Who Is Your Ideal B2B Customer: Building ICP From Conversations

ICP is often drawn in a workshop: “mid-market, 50–200 people, wants growth.” Marketing targets that portrait; sales complains about junk in the funnel. The reason is simple: the portrait came from wishful thinking and CRM demographics — not from talks with people who already bought, or already said no.

A real ICP is audible on calls: how they name pain, how they talk about budget, who actually decides, what trigger makes it “now,” and which phrases almost never end in a closed deal. Below — how to build ICP from conversation signals, validate it on a closed-won cohort, and how SENSO aggregates language patterns without “five wins from memory.”

Why marketing personas and CRM demographics often lie to sales

• The persona describes demographics (“director, IT, Moscow”), not readiness to pay or how they buy.
• CRM holds industry and company size — useful, but weak predictors if the call has no pain, budget frame, or decision process.
• The slide has no anti-profile: whom not to drag into a demo.
• Pain language is invented by a copywriter, not taken from transcripts.
• Script qualification does not match how customers actually describe the job.

Result: leads “look like ICP,” but the conversation is another world. Salesforce State of Sales 2026 materials stress data hygiene and a full picture of the deal: AI agents and coaching are useless if the system only has demographics without conversation context (vendor research — not an industry benchmark). The same applies to your ICP: without call language, the portrait is a polished hallucination.

Start with a cohort, not one lucky call

One vivid win does not make an ICP. Build a sample:

1. Closed-won from the last 60–90 days (or the last full quarter) — dozens of deals if volume allows.
2. Separately — closed-lost / disqualified for the same window.
3. Listen and tag repeating phrasing, not “the best case the sales lead remembers.”
4. Only then cross-check CRM fields: where customer language matches revenue, that is where ICP lives.

If a phrase appears in one won call and nowhere else — it is an anecdote, not a qualification criterion.

ICP signals that live in calls

1. Pain in their words

Not “they want automation,” but concrete lines:

• “we lose leads overnight, nobody calls back”;
• “the sales lead can’t listen to calls”;
• “new hires learn for a month on live customers”;
• “‘too expensive’ kills half of demos.”

If you cannot quote the pain, it is still a hypothesis — not ICP.

2. Budget frame

Ideal customers do not always name a number on call one, but they give a frame:

• compare to rep payroll or the current tool’s cost;
• ask about rollout, pilot, payback;
• name a range or “we’ll pay if it closes X.”

No budget frame with long “interest” is a caution signal, not a verdict. But the pattern “we never talk money” almost never shows up on won deals.

3. Decision process and buying committee

Listen to process, not only the CRM title:

• “I’ll align with finance”;
• “we need a 2-week pilot, then a board”;
• “four people decide: sales, IT, security, procurement”;
• “I can sign up to N, then it’s a tender.”

ICP includes how they buy, or you optimize demos for someone who cannot move the deal. Gong research (correlations on their customer base — not an industry average) links multi-threading and more buyer contacts to higher win rates on larger deals — e.g. on the order of +130% win rate with multi-thread on deals >$50K, and roughly 2× buyer contacts on successful deals. For ICP that means: “one champion, no committee” is often not your ideal profile if you sell complex B2B. Science of Scaling (leader survey, Jan 2026) separately highlights weak next steps, skipped qualification, and one-threaded deals as common discovery failures — mutual fit and decision dynamics, not only “the right questions.”

4. Buying trigger

Why now, not “someday”:

• new sales lead / founder demands control;
• team growth and onboarding collapse;
• a missed quarter on stage conversion;
• current vendor exit or license limits.

Without a trigger, an “interesting lead” is often a research call. Public aggregators such as Gartner (B2B Buying Survey / buyer-enablement materials often cite ~17% of purchase time with all suppliers combined — verify the current publication) remind us: buyers already spend little time with vendors. Your ICP must confirm problem and process quickly — or you burn a scarce window on anti-fit.

5. Anti-signals (not our customer)

ICP is useless without a disqualification list. Typical lines:

• “we just need call recording, no review”;
• “zero budget until year-end, just researching”;
• “procurement-only tender, no pilot”;
• “we want AI to sell with no people” while no process exists.

Lock anti-phrases as hard as pain points. Otherwise qualification stays polite and leaky.

How SENSO turns language into ICP

Instead of five “wins from memory,” look at patterns across the base:

1. Filter won deals from the last 60–90 days.
2. Extract repeating pain, budget, DM/committee, and trigger phrases.
3. Review lost deals separately: which lines appear before a no.
4. Cross-check CRM fields (industry, size, source) — where customer language matches revenue.
5. Compile top quotes into an “ICP dictionary” for marketing and the script.

SENSO is built for this aggregation: consistent labeling, repeat detection, link to deal outcome. You get testable traits, not a slide — and no invented “+X% conversion” from thin air.

Mini ICP template from conversations

• Who: role + size of sales/support team + industry (only if proven by revenue).
• Pain: 2–3 verbatim quotes from closed-won calls.
• Proof of readiness: what they tried / what they spend / which trigger.
• How they buy: pilot, timeline, who signs, who blocks, how many threads in the committee.
• Not our customer: anti-phrases and anti-scenarios.
• Qualifying questions: 3–5 questions that cut anti-fit in the first minutes.

What to do after you draft ICP

1. Update qualification in the script and SENSO checklist for this ICP.
2. Give marketing the pain dictionary and anti-profile — so creatives speak the same language.
3. Stop pushing anti-fit profiles for CRM activity.
4. If PULSO handles leads — put the same qualifying questions and escalation rules into the agent scenario from minute one: the agent should check pain, budget frame, process, and trigger with the same criteria as a human.
5. Rebuild ICP every quarter from new wins/losses — markets and wording drift. An old pain dictionary without a rebuild goes stale fast.

Caveats on external stats

Figures from Gong, Science of Scaling, Salesforce State of Sales, Gartner, and similar sources are vendor research, leader surveys, or public aggregators. They are directional correlations from other bases — not a guarantee for your funnel and not NURREL metrics. Use them to interrogate your own closed-won cohort, not to copy someone else’s slide into your ICP.

Bottom line

Your ideal B2B customer is not whom you wish you sold on a slide, and not an “industry + size” CRM row. It is whose language already repeats across a winning cohort: pain, budget frame, clear decision process (often a committee), a live trigger, and no hard anti-signals. SENSO helps aggregate those patterns across the base. Then ICP must live in the script, in qualification, and — when needed — in the PULSO scenario. Quarterly rebuild — non-negotiable.