Sketches / Causal ML

Dynamic Pricing

Faster pricing cycles, tighter inventory pacing, a 15x improvement in ML processing time, and a reusable way to talk about treatment effects where randomized prices are rarely available.

A catalog this large cannot be priced by hand, and a naive demand model will happily recommend prices that look smart in a notebook and fail on the floor. Observational price and quantity data are confounded: popular titles get different prices for reasons that also affect demand. If you regress quantity on price, you often estimate the merchandising policy, not elasticity. The business needed prices that moved with elasticity, velocity, seasonality, and stock — without reckless discounting.

Python · Causal DML · Causal Forests · Clustering · Optimization · MLOps