account is tagged.

When a narrow ICP is the wrong move

A narrow profile is the wrong move when your product only works with network density inside a large group, when the buying decision sits with a committee you cannot narrow to one role, or when you have paying customers from several segments and no evidence yet about which repeats. In those cases, pick the smallest group that still produces the effect, or run the segments as separate hypotheses until one repeats.

Even then, the test is the same: can you name the list, can you say no, and do the stories repeat. The answer decides the width, not a preference for big or small.

A fill-in record

Profile name: <wedge name>
Problem, in the buyer's words: "<phrase heard on a call>"
Workaround they use today: <tool, spreadsheet, person>
Trigger that makes it urgent: <event>
Buyer who owns pain and budget: <role>
Firmographics that predict the problem: <only the ones that do>
Exclusions and reason: <who you refuse, why>
Named list this cycle: <count>, source <where the names came from>
Test slices: <wedge> vs <adjacent slice>, same message, <contacts or interviews per slice>
Pass marks: recognition <count>, next step <count>, refusal heard <yes/no>
Readiness, kept apart from fit: trigger seen on <count> of the named accounts
Review date: <date, 6 to 8 weeks out>  Owner: <name>
Decision options: keep / loosen <one line> / tighten <one line> / split

How Piet works on ICP width with post-seed founders

Piet Baudoin is an AI-native growth marketer at Poldermarketing, a one-person practice working remotely in Dutch and English. The work he enjoys most is this stage: finding the first customer, then the second, then building a process for the third, until there is a repeatable way to bring the product to the customers who are genuinely glad it exists. Not as many customers as possible, the right ones.

He does not only advise on the profile; he builds the list, drafts the messages per slice, sets up the CRM fields so every account carries its slice and its result, and reads the evidence back with you each cycle. The wider plan this fits into is described in first go-to-market strategy for a seed startup and in positioning and messaging. If you want an outside read of whether your current profile can pass the four tests, start with a free growth scan.