The Account Scoring Model Nobody Ships: A Weighted Template You Can Use

Search "account scoring model" and you'll land on five or six posts that all give you the same two axes: fit and signal. Almost none of them give you a number. And none of them show the arithmetic: they explain the theory, gesture at a spreadsheet, and stop right before the part where you'd need it.

Here's a full weighted model instead: the dimensions, sample weights, a decay rule, and three fictional accounts scored end to end so you can see how the arithmetic resolves to a tier. The spreadsheet structure that operationalizes all of it is documented tab by tab further down, so you can build it yourself in whatever tool you already use.

Why most account scoring advice stops at the framework

Demandbase's account scoring guide hands you illustrative point values, fintech as an industry worth 20 points, a headcount over 500 worth 15, a pricing-page visit worth 30, and then never assembles them into one account with a final score. Salesmotion is the one guide in this space that states a decay curve. But its weighting stays a gut call: reasoning like intent is 3x more predictive, so weight it 40 percent, skips the step of testing whether 40 is right. And Demand Metric built a real downloadable spreadsheet, then put it behind a membership signup wall, so the weights and formulas never see daylight.

None of that is malicious. Account scoring is hard to make concrete without picking numbers someone can argue with. But a framework nobody can run stays a slide deck, not a model. This is account scoring, not lead scoring, and if you're after the individual-level version of this problem, the sibling piece on building a lead scoring model in Clay covers that ground directly.

Account scoring is not lead scoring

Lead scoring ranks a person: their title, their engagement, whether they opened the last three emails. Account scoring ranks the whole buying unit, the company as a single object with its own fit and its own signal, independent of which individual happens to be replying this week.

The distinction matters because B2B deals close on committee consensus, not one inbox. A team that only runs lead scoring will watch five stakeholders at the same account go quiet and read that as five cold leads. But score at the account level and that pattern becomes visible: five quiet contacts with rising engagement in aggregate is a very different account than five quiet contacts with none.

None of this means dropping lead scoring. Run both, at different layers, feeding different decisions. Lead scoring tells you who to call first inside an account. Account scoring tells you which accounts deserve that attention at all.

The two-axis model: fit and signal

Every dimension in this model belongs to one of two axes, never both.

Fit measures whether an account could ever be a good customer, and it's structural and slow-moving: revenue band, tech stack, org shape, none of which shift because someone visited your pricing page yesterday. But signal measures whether this account is in-market right now. It's behavioral and fast-moving, and unlike fit, it decays. A signal from six months ago is not the same signal as one from six days ago, and a model that treats them the same is lying to whoever reads the score.

Keep the two axes separate all the way through the model. And blending them, the way a few of the "free scorecard" templates out there do, quietly turns account fit into deal-readiness scoring without ever renaming it.

Fit dimensions: the structural axis

Three dimensions make up fit. Each gets scored once per account and refreshed quarterly, not daily, because none of them move fast.

Firmographic fit

Revenue band, headcount, growth stage, geography if it matters to your delivery model. This is the dimension most teams already have some version of, usually as a hard filter (in ICP or not) rather than a graded score. But grading it 1 to 5 instead of pass or fail matters because plenty of accounts sit close to your ICP band without sitting inside it, and those near misses deserve a different tier than a company three segments away.

Technographic fit

Stack compatibility and build-versus-buy posture. Does this account already run tools your product needs to sit next to, or would you be the first piece of infrastructure they've had to integrate in that category? So a company already running adjacent tooling scores higher here than one that would need to build the surrounding process from nothing.

Structural fit

Buying committee shape and org complexity. A flat, ten-person company and a matrixed enterprise with six functions touching the decision are not the same fit, even at identical revenue. So committee shape is itself a scorable attribute, and it's the one most models skip entirely. The buying committee roles breakdown and the stakeholder checklist for multithreading enterprise deals are both worth reading before you set your structural fit rubric, since that's what you're grading against.

Signal dimensions: the dynamic axis

Three more dimensions cover signal, and every one of them is only useful if it's timestamped. An intent spike from March means nothing in August. So this is the axis that separates static account tiering from anything you'd call signal-based outbound.

