Lead scoring was built for a world where one person found you, downloaded something, and eventually talked to sales. B2B deals no longer work that way. Forrester's 2024 research found that on average 13 people are involved in a purchasing decision. When you score each of them separately, you get thirteen lukewarm leads instead of one hot account. Account scoring fixes that by rolling every signal up to the company that will actually sign the contract.
This post explains how to build an account scoring model that sales trusts: what goes into it, why fit and engagement should never be added together, how to count buying-committee breadth, and a worked example you can adapt.
Account Scoring vs Lead Scoring
The difference is the unit, and it changes almost everything downstream.
- Lead scoring rates one person. It rewards that person's own behavior: emails opened, assets downloaded, pages viewed. It is good at deciding when a known contact is ready for a call.
- Account scoring rates one company. It aggregates behavior across everyone from that company, identified or not, and combines it with how well the company fits your ideal customer profile. It is good at deciding which companies deserve sales and marketing attention right now.
You need both. Account scoring decides where to focus, lead scoring decides who to call inside that account. If you only have lead scoring, read our post on lead scoring for high-intent B2B accounts first, then come back here.
The Mistake: One Number for Fit and Engagement
Most account scoring models we audit add fit points and engagement points into one total out of 100. It looks tidy. It produces bad decisions.
Here is why. A perfect-fit enterprise account that has never visited your site and a poor-fit startup whose intern read twenty blog posts can land on the same score. Sales sees two "65s" and has no idea that one is a strategic target with no intent and the other is noise with lots of activity. The single number hides the only thing the rep needed to know.
Keep two scores on two axes instead:
- Fit score (0-100): how closely the company matches your ICP. Changes slowly. Based on firmographics and technographics.
- Engagement score (0-100): how actively the company is researching right now. Changes daily. Based on website behavior, intent data, and campaign responses, with decay.
Then act on the combination, not the sum. A simple grid does the job:
| Low engagement (0-39) | Medium engagement (40-69) | High engagement (70-100) | |
|---|---|---|---|
| High fit (70-100) | Target list: ads, outbound, personalized site experience | Nurture hard: account-specific content, SDR research | Sales now: route to owner same day |
| Medium fit (40-69) | Ignore for now | Marketing nurture | Qualify: SDR checks fit before AE time |
| Low fit (0-39) | Ignore | Ignore | Self-serve path or disqualify |
This grid also gives marketing a clear job for the high-fit, low-engagement corner, which is where most account-based programs should spend their budget.
What Goes into the Fit Score
Fit comes from your ideal customer profile, and it should be built from your closed-won data, not from what leadership hopes the ICP is. Typical inputs:
- Industry (weighted by your win rate per vertical)
- Employee count or revenue band
- Geography (where you can sell, support, and comply)
- Tech stack (uses a platform you integrate with, or a competitor you displace)
- Business model signals (for example, has a sales team, runs paid acquisition, sells B2B)
Weight each input by how strongly it separates won deals from lost ones. If 60% of your closed-won revenue comes from software companies with 200-2,000 employees, those two attributes deserve most of the fit points. Our guide to building an ICP for website personalization walks through the analysis.
What Goes into the Engagement Score
Engagement should measure buying activity, not general interest. Three kinds of input matter:
Depth: what the account looked at
Weight late-stage pages heavily: pricing, comparison pages, integration docs, security pages, case studies from the account's industry. Weight blog traffic lightly. A company reading ten blog posts is learning about a topic. A company reading pricing and your SOC 2 page is evaluating a vendor.
Breadth: how many people are involved
This is the input lead scoring can never capture, and it is the most predictive one we see. Two people from the same company on your site in the same week means a conversation is happening internally. Three people from different functions means a buying group has formed. Our 2026 B2B personalization research found that buying committees average 6-11 stakeholders per deal at the mid-market level and above, so even partial visibility into that group is valuable.
Score breadth by distinct visitors, with a bonus when you can infer different roles (for example, one visitor reading technical docs and another reading pricing).
