Lead Scoring Models Sales Teams Actually Use

Lead scoring fails when it is built in isolation by marketing and handed to sales as a verdict. The models that survive contact with a real pipeline separate two questions: does this lead look like our best customers, and is this lead showing buying behavior right now? Call the first fit and the second intent. Fit changes slowly and comes mostly from firmographic and demographic data. Intent changes daily and comes from actions. Keeping them apart lets you reason about a lead instead of arguing about a single number.

Start with fit. Pull a sample of closed-won and closed-lost deals from the last year and list the attributes that actually correlate with winning: industry, company size, revenue band, role of the person who contacted you, region, and the systems they already use. Turn the most predictive attributes into scoring rules with small, capped point values. Fit should be mostly static, updated by enrichment or form data rather than by behavior, and it should never spike because someone opened an email.

Intent comes from behavior, and not all behavior is equal. A pricing page visit, a demo request, a reply to an outbound email, or a repeat visit from the same company usually signals more than a blog page view or a newsletter open. Give deliberate actions meaningful points and passive actions few or none. Track depth as well as recency: three visits in a week matter more than one visit three months ago. Where possible, score the account as well as the person, because B2B purchases involve multiple people.

Build the first version in a spreadsheet before you touch the CRM. Take a hundred to a few hundred recent deals, apply your proposed weights, and sort. If the top of the list is not visibly better than the bottom, the model needs different inputs, not more inputs. Keep the first version to a handful of criteria; a model with thirty rules is impossible to explain and nearly impossible to tune. The goal is a ranking that sales agrees is directionally right, not a perfect prediction.

Thresholds turn scores into action, and they need a shared definition. Decide what score marks a marketing-qualified lead and what marks a sales-qualified opportunity, and make sure both teams use the same words. A common approach is a score band rather than a single cut-off, with the bottom band staying in nurture, the middle getting targeted content, and the top going to a rep. Revisit thresholds with sales monthly, because a threshold that sends too many weak leads will be ignored no matter how good the model looks.

Decay keeps intent scores honest. A visitor who downloads a guide and then disappears for two months is not the hot lead they were in week one. Apply time-based decay so older activities lose weight, and add negative points for long inactivity. Reset or boost scores when a fresh high-value action occurs. Decide whether decay is linear or stepped, and write the rule down so sales can understand why a score dropped. A score without decay gradually inflates until everything looks warm.

Negative scoring is as important as positive scoring. Competitors researching your pricing, students downloading templates, recruiters, and job seekers can all trip behavioral triggers. Maintain a list of disqualifying domains and roles, and subtract points or suppress them from routing. Be careful not to over-filter: a consultant researching on behalf of a client may be a real opportunity. Review suppressed records periodically to confirm the filters are not hiding legitimate demand.

The feedback loop is what separates a model that improves from one that decays. Ask sales to record disqualification and closed-lost reasons in structured fields, then review a sample of scored leads every month. Compare leads above and below the threshold for meeting rates and win rates. If high-score leads are not converting, examine which criteria drove the score and adjust. If sales consistently ignores scores, the model or the handoff is wrong, and both are worth fixing before adding more automation.

Common failure modes repeat across teams: too many criteria, no decay, scores that only go up, no negative scoring, thresholds set by opinion, and zero sales involvement in design or review. Another is scoring the wrong entity, giving a person a high score when the account is a poor fit or the reverse. Document the model in plain language, including each rule's weight and rationale, so anyone can audit why a particular lead landed where it did.

Roll out in stages and keep it simple. Launch with a small set of fit and intent rules, train the team on what the numbers mean, and review at thirty, sixty, and ninety days against actual outcomes. Treat the first version as a baseline to tune, not a final answer. A useful model is not the most sophisticated one; it is the one reps trust enough to change their daily priorities around, and that trust comes from transparency and iteration.