Lead scoring is meant to tell sales which enquiries to call first. In many companies the score exists in the CRM and nobody looks at it. The usual reason is simple: the people expected to act on the number do not believe it, because they cannot see where it comes from and it has been wrong in front of them.
Why scores get ignored
A typical scoring setup adds points for many small actions. Ten for opening an email, five for visiting a page, twenty for downloading a document. After a few months a student researching a project has a higher score than a buyer who sent one precise enquiry. A salesperson calls three high scorers, finds no real interest, and stops trusting the system.
Separate fit from intent
Two different questions are hidden in one score. Fit asks whether this is the kind of customer you can serve: industry, company size, location, role of the person. Intent asks whether they are trying to buy now: they requested a quote, booked a call, visited the pricing page several times this week.
Keep the two apart. A good fit with no intent belongs in nurturing. Strong intent with poor fit deserves a polite, quick answer. Good fit and strong intent is the call to make today. One blended number hides exactly this distinction.
Start with rules people can read
Begin with five to ten rules, written in plain language and agreed with the sales team. For example: requested a demo, company in a target industry, more than fifty employees, used a business email address. Give each a weight that sales considers reasonable.
Next to the score, show the reasons. "High: requested pricing, target industry, returned three times this week." A salesperson who can read why a lead is rated highly can judge it in seconds, and will tell you when a rule is wrong. That feedback is what improves the model.
Let intent fade
Interest is temporary. A visit to the pricing page last month means little today. Reduce the weight of behaviour over time so that the score reflects current activity. Without this, old leads sit at the top of the list for ever.
Check against what really happened
Once a quarter, take the deals that were won and the leads that went nowhere, and look at what they had scored at the start. If won deals scored no higher than lost ones, the rules are not measuring anything useful. Find which signals the won deals shared, and adjust. This step is what turns a set of opinions into something that earns trust.
Where AI is useful
Two tasks suit it well. The first is reading the free text of an enquiry. "We need to replace our system before the contract ends in March" signals urgency that no checkbox captures. A language model can classify intent and extract details such as timeline and budget hints from that text.
The second is filling in missing company information so that fit rules have data to work with. In both cases the model supplies inputs to transparent rules. It does not replace them with a number nobody can explain.
Common mistakes
- Scoring email opens, which are unreliable since mail clients load images automatically.
- Too many rules, so that no single one can be evaluated.
- Building the model without the sales team and announcing it afterwards.
- Never revisiting the weights after launch.
Agree what the score triggers
A score matters only if something follows from it. Define it: high scores are contacted the same day by a named person, medium scores enter a follow-up sequence, low scores receive a helpful reply and no further effort. Then measure whether the high group converts better. If it does, trust follows.
Summary
Keep fit and intent separate, use a small set of readable rules, show the reasons, let old behaviour fade and test the score against real outcomes. Scoring is one part of our AI marketing automation service. If your team ignores its current scores, let us look at why.