TL;DR
- An ICP is finished when it makes decisions: what to say, and which lead to turn down on Monday morning.
- Describe a buying situation, the trigger, the job, who feels the pain, what done looks like. Industry and headcount are a consequence you back into.
- Stuck between audiences? Score each on pain, willingness to pay, alternatives, and reach. Run the winner for a fixed window.
- Zero customers? Build the V0 from five 20-minute conversations, no selling, and write it in their words.
- AI can draft a polished profile in 30 seconds. The scarce part is making it narrow enough to say no with, and that part is still yours.
You already have an ICP. It just doesn't decide anything.
Here is who types this question. Not the beginner asking what the letters stand for. It's the founder with a profile that sounds finished, "startups, 10 to 200 employees, major tech hubs, seed money in the bank", and a calendar that looks exactly like it did before the profile existed. Same outreach, same everyone-shaped homepage, same guessing.
Or it's the other version, five candidate audiences and no way to pick. Lawyers or dentists or Shopify stores. Fractional CMOs or product marketers. Every week a different one feels right, and the debate itself has started to feel like work.
Six months of LinkedIn outreach aimed at all five reads as activity. Zero booked calls is what it converts to, because a message written for five audiences is written for none.
In 2026 the embarrassing part is that producing the document is free. One prompt and an AI writes you a polished firmographic card, industry, size, revenue, tech stack, personas with stock-photo faces. Free to generate, and it still can't tell you which of this morning's two inbound leads deserves the reply.
The bottleneck was never the document.
The document parks you at this mark. One notch right, a named trigger event, is where the deciding starts.
A usable ICP is a buying situation and a job
The sharpest operators run a blunt test on any profile you hand them. Demographic, or behavioral. "Active adults 30 to 60" is demographic. "Corporate professionals past 40 who just got scared about their heart" is behavioral, it names a situation someone is standing in. The first group is a census. The second group is a queue.
"We can help anyone" is the sentence that keeps profiles broad, and it's usually true, which is what makes it expensive. Being able to help anyone is a capability. Choosing who to be found by is a strategy.
So the working definition. An ICP you can use names the buying situation and the job, not the industry and the headcount. Four parts. The trigger event that started the search this week. The job they are hiring you to do. The person who feels the pain, who is often not the person who signs. And what done looks like, the state where they'd renew and refer you.
a16z's version of why this matters is worth taping over the desk, "no amount of sales or marketing can force a product to durably grow in a market where it isn't needed by the buyers." The widely cited 68% higher win rates for teams with a defined profile trace back to the same mechanism. The profile stops you paying to talk to the wrong queue, and that is the entire trick.
Every section below earns the profile one more decision.
the parts that decide
Fashion e-commerce brands past $100K a month1 whose first media buyer starts next quarter,2 still running ads off a founder-built spreadsheet.3 The founder signs, the new media buyer feels the pain.4 Done looks like the weekly report building itself before Monday standup.5
The threshold filter
A number you can screen a lead against. An industry can’t say no to anyone, a spend level can.
The trigger event
The thing that changed and started the search. It tells you when to show up, and it ages in weeks, which is why it works.
The workaround
What they run today instead of you. Proof the pain is real, already costed, and already annoying someone.
Buyer and user, named apart
The line most profiles blur. Who hurts and who signs need different sentences, sometimes different pages.
What done looks like
The state where they renew and refer. If you can’t describe it, you can’t promise it.
Five parts, and every one can change what you say or who you skip. An industry and a headcount can do neither.
Start from the job. The demographics are a consequence.
Jobs-to-be-Done is the tool, and good mentors reach for it by name. What job is the customer hiring you for. Split it three ways, the functional job, the emotional job, the social job. An acai bowl is "I'm hungry," "a guilty pleasure after a hard day," and "being seen ordering the healthy thing," all at once, and the job you lead with changes who shows up.
The reframe lands with a thud. A product person runs the exercise and admits the buyer is people exactly like her, six years in the role without once considering it. The job names the person.
The job matters because the trigger lives inside it. GrowthMentor exists because in 2018 I was hiring freelancers off Upwork just to have someone to talk a problem through with. The demographic version of me was "founder, Athens, bootstrapped." The situation was professionally alone, at 2am, with a decision I couldn't sanity-check. Only one of those two descriptions was ever going to build a product.
