Frequently Bought Together

Frequently bought together (FBT) is the recommendation pattern that surfaces items co-occurring in real purchase histories — the camera with the memory card and case, mined rather than guessed. It is market-basket analysis turned storefront feature: Amazon's item-to-item collaborative filtering made it e-commerce's most familiar module, and its honesty is its power — the suggestion carries the social proof of ten thousand prior baskets.

The engine is co-occurrence statistics: which items appear in the same baskets beyond chance (the association-rule lineage — support, confidence, lift — and the item-to-item collaborative filtering Amazon published in 2003 that made it scale), with the craft in the thresholds (rare-but-real pairings versus popularity noise — everything 'pairs' with bestsellers unless lift corrects for it) and the catalog hygiene (variants deduplicated, accessories mapped, the complement-versus-substitute line held: FBT recommends completions, not alternatives — the substitute slot belongs to 'similar items'). The placement decisions price the pattern: product pages (the classic bundle module with one-click add-all — the CROSS-SELL-RATE machinery at its most native), cart and checkout (gap-aware additions sized to thresholds per the FREE-SHIPPING entry's nudges), and post-purchase (the order-confirmation add-on window the fulfillment timing allows). The measurement runs the recommendation family's honest stack: attach-rate per placement, AOV lift against holdout (the recommendation that displaces a planned purchase moved nothing — incrementality's question, in module form), and margin-weighted ranking where the business case justifies it (the highest-lift pairing isn't always the highest-margin one — the ranking decides whose interest the module serves). The cold-start and small-catalog boundaries are real: FBT needs basket volume; below it, rules-based pairing (the merchandiser's complements) and content-based methods carry until the data arrives.

FBT matters wherever catalogs have genuine complements and basket volume to mine — electronics, beauty routines, DIY projects, grocery — as the AOV lever with the least persuasion required: the data did the arguing. It matters most at the placement-and-ranking decisions where honesty meets margin. The discipline is lift-corrected mining, complements kept distinct from substitutes, holdout-measured incrementality per placement, and the ranking's interests declared — the module works because it tells the truth; rank it into a margin billboard and the truth-telling stops converting.

Market-basket analysis predates the web — the association-rule literature mined retail transactions in the early 1990s — and Amazon's item-to-item collaborative filtering (published 2003) industrialized it into the module every storefront now copies: the basket the data already built, offered back.

Etymology: source.

Search interest for this term over the last five years:

View interest-over-time on Google Trends →

Areas of marketing where frequently bought together is a core concern: