An ideal customer profile defines the accounts and buyers you are for, scored on real fit. Jobs-to-be-done defines the progress those buyers are trying to make in the moment they reach for a solution. Done together, an ideal customer profile B2B teams can defend reads as one sentence per segment: this account adopts this service to make this job progress, and we capture value through a named mechanism. A segment that can't finish that sentence belongs somewhere else: an audience you route, or an orphan you cut. Most teams write the ICP as a firmographic wish list and stop, which produces a fit map, not a revenue plan, and the gap between the two is where the budget leaks.
Is the ICP really your constraint, or is it positioning?
The constraint is rarely "we don't have an ICP." Almost everyone has a slide. The constraint is that the ICP isn't evidenced, isn't scored against real wins and losses, and isn't ranked by money. Three failures usually travel together: the ICP is deduced from firmographics instead of grounded in what switchers said, there's no negative ICP so the funnel keeps feeding high-fit-looking accounts that never expand, and segments get ranked by how much the team likes them rather than by expected revenue.
Before you accept ICP as your bottleneck, confirm the failure sits upstream of positioning rather than being a positioning problem in disguise.
| You see this | Probably ICP/JTBD | Probably something else |
|---|---|---|
| Win rate is fine but net revenue retention sags | Winning accounts that can't get value | Not positioning |
| Demos land, then the deal goes quiet | The buyer isn't the value-getter | Often positioning |
| Cost to acquire is healthy, lifetime value by segment is not | Scaling a segment under 2:1 | Check pricing |
| Sales, marketing, and product each name a different best customer | The classic tell | Rarely anything else |
The cleanest single test: pull your last forty closed-won accounts and ask whether more than a third sit outside the cluster your ICP slide describes. If they do, the ICP moved under you and nobody updated the document. That's a binding ICP constraint. If your wins cluster tightly but buyers still can't tell you apart from a competitor, that's positioning, a different play.
What does the fix look like?
A dated, confidence-tagged pack: a scoreable ICP (firmographic plus behavioral plus pain-trigger), one to three evidenced job statements per segment, a few personas, an explicit negative ICP, and a segment-to-revenue ledger that ranks on money. You write it once and everything reads from it, so getting it wrong is expensive: positioning, content, and pricing all inherit the error and run against it at full speed.
Two declarations sit on the cover, because everything downstream re-keys off them. The first is the data tier. Tier 0 is public-only: you infer from the live site, case studies, and job postings, and cap behavioral and money claims at indicative. Some fields have no public signal at all, like the activation threshold and retention by segment. Stamp those "blocked, needs client data" and carry a labeled range, never a point estimate. Tier 1 is the engaged client with closed-won and billing data, where you re-derive everything at full confidence.
The second is the buying motion. A committee that signs off before anyone sees value is account-mode, scored on firmographics and defended by retention. One person checking out on their own card is individual-mode, scored on behavior and defended by re-buy. That choice re-keys the personas and the negative ICP too.
How much evidence do you actually need?
Evidence is the constraint inside the constraint. The two best methods are switching interviews (people who recently moved to you, away from you, or off doing nothing) and win/loss interviews. Run eight to fifteen per segment and stop at saturation: two interviews in a row that surface no new push, pull, anxiety, or habit. Anchor on a switch in the last thirty to ninety days, because memory degrades fast. Over-recruit losses, because rejected buyers have little reason to talk yet carry the highest-signal anxiety, and let a neutral party run the call, not the rep who worked the deal. This is where Sensing and Intelligence pays off: sales-call mining feeds the job statements directly.
When live interviews are impossible, you descend a ladder and the ceiling drops with you. Each tier earns a hard cap on what it can claim.
- Live switch or win/loss interviews: high confidence, can anchor a primary ICP.
- Mining recorded calls: medium-high, biased toward what reps asked.
- Coding 50 to 100 reviews: medium, supports a secondary persona only.
- Support tickets, search, community, or proxy verbatim: low, seeds an experiment backlog but never defends a price.
Build the scorecard as a 0 to 100 score, not a binary yes/no, across three layers: firmographic (a pass/fail gate, then a score), behavioral (what accounts run and do, including hiring signals, which forecast intent better than current tooling), and pain-trigger (the dated "why now"). Down-weight any attribute that shows up just as often in churned accounts, because it has no discriminating power.
Write the jobs from switcher language, in job-story form: "When [situation], I want to [motivation], so I can [outcome]." The functional job is rarely why people switch. The canonical tell: the stated job was "manage projects efficiently," the real one was "look in control when my manager asks for a status update." Name the fired alternative for each job, because that, not a competitor, is usually what you're beating, and it's often a spreadsheet or doing nothing. Then validate the pack's messaging through Customer Intelligence and Synthetic Testing.
