How systems suggest what you might want.
WHAT THEY DO
Predict what a user will find relevant, based on behaviour.
THE MAIN APPROACHES
- Collaborative filtering: people similar to you liked this
- Content-based: this resembles things you liked
- Hybrid: both combined
WHERE YOU ENCOUNTER THEM
Shopping sites Streaming services Social media feeds Search results Advertising
FOR A SMALL BUSINESS
Simple versions are achievable: related products, frequently bought together, recently viewed.
Most shop platforms include basic recommendation features.
WHAT THEY REQUIRE
Data about behaviour. A shop with few orders has little to learn from.
THE COLD START PROBLEM
New products and new customers have no history, so recommendations are poor until data accumulates.
THE RISK
Optimising for engagement is not the same as optimising for value.
A system recommending whatever produces clicks may not serve the customer or, ultimately, the business.
WHAT TO MEASURE
Whether recommendations lead to satisfied purchases, not just to clicks.