So write the trigger first. Something shifts in a normal week, a boss demands reporting that doesn't exist, a funding round closes, the first sales hire starts, and a person who wasn't looking on Friday is looking on Monday. The sales world now calls this signal-based targeting, and the numbers behind it are blunt. Trigger-based outreach replies at 8 to 15%, against 2 to 5% for cold lists, and reaching someone within 48 hours of the trigger lifts meetings booked by about 40%.
Back the demographics out afterward. They are the shadow the situation casts.
Write it as a filter you can screen a lead against
Mentors assign the same homework so often it deserves its name, the one-page positioning doc. Five boxes. ICP, trigger events, personas, pricing, differentiators. One page, because the ICP has to survive contact with the other four boxes to be real. If the pricing line contradicts the ICP line, one of them is fiction.
The tell of a finished profile is that it reads as a threshold, and the fastest way to feel it is the before-and-after. One IndieHackers worksheet calls the left column the blob. The rewrites below are the ones operators keep making, in every industry, in both directions of the funnel.
The rewrite operators keep making
| The blob | The filter |
|---|---|
| Companies, 50 to 500 employees | Managers who lead 5+ people and hold discretionary budget |
| E-commerce store owners | E-commerce brands doing $80–100K+ a month |
| B2B SaaS companies | Series A teams that just hired their first sales lead |
| Broad millennials, 30 to 45 | Women in their 30s training against muscle loss and injury |
- The filter
- Managers who lead 5+ people and hold discretionary budget
- The filter
- E-commerce brands doing $80–100K+ a month
- The filter
- Series A teams that just hired their first sales lead
- The filter
- Women in their 30s training against muscle loss and injury
Notice what the right column is made of. A number, a role, a moment, a fear. Gusto's launch profile is the canon example of how narrow winning looks, California businesses, under five employees, no benefits offered, salaried staff, on a specific payroll timing. Comically narrow used to be a contrarian take. Now it's the quality bar.
And put the tools in the right order, because founders reach for them too early. Clay, Apollo, and Sales Navigator are list builders. Once the filter is real, they'll find you a thousand people who match it. Point them at a blob and they'll find you a thousand strangers. The profile decides, the tools fetch.
Then run the Monday test. Two inbound leads, one fits the trigger and the threshold, one is a household name that doesn't. If the profile can't make you reply to the first and pass politely on the second, it isn't narrow enough yet. Keep cutting until it picks for you.
Torn between audiences? Score them.
The most common arrival state I see is five candidate audiences and a year of cycling between them. Agencies or SMBs. Creators or founders. The debate feels like diligence, and it burns a quarter each time it reopens.
Score the candidates instead of re-arguing them. Four axes, one to five each. How intense is the pain. How able and willing are they to pay. How weak are the alternatives they'd otherwise reach for. How easily can you reach them. Add it up, and let a boring number end a six-month argument.
two audiences, scored
Candidate A
Creators selling courses
Candidate B
Agencies, 5 to 20 people
Illustrative scores, real method. The higher total is a bet you run first, and a boring number ends a six-month debate.
When two candidates tie, fork. Run both for a fixed window, a few weeks of hand-written outreach at five prospects a day, and pitch each segment live. The sales conversation is the positioning test, and whichever buyer leans in first wins the next quarter. You're scoring a bet to run first, and narrowing is reversible, which is the whole case for picking one and then leaving the decision closed.
The fear under all the cycling is the same, narrowing means turning away someone you could have served. It doesn't. The profile decides who you pursue, and revenue that walks in on its own still gets a handshake. Focus governs your outbound effort, never your ability to say yes.
Rank, test, commit. The agonizing was the expensive part.
Build it from five conversations, not one prompt
Every template on the internet gives the same instruction, analyze your last 12 to 24 months of closed-won deals. Great advice for the reader who doesn't need this post, and useless to the founder with zero to five messy customers, which is exactly who types this query. If you have a base, mine it. Segment the customers, read the churn, read your competitors' one-star reviews. If you don't, you're allowed to start anyway.
At zero customers the correct output is a hypothesis, a V0, and you build it from conversations. Five people who fit your best guess. Twenty minutes each. No selling on the first call, you're there for the trigger, the workaround they're running today, and the words they use for the pain. Then write the profile in their vocabulary, because their words become your homepage and your subject lines.
