Lead Scoring Is Broken: Score Accounts, Not People

Your reps ignore the lead score because it is measuring the wrong object. A B2B purchase is made by a group, so the unit of intent is the account, and the score should be built from what the whole buying group did in the last 30 days, not from lifetime points a single contact accumulated since 2023.

That is the whole argument. The rest of this is how to rebuild it so sales actually looks.

Why your reps stopped looking at the score

Three failure modes, and almost every point grid has all three.

No decay. Points are awarded forever. An ebook download from eighteen months ago still carries full weight today, which means your highest scoring contacts are often your oldest, not your hottest. That is a commonly cited failure of point based models.

Demographic points inflate the number. Title, company size, industry and country all add points. A VP at a 500 person company who has done nothing scores higher than a manager at a 40 person company who has visited pricing four times this week. The score now says fit, but the rep reads it as urgency. Those are different questions and the model has blended them into one unreadable number.

Engagement measures curiosity, not intent. Email opens, one page views, a webinar registration. These are information gathering. 6sense's 2024 buyer experience research, a survey of 2,509 recent B2B buyers, found that 81% had already chosen a winner before speaking to a sales rep, and that nearly 70% of the purchase process happens before buyers engage with sellers. That is vendor research from a company that sells account intent data, so read it with that in mind, but the direction matches what reps feel every day: early engagement is not readiness, and most of the real evidence happens before anyone fills in a form.

There is a fourth failure, and it is the one that actually kills adoption. Models fail on belief, not accuracy. A lead scores 91, the rep asks why, and the only available answer is that the model decided. Explainability is the adoption variable. If your score cannot be explained in a sentence, it is decoration.

Three people from one company read your pricing page. That is one signal, not three leads

Gartner puts the median buying group for a complex B2B solution at six to 10 decision makers, each armed with four or five pieces of information they have gathered for themselves. The same body of work found 77% of B2B buyers describe their latest purchase as very complex or difficult.

Then look at how those buyers spend their time. Gartner's breakdown of the buying journey: 27% on independent online research, 22% meeting with the buying group internally, 18% on offline research, 17% meeting with all potential suppliers combined, and 16% on other activities. When buyers are comparing multiple suppliers, the time spent with any single sales rep may be only 5% or 6%.

Read the 22% again. Almost a quarter of the buying process is the committee talking to itself, with no contact record attached to any of it. The only trace it leaves on your side is a pattern: several people from one domain, poking at the same pages, in the same week.

A contact level model cannot see that pattern. It produces three separate mediocre scores, none of which clears the threshold, so nobody is alerted. An account level model produces one obvious signal: this company has a buying group forming.

This is not a scoring refinement. It is a change of unit. Score the account. Keep contact scores if you like, but demote them to a routing detail, which rep calls which person once the account lights up.

Fit and intent are different questions, so stop adding them together

Never sum them. Two axes, always separate.

Fit is static and firmographic. Industry, employee count, tech stack, region, whether they match your ICP. It changes slowly, it does not decay, and it answers one question: should we want this company as a customer?

Intent is behavioural and perishable. Repeat visits to pricing and product pages, multiple roles from one domain, a return after a gap, speed of reply to an inbound. It answers a completely different question: are they in market right now?

The reason summing them destroys trust is simple. A perfect fit account doing absolutely nothing and a poor fit account in heavy active research land on the same total. The rep works both, finds one is nowhere near buying and the other is never going to be a good customer, and concludes the score is noise. They are not wrong.

With two axes you get four quadrants and four different plays. High fit plus high intent means call today. High fit plus low intent is a nurture and demand creation job. Low fit plus high intent gets self serve or a polite qualification call. Low fit plus low intent gets nothing. Each one is explainable in a sentence, which is the whole point.

Give intent a rolling 30 day window with decay inside it. A pricing visit from yesterday should outweigh one from three weeks ago. If you cannot decide on a window, start at 30 days and tune it against your own sales cycle rather than copying someone else's number.

