What a recommendation strategy is
A recommendation strategy is the logic the system uses to pick products for a specific widget. The
same widget container can run different strategies on different pages: personalized products on the
homepage, similar items on the product page, frequently bought together in the cart.
The choice of strategy moves widget metrics more than its design or its position on the page.
The main classes of strategy
| Class | Strategies | Where it performs |
|---|---|---|
| Behavioural | Personalized, Recently Viewed | Returning shoppers with history |
| Social | FBT, Trending, Bestsellers | New visitors, fallback |
| Contextual | Similar Items, Complete the Look, Cross-sell | On a PDP or in the cart, anchored to an item |
| Manual | Merchandising rules, editorial | Business priorities outside the data |
Strategies by page type
Homepage: personalized recommendations for returning visitors plus trending or bestsellers for
new ones. The goal is engagement and a route into the catalogue.
Product page: similar items plus frequently bought together. Two widgets covering two different
intents — I want something like this but different, and I want to complete this purchase.
Cart: cross-sell (what is bought alongside the items in the basket) plus upsell (a more expensive
version). The goal is average order value before payment.
Empty pages and 404s: trending or recently viewed, so a dead end does not cost you the session.
Combining the algorithm with merchandising rules
A pure algorithm optimises against historical data but knows nothing about business priorities.
Merchandising rules add overrides:
- Boost — raise the priority of specific items or categories (promotions, new arrivals)
- Filter — exclude items or categories (out of stock, a competing brand)
- Pinning — fix a specific product in the first position regardless of the algorithm
Tip: do not overuse manual overrides — they cost relevance. Merchandising rules should be the
exception that encodes a business priority, not a replacement for the algorithm.
How to choose and verify a strategy
- Define the widget’s job: retention (move the shopper to another item), cross-sell (add to the
purchase) or upsell (raise the order value). - Pick two candidate strategies.
- Run an A/B test with revenue per visitor or average order value as the primary metric.
- Read the result after two or more weeks, accounting for a 7–14 day attribution window.