The challenge
Sales of some products were affected by changes to others. A new version of a product hurt sales of the old one; a price cut on a substitute hurt sales; a price cut on a main product lifted sales of its attachments.
The client wanted to identify substitute and attachment products, and to understand how price changes affect them so price models could be built at category or sub-category level.
What we did
- Used cross-price elasticity and attribute-based product classification to identify substitutes and attachments.
- Studied historical price changes to measure their effect on related products.
- Built category-level price models to measure the overall effect of a price change on one product.
The result
The client now makes better pricing decisions by understanding the effect on substitutes and attachments.
It also learned that pricing decisions should be measured at category level rather than product level.
How we would build it today
- Product similarity from catalogue text and images using embeddings, combined with basket affinity.
- Substitution and attachment scores fed straight into the pricing engine.
- Assortment simulations before a new product launch.
This describes our current toolkit for the same problem. It is not part of the original engagement.