Predictive Lead Scoring

Overview

Predictive Lead Scoring is a method used to analyze lead behavior in historical customer data to find patterns resulting in a positive business outcome, such as a closed deal with a client. In this study, I developed a lead scoring model using the Bank Marketing dataset, which contains the outcome of clients subscribing to a term deposit or not based on a direct marketing campaign performed by a Portuguese bank.

Model Design

I assumed (Losing a potential customer cost) > (Sales Resource Cost) as a business objective as evaluating model outcomes, which statistically translates to developing a model that gives Low False Negatives and High True Positives, with balancing False Positives.

Model Prediction

To mimic a real-time model evaluation, I separated ~10,000 observation points from the dataset and trained on ~30,000 observation points. The following is the result of my trained LightGBM model on the hold-out dataset.

75.04% of Leads predicted by the model have resulted in conversion. And, a 29.56% False Positive rate, 24.95% False Negative Rate is observed.

Segmenting Leads based on Model Predictions:

Data Source

I used UCI Machine Learning Repository Bank Marketing dataset.