Lead Scoring: Where the Points Come From

Scoring ranks leads on fit and engagement

A lead score is a number attached to a person or a company that says how worth pursuing they look. It splits into two questions: whether these are the right people, and whether those people are showing the right level of interest.

The first is explicit scoring, built on data a prospect hands you: job title, industry, company size. The second is implicit scoring, built on behaviour: visits, downloads, replies. The two questions are independent, and most published models keep them on separate axes.

Gartner says the sequence you would score against is not there. Buying jobs happen without a consistent order, and most buyers revisit at least one, so a returning prospect reads as a fresh one to any score built on stages.

One vendor publishes a grid for combining them. Fit runs A to D, engagement runs 1 to 4, and A1 is the most qualified while D4 is the least. Sixteen cells, and each cell can carry its own follow-up rule.

HubSpot ships the same idea at a different resolution. Its lead scoring tool offers engagement scores, fit scores and combined scores, and the combined score grades on a three by three matrix from A1 to C3. Nine cells instead of sixteen.

The same two questions, shipped as a product

1. 1Engagement and fit are separate score types, not two inputs to one number. A record can carry both.
2. 2The combined score is where they meet, and it produces a letter and a digit instead of a total.
3. 3A grade tells you which of the two halves is weak. A single total does not, which is the argument for keeping the axes apart.

Four named methodologies get you to a score: ideal customer profile, lamb or spam, rule-based, and predictive lead scoring. The first three are ways of choosing criteria by hand. The fourth learns them from outcomes, and it is a different animal.

The labels either side of the score are worth keeping straight. A score crossing a threshold is what usually produces a marketing qualified lead, and what happens after that is a separate decision covered in MQL vs SQL.

One published method derives its numbers

Point values come two ways: assigned, or derived from what closed. Three of the vendor guides show the values; a fourth shows the derivation. That gap is easy to miss, because a table of invented points and a table of derived points look equally confident.

Salesforce publishes the derivation in four steps. Work out your baseline conversion rate. Define the ideal customer. Calculate the close rate for each attribute. Assign points to attributes whose close rate beats the baseline.

The worked example on its page: 100 customers from 200 leads gives a 50 percent baseline. Leads who watched a webinar close at 75 percent, and leads with a CTO title close at 65 percent. Webinar attendance gets 25 points. CTO title gets 15 points.

The other kind of page gives the numbers and skips the derivation. One platform puts a newsletter form at 5 points, a report request at 10, a consultation goal at 20, an email open at 2 and a link click at 5. Email opens and clicks expire after three months. A lead becomes sales-ready at a threshold, and the page gives 50 in one example and 300 in another.

Neither page is doing anything wrong. They are answering different questions: one shows a method, the other shows a shape. The problem is that a reader looking for numbers copies the second and gets a model with no evidence in it.

One of them writes out the part the rest leave implied.

It's up to you to determine your lead scoring model, i.e. how many points each action is worth, and at what scoring threshold you consider a lead to be marketing qualified.

Agile CRM, What is Lead Scoring, read 5 September 2026

What an assigned model looks like in the wild is worth seeing, because the arbitrary part is not hidden. You can see it in the last row of the template.

A template with the decisive number left blank

1. 1Three score types, each with its own points table, and every value chosen, not measured.
2. 2The decay table is the part most models skip, and it does real work: without it a five year old contact outranks a live one.
3. 3The last line is the one that decides everything, and the template leaves it as xx.

Building a score from close rates

The derived method needs your closed-won history and nothing more exotic. If you have a few hundred closed deals with attributes attached, you can compute this in a spreadsheet. If you do not, use the assigned method and label your numbers as provisional.

1. Compute the baseline

2. Compute close rate per attribute

3. Keep the attributes that beat the baseline

4. Split fit from engagement before totalling

5. Set the threshold yourself, because no vendor sets it

6. Attach a follow-up rule to each grade

One consequence of step five deserves its own line. HubSpot documents that when a score is updated, property values and actions are re-evaluated retroactively. Change a rule today and yesterday's records get rescored, so a comparison across a rule change is not a comparison.

Step six needs somewhere to write its answer, and in most systems that place already exists. A score sits in its own property, and the handoff is a change to a different field: the lifecycle stage. Keeping them apart is what lets you ask how many scored leads went on to move.

Where the handoff gets recorded, separately from the score

1. 1The score is a number. The stage is a state, and only the stage change is an event another team can see.
2. 2Marketing qualified lead and sales qualified lead are two of the default stages, so the boundary in the funnel already has a field waiting for it.
3. 3A score that never moves a stage has produced no observable outcome, which is the cheapest test of whether a scoring programme is doing anything.

