Ideal Customer Profile and Lead Scoring, Without the Fantasy
Key takeaways
- An ICP is derived from your own closed-won data, not chosen in a planning session.
- Include retention and referral, not just closing. Customers who buy quickly and churn are the most expensive kind.
- Lead scoring fails when it rewards engagement without intent — newsletter openers outranking buyers.
- Scoring models decay. Without a scheduled review against actual outcomes, they become superstition.
Deriving an ICP from evidence you already have
The failure mode is familiar: a room agrees the ideal customer is a mid-market company with a progressive attitude to technology and a real budget. That describes an aspiration, and it filters nothing — every prospect can be argued into it.
A useful ICP is derived. Take your closed-won accounts from the last 12–24 months and score each on four outcomes:
- Closed — did they buy?
- Closed fast — sales cycle below your median?
- Retained — still a customer, or completed successfully?
- Referred or expanded — did the relationship grow?
Then look for the attributes shared by accounts that score on all four, and the attributes shared by accounts that closed but churned. The second list is more valuable and almost nobody compiles it.
What belongs in the profile
| Attribute type | Examples | Use |
|---|---|---|
| Firmographic | Size, sector, geography, structure | List building — cheap to filter on |
| Situational | Recent change, tooling in place, team shape | Timing — the strongest predictor |
| Behavioural | How they evaluate, who they involve | Process fit — predicts cycle length |
| Disqualifying | Hard requirements you cannot meet | Speed — the most underused field |
Situational attributes usually beat firmographic ones. "Just lost their bookkeeper" predicts a purchase far better than "50–200 employees", and it is the attribute a scoring model can act on immediately.
Building a scoring model that means something
Score fit and intent separately. Collapsing them into one number is the most common design error, because a poor-fit account that opened five emails outranks a perfect-fit account that has not visited yet — and the rep works the wrong one.
- Fit comes from the ICP attributes above. It is largely static and knowable before any engagement.
- Intent comes from behaviour with a time decay: pricing page views, repeat visits, demo requests, replies. A pricing visit last week means something; one from March does not.
Then set thresholds by action, not by score: which combination triggers a call, which triggers nurture, which triggers nothing. A score with no action attached is a number nobody uses.
Add negative scoring and use it. Student and competitor domains, job applicants, existing customers on a support hunt, and free-tool users with no commercial intent all inflate scores and waste rep time. Most models only add points, which is why so many of them drift upward until every lead looks warm.
Why models quietly stop working
A scoring model is a hypothesis about what predicts revenue, and hypotheses expire. Your product changes, the market shifts, the site changes and a page that meant intent no longer does.
Review quarterly against outcomes, not opinions. The test is simple: of the deals that closed last quarter, what did the model say about them at the time? If high-scoring leads did not close at a higher rate than low-scoring ones, the model is decoration.
The AI Sales Automation Vault includes an ICP generator, target account list building, lead enrichment, buying-signal identification, lead scoring and opportunity qualification among its 50 workflows — fit and intent kept separate by design. For the discovery conversation that follows, see sales discovery questions; for reaching them, cold email and cold calls. The marketing-side equivalent is personas built from evidence in the Marketing Vault, and both are in the bundle.
What is an ideal customer profile?
A description of the accounts that buy, close quickly, stay, and refer — derived from your own closed-won data rather than agreed in a planning session. If every prospect can be argued into it, it is an aspiration rather than a profile, and it will filter nothing.
How do you build a lead scoring model?
Score fit and intent separately. Fit comes from ICP attributes and is largely static; intent comes from time-decayed behaviour like pricing visits and replies. Then attach actions to combinations rather than to a single total — a score with no action attached is a number nobody uses.
Why does our lead scoring not work?
Usually because fit and intent are collapsed into one number, so an engaged poor-fit lead outranks a perfect-fit lead who has not visited yet. The second common cause is having no negative scoring, which lets students, competitors and job applicants accumulate points until every lead looks warm.
How often should a scoring model be reviewed?
Quarterly, against outcomes rather than opinions. Check what the model said at the time about the deals that actually closed. If high-scoring leads did not convert at a higher rate than low-scoring ones, the model is decoration and should be rebuilt from the last two quarters of results.
Sources
- Adamson, Dixon & Toman, HBR — “The End of Solution Sales” — research on buying groups and consensus, behind separating fit from observed intent.
- Christensen et al., HBR — “Know Your Customers' Jobs to Be Done” — why situational triggers outperform firmographic attributes as predictors.
- GDPR.eu — data protection obligations — constraints on collecting and processing behavioural data for scoring in the EU/UK.
- Nielsen Norman Group — analytics and behaviour — why behavioural signals need time decay to remain meaningful.
Keep reading
All articles →
Building vs Buying Technology Solutions
Build vs Buy Technology Solutions: The Decision That Shapes Your Digital Transformation In the middle of growth, most businesses hit a familiar wall: customer e...
May 02, 2026 · 11 min read
Buyer Personas Built From Evidence (Not From a Workshop)
A persona with an invented name, a stock photo and a favourite coffee order changes no decisions. One built from what customers actually said, structured around the job they hired
Sep 04, 2026 · 10 min read
Cold Call Scripts: Winning the First Ten Seconds
Published cold calling benchmarks disagree wildly because they measure different things. What they agree on is that the call is decided in the opening seconds, and that most reps s
Sep 04, 2026 · 10 min read