LG Electronics Customer value modeling
Put attention where it matters.
A customer value system that separates purchase likelihood from expected spend, helping teams prioritize outreach and plan audiences with a clearer view of value.
Model design, evaluation, and production delivery within the broader customer intelligence product.
Production customer model
Who is ready to buy, and where is the value?
Separate purchase-readiness and expected-value signals, supporting different decisions for reach, budget allocation, and higher-value outreach.
Conceptual workflowThe business problem
A large customer file is not a contact strategy. Teams need to distinguish customers likely to buy from customers likely to spend more, and decide where limited attention and budget can be most useful.
What I built
Developed separate predictions for purchase propensity and spend conditional on a purchase, combining them into an interpretable near-term expected-value signal.
Evaluated both components independently: whether probabilities match observed buying behavior, and whether spend estimates distinguish higher-value purchases.
Connected customer value with category preferences to support audience prioritization, cross-sell planning, and retention decisions.
How I evaluated it
Rebuilt historical predictions with only information available at the time, then compared them with later purchases. Checked buyer concentration, calibration across probability bands, and predicted versus observed spend.
Built to operate
Recurring customer scoring with automated publication checks. Model replacement gates compare candidate performance before promotion, with scheduled retraining prepared for ongoing operation.
What the work enables
Historical evaluation concentrated more future buyers in priority audiences than simple recency and untargeted baselines, while validating probability calibration and spend estimates separately.