Bayesian Cannibalization Analysis for New Product Launches

Modeling consumer choice

We built a model that represents how consumers choose between products on a shelf. Rather than looking at aggregate sales trends, the model captures the underlying decision: given everything available, which product does a shopper pick? Product availability varies by market and changes over time, so the model accounts for what's actually on shelf in each location rather than assuming universal distribution.

The model understands Colgate's brand architecture — the relationship between a parent brand, its sub-brands, and individual variants — which helps it distinguish between a new product that competes with its siblings versus one that expands into genuinely new territory. We also built in the distinction between minor tweaks, meaningful new variants, and true category entries, since the expected pattern of cannibalization differs for each.

Answering the counterfactual

The real payoff is counterfactual: for every product Colgate launched over the past five years, we can answer "if this had never existed, where would those sales have gone?" — with a full measure of uncertainty around each answer.