Retail & CX metrics glossary
Basket analysis finds what customers buy together. Here is how to read it — and how to make it happen in a physical store.
Basket analysis — market basket analysis, in the classic data-mining vocabulary — is the technique of scanning your transactions to discover which products are bought together in the same purchase. It answers questions your category reports never ask: when someone buys running shoes, how often do socks go in the same bag? When a customer takes the drill, does the bit set follow? The raw material is nothing exotic: the receipt lines you already have in your POS.
The output is a list of associations — product A tends to appear with product B — each scored by three numbers with intimidating names and simple meanings: support, confidence and lift. Read correctly, those three numbers tell you which pairs are frequent enough to matter, how reliable the connection is, and whether it is a real relationship or just two popular products crossing paths. This page explains all three without heavy math, shows what to do with an association once you find one, and names the step that separates e-commerce from the store floor: online, an association becomes an automatic recommendation; in a physical store, the recommendation is a person.
Basket analysis (market basket analysis)
Co-purchase rate = (baskets with A and B ÷ baskets with A) × 100
baskets with A and B = transactions that contain both products in the same purchase · baskets with A = all transactions that contain product A · Co-purchase rate = this is the classic confidence of the association A → B: given A, the probability of B
Co-purchase (confidence) calculator
A sporting goods chain pulls one month of receipts. Running shoes appear in 1,000 baskets; in 240 of them, performance socks are also present. Confidence of the association shoes → socks = (240 ÷ 1,000) × 100 = 24%. To know if that is meaningful, check lift: socks appear in 8% of all baskets, so lift = 24% ÷ 8% = 3.0 — a shoe buyer is three times more likely than the average customer to take socks. That is a real association, and a suggestion the team should be making on every shoe sale.
What counts as a strong association?
There is no universal table for associations — they are specific to your assortment — but these reading thresholds and classic retail pairs are a useful map:
| Classic strong pairs | phone → case, razor → blades, paint → brushes, shoes → socks |
|---|---|
| Confidence above 30% | actionable — put the suggestion in the sales script and train it |
| Confidence 10–30% | worth testing — shelf adjacency, bundle/kit or a paired promotion |
| Lift above 1.5 | real association — the pair happens well beyond chance |
| Lift around 1.0 | coincidence — both products are simply popular on their own |
Thresholds are practical heuristics, not statistics textbook rules: the right cutoffs shift with assortment breadth and average basket size. Compute the numbers on your own transactions before acting.
Support, confidence and lift — without the heavy math
Support is simply how often the combination shows up at all: the share of all baskets that contain both A and B. If shoes + socks appear together in 240 of 10,000 monthly baskets, support is 2.4%. Support filters out curiosities — an association that happens four times a year may be fascinating and still not worth a planogram change.
Confidence is the direction question: of the baskets that contain A, what share also contain B? It reads as a conditional probability — given shoes, socks follow 24% of the time. Note that confidence is directional: shoes → socks and socks → shoes usually score differently, because the two products sell in different volumes.
Lift is the honesty check. It compares the confidence of A → B with how often B is bought anyway. Lift of 3.0 means the pair happens three times more than chance would predict; lift near 1.0 means nothing is going on — you have found two bestsellers, not a relationship. Any association you plan to act on should clear a lift meaningfully above 1.
What to do with an association once you find it
Shelf adjacency is the cheapest move: put B within arm's reach of A, so the basket completes itself. It works best for low-consideration complements — batteries near toys, bags near shoes — where seeing the product is enough to trigger the purchase.
Kits and paired promotions monetize the association directly: bundle A and B at a light discount, or trigger a coupon for B when A is scanned. Because you already know the pair co-occurs, the discount subsidizes a purchase that was likely anyway — so keep it small and watch margin, not just attach volume.
The highest-leverage move is the suggestion script: teach the team that whoever takes A should be offered B, with a concrete phrase and a concrete moment in the sale to say it. This is where the association stops being a report and becomes revenue — and, as the next section argues, it is also where most physical retailers silently lose the value of the whole analysis.
The blind spot: online the finding recommends itself. In the store, the recommender is a person.
In e-commerce, basket analysis closes its own loop: the association feeds a recommendation engine, the engine shows 'frequently bought together' on every product page, and the uplift is measured automatically. Discovery and execution are the same system.
In a physical store the loop breaks in the middle. The data can discover that shoe buyers take socks three times more than chance — but no shelf hears that insight. The only recommendation engine on the floor is the salesperson, and the association only produces revenue if it becomes a sentence spoken at the right moment of the conversation.
That execution layer is exactly what transaction data cannot see. Receipts tell you when the suggestion worked; they say nothing about the thousands of conversations where it was never made. Two stores with identical assortments and identical association tables can have wildly different attach performance — and the difference lives entirely in whether the conversation happened.
The data discovers the basket. The conversation builds it.
Cognifyze instruments the missing half of the loop: it captures the in-person sales interaction itself — with consent, without identifying any individual shopper — and shows whether the suggestions your basket analysis prescribes are actually being made on the floor: which complements are offered, at what moment, and what happens when they are. The association table tells you what to say; interaction measurement tells you whether it is being said.
In measured deployments, making the interaction visible moved same-store conversion from 51.5% to 79.5% (+28pp, p<0.001), with 383% ROI and payback in 1.4 months.
Find out whether the baskets your data discovered are being built in the store — book an executive diagnostic.
Related metrics and guides
Basket analysis — frequently asked questions
What is market basket analysis?
It is the analysis of your transaction records to find products that are bought together in the same purchase more often than chance would explain. Each association is scored by support (how frequent the combination is), confidence (given A, how likely B is) and lift (how far above coincidence the pair sits).
What is the difference between support, confidence and lift?
Support measures frequency: the share of all baskets containing the pair. Confidence measures direction: among baskets with A, the share that also has B. Lift measures surprise: how much more often the pair occurs than if the two products were independent. You need all three — frequent, reliable and above chance — before acting.
How many transactions do I need for basket analysis?
Enough for the pairs you care about to appear dozens of times, not a handful. As a practical floor, work with several thousand receipts per period and per cluster of similar stores; below that, confidence and lift swing wildly and you will chase pairs that were noise. Aggregate longer windows for slow categories.
How is basket analysis different from attach rate?
Attach rate tracks one pairing you already decided to push — accessories attached to a device, for example — and measures execution over time. Basket analysis is the discovery step: it scans all combinations to find which pairings deserve to become an attach target in the first place. Discovery feeds the target; attach rate monitors it.
Does basket analysis work in physical retail or only in e-commerce?
The analysis works identically — POS receipts are all it needs. What differs is execution: online the finding feeds an automatic recommendation engine; in a store it has to travel through planograms, kits and, above all, the sales conversation. The analysis is only as valuable as the store's ability to actually make the suggestion.