- Generic recommendation widgets underperform ones that use a customer's own behavior signals.
- Recommendation quality depends heavily on the underlying event tracking, not just the model.
- Personalization should extend across product pages, email, and search, not just one surface.
- Testing recommendation changes against real revenue per visitor, not just click rate, matters.
Generic widgets leave revenue on the table
A 'customers also bought' widget based on aggregate purchase data is better than nothing, but it ignores the specific individual's own browsing and purchase behavior. Personalization that actually uses an individual customer's signals consistently outperforms the generic version on revenue per visitor.
Recommendation quality depends on tracking, not just the model
A sophisticated recommendation model fed poor, incomplete behavioral data will underperform a simpler model fed rich, accurate event data. Getting the underlying tracking right is a prerequisite for personalization to actually work well, not an afterthought to add later.
Extend personalization beyond the product page
The product page recommendation widget is the obvious surface, but personalized email content and personalized search result ranking often have just as much, if not more, impact on revenue — and are commonly under-invested in relative to the on-page widget.
Measure against revenue per visitor, not clicks
A recommendation change that increases clicks but doesn't increase actual revenue per visitor hasn't necessarily improved anything meaningful. Testing personalization changes against the metric that matters to the business, not a proxy metric, avoids optimizing for the wrong outcome.