In a 100-point lead scoring model, company revenue maps to four scores and no more. A company inside your ICP revenue band scores +25. A company one band adjacent, small enough or large enough that it sits off your centre but could still buy, scores +10. A company clearly out of band, one you cannot serve profitably at any price, scores −10. A company whose revenue you do not know scores 0. Not minus five, not "pending". Zero.
That is the whole rubric. The rest of this page is why the numbers are the size they are, how to set the band boundaries from your own closed-won data, and the evidence that the relationship between customer size and acquisition economics does not run in the direction most scoring models assume.
The rubric, and what each score asserts
| Revenue band | Score | What the score asserts |
|---|---|---|
| Inside your ICP band | +25 | The product and the price were built for companies this size. Full firmographic credit. |
| One band adjacent | +10 | Could buy, or could grow into the centre. Real fit, weaker economics. |
| Clearly out of band | −10 | Cannot be served profitably at any price. Both ends of the range live here, for opposite reasons. |
| Unknown or unenriched | 0 | Nothing. The model declines to guess, and the enrichment gap gets fixed instead. |
Four is not an arbitrary stopping point. It is the most granularity that closed-won data at normal B2B deal volumes can actually confirm. A model with nine revenue tiers implies a precision you cannot validate and adds maintenance without adding a single routing decision, because no sales team behaves differently toward tier six than toward tier five.
The values are proportions rather than constants. In a 50-point model, halve all four. The ratios are what carries.
The Revenue Bands to Lead Scores field guideTwenty-two pages: the rubric, the acquisition-cost evidence behind the band shape, a worked 100-point model, the five failure modes, a printable scoring worksheet, and the four widely quoted figures we checked and refused to print. Free, no email required.Download the field guide (PDF, 0.8 MB)
Why revenue is worth 25, and not 50
Revenue band is the strongest pre-intent signal available. A lead can download everything you publish and never buy, because the company behind the address cannot afford the product or does not have the problem at their size. That is why firmographic fit deserves real weight.
But it is a ceiling rather than a floor. Twenty-five points means an in-ICP company still has to do something before it crosses an MQL threshold: visit pricing, book a demo, come back a third time. Give revenue 40 or 50 points and the model routes companies to sales for what they are rather than what they did, and your sales team starts calling well-shaped strangers.
There is also an evidentiary reason for the ceiling, and it is the more honest one. The spread inside any single revenue band is wider than the gap between adjacent bands. In the acquisition data below, companies in the $100K to $250K contract band show a 25th-to-75th percentile range of 20 to 37 months against a median of 24: a 17-month spread within one band, against a median difference of zero between that band and the one beneath it. A signal with that much internal variance is worth tilting a decision. It is not worth making one.
Set the bands before you set the points
The numbers only work if the boundaries are honest, and the boundaries come from what you have actually sold rather than from what you would like to sell.
Pull your last twenty closed-won accounts and record the revenue of each, with cycle length and first-year retention beside it. The band where deals close fast and retain well is your +25. One step outside it on either side is your +10. Everything beyond that is the −10. Twenty is a floor rather than a target: if you have two hundred, use them, and if you have six you do not have a scoring model yet, you have a hypothesis that should be labelled as one.
Note what the procedure does not include: any number from this page, or from any other. Your band boundaries are a property of your price, your delivery cost and your sales motion. Two companies selling adjacent products into the same market routinely have different bands, and both are right.
