How to build an ICP scorecard template from closed-won deals

Most ICP scorecards are built in a room that has no data in it. A founder, a head of sales and a marketing lead sit down for a Tuesday afternoon workshop, argue for two hours, and come out with a slide called "Our Ideal Customer" that lists three industries, a headcount band and a couple of pain points. It looks defensible. It is not.

The failure is in the source, not the format. A scorecard built from opinion records who a team hopes will buy. A scorecard built from closed-won deals records who actually does. Those two lists overlap less than most revenue leaders assume, and the gap between them is where quota goes to die.

Here is how to build the second version, without a data team, using a CRM the business already owns.

A scorecard is only worth what the closed-won data says it is

Almost every B2B company has an ICP document somewhere. Far fewer have ever tested that document against their own closed-won list. The test is cheap to run, takes an afternoon, and it is the only thing separating a scorecard from a wish list. Attributes that felt obvious in the workshop routinely turn out to carry no predictive weight at all, and a few turn out to be negative signals.

Workshop scorecards capture what a team hopes about the market. Closed-won scorecards capture what the market has already paid for. That gap is why the rebuilt version usually looks nothing like the one on the wiki.

Scope matters here. Real-time buyer intent scoring is a different job, covered in the buyer intent signals guide. That system asks whether an account is showing behaviour worth acting on this week. An ICP scorecard sits underneath it and asks the slower question: if this account did show intent, would it actually close, and would it stay long enough to renew. Skip the ICP work and the intent scoring amplifies the wrong noise, so reps simply get faster at chasing deals that were always going to be lost.

What actually goes on the scorecard

An ICP scorecard is a short list of attributes with weights attached, scored against every open account, producing a single number that ranks fit. The mechanics are not the interesting part. The interesting part is which attributes make the list and how the weight gets decided.

Start with the closed-won list from the last 18 months. Not 12. Eighteen gives enough deals to see a pattern without pulling in stale ones from an old positioning. Rank them by lifetime value rather than annual contract value. LTV filters out the deals that closed loud and churned quiet, which are the deals most likely to distort a scorecard weighted on revenue alone.

Then list every attribute that can plausibly be known about a customer before the deal opens. Group them into four families:

Weight the attributes in each family by how strongly they correlate with the outcome that actually matters, which is usually renewed revenue two years out. Then hard-cap the total at no more than seven attributes across all four families. The tenth attribute is almost always noise a spreadsheet cannot hear over.

The rubric produces one number per account. Above 70, prioritise. Between 40 and 70, work but not first. Below 40, disqualify or park. Those thresholds should be set against actual win rates, never copied from a template.

Three plays for building the scorecard from real deals

The deliverable is not the scorecard itself. It is what the scorecard changes about next quarter's targeting, routing and outbound. Three plays get there.

The first play is the top-quartile diff. Pull the top 25% of closed-won accounts by LTV, then the bottom 25%. Line up every attribute side by side and mark the ones that differ meaningfully. The attributes that separate the two groups are frequently ones that never appeared in the stated ICP, and technographic readiness is the usual example: whether the account had already adopted a particular class of tooling before the first sales call. An attribute found this way often outranks everything the workshop produced.

The second play is loss-adjacent analysis. Pair the scorecard with closed-lost data. If an attribute scores high on the closed-won list but also appears across a lot of the losses, it is a fit signal masking a timing or capability problem. It stays on the scorecard, but it earns a note beside it: "high fit, needs multi-thread from week one." A scorecard without a loss overlay will confidently point reps at the same graveyard the team just walked out of.

The third play is stack-fit weighting. Technographic fit frequently predicts conversion more reliably than firmographic attributes do. A prospect running three of five ideal stack components is more likely to close than a headcount match running none of them, because the stack reveals an operational maturity that headcount alone does not. Weight it once the lift has been measured, never because the vendor selling the stack-detection tool recommended it.

A team working from a persona doc with no stack-fit attribute in it will usually see the target list change inside the first hour of the rebuild. That is the point.

Join the MeetIQ waitlist →

What to do this week

The first version of this needs no data team. It needs three hours and a CSV export.

1. Pull the export. Every closed-won deal from the last 18 months. Include ACV, cycle length, source, industry, headcount band, primary contact role and any technographic data the CRM captures. If the CRM does not capture stack data yet, add it as a five-minute update on the next handful of deals so the next audit has it.
2. Rank and diff. Sort by LTV. Compare the top quartile against the bottom quartile on every column. Ignore any attribute that does not show at least a 1.5x lift. The goal is signal, not decoration.
3. Score ten open accounts by hand. Take the weighted attribute list, apply it to ten accounts the team is already working, and check whether the ranking matches the reps' instincts. Where the scorecard and the reps disagree, that is the conversation worth having on Friday. One of them is wrong, and the answer is informative either way.

The point of the first pass is not a perfect model. It is to establish that a rules-based scorecard, run on data the business already owns, produces a different account priority than the one the team is running today.

The scorecard is only worth what it changes

A scorecard earns its place by changing what happens next: which accounts get worked, which get parked, and which get routed to the rep who closes that pattern most often. Building it from closed-won data anchors the fit call in something the market has already paid for, rather than something five people in a room agreed on a Tuesday.

Bain & Company's classic loyalty work, Prescription for Cutting Costs, put the value of a marginal retained customer at roughly five to twenty five times the value of a marginal new one. A scorecard is the mechanism that keeps sales cycles pointed at the accounts most likely to renew.