Lead scoring model: how to build one

Start from the three behaviours you would act on

Not from a points table. Ask your best sales person which three things make them pick up the phone. Usually it is a pricing page visit, a reply, and a second person from the same company appearing.

Do this: Write those three down. That is your first model, and it is already better than most.

Points without validation are decoration

A score built from guessed weights feels precise and predicts nothing. It is fine as a starting point and dangerous the moment anyone treats the total as a fact.

Do this: Take last quarter’s closed deals and check what the model said about them at the time. If it said nothing useful, the model is decoration.

Decay matters more than the weights

Somebody who visited pricing yesterday is a different person from somebody who visited a year ago, and most models treat them identically. Score decay is usually the single change that makes a model work.

Do this: Halve the value of every behaviour older than thirty days and watch the ranking change.

The score is a queue, not a verdict

The point is deciding who gets called first today. It is not a judgement about the person, and it should never block a human from calling somebody who looks interesting.

Do this: Publish the score as a sorted list, not a threshold. Thresholds create arguments, lists create calls.

This is teaching material and our own reading of standard practice, not advice for your specific setup. Check anything important with your own specialist.

Lead scoring model

Start from the three behaviours you would act on

Points without validation are decoration

Decay matters more than the weights

The score is a queue, not a verdict

Want us to look at yours? It costs nothing.