Acquisition cost by customer size is a curve, not a ramp
Most revenue-band rubrics encode one of two folk assumptions: that bigger customers are better and should score higher, or that enterprise is ruinously slow and expensive and should score lower. The only primary dataset available on the question says both are wrong.
| Average contract value | 25th pct | Median | 75th pct |
|---|---|---|---|
| Under $1K | 5 | 8 | 10 |
| $1K to $5K | 6 | 8 | 12 |
| $5K to $10K | 13 | 14 | 18 |
| $10K to $25K | 15 | 18 | 22 |
| $25K to $50K | 13 | 22 | 26 |
| $50K to $100K | 16 | 24 | 33 |
| $100K to $250K | 20 | 24 | 37 |
| Over $250K | 18 | 18 | 23 |
Median payback rises steadily with contract value, from 8 months at the bottom to 24 months in the $50K to $250K range, and then falls back to 18 months above $250K. The curve is an inverted U. The worst band is the middle, not the top. Benchmarkit's own commentary on the same chart notes that deals above $250K are "materially lower than solutions in the $50K to $100K range and even lower than in the $25K to $50K range", which suggests "an Enterprise solution that requires more time and resources to win may actually be more profitable over time".
The pattern is not an artefact of one metric. Sales-and-marketing cost per dollar of new recurring revenue traces the same shape from a separate sample of 73 companies: it peaks at $2.50 in the $50K to $100K band and drops to $1.59 in the band above it. Benchmarkit reports the anomaly a third time and notes it is not a one-year exception, with solutions in the $10K to $50K range "often more expensive to acquire" than those in the $50K to $100K range.
One caution on transferring any of this directly. Those figures are contract value, not customer revenue, and a $100K contract can come from a $5M company or a $5B one. They are evidence about the shape of the relationship between deal size and acquisition economics, not a lookup table for your bands. The shape is the transferable part: run the check on your own data and expect a curve.
What the curve does to your −10
A monotonic rubric is wrong by construction
If your scores rise in a straight line with company revenue, you are asserting a relationship the data does not show. The band that deserves your highest score is the one where your own payback and retention are best, and that is an empirical question with a non-obvious answer.
"Too big" is the least reliable −10 there is
The reflex to disqualify large accounts, on long cycles and procurement committees and heavy delivery, describes real friction. What the data disputes is the conclusion drawn from it: above $250K those deals paid back faster than the two bands beneath them.
If you are scoring large companies negative, that has to be a finding from your own book rather than a feeling about enterprise sales.
The bottom of the range is where a −10 is usually earned
A company that cannot fund the engagement is disqualified by arithmetic rather than by preference, and no amount of intent behaviour changes it. This is the one direction where the folk assumption and the evidence agree.
A worked 100-point model
Firmographics carry 40 points: revenue band 25, industry or vertical 15. Person-level fit carries 15, for role and seniority. Behaviour carries 45: engagement 25, high-intent actions 20. MQL sits at 60.
| Component | Weight | Example signals |
|---|---|---|
| Revenue band | 25 | The 25 / 10 / −10 / 0 rubric above |
| Industry or vertical | 15 | Core vertical, adjacent, excluded |
| Role and seniority | 15 | Economic buyer, influencer, neither |
| Engagement behaviour | 25 | Return visits, content depth, email engagement |
| High-intent actions | 20 | Pricing page, demo request, audit signup |
Run the arithmetic and the design intent falls out of it. An in-ICP company with a decision-maker on the form sits at 40 before anyone has clicked anything, and everything an account can score without acting totals 55: five points below the threshold, deliberately. If that identity subtotal lands at or above the threshold, the model qualifies companies for existing, and the sales team learns to ignore the score.
The negative does the work an absence of points cannot. A zero leaves an account needing 60 from everything else. A −10 leaves them needing 70 from a board that only holds 75 outside the revenue line, and 45 of that 75 requires sustained behaviour. That is the whole function of the negative score: it makes it arithmetically impossible for an out-of-band company to reach sales on curiosity alone, without hard-blocking them. An out-of-band account that clears 70 behavioural points is the most interesting lead in your database, and it should trigger a review of the band rather than a routing exception.
The mistakes that quietly break the model
Punishing unknowns
Scoring missing revenue as a negative buries every lead your enrichment provider missed. That population is not small and it is not random: it skews toward newer companies and toward the ones without a public revenue estimate, which in many books is the growth segment. Score the unknown at zero and treat the gap as a data problem, which is what it is.