Recency: when it happened
Engagement without decay is the second most common modelling mistake. An account that binged your site in March and went silent is not hot in October. Halve engagement points every 14-30 days depending on your sales cycle. Shorter cycles need faster decay.
If you buy third-party intent data, add it as a separate engagement input with a cap, so a topic surge alone cannot push an account into the "sales now" column without any first-party activity. Our post on using intent data for website personalization covers how to blend the two.
Worked Example: Three Accounts, One Week
Here is a simplified model applied to three real-looking accounts. The weights are illustrative; yours should come from your own win data.
Fit weights: target industry 35, employee band 30, integrates with your CRM 20, target region 15.
Engagement weights: pricing view 15, comparison or security page 15, industry case study 10, each additional distinct visitor 10 (max 30), blog view 2 (max 10), intent surge 10. Points halve every 21 days.
| Account | Fit signals | Fit | Engagement this week | Engagement | Action |
|---|---|---|---|---|---|
| Logistics company, 800 staff, uses HubSpot, Germany | All four | 100 | 3 visitors, pricing, security page, case study | 70 | Sales now |
| Fintech, 1,200 staff, uses Salesforce, US | Industry, size, region (no integration) | 80 | 1 visitor, 4 blog posts | 8 | Target list |
| Agency, 12 staff, UK | Region only | 15 | 1 visitor, pricing, 5 blog posts | 25 | Ignore |
Under a single-number model, accounts two and three would sit close together. On two axes, it is obvious that the fintech is a strategic account worth outbound and personalized content, and the agency is not worth an SDR's afternoon.
Using Account Scores on Your Website
Scores are usually treated as a sales tool, but they are just as useful for deciding what the website shows. A few rules we see work:
- High fit, low engagement: show industry-specific proof and a low-commitment CTA (benchmark report, ROI calculator). The goal is to earn a second visit.
- High fit, high engagement: replace content CTAs with a direct meeting offer and name the account's industry in the hero.
- Low fit, high engagement: route to a self-serve path or a lighter plan instead of a sales demo.
Markettailor's account-based marketing features apply rules like these on the page: once a company is identified, its fit and engagement tier decides which hero, proof, and CTA it sees.
How to Calibrate the Model
A scoring model is a hypothesis. Test it against history before sales relies on it.
- Export the last 12 months of closed-won and closed-lost opportunities with the account's fit attributes.
- Score each account as it would have looked 30 days before the opportunity was created.
- Check that closed-won accounts cluster in the high-fit, high-engagement corner. If they do not, your weights are wrong.
- Look at the accounts in "sales now" that never became opportunities. Those are your false positives. Find the input that pushed them up and reduce its weight.
- Re-run the back-test every quarter. ICPs drift as the product changes.
If you want a starting worksheet for the weights and thresholds, our lead scoring model template uses the same fit and behavior structure and works at the account level with minor changes.
Four Mistakes That Make Sales Ignore the Score
Scoring the whole database. Scoring 40,000 accounts produces a long tail of meaningless 12s and 17s. Score the accounts that pass a minimum fit bar, and leave the rest unscored. A smaller, cleaner list gets used.
Hiding the reasons. A score of 82 tells a rep nothing about what to say. Every score should come with its top three contributing signals ("3 visitors this week, pricing viewed twice, uses HubSpot"). Reps trust what they can explain to themselves.
Counting existing customers as prospects. A customer's employees reading your docs will light up the engagement score every week. Route customer accounts to a separate expansion or health model, or they crowd out new business.
Never retiring an input. Inputs get added after every quarterly review and almost never removed. After a year the model has 30 inputs, half of them redundant. When you add an input, remove one, and keep the total under a dozen.
Start Small
You do not need a predictive model to begin. Four fit attributes, five engagement inputs, decay, and a two-axis grid will beat most of the single-number models running in CRMs today. Build it in a spreadsheet, back-test it, then move it into your CRM or personalization platform once sales agrees the top 20 accounts look right.
To see how fit and engagement tiers can drive what each account sees on your site, book a walkthrough from our pricing page.