Which subscription tier unlocks scoring

Lead scoring is not usually bought on its own. It arrives inside a marketing automation or CRM licence, and the tier that includes it is higher than most people expect when they start planning.

HubSpot documents that contact, company and deal scoring sits in Marketing Hub Professional and Enterprise and Sales Hub Professional and Enterprise. AI-powered contact scores are Marketing Hub Enterprise only. Its published pricing puts Professional from $800 a month with 2,000 marketing contacts, plus a one time $3,000 onboarding fee. Enterprise starts at $3,600 a month with a $7,000 fee.

Salesforce publishes five Sales Cloud editions a user a month: Starter Suite $25, Pro Suite $100, Core $195, Advanced $395, Max $550. All are billed annually except Starter. Both vendors revise these figures.

So the practical floor for a scored pipeline on HubSpot is around $800 a month plus onboarding. The Starter tier that many teams begin on does not include scoring at all. A spreadsheet is a legitimate place for your scoring while the close rates are being computed, and it costs you nothing.

Predictive scoring has no public price. Sellers quote on request and no independent median is published, so there is no range here.

Suppose you want the demand feeding the score priced instead: that is our B2B PPC service.

Five ways a score stops meaning anything

Each of these is visible in published vendor material. A bad model is not a precondition for any of them, and they survive because the score keeps producing numbers the whole time.

Copying point values from a guide. Suppose you take the 20 points one guide gives a consultation goal straight into your model. You have imported someone else's guess about your buyers, and once it is a number in a field the guess becomes invisible.

Scoring the wrong half of the process. 6sense surveyed 2,509 B2B buyers in 2024. It found that 69 percent of the purchase process happens before buyers engage with sellers, and 81 percent choose a preferred vendor before speaking with sales. A behaviour score sees the last stretch of that and grades it as if it were the whole thing.

Leaving decay out. One published template includes a decay table and most models do not. Without decay, activity from two years ago counts the same as activity from Tuesday, and old records slowly float to the top of the queue.

ActiveCampaign, Lead Scoring 101, read 5 September 2026Which of your rules forget, and which never do

The five point values are the published example model, the same set charted earlier. The last column is the addition: that page states an expiry for the open and the click and states none for the other three.

Changing a rule and comparing across the change. HubSpot rescores retroactively, so the before and after of a threshold change are computed under different rules. Any improvement measured that way is partly an artefact.

Believing BANT still describes what buyers show you. Budget, authority, need and timeline are questions about a decision that has already started.

If you ask sales and marketing executives from 100 companies today how they score their leads, 99 of them will answer with the acronym BANT, that is, to qualify opportunities, a lead must have budget, authority, need, and timeline for purchase.

Oracle, What is lead scoring, read 5 September 2026

The objection is about sequence. Buyers gather information long before they have a budget or a timeline, and salespeople have no visibility into that period. A framework that requires four things a buyer has not decided yet will disqualify people who are going to buy.

Sales acceptance as the barometer

Your lead score is a prediction, so you can score it in turn. The trap is grading it on how many leads it produces. That measures where you set your threshold, not how good your model is.

The right test is your sales acceptance rate, the barometer of a scoring program's health. Acceptance is judged by the other team, and that makes it a real test instead of a self-assessment.

The second test is what accepted leads do afterwards, and the rate a score is trying to predict is MQL to SQL. First Page Sage publishes that by industry, from client data gathered between January 2022 and August 2025.

Peer-reviewed work exists on the predictive version, and its scope is narrower than the headline suggests. González-Flores, Rubiano-Moreno and Sosa-Gómez published a B2B lead scoring model in Frontiers in Artificial Intelligence in 2025. It was built on 23,154 records with 67 fields, from a single software company, covering January 2020 to April 2024.

The authors state their own limit, and it is the limit that matters for anyone hoping to copy this. They write that the recent development of the scoring model limits the evaluation of the conversion rate of prospects into real customers qualified by the application. The model separates leads well. Whether the prioritised leads went on to buy is not yet measured.

A 2026 preprint by Zhang and colleagues names three problems in rule-based scorecards. Sparse supervision, a semantic gap in unstructured CRM logs, and an inability to capture relative lead priority. Its reported improvements are not quoted here, because the abstract states them without a denominator or a control definition.

That last problem is the one worth carrying away. A score answers how good a lead looks on its own. Sales works a queue, and a queue is a question about order. Those are different questions, and a threshold quietly converts one into the other.