An ICP scoring model you can run in a spreadsheet
An ICP scoring model turns your market map into a ranked list of accounts, built in a spreadsheet with four weighted dimensions and real tier cutoffs.
ON THIS PAGE · 8 SECTIONS
- What an ICP scoring model actually does
- Why a spreadsheet beats a platform for the first version
- The four dimensions and how to weight them
- Building the model, step by step
- A worked example
- Keep the score honest: data decay and refresh
- Where scoring models fail and the limits worth knowing
- Questions about ICP scoring
THE SHORT VERSION
- An ICP scoring model is a point system that ranks every account in your market by fit, so reps work the best accounts first instead of the loudest ones.
- Four weighted dimensions cover most B2B markets: firmographic fit, buying triggers, technographic fit and behavioral engagement.
- Build the first version in a spreadsheet against your own closed-won accounts, not a template from a blog post.
- Firmographic fields are modeled estimates, not verified facts, so your model needs a refresh rule or the score rots within a year.
- Recalibrate the weights against new closed-won and closed-lost data every quarter, not after a single deal.
An ICP scoring model is a weighted point system that scores every account in your market against the traits of your best customers, so your team can rank a list of a few thousand accounts and work the top of it first. It replaces the usual method, where a rep picks accounts based on how familiar the logo looks, with a number built from the same inputs for every account. You do not need a platform to run one. A spreadsheet, a market map and 90 minutes with your closed-won list are enough for a working first version.
This only works if you already have a market mapped. If you have not built that list yet, start with how we map a market before the first email goes out or, for a free-source version, how to build a TAM list without buying one. Everything below assumes you have a list of accounts and need to decide which ones get worked first.
What an ICP scoring model actually does
Most B2B teams can describe their ideal customer in a sentence. Mid-market manufacturing companies, 100 to 500 people, using a legacy ERP. The problem shows up once that sentence meets a list of 4,000 accounts. Every rep reads "mid-market manufacturing" a bit differently. Without a number attached the accounts that get called first are the ones a rep recognizes, not the ones most likely to close.
A scoring model fixes that by breaking the sentence into fields: industry, headcount band, revenue band, a buying trigger, a technology signal. Each field gets a weight. Each account gets a score built from those weights. The score sorts the list. The sales motion changes with the score too. A top-tier account gets a trigger-based email and a call within the week. A bottom-tier account gets a slower nurture sequence or gets skipped entirely.
- One number per account that every rep reads the same way
- A sort order for the list, not just a filter
- A record of why an account was worked, which you can check against the deal it became
Why a spreadsheet beats a platform for the first version
Predictive scoring platforms promise a model trained on your data from day one. In practice they need months of closed-won history to train against. Most teams buy one before they know which fields actually predict a close in their own market. A spreadsheet model costs nothing, takes an afternoon to build and forces you to state your assumptions in plain columns instead of a black box.
Run the spreadsheet version for one full sales cycle first. Once you know which fields actually separate winners from the rest, a platform model has something real to learn from instead of guessing at your ICP from a template. If you want the system view on where this step sits, see the outbound system page at ../../#system.
A score is only as good as the ICP behind it. Our founder Hlib Storchak has a guide to building the ICP itself on storchak.eu.
The four dimensions and how to weight them
Most B2B markets score cleanly across four dimensions. The split below works for a mid-market B2B seller with a defined ICP and at least some closed-won history to check it against.
| Dimension | Weight | What it captures |
|---|---|---|
| Firmographic fit | 40 pts | Industry, headcount band, revenue band, region |
| Buying triggers | 25 pts | Funding round, leadership change, hiring surge, tech migration |
| Technographic fit | 20 pts | A stack signal that predicts need for your product |
| Behavioral engagement | 15 pts | Site visits, content opens, reply history |
Adjust the split to match what actually predicts a close in your own closed-won data, not this table.
Firmographic fit carries the most weight because it is the hardest to fake and the most stable over a sales cycle. Buying triggers come second because they tell you when to act, not just who to target, which is the same logic behind the buying signals that actually predict a deal. Behavioral engagement sits last on purpose. A visit to your pricing page means little from an account that fails firmographic fit.
Building the model, step by step
- Pull your closed-won list. Take every deal closed in the last 12 to 18 months and list the firmographic and technographic traits each account shared.
- Pick 4 to 6 fields that actually separate winners. Drop any field where your closed-won and closed-lost accounts look the same. A field that sounds good at an offsite is not the same as one that predicts a close.
- Set the weights so they add to 100. Split firmographic fit across its 2 or 3 sub-fields instead of treating it as one lump score.
- Set tier cutoffs from your own close rate. An A tier (roughly 80 to 100 points), a B tier (55 to 79), a C tier (30 to 54) and a disqualify band below that, adjusted until each tier's historical close rate actually differs from the one below it.
