What scoring is for
Scoring exists because sales capacity is finite and leads are not equal. Without a score, the leads that get attention are the loudest, the most recent or the ones whose company name a salesperson recognises. With a score, attention follows evidence: the contacts that look like past customers and are behaving like buyers rise to the top of the queue, and the rest stay in marketing's care until they do.
That is the whole purpose, and it sets the standard a model must meet. A score is good if the people above the line close at a visibly higher rate than the people below it, and if sales believes that enough to work the queue in order. Everything else — the elegance of the point system, the number of signals, the sophistication of the tooling — is secondary to those two tests.
Scoring sits between two neighbours. Upstream, the intent signals guide explains how behaviour reveals where someone is in the FunnelModel of the customer journey stages.; scoring is the operational version of that reading. Downstream, the lead stages and pipeline guide defines what a marketing-qualified or sales-accepted lead means; scoring decides who crosses those lines but does not define them. This guide stays on the model itself.
Fit and intent signals
A score mixes two different questions. Fit asks whether this person and their organisation resemble the customers you want: industry, company size, role, region, technology, budget authority. Fit is mostly static, comes from form fields and enrichment, and does not change because the person read a blog post. Intent asks whether this person is showing signs of buying now: which pages they viewed, what they downloaded, whether they requested pricing or a demo, how recently and how often. Intent is behavioural, decays quickly and is what most platforms capture automatically. Fit points are only as good as the profile behind them, and the tutorial on building an ideal customer profile from CRM data derives those fields from the accounts that renewed and grew rather than from the ones the business would like to win. Role fit is a separate question from company fit, and the comparison of an ideal customer profile and a buyer persona shows which of the two documents each kind of fit point should come from.
| Signal | Type | Where it comes from | Weight in practice |
|---|---|---|---|
| Job title or function matches the buyer or user profile | Fit | Form field, enrichment, lead gen form | High; a wrong title rarely closes |
| Company size and industry in target | Fit | Enrichment, form, CRM account record | High for enterprise sales, lower for self-serve |
| Region and language served | Fit | Form, location lookup, CRM | Gate rather than points: out of region should fail outright |
| Pricing or demo page viewed | Intent | Web tracking | High, and highest when repeated within days |
| Demo or quote requested | Intent | Form | Very high; often an automatic hand-off |
| Case study or comparison content viewed | Intent | Web tracking | Medium; signals the evaluation stage |
| Top-of-funnel content downloaded | Intent | Form | Low; interest, not intent |
| Email opened | Intent | Email platform | Very low or zero; mailbox privacy features make opens unreliable |
| Email clicked | Intent | Email platform | Low to medium depending on what was clicked |
| Webinar attended live | Intent | Event platform | Medium to high; registrations without attendance count for less |
The practical rule is to keep the two lists separate and report both. A single blended number cannot distinguish a perfect-fit director who has done nothing from a student who has read every page, and that ambiguity is the first reason sales stops trusting a score. Where fit data arrives through paid channels, the quality of the form matters: the LinkedIn targeting and lead gen forms guide covers why pre-filled forms produce accurate fit fields and why that beats volume.
Points, weights and decay
Points are a way of saying "this matters more than that". The temptation is precision: a weighting for every event, tuned to two decimal places. Resist it. A first model needs only a few weight levels — gate, high, medium, low — applied to a short list of signals, because that is what the team can explain and what the data can support. Precision comes later from calibration, not from guessing at launch.
Gates are not points at all. Out of region, competitor domain, personal email address where business email is required, or a declared student status should fail fit regardless of what else is true. Modelling these as large negative points works but hides the logic; a plain condition is clearer.
Repetition needs a rule. Viewing the pricing page several times in a week is more meaningful than once, but viewing it dozens of times is not proportionally more meaningful; it is probably a colleague, a bot or a bookmark. Cap the points any one signal can contribute within a window.
Decay is what separates a useful score from a historical record. Intent that was shown months ago says little about intent today. Most teams implement decay in one of two ways: a scheduled workflow that subtracts a portion of intent points after a period of inactivity, or a time-boxed score that only counts behaviour within a rolling window. The rolling window is simpler to explain ("intent in the last period") and easier to audit; subtraction gives finer control. Fit should not decay, though it should be refreshed when enrichment data changes.
Write the point values and the decay rule down somewhere outside the platform, with the reasoning. The governance guide covers why; the short version is that the person who set the values will leave, and the person who inherits them needs to know whether "pricing page: high" was a deliberate choice or a default. The lead scoring model template is a ready sheet for that record, with fields for decay, negative signals, thresholds and calibration.
Thresholds and stages
A threshold is the score at which something happens: a contact enters a nurtureKeeping in useful contact with leads who are not yet ready to buy until they are., is flagged for sales, is routedRules that assign each new lead to an owner, queue or next step based on territory, segment, score or source. to an owner or is pulled out of marketing sequences altogether. Thresholds only mean something when they are tied to stage definitions that sales and marketing have agreed, which is why the pipeline guide comes first.
The common arrangement is two lines. Below the first, the contact is a lead in marketing's care, receiving nurture matched to what they have shown. Between the lines, the contact is engaged and worth a lighter sales touch or a more direct sequence. Above the second, the contact meets the definition of marketing-qualified and is handed over with a service-level expectation on response time. Because fit and intent are separate, each line is a combination: fit at or above its bar and intent at or above its bar within the decay window. A high-fit contact with no intent stays in nurture; a high-intent contact with poor fit is reviewed rather than routed. For products people can use before they pay, the product-qualified leads guide keeps the same two lists, with product usage in place of web behaviour on the intent side.