What actually precedes a deal

Behaviours worth weighting heavily, in rough order of how much they tell you:

• Repeat visits to pricing or a specific product page. Once is curiosity. Three times in ten days is an internal evaluation.
• Multiple roles from one domain. A practitioner plus a manager plus someone from finance is a buying group assembling. This is the single strongest account level signal you have and the one contact scoring is structurally blind to.
• Returning after a gap. A company that went quiet for six weeks and came back to the same pages has restarted a process, usually because budget or a trigger event moved.
• Inbound reply speed. How fast they come back to your first response says more about urgency than any form field they completed.
• Depth over breadth. Documentation, integrations, security and implementation pages beat another blog read, every time.

Weight down the ones that only look like engagement: opens, single visits, generic content downloads, anything from an existing customer's IP that is really a support lookup.

Backtest it before you trust it

Do not switch on alerts and hope. Take your last 20 to 50 closed won deals and run the model backwards over them.

For each one, ask: in the 30 days before the first real sales conversation, what did the account actually do? How many people from that domain were active? Which pages? What would this model have scored them, and would it have fired an alert before the rep found out some other way?

Then do the same with closed lost and with deals that went dark. You are looking for the threshold where won deals cluster and lost ones do not. That threshold is your alert trigger. Everything above it is a promise to your sales team, so set it where the evidence sits, not where the volume feels comfortable.

Then apply the adoption test. Pick any account the model has flagged and ask a rep why it is hot. If they can answer in one sentence from what is on screen, four people from this company hit pricing twice this week, the model works. If they shrug, the model is decoration no matter how elegant the maths.

How to run this in Payani

Lead scoring and website visitor and company activity tracking are built to work together for exactly this reason. The activity tracking surfaces which companies and visitors are engaging, with ISP, crawler and bot noise filtered out, so what reaches the score is real buying behaviour rather than traffic. That is the account level view the contact model cannot produce.

From there, visual workflows built from plain language fire the alert on an account threshold instead of a contact threshold, so the rep hears about the company, not about one person's email open. Forms and meetings supply reply speed and booking behaviour. Reporting that fuses GA4 with your CRM is where the backtest against closed won history lives. And when the rep asks why this account is hot, Ask Payani answers it from the account's own data, in a sentence, which is the test the whole model has to pass.

If your scores exist but nobody trusts them, the fix is not a better formula on the same wrong unit. It is scoring the buying group, separating fit from intent, and proving the thresholds against deals you already won. Payani is built to do that in one place, which is the only way the number ever becomes something a rep acts on.

Frequently asked questions

Should we delete contact level lead scores entirely?

No, demote them. The account score decides whether anyone acts, and the contact score decides which person inside that account the rep calls first. The problem is not that contact scores exist, it is that they are used as the trigger for sales action when a purchase is made by a group of six to 10 decision makers.

Why 30 days for the intent window?

It is a sensible default, not a law. Intent is perishable and a pricing visit from yesterday should outweigh one from three weeks ago, so you need a rolling window with decay inside it. Start at 30 days, then tune it against your own sales cycle using your closed won history rather than copying another company's number.

What if most of our traffic is anonymous and never fills in a form?

That is normal and it is an argument for account level scoring, not against it. 6sense's survey of 2,509 B2B buyers found nearly 70% of the purchase process happens before buyers engage with sellers, so contact records capture only the late minority of the evidence. Company level website activity, with bot and ISP noise filtered out, is where the earlier signal lives.

How do we know the new model is better before we rely on it?

Backtest it against your last 20 to 50 closed won deals. Look at what each account did in the 30 days before the first real sales conversation, score it retrospectively, and find the threshold where won deals cluster and lost ones do not. Run the same pass over closed lost deals so you can see the false positives before your reps do.

What is the fastest way to tell whether our current score is worth keeping?

Pick any account the model flagged as hot and ask a rep why. If they can explain it in one sentence from what is on screen, the model works. If the only available answer is that the model decided, the score is decoration, because adoption depends on explainability rather than accuracy.