- Score a test batch of 50 accounts. Compare the ranked list against how your best rep would have ranked the same accounts by hand. Large gaps mean a field or weight is off.
- Recalculate every quarter. Feed new closed-won and closed-lost accounts back into the weights so the model tracks your market instead of the quarter you built it in.
A worked example
Example: a 220-person manufacturing software company in the Nordics matches the target industry and headcount band, scoring 34 out of 40 on firmographic fit. It raised a funding round four months ago and posted three senior operations hires last quarter, worth 20 out of 25 on buying triggers. It runs a legacy ERP your product replaces, worth 16 out of 20 on technographic fit. Nobody at the account has opened an email or visited the site yet, so behavioral engagement sits at 2 out of 15.
The total comes to 72, inside the B tier (55 to 79). That account gets a trigger-based email referencing the funding round this week, instead of the slower nurture sequence a C-tier account would get. Six weeks later, once the account has opened two emails and visited the pricing page, behavioral engagement moves it to 83 and an A tier, which is the signal to loop in an account executive for a direct call.
A score that never moves is not a scoring model. It is a static filter wearing a spreadsheet.
THE USUAL WAY
- Reps pick accounts by gut feel and familiarity
- Every rep defines "good fit" differently
- Good accounts and weak accounts get the same number of touches
- Nobody agrees why a deal was lost until the pipeline review
THE SYSTEM WAY
- Every account carries one score built from the same fields
- The list sorts itself, so the best-fit accounts get worked first
- Outreach motion and cadence change with the tier
- The weights get checked against real closed deals every quarter
Keep the score honest: data decay and refresh
A scoring model is only as good as the fields feeding it. Firmographic fields decay faster than most teams assume. Apollo found in 2026 that most B2B data providers average only around 50% accuracy on modeled fields such as revenue and headcount bands, because these are estimates built from job postings and public signals rather than verified facts (Apollo, 2026). Datamagnet's 2026 benchmark report puts overall B2B contact data decay at 22% to 30% a year, driven by hiring, promotions and company restructuring (Datamagnet, 2026).
Source: Apollo, 2026 and Datamagnet, 2026
That decay means a firmographic field scored six months ago may already be wrong. Pair the scoring model with a refresh cadence: re-pull headcount and revenue bands quarterly for A and B tier accounts. Verify any contact-level field before it reaches a rep, the way we cover in catch-all email verification and in the provider cascade behind waterfall enrichment.
FROM THE YARD
The model breaks the first time a rep closes a C-tier account nobody scored high. The instinct is to patch the weights that same afternoon. Do not. Wait for 10 to 15 closed accounts in a tier before you touch the formula. One deal is a story, not a pattern. A model rebuilt around the last good story stops meaning anything within a quarter.
Where scoring models fail and the limits worth knowing
A scoring model is a ranking tool, not a guarantee. It tells you where to spend attention first. It does not replace a rep reading an account's context or a signal your fields were never built to catch, such as a champion who just moved companies, which is its own system covered in champion tracking. Three failure modes show up most often:
- Too many fields. Past 6 fields, the model gets hard to explain to a rep and harder to debug when a tier stops predicting anything.
- Weights copied from a template. A split that works for a different market tells you nothing about yours. Build the first version from your own closed-won list.
- No disqualifiers. A high score with a hard blocker, such as a company too small to ever buy or a competitor already in the stack, still wastes a rep's week. Build a disqualify rule that overrides the score.
Say it plainly: a spreadsheet model will not catch everything a predictive platform catches once you have the volume and history to train one. It will catch the obvious misses that cost most teams the most time, which is most of the value for a fraction of the cost.
Questions about ICP scoring
What is an ICP scoring model?
It is a weighted point system that scores every account in a market against the traits shared by a company's best customers, usually across firmographic fit, buying triggers, technographic fit and behavioral engagement. The score sorts a target list so the highest-fit accounts get worked first.
How many fields should an ICP score use?
Most working models use 4 to 6 fields across 4 dimensions. More fields make the model harder to explain to a rep and harder to debug when a tier stops predicting outcomes, without adding much accuracy past that point.
How often should you recalculate ICP scores?
Recalculate the weights every quarter against new closed-won and closed-lost accounts. Refresh the underlying data, such as headcount and revenue bands, on the same cadence for A and B tier accounts, since those fields decay within a year.
Can you build ICP scoring without a data platform?
Yes. A spreadsheet with a market map and a closed-won list is enough for a first version. Run it for a full sales cycle before buying a predictive scoring platform, so the platform has real data to train on instead of a guess at your ICP.
What is the difference between an ICP score and a lead score?
An ICP score ranks a company by fit: whether the account looks like a good customer at all. A lead score ranks a person's engagement within an account that has already qualified on fit. Score fit first, then layer engagement on top.
Sources: Apollo, 2026, Datamagnet, 2026