What do the numbers do to the ranking?
Take a software company selling cross-warehouse inventory-sync automation at 18 to 30K per deal to ops leaders at mid-market retailers. Account-mode, twelve switch interviews on recent wins and losses. The evidenced job: "When I can't trust stock counts across warehouses mid-rollout, I want a provably correct count, so I can stop overtime recounts and survive quarterly audits." The fired alternative is a spreadsheet and a temp staffer. The champion is the ops lead; finance validates, anxious about audit risk.
Size three candidate segments by expected revenue (accounts times win rate times deal size), then clamp to capacity.
| Segment | Opportunities | Win rate | Deal size | Expected revenue |
|---|---|---|---|---|
| SMB | 200 | 30% | 5K | ~300K |
| Mid-market | 50 | 24% | 30K | ~360K |
| Enterprise | 15 | 15% | 150K | ~337K |
Three segments that look wildly different on win rate and volume land within spitting distance on revenue. This is the win-rate paradox: win rate falls as deal size rises, so fit and count alone never rank segments correctly. Now clamp by capacity, where required coverage runs at roughly 1 divided by win rate. Rank on revenue per unit of sales capacity, and enterprise reaches a similar number with 93 percent fewer accounts to manage. That makes it the beachhead and the first money page, even though SMB has the prettier win rate.
The negative ICP falls out of the same work. Single-warehouse accounts asking for the enterprise product are a hard disqualifier: auto-route them to self-serve, no sales touch. Retailers wanting a feature the company doesn't sell are routed-adjacency: send them to an explainer with a re-qualification trigger written as a data event ("when they add a second warehouse"), not a calendar date. Matching the motion to deal size is its own decision; see GTM Motion.
What quietly poisons the pack?
- Skipping the negative ICP. Win-rate lift hides in subtraction. A working anti-profile has a signature: volume down, win rate up. Pre-sell that volume drop or the org rejects the discipline.
- Blending the numbers. A 105 percent blended retention figure can hide a 90 percent SMB cohort and a 125 percent enterprise cohort that need opposite motions. Segment before you average, then stop slicing once cells fall into single digits.
- Writing jobs at the wrong altitude. "Boil water" is too low, bound to one solution. "Be a good host" is too high, with no circumstance. The job lives where progress meets a circumstance.
- Treating it as a one-time deliverable. The ICP is a living spec. Rebuild it on a trigger: a new or retired service, repeated wins outside the ICP, or a segment crossing below the value-to-cost guardrail.
How does the fix show up in revenue?
Three numbers move. Win rate stops hiding inside a misleading average, because you score fit and readiness as separate axes and can see which segment converts. Net revenue retention climbs, because the accounts you acquire are built to expand. And sales capacity stops getting spent on accounts that cost more to win than they're worth, because the negative ICP suppresses them before a rep ever touches them.
The reason is that bad-fit wins distort the exact metrics you use to justify your price, so cutting them lets the metrics tell the truth again. One fintech reportedly cut leads by roughly 40 percent while lifting win rate by about 22 percent, a directional anecdote rather than a law, but the shape repeats. To test whether ICP is your first move, run the Marketing OS diagnostic.
FAQ
What is the difference between an ICP and a buyer persona?
The ICP is the account or buyer you're for, scored on fit. The persona is the person-level detail inside it: the role in the decision, the job they own, and the anxiety that could kill the deal. In B2B you build the ICP first, then derive two to four personas from the buying group reconstructed from won and lost deals, not from the org chart. Two people with the same title can play opposite roles in a purchase.
Can I build an ideal customer profile B2B teams trust without CRM data?
Partly. A public-only pack can carry high confidence on observable firmographics, but behavioral and money fields cap at indicative, and some have no public signal at all: the activation threshold, win rate by tier, and retention by segment. Stamp those as blocked and carry a labeled range. Label the whole pack "externally inferred, validate against client CRM," then promote it to full confidence once closed-won data exists.
What is a negative ICP and why does it matter?
The negative ICP is the explicit set of accounts or behaviors you refuse, written as executable CRM rules ("industry = staffing, auto-nurture") rather than adjectives. It matters because win-rate lift hides in subtraction: the disqualifying signal lives in churn, refunds, and lost deals, a different dataset from your winners. Split hard anti-profile (route to nothing) from routed-adjacency (real buyer, wrong timing, carries a named route), and the conversion gains come from that second bucket.