Your ICP Is Not 'B2B Companies With Budget'

Ask a founder to describe their ideal customer and you usually get a TAM slide: industry, headcount band, geography. That is a market, not an ICP. An ICP is the narrow slice of that market where you win disproportionately (faster cycles, higher win rates, better retention), and it is defined as much by what it excludes.

Why vague ICPs feel safe and cost the most

A broad ICP feels like keeping options open. In practice it is a tax on every downstream function. Marketing writes messaging generic enough to fit everyone, which moves no one. Sales works accounts that were never going to close, which quietly caps win rates and burns the calendar of your most expensive people. Product roadmaps get pulled in six directions by prospects who were never the business. And forecasting turns to noise, because a pipeline full of poor-fit deals converts unpredictably by construction. The uncomfortable truth is that narrowing does not shrink your revenue; it shrinks your waste. You still take good off-profile deals when they arrive. You just stop spending proactive effort acquiring bad ones, which is where outbound economics are won or lost.

Mine your closed-won

Take your last 30 wins and 30 losses. Look past firmographics to situational markers: what was happening inside the account when they bought? A new leader, a failed incumbent, a compliance deadline, a growth target someone's job depends on? The pattern in those situations is your real ICP.

Running the 30-30 review well

The review works best as a structured half-day with sales, marketing, and success in one room. For each account, answer five questions from evidence, not memory: what triggered the evaluation, who drove it internally, what alternative did they seriously consider, why did we win or lose in the buyer's words, and what happened after go-live. Patterns emerge fast and they are rarely the firmographic ones on the website. You will typically find two or three situational triggers that precede most wins, one buyer role that consistently sponsors, and a loss pattern that clusters around a segment everyone privately suspected. Write the findings as sentences, not tags. 'Companies mid-consolidation after an acquisition, sponsored by a newly hired operations leader' is an ICP. 'SaaS, 200 to 2,000 employees' is a mailing list.

The exclusion test

A working ICP lets your SDR team disqualify quickly and without guilt. If nobody can name the accounts you refuse to pursue, you do not have an ICP. You have a wish. Every hour spent on a poor-fit account is an hour taken from a great-fit one; discipline here is compounding.

Writing the do-not-pursue list

Make exclusion explicit by writing the anti-profile with the same care as the profile. Name the segments where you lose on structure: the deal sizes that cannot fund your sales motion, the industries whose compliance you cannot yet clear, the buyer situations where an incumbent always wins. Give reps a script for declining gracefully, because a respectful no preserves the relationship for the day the fit changes. Then enforce it where behaviour actually changes: in territory planning and comp, so nobody is paid to chase what the strategy excludes. Teams that skip this step end up with an ICP in the deck and a wish in the CRM, and then wonder why SQL volume debates never converge: the arithmetic was being run on two different markets.

Segmentation is not the same thing

Teams often answer the ICP question with a segmentation model, which is a different instrument. Segmentation divides the whole market into tiers for coverage planning. An ICP names the slice where you win on structure, and inside it you can still tier: a primary profile where the situational trigger is live right now, and a secondary profile that fits but has no trigger yet, which belongs in nurture rather than outbound. The practical output is a two-axis score, fit and situation, applied to every account before it enters a sequence. Fit without situation gets content and patience. Situation without fit gets a polite decline. Both together get the full research-led motion. Collapsing those axes into one 'score' is how good accounts get spammed early and bad accounts get worked forever.

When to deliberately break the profile

Discipline does not mean rigidity. Reserve a small, explicit share of effort (ten to fifteen per cent is typical) for accounts outside the current definition: a new vertical you suspect is emerging, a company size the product just grew into, a geography a partner opened. The rules that keep this honest are the same as any experiment. The bet is written down before the outreach starts, it gets a fixed window and a fixed account count, and the result updates the definition in the quarterly review rather than in the heat of a promising first call. What you are protecting against is silent scope creep, where every rep's pet account becomes an 'experiment' retroactively. Deliberate exceptions are how ICPs evolve. Accidental ones are how they dissolve.

Revisit quarterly, not annually

ICPs drift. Products mature, new segments emerge. Put a quarterly 30-minute review on the calendar: what did we win, what did we lose, does the definition still hold? The teams that treat ICP as a living instrument consistently out-convert the ones that laminated it.

ICP as the system's keystone

The reason ICP discipline pays compound interest is that every other GTM asset inherits it. Messaging sharpens because it finally addresses someone specific. Outbound research gets cheaper per account because the accounts resemble each other. Handoffs improve because a shared pipeline definition is easy to agree when both sides agree on who the buyer is. Even pricing conversations simplify, because you are anchoring against a known alternative rather than guessing. That is why, when we run a full-system diagnostic, ICP evidence is one of the first artefacts we build: it is the one input that quietly sets the quality ceiling on everything downstream. Narrow until it hurts a little. The win rate will tell you when you have gone far enough.