Intent signals

Third-party research activity: topic surges, comparison-page visits on review sites, the kind of behavior an account exhibits before they've told anyone they're shopping. Sourcing this well is its own problem, and intent data providers compared is the deeper read if you haven't picked a vendor yet. A buying signal is only worth scoring once you can date it, which is the whole argument for the decay section below. And intent data specifically comes with its own freshness and coverage tradeoffs worth understanding before you weight it heavily.

Behavioral signals

First-party engagement, but across the whole committee, not one contact. Three people at an account opening your last email is a different signal than one person opening it three times. Most CRMs already have this data sitting in activity logs. So the work is aggregating it to the account level instead of leaving it scattered across individual contact records.

Trigger events

Job changes, funding rounds, hiring surges in a relevant function. These are the loudest signals and also the shortest-lived. A new VP of Revenue Ops is a real trigger for as long as they're still setting up their stack, which is typically a matter of weeks, not quarters. So the job change and funding signals playbook goes deeper on sourcing and timing these specifically.

Weighting the model without faking precision

Most of these guides skip weighting, and weighting is the part that decides whether the model works. A split like "40 points firmographic, 30 behavioral, 30 intent" looks precise. But it isn't. Numbers like that get chosen because they sum to something clean, not because anyone tested whether firmographic fit separates winners from losers at 40 percent and not 25.

I don't hand anyone a fixed weight split anymore, and I'd be skeptical of any guide, including this one, that tells you its weights are the right ones for your business. Start even: divide 100 percent across your six dimensions, roughly 16 or 17 percent each. Then run the model, as-is, against the 15 to 20 closed-won and closed-lost accounts you already have sitting in your CRM. And look at which dimension moved with the outcome. If firmographic fit was nearly identical across your wins and your losses but intent signal wasn't, that's your evidence to shift weight toward intent. Not a benchmark from a blog post. Your own outcomes.

Here's the weight set used for the worked example below, arrived at the same way, for a hypothetical mid-market SaaS motion where intent turned out to separate winners from losers more than firmographics did:

Fit's three dimensions sum to 40 percent, signal's three sum to 60. But that split reflects one hypothetical business's calibration, not a rule. A company selling a six-figure platform into enterprise procurement would probably land somewhere closer to the reverse, since fit disqualifies far more of that pipeline than any signal ever will.

Score decay: why a static score lies to you

A fit score barely moves month to month. But a signal score without a decay rule is actively misleading, because it makes a company that visited your pricing page six months ago look identical to one that visited yesterday, and those are not remotely the same account.

The fix is a decay multiplier applied to every signal before it feeds the composite score. Full weight inside a freshness window, a stepped-down weight as the signal ages, zero past a cutoff. And the windows should be inputs you set, not a fixed number borrowed from someone else's sales cycle. A company with a 14-day sales cycle and a company with a 9-month enterprise cycle should not be running the same freshness window, and any template that hardcodes one is quietly built for a business that isn't yours.

Here's an illustrative decay curve, the kind you'd set as input cells in the spreadsheet rather than bake in as constants:

And if you're also running intent scoring at the buying-signal level rather than just the account level, scoring buying intent without black-box AI covers freshness from that angle. The two models should use compatible windows so a signal doesn't read as fresh in one system and stale in the other.

Worked example: three fictional accounts, scored end to end

These are three fictional accounts, invented for this walkthrough, run through the model above using the sample weights and decay logic from the two sections before this one.

Northfield Analytics (fictional) is a fast-growing mid-market SaaS company sitting squarely in ICP. Firmographic subscore 5, technographic 4, structural 3, since the buying committee is still small and forming. On signal: intent 5 off a fresh comparison-page visit, behavioral 4 with three stakeholders engaging, trigger 3 off a recent funding round. Run the fit dimensions through their weights and rescale to 0 to 100, and Northfield's fit composite lands at 80. Run signal the same way and it lands at 83. So the composite score, at 40 percent fit and 60 percent signal, comes out to 82.

Vantage Industrial Group (fictional) is a large enterprise with a stalled buying committee. Firmographic subscore 5, the revenue band is right, but technographic drops to 3 and structural to 2, since there's no identified executive sponsor and the committee is fragmented across three departments that don't talk to each other. Signal is where this account really falls apart: intent 2, mostly decayed research activity from months back, behavioral 2 with a single contact opening two emails and nobody else engaging, trigger 1, nothing recent. Fit composite comes out to 68, but signal composite drops to 35, so the composite score lands at 48.

