ICP Scoring Model: How to Weight Fit Criteria (With Template)

An ICP scoring model gives each fit criterion a weight, sums to a score out of 100, and sorts accounts into tiers such as A, B and C. Build it by choosing five to eight criteria from your wins and losses, weighting the most predictive highest, and then calibrating the thresholds against past deals.

Why score at all

A profile with ten criteria produces arguments: this account meets six, that one meets seven but misses the one that matters. A score ends the argument by making trade-offs explicit. It also lets you sort a list of 2,000 accounts in one step instead of reviewing each manually.

Scoring has costs. It can look more precise than it is, and a badly calibrated model makes confident mistakes. The fixes are simple: keep the model small, explain every weight, and test it on history.

Step 1: choose criteria that separated wins from losses

Start from the contrast you found when you analysed deals, as described in how to create an ICP. Keep only criteria that were more common in wins than in losses. Typical candidates:

• Employee count band
• Industry (only if it separated wins from losses)
• Uses a specific tool or category of tool
• Region you can support
• A situation or trigger (recent hire, funding, migration)
• Buyer role reachable

Exclude anything you cannot determine for most accounts. A criterion with 70 percent missing data adds randomness.

Step 2: weight to 100

Distribute 100 points across the criteria in proportion to how strongly each predicted winning. A sample rubric for a fictional analytics product:

Criterion	Points	How to score
Employees 20-200	25	Full points in range, half within 25 percent of range, else zero
Runs a CRM	20	Yes 20, unknown 5, no 0
In supported region	15	Yes 15, no 0
Recent sales-leader hire	20	Last 12 months 20, older 0
Reachable buyer title	10	Found 10, not found 0
Industry in core two	10	Core 10, adjacent 4, other 0
Total	100

BusinessMCP’s ICP generator includes a scoring rubric skeleton in its output, with weights that total 100, and its page suggests example thresholds of A at 70 or above and B at 40 or above that you set yourself. Treat those as a starting point rather than a rule.

Key takeaway: the weights are hypotheses. Write one sentence explaining each, so that when results disagree with the model, you know which assumption to revisit.

Step 3: handle disqualifiers separately

Do not subtract points for disqualifiers. A company outside your supported region should not score 60 and sit in tier B; it should be out. Apply disqualifiers first as a hard filter, then score what remains.

Step 4: set thresholds and calibrate

Score your last 20 wins and 20 losses with the draft model and plot the totals.

1. Wins should cluster high, losses low.
2. Choose the A threshold so that most wins land above it and few losses do.
3. Choose the B threshold where the overlap begins.
4. If wins and losses score alike, the model is not predictive and needs different criteria.

With only a few dozen deals the thresholds will be rough. That is fine; label them provisional and review after the next quarter of data.

Step 5: wire it into the process

A score is useful only if it changes behaviour.

Tier	Action
A	Personal outreach within a day, senior rep
B	Sequenced outreach, standard rep
C	Marketing nurture or self-serve path only
Disqualified	No sales time

If you have visitor or product signals, combine them with the fit score so that timing and fit both count. BusinessMCP’s sales triggers guide argues first-party signals such as repeat pricing-page visits rank highest on precision; whether you agree, keep behaviour as a separate dimension from firmographic fit so you can see which one is driving a score.

Getting the data to score

Firmographic criteria need data on each account. For a list of domains, a company lookup can supply industry, size and location. The BusinessMCP companies data API returns those fields for a domain, up to 100 domains per bulk call per its page; always check a sample for accuracy before scoring a big list. You can then feed the scored list into a target account list.

Mistakes to avoid

• Too many criteria. Past eight, the model becomes a black box.
• Equal weights. If everything matters equally, nothing was prioritised.
• Scoring on unknowns as if they were zero. Treat missing data deliberately, for example with a small default.
• Never recalibrating. Check score against outcome every quarter.
• Confusing fit with intent. A perfect-fit company that is not buying is still a long shot.

A simple recalibration routine

Each quarter, export closed deals with their scores. Note how many A accounts closed, how many C accounts closed and whether any criterion’s weight looks wrong. Change one or two weights at a time and record why. After a few cycles you will have a model you trust and a history of why it looks the way it does.

A worked calibration example

Suppose you score 20 wins and 20 losses with the sample rubric above. The wins land between 55 and 95, with a middle of about 78. The losses land between 10 and 70, with a middle of about 35. You set A at 70 and B at 40. Looking at the results, 16 of the 20 wins sit in A or high B and only three losses reach A. That suggests the model separates well enough to use. If instead the losses averaged 65, you would suspect the criteria were shared by both groups and go back to the contrast step.

(The figures in this example are invented to show the method; use your own deal data.)

Fit score and timing score together

Many teams keep two numbers per account: a fit score from firmographic criteria and a timing score from signals such as recent visits, replies or hiring events. A simple grid keeps both visible.

	High timing	Low timing
High fit	Act today	Nurture and watch triggers
Low fit	Check manually, may be a one-off	Ignore

The grid prevents two classic errors: chasing an active but unsuitable lead because it replied, and neglecting a perfect-fit account because it has been quiet.

Explaining the model to sales

Sales teams distrust scores they cannot explain. Share the rubric, show three scored accounts and invite challenges. Collect cases where reps believe the score is wrong, and review them each quarter. If the same disagreement recurs, a weight probably needs to change. A transparent model that reps can argue with usually outperforms a clever one they ignore.