Use case assessment
Is AI lead scoring worth building?
Usually rules wearing a model. And the data underneath is normally the actual problem.
Most lead scoring runs on company size, industry and behaviour, all of which are rules. The model appears when you have enough closed-won history to learn from, and most companies do not.
What it usually means
Ranking inbound leads so sales works the promising ones first.
Is it genuinely AI?
Rarely, in the form it is usually proposed. Company size, industry, job title, pages visited, demo requested — these are attributes, and weighting attributes is a rules exercise your sales team can articulate in an hour.
A model genuinely helps when you have enough closed-won and closed-lost history for patterns to emerge that nobody has noticed. The threshold for that is higher than most companies assume: a few hundred deals is not a training set, it is an anecdote.
The problem underneath
Lead scoring projects usually surface a data problem rather than solve a ranking one. Duplicate records, enrichment that is two years stale, opportunity outcomes recorded inconsistently because the field was never mandatory.
A model trained on that learns your record-keeping habits. It will be confident about them.
What the simpler version looks like
Ask three salespeople to describe the leads they wish they got more of. The overlap is your scoring model. Write it as rules, run it for a quarter, and measure whether high-scored leads actually close better.
If they do not, you have learned something important before spending anything. If they do, you now have a labelled baseline that a model would need to beat.
What it costs to run
Data hygiene, continuously. Retraining as your market shifts. And a feedback loop that requires sales to record outcomes consistently, which is an organisational cost rather than a technical one and the reason these systems decay.
Where it sits under the EU AI Act
B2B lead scoring is not an Annex III high-risk use.
Two edges worth knowing. Creditworthiness assessment of natural persons is Annex III, so scoring that influences credit terms for consumers is different. And scoring on personal data has a GDPR profiling dimension independent of the AI Act.
When it is worth building
When you have thousands of recorded outcomes, clean data, and a rules baseline that has stopped improving. When lead volume is high enough that prioritisation order genuinely changes revenue.
Not when sales are slow. That is usually a product, pricing or positioning problem, and lead scoring is a comfortable place to look instead.
Common questions
Does lead scoring need AI?
Usually not. Company size, industry, title and behaviour are attributes, and weighting attributes is a rules exercise your sales team can describe in an hour. A model helps when you have enough closed-won history for unnoticed patterns to emerge.
How much data do you need for a lead scoring model?
More than most companies have. A few hundred deals is an anecdote rather than a training set, and a model trained on sparse, inconsistently recorded outcomes learns your record-keeping habits rather than your market.
Is lead scoring regulated under the EU AI Act?
B2B lead scoring is not an Annex III high-risk use. Creditworthiness assessment of natural persons is, so consumer scoring that influences credit terms is different. Scoring on personal data also has a GDPR profiling dimension independent of the AI Act.
Scoring 40 AI use cases in a regulated enterprise — the question that separated genuine AI from weighted rules.