Petal & Co. (fictional) is a poor-fit small account riding a hot intent spike. Firmographic subscore 1, well under the revenue band this model is built for. Technographic 2, structural 2, a single-threaded account with no real committee to speak of. But signal tells a different story: intent 5 off a fresh pricing-page visit inside the freshness window, behavioral 4 from one very engaged contact, trigger 2. Fit composite lands at 33, signal composite at 78, for a composite score of 60.

Northfield is the clean case: good fit, hot signal, an obvious A. Vantage is the harder read, strong fit but dead signal, a company worth staying in touch with but not worth an AE's time this week. Petal is the one that should actually make you nervous. A raw composite score of 60 puts it in the same tier as an account with real fit behind it, purely because signal is weighted heavily enough to drag a bad-fit account up with it. The fix is a fit floor: a rule stating that no account below a minimum fit composite, regardless of signal, qualifies for anything above a "monitor" tier. A hot intent spike from a company that will never buy is still zero pipeline, and a composite score alone won't catch that.

The weighted account scoring spreadsheet: what is inside

Last reviewed: August 2026. The structure below is a starting point to calibrate against your own data, not a plug-and-play formula. And the decay defaults in particular are worth revisiting as your sales cycle changes.

Build it as a four-tab workbook, no macros, so it opens the same in Google Sheets or Excel.

Tab 1, Weights and thresholds

The only tab most users need to touch after setup. Four columns: dimension name, axis, weight percent, notes. One row per dimension, six rows total, plus a validation cell that shows the running total of all weights and flags in plain language if it doesn't sum to 100 percent. So every other tab pulls its weight values from here: a mid-quarter recalibration means editing six cells, not rebuilding formulas across the workbook.

Tab 2, Fit scoring

One row per account. Columns for account name, and a 1 to 5 subscore for each fit dimension, with dropdown validation so scorers pick from a defined scale instead of typing free numbers, which is what keeps the model consistent across however many people are scoring accounts. The scoring key for that dropdown looks like this:

A fit composite column does the weighted-and-rescaled math automatically once the subscores and Tab 1 weights are both filled in.

Tab 3, Signal scoring and decay

This tab expects multiple rows per account, one row per logged signal, since accounts generate signals continuously and fit doesn't. Columns: account name, signal type, signal date, raw signal weight pulled from Tab 1, days since signal calculated automatically against today's date, a decay multiplier pulled from a lookup table further down the tab, and the resulting decayed signal score. The three age-band cutoffs in that lookup table are input cells you set once, based on your own sales cycle, not hardcoded constants. And a small rollup at the bottom sums each account's decayed signal scores into one signal composite per account, which feeds the final tab.

Tab 4, Account register and output

This is the tab that gets used day to day, filtered and sorted and shared with sales, while everything above it stays scoring machinery nobody but the model owner needs to open. Columns: account name, fit composite pulled from Tab 2, signal composite pulled from Tab 3, composite score, tier, a recommended action per tier, and a last-updated date. Tier is a simple threshold formula: composite at or above the A-tier cutoff gets an A, the next band gets a B, everything else gets a C, with all three thresholds as input cells rather than numbers baked into the formula.

Rolling it out without breaking sales trust

The fastest way to lose a sales team's trust in a scoring model is to hand them a finished tier list and tell them to work it.

So pilot first.

Most of the disagreements in that pilot conversation won't be about the math. They'll be about a dimension the model didn't capture at all, which is useful information, not a failure of the model. This scoring layer usually ends up wired into whatever CRM or GTM tooling a team already runs, as one piece of a broader engineered pipeline rather than a standalone spreadsheet someone updates by hand forever. So getting the weights right before that wiring happens saves you from automating a bad rubric.

Where account scoring fits in the bigger signal stack

Account scoring is the account-level half of a pair, sitting alongside lead scoring and intent scoring on the individual side. On its own it answers one question: which accounts deserve attention. But it doesn't replace the work of scoring the people inside those accounts. It's a prerequisite, not a substitute, for ABM with GTM engineering built on top of it.

A scored account list is what feeds real ABM tiers, and a scored, tiered list is what a target account list is once you strip away the term. Skip the scoring step and a target account list is just a list someone made up in a spreadsheet meeting.