Recommendation Engines: Beyond "Customers Also Bought"
E-commerce
Most e-commerce sites have basic recommendations: "customers also bought" or "similar items." These are fine but leave a lot of value on the table. Better recommendations can meaningfully impact revenue.

The simplest approach is collaborative filtering—people who liked X also liked Y. It works but has limitations. New items have no interaction history (the cold start problem). Popular items get over-recommended. It can't explain why something is recommended.

Content-based filtering uses item attributes instead of user behavior. If you liked sci-fi movies, here are more sci-fi movies. This helps with cold start but tends to over-recommend similar items and miss unexpected connections.

Modern systems combine both approaches in hybrid models. They also incorporate context (time of day, device, current session behavior), business rules (promote high-margin items, don't recommend out-of-stock), and diversity (don't show ten versions of the same thing).

Real-time recommendations are increasingly important. What someone looked at two months ago is less relevant than what they looked at two minutes ago. Session-based models that update in real-time outperform batch-processed recommendations.

Measure the right things. Click-through rate is easy to track but can be gamed. Better metrics: did they buy? did they return? did they come back? Short-term engagement doesn't always correlate with long-term value.

Finally, recommendations should feel helpful, not pushy. There's a fine line between "you might also like" and "we're tracking everything you do." Transparency and user control matter.
Sarah Rodriguez

Written by

Sarah Rodriguez

AI Engineer at APPTAILOR

Share this article:

Questions about this?

Send us a message, we're happy to chat.

Get in touch