Product recommendations account for up to 35% of revenue on Amazon. For most other ecommerce sites, it is 5-15%. The gap is about recommendation quality. Most engines are stuck showing customers variations of what they already looked at.
Why Basic Recommendations Fail
Recently viewed is not a recommendation. It is a browser history with better styling.
Category-based suggestions are too broad. You bought a laptop, so here are more laptops. You do not need another laptop. You need a bag, USB hub, or monitor.
Best-sellers are generic. Showing everyone the same top sellers is a popularity contest, not personalization.
What Good Systems Do
Collaborative filtering. People who bought X also bought Y. The most powerful signal. Needs 50,000+ orders to work well.
Sequential pattern mining. After buying a camera, next purchase is typically a memory card (1 day), then camera bag (1 week), then lens (1 month). Timing recommendations to purchase journey improves relevance.
Context-aware recommendations. On the product page: complementary items. In the cart: accessories. On the homepage: profile-based. In post-purchase email: logical next purchase.
Intent detection. High-intent shoppers (adding to cart, comparing, reading reviews) benefit from decision-helping recommendations. Low-intent browsers benefit from discovery recommendations.
Placement Matters
- Cart page: Complementary accessories. Highest purchase intent moment.
- Product page: Complementary products and alternatives at different price points.
- Post-purchase: Logical next purchases.
- Homepage: Personalized for returning visitors, trending for new visitors.
Measuring Performance
Click-through rate, add-to-cart rate, revenue attribution (10-30% of total for good engines), and AOV lift (10-15%).
See our ecommerce and retail industry page.