The job and the place in the funnel
The Similar items block exists for retention: if the current product does not fit on price,
availability, colour or specification, alternatives on the same page give the shopper a reason to
stay rather than go back to the search engine or to a competitor.
The main placements:
- Product page — under the description or in the sidebar
- Out of stock — an automatic substitution when the item cannot be bought
- Search page — when the query produced no exact match
- Category — while scrolling among comparable items
Selection algorithms
Content-based (attribute similarity). The system compares item attributes: category, brand,
price band (usually within 20–30%), technical specifications. It needs no historical data, which is
why it handles new arrivals and low-traffic categories well.
Collaborative filtering. People who viewed X most often also viewed Y. This captures
non-obvious relationships — that certain sneaker models are perceived as alternatives even when
their attributes diverge. It requires a sufficient volume of behavioural data.
Hybrid. Modern recommendation engines combine both and rank the candidates through a reranking
model that accounts for personal context.
Tip: in fashion, attribute similarity (style, price, brand) often matters more than behavioural
similarity. In electronics it is the reverse: shoppers look for functional equivalents that are
far apart on formal attributes.
Merchandising and control over the block
A purely algorithmic block can recommend items that do not suit the business — low margin, excess
stock, or conversely items in short supply. Merchandising rules allow:
- Boost — raising the position of priority SKUs (partner products, promotions)
- Exclusion — removing specific SKUs or brands from recommendations
- Availability filter — never showing out-of-stock positions
Balancing algorithmic relevance against business goals is the central configuration question for
this block.