Set the first thresholds from the distribution, not from instinct. Look at where the leads that closed in the last period sat on each score the week before they were accepted, and set the hand-off line where most of them would have crossed. Then accept that the first line will be wrong and plan the calibration described below.
One threshold deserves its own workflow: the automatic hand-off on a declared action. A demo or quote request is intent at its clearest, and routing it immediately regardless of score avoids the absurdity of a buyer waiting for a nurture email because their fit field was blank. Once a lead has an owner, single events on the record are better handled by alerts than by a rising score, and the guide to sales alerts from marketing activity decides which events deserve one.
Where the threshold should sit can also be tested directly: the lead scoring threshold test randomises leads in a band just below it between sales and nurture and reads pipeline per lead. A line that sits too high shows up in nurture rather than in the sales queue, and the diagnosis for leads stalling in a nurture sequence traces it beside the other reasons a sequence's entrants rarely reach its goal.
Negative scoring
Negative scoring subtracts points for signals that make a contact less likely to buy or less worth a call. It is the part of a model most often left out and the part that most improves sales trust, because it removes the leads a salesperson would have dismissed on sight.
Typical negative signals fall into three groups. Disqualifying fit: competitor domains, agencies and consultants where you sell direct, students and job seekers, free email addresses in a business-only market, roles with no influence on the purchase. Disqualifying behaviour: careers page views, support documentation visits from existing customers, unsubscribes, bounced addresses. Status: an existing customer, an open opportunity already owned by a salesperson, a closed-lost deal within a recent period, or an explicit "not interested" logged by sales.
Decide for each whether it is a subtraction or a gate. A careers page view is a subtraction; a competitor domain is a gate. Existing-customer status is usually a gate out of prospect scoring entirely and into a separate expansion or retention model, which is a different question with different signals.
Negative scoring also has to be explainable. If a contact's score drops, the salesperson looking at the record should be able to see why. Keep the negative list short and visible, and log score changes with their cause where the platform allows it.
Calibrating against outcomes
A model is calibrated when the score predicts what it claims to predict. The test is retrospective and simple: take every lead that reached the hand-off threshold in a past period, find out what happened to them, and compare with leads that did not reach it. If the scored leads were accepted and closed at a clearly higher rate — their conversion rate (CVRShare of visits or clicks that convert.) to opportunity and to won — the model is doing its job. If the rates are similar, the score is noise and the thresholds or signals need to change. If a large share of closed deals came from leads that never crossed the line, the model is missing a signal or the fit bar is too tight.
Calibrating a scoring model after one quarter
A business software company launched scoring with a hand-off at a fit score of 60 and an intent score of 40 within a 30-day window. After one quarter, the operations lead pulls every lead that crossed the line: 412 leads, of which sales accepted 290 and 31 became closed-won. Below the line, 2,900 leads produced 1,100 that sales reviewed anyway and 19 closed-won. The scored group closed at roughly 7.5% against 0.7% for the rest, so the model separates. But 11 of the 19 wins from below the line shared one trait: a job title containing "head of" that the title mapping treated as medium fit. Raising that mapping to high moves 160 historical leads over the line and would have captured 9 of those 11. The team also finds that webinar registrations earned 15 intent points whether or not the person attended; attendance is now worth 15 and registration alone 5. Both changes are made, documented and dated, and the same review is scheduled for the next quarter.
Calibration also feeds budget decisions. The cost to acquire a customer (CACCost to acquire one new customer.) by source changes once scoring shows which sources produce leads that cross the line and close, rather than leads that merely arrive. A source that generates a large share of low-fit volume looks productive on lead count and poor on scored leads; that is the number the scoring model exists to surface.
Review cadence
A scoring model drifts for ordinary reasons: the product changes who the ideal customer is, a new content programme generates behaviour the model never weighted, a field is renamed and silently stops feeding the score, sales learns something about which leads convert. Three reviews keep it honest.
Weekly, briefly: look at the leads that crossed the hand-off line and the ones sales rejected. Rejections with a reason are the fastest feedback on fit signals. A sudden jump or collapse in the number crossing the line is usually a broken field or a new form, not a change in the market.
Monthly: check the inputs. Are enrichment fields still populating, is web tracking still firing on the pricing page, has an integration dropped a field? Scoring fails quietly when its inputs do, and the symptom is usually a slow decline in hand-offs that gets blamed on demand.
Quarterly: run the calibration above, adjust weights and thresholds with sales in the room, and record what changed and why. Tie this review to the pipeline KPIA metric tied directly to an objective. review so that changes to stage definitions and changes to the score happen together. After any change to the rules, the lead routing QA checklist confirms with test leads that scores, stages and the hand-off to sales still update as intended. To judge the model alongside routing, nurture and the sales agreement in one review, the lead management audit scorecard rates each and ranks where leads leak most.
The nurture sequences guide is the other half of the loop: what happens to contacts that the score keeps on marketing's side of the line. For each sequence, the nurture sequence plan template records which actions change the score and the point at which a lead is handed to sales. The workflows hub shows where scoring sits among the other automations.
If leads above the hand-off line are routinely rejected by sales, the page on marketing-qualified leads that never become sales-qualified checks whether the score is counting reading rather than buying before any other cause.
Common mistakes
- One blended score. Fit and intent mixed into a single number that cannot be explained or acted on.
- Points for opens. Mailbox privacy features make opens an unreliable signal; a model built on them rewards nothing.
- No decay. A contact who downloaded a guide last year still sits at the top of the queue.
- Thresholds without stage definitions. The hand-off line means nothing to sales because nobody agreed what it represents.
- No negative signals. Competitors, students and existing customers routed to sales as new leads.
- Launched and never calibrated. The model is judged by whether it looks sensible rather than by what closed.
- Hand-off without a service level. Scored leads sit untouched, and the score gets the blame.
