Category adjacencies: what sells together

Nobody buys salsa by itself. Product adjacency analysis is the study of which items land in the same transaction, read from line-item point-of-sale (POS) data, and it exists to answer a plainly commercial question: what should sit near what? Co-purchase patterns are the closest thing retail has to hearing shoppers think out loud.

What is product adjacency analysis?

It's market basket analysis pointed at a merchandising decision. Market basket analysis examines the combinations of items that appear together in transactions; a basket, in this context, is simply everything one customer bought in one visit. Adjacency analysis takes those co-occurrence patterns and asks where products should physically live relative to each other.

The distinction matters. The basket pattern is data; the adjacency is a decision. Chips and salsa appearing together in transactions is an observation. Moving the salsa display next to the chip rack is a bet informed by it, and the bet still has to be tested.

What do you do with a pairing once you find it?

Three uses come up constantly at a brand manager's desk. Placement is the obvious one: if the data in a store set shows your energy drink frequently riding along with a protein bar, arguing for rack position near the cooler is a grounded ask rather than a hunch. Promotion is the second: a combined price on a pair the data already says travels together pushes on an open door. The sell-in story is the third, because a store owner or distributor hears "these two items appear in the same baskets in stores like yours" very differently from "please stock my product." None of the three requires certainty, only better odds than guessing, which is the standard merchandising decisions actually face.

How do you test an adjacency?

Cheaply and in a handful of stores, which is one of the independent channel's quiet advantages. Move the items near each other, hold price and stock steady, and read the pair's combined velocity against stores where nothing moved. Placement decisions in this channel take a conversation rather than a reset calendar, so the cost of learning is low.

Keep the read same-store and give it a few weeks, and read urban and residential clusters separately, since their trip missions differ. A pairing that only shows up during one hot weekend has told you about the weekend, not the shelf.

One more discipline: decide in advance what would count as success. A combined-velocity target written down before the move keeps the test honest, because after the fact almost any number can be argued into a win. Placement tests are cheap in this channel; self-deception is expensive everywhere.

Where does adjacency logic go wrong?

Co-occurrence isn't causation. Two items can share baskets because of the trip, not each other: coffee and a breakfast sandwich ride the same morning errand, and putting them side by side may add nothing the daypart wasn't already doing. Time of day, payday cycles, and weather all create co-purchase patterns with no merchandising meaning at all.

The popularity trap is the other classic. A store's best-selling items co-occur with everything, the way a busy intersection sees every car in town. The fix is to ask a sharper question: do these items appear together more often than their individual popularity predicts? That's the idea analysts call lift, and any pairing that can't clear it is noise wearing a pattern's clothes.

Small samples mislead too. One store's baskets over one week can make a coincidence look like a strategy. Aggregate across many stores and weeks before believing a pair, which is the kind of channel-wide view NRS Insights works from, with the monthly report archive as its public cadence.