How to define your first ICP, uncomfortably narrow, then iterate

The short version. Define your first ICP by choosing the smallest buyer group that shares the same painful event, visible account evidence, buying role, and urgency. Go narrower. If the sentence makes you worry the market is too small, it is finally useful, because founder-led sales needs repeated conversations with similar buyers before any broader go-to-market story deserves trust.

Before you start:

• Your first ICP is a sales instrument, so define it by a painful buying event and visible account evidence.
• Make the segment narrow enough that the same proof asset and discovery questions work across the account list.
• Reject nearby personas early; polite curiosity from adjacent roles muddies the signal.
• Iterate by moving the smallest variable, then keep the rest of the test stable.

Step 1: ask how painful the event is, then define what done looks like?

Begin with the event that makes the buyer admit the problem in plain English. Stay close. A VP Sales who just missed pipeline because rep notes live in scattered calls will behave differently from a founder shopping for "AI insights" after a podcast episode. I want the moment when pain has a calendar date and a boss watching.

The first ICP is the slice where that event repeats across accounts you can actually identify before a call. If the trigger lives only in your imagination, your outbound will turn into therapy for strangers. Done means you can write a buyer sentence like this: "US cybersecurity teams that just hired a RevOps lead after a public forecast miss", and every noun can be found or verified.

• Painful event
• Named buyer
• Visible trigger
• Existing workaround

Step 2: draw a hard border around the buyer, then define what done looks like

Now separate the person who feels the pain from the person who signs the order. Be rude. Early ICP work dies when "operations leader" includes the COO at a logistics firm, the RevOps manager at a SaaS company, the founder doing finance at midnight, and the analyst blamed for cleanup. Those people buy with different fear, language, budget rituals, and peer pressure.

For US selling from abroad, the border matters because Building US trust and presence remotely on a seed budget sits inside the same trust problem, long before any meeting starts. A narrow ICP keeps your LinkedIn comments and event choices aimed at the same room. Done means an account can be rejected in a spreadsheet without a founder debate.

• Role that owns urgency
• Role that signs
• Budget path
• Geography
• Exclusions

Step 3: build a searchable account pool, then define what done looks like

A narrow ICP still needs enough named accounts for learning, so build the pool before you celebrate the sentence. No spreadsheet romance. LinkedIn describes Sales Navigator filters for seniority, function, geography, company size, and recent activity at https://business.linkedin.com/sales-solutions/sales-navigator; those fields turn a fuzzy buyer into a row you can sort.

Tool signals matter when the problem attaches to a stack, because a buyer running Salesforce and Gong has a different daily mess from a team running a spreadsheet and WhatsApp. BuiltWith publishes website technology profiles at https://builtwith.com/, which helps when your wedge depends on Shopify, HubSpot, Snowflake, NetSuite, or another installed system. Done means every account entered the sheet because visible evidence existed before founder instinct showed up.

• Firmographic filter
• Trigger evidence
• Stack signal
• Recent hiring

Step 4: run discovery inside the border, then define what done looks like

Discovery should prove whether the border is real. Keep the lane tight. When every call comes from the same painful event, you stop collecting random feature requests and start hearing whether the same memo, objection, procurement path, and competitor names repeat. That repetition is the seed of a sales motion.

I would rather hear a harsh answer from the exact buyer than polite curiosity from a nearby role with no budget. This is where founder-led sales matters, because the founder can change the product promise while the call is still warm; I wrote the stop line inside Founder-led sales: why you must sell first, and when to stop for the moment when repetition turns into a motion. Done means your notes show repeated phrases you can copy into outreach and homepage copy.

• Trigger quote
• Current workaround
• Buying committee
• Budget owner
• Urgency clock

Step 5: turn the border into messaging, then define what done looks like

Your ICP should change the words on the homepage the same week it changes the outbound list. Make it visible. If the hero line still sounds like a pitch deck after discovery, keep the homepage message beside the call notes while you rewrite.

ICP proof lives in nouns: the job title, workflow, broken system, board pressure, and business consequence the buyer already recognizes. A vague page attracts vague meetings. Done means the top screen names the buyer's job with enough bite that the right visitor feels accused in a useful way and the wrong visitor leaves without a follow-up question.

What goes wrong

Founders usually widen the ICP when early replies feel thin. Bad move. A wider list hides the pattern you need, because each meeting now demands a different problem story, proof asset, sales question, and pricing rationale. The spreadsheet feels busy, yet the founder learns less with every call.

Another failure mode is naming an industry while ignoring the buying event. "Healthcare" is a warehouse label, while "claims operations teams replacing manual denial reviews" is a route to a real conversation. I also see founders borrow India proof and ship it unchanged to US buyers, which is why India PMF transfer belongs inside ICP work before the deck hardens.

The ugliest mistake is treating the ICP as a branding workshop. It is a sales instrument. If the account list, call script, proof page, price story, and owner questions stay the same, the ICP sentence is theater wearing a clean shirt again. Nobody needs more theater during seed-stage selling.

• Wide lists
• Persona labels
• Borrowed proof
• Unchanged pricing

Step 6: iterate from evidence without widening, then define what done looks like

Iteration should start with a score sheet rather than a founder mood swing. Guard the scar tissue. Keep the ICP narrow until the same trigger, objection, proof gap, buying path, and urgency clock repeat clearly enough that a new account feels familiar before the call starts. That familiarity is the signal.

When the adjacent segment pulls harder, change only the smallest variable you can name: role, trigger, stack, region, budget path, or urgency clock. If every variable moves, the test resets. Done means the next ICP version has a narrower sentence, a cleaner exclusion list, a matching account sheet, and a reason written from calls rather than hope.

Common questions

How do I figure out my ICP if I have almost no customers?

Start with the pain event you can observe before the call, then match it to a buyer with budget and a reachable account pool. Use founder conversations to test whether the same words appear across calls. Your first ICP can come from sharp discovery, lost deals, pilot usage, public triggers, and support tickets.

How narrow should my first ICP be?

Narrow enough that the same outreach angle and proof asset feel natural across the account list. A stranger inside the segment should recognize the problem without a vocabulary lesson, and a stranger outside it should feel excluded. If the sentence sounds almost embarrassing in its specificity, you are probably close.

Should I define ICP by industry or persona?

Define it by the painful buying event, then use industry and persona as filters. An industry label alone can sprawl across too many workflows, and a persona without a trigger becomes a directory of job titles. The usable ICP combines visible account evidence, a buyer role, a current workaround, budget path, and urgency.

When should I change my ICP?

Change it when calls and deals show the same adjacent pattern with clearer urgency than the current slice. A loud investor suggestion or a random inbound logo should earn a note before it earns a new market. Move the smallest variable you can name, then keep the rest of the test intact.