Lead Scoring
A model assigning a numeric value to leads based on fit and behaviour, so attention goes to the ones most likely to convert.
What it does
Assigns a number to each lead so limited attention goes to records most likely to be worth it. Two ingredients are usually combined: fit — attributes of the account — and behaviour — what the person actually did.
Check the inputs before the model
Most scoring problems are not modelling problems. A rule awarding points for company size does nothing on records where company size is empty, and on typical inbound forms most enrichment fields are empty at the moment scoring runs.
The five-minute diagnostic
Fitting it to outcomes
Assemble at least six months of leads with known outcomes — converted or not. Without labelled outcomes there is nothing to fit and the weights are opinion.
For each candidate attribute, compute conversion rate with and without it. Attributes that do not separate the groups get no points, however intuitive they seem.
Weight by observed separation rather than by stakeholder preference. This is the step most often skipped and the one that makes the model a model.
Score fit and behaviour separately and keep both visible. A high fit score with no behaviour is a target; high behaviour with poor fit is a support query or a competitor.
Hold out a validation set and check the ranking actually predicts. Re-fit quarterly — the population shifts.
When scoring is the wrong tool
At low volume, where a person can read every lead, a score adds latency and a layer to argue about.
In account-based motions, where the account is targeted deliberately and an individual's score is beside the point.
Where the real constraint is response time. A better-ranked queue worked four days late converts worse than an unranked queue worked in an hour.
RELATED TERMS
Lead Routing
The rules assigning each inbound lead to a specific owner — by territory, segment, product, account ownership or round-robin.
Account Scoring
Scoring at the account level rather than the individual, reflecting that B2B purchases are made by groups.
Buyer Intent
Signals suggesting an account is actively researching a purchase, drawn from on-site behaviour or third-party data.
Marketing Qualified Lead (MQL)
A lead that marketing considers ready for sales attention, according to a definition both teams have agreed. Without that agreement it is a marketing activity metric wearing a pipeline costume.
COMMON QUESTIONS
- How does lead scoring work?
- Attributes are weighted and summed into a single number used to prioritise follow-up. Fit attributes describe the account, behavioural attributes describe what the person did. Combining them into one score is convenient and hides which half is driving the ranking.
- Why do lead scoring models fail?
- Usually at the input layer rather than the model. Rules read fields that are empty on most inbound records, so the score is computed from absence and the model's precision is irrelevant.
- Should you use points-based or predictive scoring?
- Points-based models are transparent and can be debugged by the people who own them. Predictive models fit better where you have enough labelled outcomes — typically thousands. Below that volume, a fitted points model usually performs comparably and can be explained.
- How do you validate a lead scoring model?
- Hold out records, then compare score against actual outcome. If high-scoring records do not convert better than low-scoring ones, the model does not work regardless of how sensible its rules read.
Learn how to apply this: CRM Admin 101
Definitions are the vocabulary. The courses are where you learn to operate it, with the interactive audit tools.
See the course