Loan officers have limited hours. Lead scoring helps them spend those hours on the prospects most likely to become active members. Done well, it is simple, explainable and fair. Done badly, it is a black box nobody trusts.
Lead scoring is not credit scoring
A lead score answers “who should we contact first?”, not “should we lend to this person?”. Credit decisions still follow your credit policy, with proper assessment. Keeping the two separate avoids turning a marketing tool into an unregulated credit decision.
Use signals about the relationship
The most useful and least controversial signals describe the prospect’s relationship with your institution:
- Source: a member referral or walk-in usually converts better than a cold list.
- Consent: people who asked to hear from you respond more often.
- Defined need: a known product and amount means a real conversation.
- Engagement: recent calls, visits and replies.
- Recency: interest cools quickly; a week-old lead is worth more than a three-month-old one.
Leave out protected characteristics
Do not use gender, age, religion, ethnicity or tribe — not even indirectly. They raise fairness and legal concerns and add little that the relationship signals do not already capture.
Make every score explainable
Officers should be able to see why a lead scored 82 instead of 45, and what would raise it. A score that tells an officer “log a call to add engagement points” becomes a daily to-do list rather than a mystery.
Tune, then check
Start with sensible default weights. After a few months, compare scores with actual conversions. If referrals convert far better than other sources in your market, give the source factor more weight. Re-check regularly for unintended bias between branches or groups.
When to add machine learning
Once you have hundreds of converted and lost leads, a statistical model can find patterns the simple rules miss. Insist on explanations for each prediction (for example SHAP values) and on fairness testing before using it.