How many points should each signal be worth?

Start with a simple rubric that anyone can explain — big signals worth thirty, medium ten, small five — then adjust by looking at which signals the leads that converted actually had. The exact numbers matter less than whether the top band converts better than the middle.

Does lead scoring need machine learning?

Not at first, and often not ever. A rules-based score that is explainable and updated monthly from conversion data serves most teams. Machine learning helps when there are thousands of leads a month and many signals; before that, it mostly adds opacity.

What should happen to low-scoring leads?

A nurture track rather than silence: useful content, a periodic check-in, and a rescoring when their behaviour changes. Many of them are right-company, wrong-time; the score should rise when the timing signals appear, and the rep should be told.

← Blog·Inbound & speed-to-lead

Lead scoring that sales actually uses: explainable, current and tied to routing

Scores that nobody trusts are ignored. The scoring model that combines fit and behaviour, explains itself in one line, updates as signals arrive, and changes who gets called first.

12 Jan 2026 · 2 min read · by HeiSyn

Lead scoring has a reputation problem: a number appears on the record, nobody knows where it came from, and reps sort by recency instead. A score is only useful when it changes behaviour, and it only changes behaviour when the rep believes it. Belief comes from explanations, freshness and visible results.

Two scores, not one

Fit: how closely the company and role match the ICP — size band, industry, role seniority, geography, technology. Behaviour: what they have done — pages viewed, content downloaded, emails opened, trial actions, replies. Keep them separate on the record. A high-fit, low-behaviour lead needs outreach; a low-fit, high-behaviour lead needs a polite redirect. One combined number hides that difference.

Explain every score in one line

Next to the number, the reasons: '+30 mid-market, +20 director-level, +25 visited pricing twice this week, −15 no company website'. A rep who can read the reasons will trust the ranking and will also tell you when a reason is wrong, which is how the model improves.

Decay and refresh

Behaviour points should fade: a pricing visit last week matters, one last quarter does not. Recompute nightly at least, and on the event for the high-value signals. A score that was right in March and never updated is worse than no score.

Wire it to action

The score should decide something: who is called first, which sequence they enter, whether an SDR or an AE takes them. Report conversion by score band monthly. If the top band does not convert better than the middle, the model is wrong and should be changed — visibly, with a note to the team.

Questions people ask next

How many points should each signal be worth?: Start with a simple rubric that anyone can explain — big signals worth thirty, medium ten, small five — then adjust by looking at which signals the leads that converted actually had. The exact numbers matter less than whether the top band converts better than the middle.
Does lead scoring need machine learning?: Not at first, and often not ever. A rules-based score that is explainable and updated monthly from conversion data serves most teams. Machine learning helps when there are thousands of leads a month and many signals; before that, it mostly adds opacity.
What should happen to low-scoring leads?: A nurture track rather than silence: useful content, a periodic check-in, and a rescoring when their behaviour changes. Many of them are right-company, wrong-time; the score should rise when the timing signals appear, and the rep should be told.

lead scoringinboundCRM

Want this built for your team?

A free scoping call: we map your motion, name the number that should move, and quote a fixed-scope build.

blog/lead-scoring-that-sales-actually-uses