Lead Scoring: Score Fit and Intent Separately

Two axes, four decisions — and why one combined score hides half of them.

Most pages on this subject hand you a points table and stop. The table is the least interesting part.

What decides whether scoring works is whether the model separates the two questions sales actually asks, whether it subtracts as well as adds, and whether anyone agreed what happens when a lead crosses the line.

This assumes leads are already arriving. If the problem is volume rather than triage, start with our guide to B2B lead generation and come back — scoring a thin pipeline only sorts a small list into a smaller one.

What is lead scoring?

Lead scoring is a ranking system that assigns values to each lead based on who they are and what they have done, so your team works the list in order of probability rather than order of arrival.

Two design choices matter more than everything else: what you score, and what the score triggers. Get either wrong and you have built a sorting field nobody reads.

Why one combined score hides the leads that matter

Collapsing fit and intent into a single number is the commonest design error, because the two say different things and cancel each other out when added.

Fit answers: are they the kind of company we sell to? Industry, headcount, revenue band, job role, geography.

Intent answers: what have they actually done? Pages viewed, emails replied to, pricing checked, demo requested.

A perfect-fit lead who has done nothing scores 60. A terrible-fit lead who downloaded everything also scores 60. They need opposite treatment, and one number makes them identical in the queue.

So use two axes and route by quadrant.

Two of the four quadrants are explicitly not sales' problem. That is the point of the exercise.

Lead scoring criteria: the signals worth scoring

Weight directions below, not point values — exact points are company-specific, and anyone publishing a universal number is guessing on your behalf.

Negative scoring and the decay rule

Roughly half the articles on this topic never subtract a single point, which is why so many models eventually rank a competitor's intern at the top.

Subtract for the people who behave like buyers but cannot buy:

• Competitors — maintain the domain list; it is ten minutes a quarter.
• Students and researchers — "student", "intern", "research scholar" in a title or company field.
• Job applicants — careers-page visits, "looking for opportunities", CV attachments.
• Free-mail domains — only where your product is genuinely enterprise, not in a market full of real businesses on Gmail.
• Unsubscribes and hard bounces — a dead address with a high score is pure queue pollution.

This is not an exotic requirement. HubSpot's documentation for lead scores describes models built from manual criteria with inclusion and exclusion lists (HubSpot Knowledge Base, "Edit, turn off, or delete lead scores", last updated 25 August 2026) — the subtracting side is part of the design, not an advanced extra.

Then add decay, because a pricing-page visit from four months ago is history, not intent.

A simple version works: halve activity points after a month of silence, zero them after a quarter. Fit points do not decay — a company's headcount is no less relevant because nobody clicked anything.

Without decay, your highest scores drift towards whoever has been in the database longest. Those are the oldest leads, not the hottest.

Where the points actually come from

The first version of any lead scoring model is a guess dressed up as arithmetic. The weights come from opinion, and the opinion belongs to whoever ran the workshop.

That is fine as a starting point. It stops being fine when nobody ever checks it against outcomes you already own.

1. Pull your last 20–30 won deals. For each, list what the lead was (fit attributes) and what they did before the first sales conversation.
2. Pull your last 20–30 lost or disqualified ones. Same two lists. Include the ones sales refused to call — those are the most informative records you have.
3. Mark the signals that separate the groups. Present in won, mostly absent in lost. Those earn weight.
4. Delete the signals present in both. A behaviour that shows up equally in winners and losers carries no information, however satisfying it felt in the workshop.
5. Re-run it next quarter with fresh closed deals, and change the weights when the evidence changes.

Step four is where most published point tables die. Gated downloads and blog reads usually appear just as often in the deals you lost.

All of this needs outcome history in a system people actually update, so choosing a CRM for a small business is the upstream decision, and marketing automation is what applies the points once the rules exist. Neither designs the model for you.

MQL vs SQL: what the score can and cannot decide

A score can create an MQL. Only a conversation can create an SQL. Confusing the two is how teams end up arguing about lead quality for a year.

If those stage names are fuzzy in your organisation, fix the definitions first — our marketing funnel guide covers where each stage begins and ends.

The threshold is an agreement, not a number

This is the organisational part that decides whether any of it works, and it gets skipped because it needs a meeting rather than a dashboard.

Sales has to agree the threshold. Not be informed of it — agree it, with the standing ability to say the bar is too low.

Then write down what happens when a lead crosses it. Four things:

• Who owns the follow-up, by name or by rota — not "the sales team".
• How fast — a stated first-touch window, and what counts as a touch.
• How many attempts before the lead returns to nurture.
• What happens on rejection — returned to marketing with a reason code that feeds the next calibration round.

Without that document, scoring just relabels the same backlog: the same leads in a different order, and the reason codes that would have improved the model never get recorded.

So track the handover, not only the score — acceptance rate, first-touch time and rejection reasons by category. When we build demand programmes as part of our performance marketing work, that definition gets written before anyone configures a point value.

Predictive lead scoring: what it actually needs

Predictive lead scoring replaces hand-set weights with a model trained on your historical outcomes. It is genuinely better than a rules table — once there is enough history to learn from.

That requirement is not vague, and vendors publish it. Salesforce's data requirements for Sales Cloud Einstein state that Einstein Lead Scoring needs at least 1,000 leads created in the last 200 days, of which at least 120 must have converted to an account and contact (Salesforce Help, "Data Requirements for Sales Cloud Einstein"; the page displays no last-updated date — read 8 October 2026).

Roughly 150 new leads a month, with more than half a conversion a day, sustained for six months.

A business closing a handful of deals a month does not have that, and no licence fee creates it. Below that volume a two-axis rules table calibrated on your own closed deals will beat a model starved of examples — and you can explain it to the rep who disagrees with it.

One case changes the question entirely. If you sell to buying committees and run account-based marketing, score the account rather than the person: eight mild signals from eight people at one target company matter more than one strong signal from a lone contact.

What we check before we build a scoring model

Digital Hangover's own observation, from client work. It carries no measured figure — this is a pattern we keep meeting, not a study, and we are not attaching a number to it.

Before designing anything, we check whether past deals in the CRM carry an honest outcome: closed-won and closed-lost actually set, and a disqualification reason filled in on the ones sales refused.

When that history is missing or half-entered, there is nothing to calibrate a lead scoring model against, so we fix the record-keeping before touching point values. Building the model first only produces a prettier guess.

And the commonest reason we find a client's existing model ignored by their sales team is almost never the weights. It is that nobody wrote down what happens when a lead crosses the threshold.

The score exists, visible on the contact record, owned by no one, triggering nothing. So reps carry on working from their own mental list of accounts worth calling — often reasonably — and the score becomes decoration that marketing reports on and sales never opens.

When lead scoring is not the answer

Scoring is triage. Without a triage problem, building one creates work and solves nothing.

• Low lead volume. If your team can personally contact every inbound lead within a day, call everyone — a human reading each enquiry beats any points table at that scale.
• A broken definition of a lead upstream. If "lead" currently means anyone who downloaded anything, scoring will faithfully rank a list that should not exist. Fix what qualifies at the form, then score.
• A generation problem in a quality costume. When too few of the right companies are arriving, no ranking fixes it — that is a channel and targeting question.
• No agreed owner for the output. If sales will not commit to the threshold, do not ship the model.

Where to go from here

Build the smallest version that can be wrong in a useful way: two axes, a short list of signals you can actually observe, negative points for the obvious disqualifiers, decay on activity.

Then write the handover rule, agree the threshold with sales, and diarise a date to test the weights against your next batch of closed deals. Organic demand takes roughly three to six months to produce meaningful volume, so expect that first honest calibration round about a quarter after launch.