Account Scoring: How to Build a Model, With a Worked Example

Checked October 30, 2026. Every vendor fact and price below was read on the vendor's own page or document that day. The worked example uses made-up accounts to show the math. It is not customer data, and we did not run a test for this page.

What is account scoring?

Account scoring is a model that gives each company in your market a score, usually from 0 to 100, for how likely it is to buy and how good a customer it would be. Sales uses the score to decide who to call this week. Marketing uses it to pick accounts for ads and account-based campaigns.

It exists because B2B purchases are made by groups. One enthusiastic contact at a company with no budget looks hot in a lead score. Three quiet people at a perfect-fit company look cold. Scoring the account puts both in context.

Account scoring vs lead scoring

Most teams run both. Score accounts first, then rank the people inside your top-tier accounts by role and engagement. Our AI lead scoring guide covers the person-level side.

The three parts of an account score

Fit: could they buy, and would they be a good customer? Industry, headcount, region, the tools they run and whether the team you sell to exists and is big enough. Fit changes slowly, so it usually carries the most points. It comes from enrichment and people search, and it is only as good as your ideal customer profile.

Intent: are they shopping now? Research surges on third-party sites, review-site comparisons and company events such as a new leader, a funding round or a hiring push. Intent moves fast and should lose its points within weeks. Our guide to buying signals lists 14 of them, with the tools to track each.

Engagement: are they talking to you? Identified website visits, content and webinar activity, email replies and meetings, rolled up from every contact at the account. Count distinct engaged people, because one busy intern clicking everything is not a buying group.

A common starting split is 50 points for fit, 25 for intent and 25 for engagement. Treat it as a first guess and let your won deals correct it.

How to build an account scoring model in 6 steps

1. Pull your won and lost deals. Export 18 to 24 months of opportunities with their account fields. More history gives steadier weights. Salesforce's Einstein Key Accounts Identification, which builds a model from the same kind of data, recommends at least 200 accounts with won opportunities and 200 with lost ones from the last 2 years (Salesforce's Einstein implementation guide, last updated July 22, 2026, checked October 30, 2026). A rules-based model can start with 30 to 50 of each.
2. Find what changes your win rate. For each value of each attribute, such as a headcount band or an industry, divide its win rate by your overall win rate. That ratio is lift. If 51 to 200 employee accounts win 30% of the time and you win 15% overall, that band has a lift of 2.
3. Turn lift into fit points. Give the highest-lift value of each attribute the most points and cap fit at 50. Give negative points to values that lose, and handle hard no's, such as current customers, competitors and regions you cannot serve, as exclusions.
4. Add intent and engagement, with decay. Give each signal points, then halve them after 14 days and drop them after 30, or match your sales cycle. Score engagement by distinct people, plus a bonus for a pricing-page visit.
5. Roll people up to the account. Use the count of engaged people and the strongest single action. Summing every click lets one noisy contact outrank a whole buying group.
6. Set tiers and the action for each. For example: A at 70 or more goes to an AE this week, B at 50 to 69 gets an SDR sequence, C at 30 to 49 stays with marketing, and D below 30 waits. A demo request skips the score and goes straight to a rep.

Then check it every quarter. Tier A should win clearly more often than tier C. If the tiers win at similar rates, the weights are wrong, so run step 2 again. If more than a fifth of your accounts land in tier A, raise the threshold until reps can work them all.

A worked example: five accounts, scored end to end

Say you sell sales software to B2B software and fintech companies with 51 to 500 employees in North America. Your won deals showed lift for those industries, for 51 to 500 employees and for companies hiring sales reps. Here is a model built from that. All five accounts are made up.

The rules

The scores

What the example shows:

• A wins on all three parts. An AE should call this week, and the G2 comparison tells them which competitor to prepare for.
• B has no intent at all and still ranks second, because it fits and its people are engaging. Intent is a bonus on top of fit.
• D is researching the topic this week and still only just makes tier C, because it does not fit. Without fit points, intent sends reps to accounts that will not buy.
• E fits on paper, and the negative rule drops it to tier D. Negative points keep bad accounts from floating up.

The data an account score needs

Score only complete records. A model run on blank fields scores your data gaps: an account with no headcount gets 0 fit points and drops out of tier A for a missing field. Enrich the fit fields first, then score.

Account scoring tools, with prices

If you already pay for HubSpot or Salesforce, build the rules there before you buy anything. Buy a platform when you need third-party intent and account identification in the same model. For a ranked look at scoring software, see our lead scoring software comparison.

How Cleanlist AI handles the fit score

Disclosure. We make Cleanlist AI, and this entry is our own description of our product.

Cleanlist AI covers the fit part of an account score. Describe your best customer in a sentence, and every lead comes back with a score from 0 to 100 and the reasons behind it. Enrichment fills headcount, industry, title and location. The Research skill reads each company's site and careers page for the rest, such as the tools it runs or a hiring push, and cites the page it read. Scoring is 5 credits a lead, on plans from $49 a seat a month. Inside an agent, on Pro, it can run before enrichment, so you only pay to enrich the leads that score high. Cleanlist AI has no intent feed and no technographic database, so intent and engagement come from your CRM and your intent vendor. See ICP scoring.

Frequently asked questions

What is a good account score?

There is no universal number, because the scale is yours. A good model is one where tier A wins clearly more often than tier C, and where tier A holds about as many accounts as your reps can work in a quarter.

How often should account scores update?

Engagement and intent should update daily or weekly, because they decay fast. Fit can refresh monthly or when you enrich again. Revisit the weights every quarter, and after any change to your pricing or product.

Can you do account scoring in a spreadsheet?

Yes, for a few hundred accounts. Put each rule in a column and add them up. Once the tiers predict wins, move the rules into your CRM so the score updates itself.

What is predictive account scoring?

A model that learns the weights from your past wins and losses instead of rules you write. It needs a lot of history, and it is harder to explain to reps. Start with rules, prove the inputs, then add a predictive model once you have the deal volume.

What is the difference between account scoring and account tiering?

The score is the number. The tier is the bucket and the action that goes with it, such as "AE calls this week". You can also hand-pick your top 20 to 50 strategic accounts as a tier above the score.

Does account scoring work without intent data?

Yes. Fit plus engagement can rank accounts on its own, and intent adds timing. Start without it, and buy intent once you know which topics your won deals research.

Sources

• Salesforce, Implementation Guide: Einstein Features in Account Engagement, Winter '27, last updated July 22, 2026, read October 30, 2026
• Salesforce, Account Engagement pricing, read October 30, 2026
• HubSpot, Build lead scores, updated August 14, 2026, and Sales Hub pricing, read October 30, 2026
• Factors, pricing, read October 30, 2026
• 6sense, pricing, read October 30, 2026
• Demandbase, pricing, read October 30, 2026