Background and challenge
L.E.K. was asked to design, build and deploy a pricing engine that could iteratively reach optimized price points and therefore maximize the lifetime value of our client’s product portfolio.
The client had a portfolio of 250 product categories across 200 channels, with the ability to price each combination independently. As a result, there were approximately 50,000 possible product-channel combinations to optimize.
Approach
L.E.K. performed a range of analytical assessments to support the client’s goals:
-
Built a robust pricing optimization model that would maximize lifetime value (LTV) of each of the approximately 50,000 product-channel pairs
-
Segmented product-channel pairs into archetypes that informed the optimized price points (e.g., high levels of customer demand, low price sensitivity, etc.)
-
Built a live model engine that queries millions of datapoints from the client’s data warehouse, and runs through a demand forecast model, a customer propensity (renewal) model, and lastly an optimization model
-
The demand forecasting model was based on a Bayesian time series that deployed across all approximately 50,000 product-channel combinations
-
The customer propensity model, which was a light gradient boosting model that used around 30 input variables, predicted the likelihood each customer would renew within about a 1% margin of error
-
Using data from the above models, L.E.K. created a pricing optimization engine that output the optimal price and promotion for each of the nearly 50,000 product-channel combinations
-
-
Built-in advanced machine-learning capabilities, so that the model becomes increasingly smart over time, and can recommend increasingly significant price increases as it gains in confidence
-
Designed and built a front-end dashboard to intuitively display pricing information and outputs for the client’s staff
Results
-
Within 6 months of go-live, the model has informed an average price increase of about 9% across product-channels
-
Client uses the pricing optimization engine model on a weekly basis to re-assess optimal pricing and promotion across its approximately 50,000 possible product-channel combinations
10242023151000