What it means
The name comes from the supermarket basket. Each completed transaction is treated as a basket of items, and the analysis scans thousands or millions of baskets for items that turn up together more often than chance would suggest.
A classic finding might be that customers who buy pasta also tend to buy a particular sauce. Three measures describe each pattern.
Support is the share of all baskets that contain the combination, confidence is the share of baskets containing the first item that also contain the second, and lift compares the actual frequency of the pair with what independence would predict. A lift above 1 suggests the items are bought together more often than random chance.
Finance and commercial teams care because the results change how money is made and spent. They guide which products to bundle, which to discount as loss leaders (items sold at a thin margin to draw customers in), and where to place stock.
They also help estimate the profit effect of a promotion beyond the single product being promoted. The technique is also used outside retail.
Banks apply the same logic to see which products customers hold together, such as a current account with a credit card, to target cross-selling. Insurers, telecom providers and software companies do the same with their own product sets.
A common caution is that association is not cause. Two products may appear together because both are bought for a holiday, not because one drives the other.
Results should be tested with small controlled trials before major changes are made. Data quality decides how useful the answer is.
Baskets need to be linked to a time period, a store or channel, and ideally a loyalty card, so that patterns can be compared across customer groups. Without that detail, the analysis tends to find only obvious pairs that staff already knew about, such as bread and butter.
In practice
Real-world examples
Example 1
A grocery chain finds that customers who buy barbecue charcoal are far more likely than average to buy firelighters and sausages. It places the three items together and runs a bundle price on summer weekends. The average basket value rises by $4.
Example 2
An online bookseller discovers that readers of a popular personal finance title frequently add a budgeting planner. It shows the planner as a recommendation at checkout. Revenue per order increases without extra advertising spend.
Example 3
A retail bank looks at which products customers hold together and finds that holders of a mortgage often also open a home insurance policy within three months. The marketing team times its insurance offers to the weeks after the mortgage completes.
Think of it
It is like a seasoned chef knowing that whenever someone orders a rich tomato pasta sauce, they are almost certain to want fresh garlic and grated parmesan cheese nearby.
Formula
Calculation
Support(A and B) = Baskets with both / Total baskets
Confidence(A to B) = Baskets with both / Baskets with A
Lift(A, B) = Support(A and B) / (Support(A) x Support(B))
Suppose a shop recorded 1,000 baskets. Of these, 200 contained bread, 100 contained butter and 50 contained both. Support = 50 / 1,000 = 5%. Confidence = 50 / 200 = 25%, meaning one in four bread buyers also bought butter. Lift = 0.05 / (0.20 x 0.10) = 0.05 / 0.02 = 2.5, so the pair appears together two and a half times as often as chance alone would predict.Case study
Seen in the real world
Greenacre Pantry is an illustrative, fictional chain of 30 neighbourhood food stores. The finance team analysed 480,000 baskets from one quarter to decide which products should carry the next promotion budget of $60,000. They found that fresh coffee beans had a lift of 3.1 with premium biscuits, while a heavily discounted breakfast cereal had a lift of only 1.0 with anything.
The cereal discount was costing about $15,000 in margin and pulling in shoppers who bought nothing else. The team moved the money to a coffee and biscuit bundle and displayed the two items on the same shelf. After two months, average basket value in the trial stores was $2.60 higher than in the control stores.
The illustrative lesson is that promoting products with strong partners can pay for itself, while promoting items that stand alone often just gives margin away.
Watch out
Common mistakes
- Treating a high lift as proof that one product causes the purchase of another, when both may be driven by a season or event.
- Ignoring the margin on each product, so the analysis recommends bundles that sell well but earn little.
- Using too few transactions, which makes rare combinations look more meaningful than they are.
Questions
People also ask
What is the difference between support, confidence and lift?
Support measures how common a combination is, confidence measures how often the second item follows the first, and lift compares the pair with what chance alone would produce.
Does market basket analysis only work for supermarkets?
No, it works wherever there are transactions containing several items, including banking products, software subscriptions and online services.
What does a lift below 1 mean?
It means the two items are bought together less often than chance would suggest, which can indicate substitutes that customers choose between.