"I can't even find five people" has a concrete fix. Pick an adjacent paid behavior as the qualifying signal. Someone who already pays for ChatGPT has proven they'll pay for software that saves them time. Someone who buys on AppSumo has proven they try new tools.
Recruit the interviews from that pool, and the unreachable ICP becomes a filter you can act on today.
Use the AI where it's strong. Feed it real context and let it mine pain language on Reddit, G2, and Capterra, asking for pain-point patterns rather than tidy persona cards. What it can't do is stand in for the five conversations. A profile built from prompts alone is confident, polished, and untested, and the corpus of founders who tried is not encouraging.
How-to guide
The V0 ICP, built from five conversations
The pre-revenue version. No closed-won deals required, and the whole thing fits inside one week.
Write the best guess
One sentence, a situation and a job. "Series A teams that just hired their first sales lead and need pipeline reporting off the founder's plate." Wrong is fine. Vague is the enemy.
Recruit five people who fit it
If your ICP feels unreachable, pick an adjacent paid behavior as the qualifying signal, someone who already pays for ChatGPT, someone who buys on AppSumo, and recruit from that pool.
Run twenty minutes each, no selling
You're there for three things. The trigger that would start a search, the workaround they run today, and the words they use for the pain.
Write the profile in their vocabulary
Trigger, job, who hurts and who signs, what done looks like. Their phrasing becomes your homepage and your subject lines, so don't translate it into marketing.
Pressure-test it
Score it against five buyers you'd love and five you'd dread. If it can't sort those ten cleanly, narrow once and test again.
ICP, persona, buyer. Which is which.
The vocabulary confusion keeps people stuck, so here is the clean mapping. The ICP is the account and the moment, the company standing in the buying situation. The persona is the human inside it you write copy for. The buyer signs the contract, the user feels the pain, and in B2B they are usually different people and different documents. A team selling developer tooling writes one page for the engineer who hurts and another for the VP who pays.
Situation, then the human, then who pays versus who hurts.
All three exist for the same reason, so the profile can make decisions. If a document can't change what you say or who you skip, it doesn't matter which name it carries.
%20copy%20Small.jpeg)
Pressure-test your ICP with someone who does this weekly
One membership, unlimited 1:1 calls. Book a positioning operator, bring your draft profile and your five candidates, and leave with a filter that can say no.
Your ICP is done when it can say no
Run the finish-line test. Can the profile make you turn down a plausible-looking lead this Monday, and can a stranger repeat who you're for in one line? Those two are the same muscle. A filter sharp enough to exclude is a sentence sharp enough to repeat.
The move this week stays small. Five conversations, twenty minutes each, one of them today. Write the profile from their words. Score it against your last five wins and five losses. Set a date to re-check it, quarterly is plenty, because profiles rot as markets move. Chase the filter and the date, and let certainty arrive later.
And if you've been circling this for a year, the fastest exit is borrowed pattern recognition. A positioning operator who runs ICP definition every week will hear your five candidates and rank them inside one call, the same way they can tell a wrong-audience problem from a wrong-page one on a live ad account. You bring the drafts. They've met a hundred previous versions of you. Setting it up takes about as long as writing one more candidate list. On GrowthMentor that operator is one membership away, unlimited 1:1 calls included, and booking one runs three screens with no checkout at the end. Here is the whole flow.
My own profile took years to say out loud, the founder with nobody to talk to. It never once mentioned company size.
The mentors below run ICP definition and discovery for a living, on the live roster now.
Suggested mentors
Operators who run ICP definition and discovery every week:
%20copy%20Small.jpeg){kind=small}
Defining your ICP FAQ
Vetted mentors, every one included
Stop describing a company.
Name the moment they buy.
Browse vetted positioning and GTM operators, book a 1:1 call on-demand, and bring your draft profile. Membership is unlimited calls, every mentor included.
Talk to a mentorKeep reading
More from the GrowthMentor blog
Marketing · May 29, 2026
Why Your Ads Get Clicks but No Conversions
Foti Panagiotakopoulos
Marketing · Jun 05, 2026
How to Explain What Your Startup Does in One Sentence
Foti Panagiotakopoulos
Positioning · May 31, 2026
You already picked your niche. Stop reopening the decision.
Foti Panagiotakopoulos
Marketing · Jun 14, 2026
How to Find Your First Customers When You're the Only Marketer
Foti Panagiotakopoulos