Too many bands
Nine revenue tiers imply a precision that closed-won data at normal deal volumes cannot confirm. If nobody behaves differently toward tier six than tier five, the two tiers are one tier with extra bookkeeping.
Set and forget
Bands drift as pricing and the product move. If accounts you scored −10 keep closing and retaining, the bands are wrong rather than the buyers. Re-validate quarterly, and treat a run of out-of-band wins as data rather than as luck.
Trusting the form
Self-reported revenue is answered optimistically often enough that a serious model treats an enriched figure and a form answer as different fields with different confidence, and prefers the enriched one. If you only have the form answer, that is closer to an unknown than to a fact.
Prove the rubric against revenue, not opinion
A scoring model is a forecast, and forecasts get graded. Once a quarter, pull score-at-MQL against close rate in three buckets: leads that crossed at 60 to 75, leads that crossed above 75, and, the revealing one, everything sales worked that never reached 60 at all.
| Bucket | Result that confirms | Result that falsifies, and the response |
|---|---|---|
| Above 75 | Closes clearly better than the 60 to 75 group | If it does not, the weights are not separating anything. Re-derive them from closed-won rather than adjusting by feel. |
| 60 to 75 | Closes better than sub-60 | If it does not, the threshold is in the wrong place. It is an output of this test, not a setting you pick once. |
| Below 60, worked anyway | Closes materially worse | If it closes the same, the model is not informing routing. Sales has already routed around it. |
Run it on a full quarter of closed outcomes rather than on open pipeline, and hold the MQL definition constant across the window. The third row is the one worth running first: it is the cheapest diagnostic in the list, it needs no change to the model to perform, and a failure there makes every other question moot. This validation loop is one of the checks inside the 47-point funnel scorecard, because lead qualification is where more pipeline quietly dies than in any ad account.
Frequently asked questions
Why 25, 10, −10 and 0 specifically?
The numbers are proportions rather than magic. In a 100-point model, 25 is the most weight a single firmographic attribute should carry: enough to matter, not enough to qualify a company on identity alone. The −10 exists to actively pull disqualified accounts below the threshold rather than merely failing to help them. If your model totals 50 points, scale everything down. The ratios are what you are keeping.
Should unknown revenue score negative?
No. Zero. A negative score for missing data punishes leads for your enrichment provider's coverage gaps, and that population is not random: it skews toward newer companies without a public revenue estimate, which in many databases is the growth segment. Treat unknowns as an enrichment problem to fix, and let the model score what it actually knows.
How many revenue bands do I need?
Four scores: in-ICP, adjacent, out-of-band, and unknown. That is the most granularity you can validate against closed-won data at normal B2B deal volumes. A nine-tier model implies precision you cannot prove and adds maintenance without adding a routing decision, because no sales team behaves differently toward tier six than toward tier five.
Should big companies always score higher than small ones?
No, and this is the most common error in the rubric. Median CAC payback in the Benchmarkit 2025 sample rises from 8 months at the bottom of the range to 24 months in the $50K to $250K contract band, then falls back to 18 months above $250K. Acquisition economics by customer size is an inverted U rather than a ramp, and the worst band is the middle. Score the band where your own payback and retention are best, wherever it happens to sit.
Where does the revenue data come from?
A firmographic enrichment provider appending revenue, or employee count as a proxy, at form-fill or list-load, with a revenue-range dropdown on the form as a fallback. Enriched data beats self-reported nearly every time, because forms get answered optimistically. Measure your own fill rate before trusting any vendor's published match rate; it is a number you already hold, and it is the only one that governs your model.
What MQL threshold should I use?
Start at 60 in a 100-point model, which is an in-ICP company plus a decision-maker plus one real intent action, and then let score-to-close correlation move it. The threshold is an output of validation rather than a setting you pick once. Check that everything an account can score without acting totals just below the threshold, so no company qualifies on identity alone.