Case Study · Predictive Analytics

Lead Scoring Model for the Automotive Industry

A machine learning model that scores every inbound sales lead by how likely it is to buy, so sales teams call the hottest leads first and can see why each lead scored as it did.

← All case studies

Client

Multi-brand automotive dealer group

Industry

Automotive

Delivers

Every lead ranked by likelihood to buy

The problem

Dealers receive leads from their website, OEM portals, marketplaces, social campaigns and walk-ins. Sales teams treat them all alike, so ready-to-buy customers wait while time goes on browsers. Speed-to-lead matters most for the leads that are hardest to spot.

Data and features

Lead attributes: source, campaign, vehicle of interest, new or used, and budget or finance enquiry.
Digital behaviour: vehicle detail page views, configurator and payment-calculator use, return visits and test-drive requests.
Relationship history: previous purchases, service visits, trade-in vehicle age and mileage.
Engagement: response time, replies and appointment bookings.
Label: whether the lead became a sale within 90 days.

Modelling approach

Gradient-boosted classifier with calibrated probabilities, so a score of 0.7 really does mean about a 70% chance of buying.
Time-based validation, where the model is trained on past months and tested on later ones, to avoid over-optimistic results.
Evaluated on AUC, lift in the top decile and calibration, not accuracy alone.
SHAP explanations give each lead a plain-language reason, for example "used payment calculator 3× and booked a test drive".

Integration and operations

Scores and reasons are written back to the CRM or DMS in near real time.
Routing rules send top-tier leads to senior sales staff with immediate alerts.
Monthly retraining and drift monitoring as models, seasons and campaigns change.