A score is a prioritisation hypothesis
Lead scoring helps teams decide where to focus when demand exceeds immediate capacity. It does not determine who deserves service or guarantee that a person will buy. The model should be understandable enough that sales can challenge it with evidence.
Separate fit signals, such as service need or supported market, from intent signals, such as a relevant request or repeat product evaluation. A good fit may not be ready now, while strong activity from an unsuitable account should not become a priority.
Use signals with a clear reason
Use a small number of reliable signals with documented weights or rules. Add negative signals for invalid data, incompatible needs or completed actions, and avoid proxy attributes that create unfair or discriminatory effects.
Define how scores change over time, what threshold triggers an action and what happens below it. Do not let an invisible number replace the original context that helps a person conduct a useful conversation.
- Define fit and intent separately.
- Use explainable, action-linked signals.
- Set decay, thresholds and override rules.
- Review outcomes and unintended bias.
Calibrate against outcomes and fairness
Compare bands with accepted leads, opportunities and outcomes, not only engagement. Review false positives and false negatives with sales, then adjust the model when behaviour, offer or data collection changes.
Monitor whether the model systematically deprioritises legitimate groups or rewards easily manipulated activity. For predictive scoring, document training data, drift, explainability and human override.