Syntha Lab RU

Market basket affinity: support, confidence, lift

From the number of receipts with each item and with both at once, counts for pairs of items the support (how often the pair occurs), the confidence (what share of those who bought A also took B) and the lift (how many times more often the pair occurs than with an independent choice). Ranks the pairs by lift among those with enough data and suggests what to do: a bundle, adjacent placement or nothing.

Store sales and receipts: traffic, conversion, ticket For whom: Category managers, merchandisers and analysts who look for items bought together for bundles and placement.

Calculate on your data →

At a glance

What question does it answer?
Which pairs of items occur in receipts more often than by chance and what to do about it.
What data is needed?
The number of receipts in the period.
How to read the result?
Rank by lift among pairs with enough support; confidence shows the direction, lift shows the strength of the link. Example on the default values: Best pair by lift: Dress + Belt; Support of the best pair: 4%; Confidence A → B: 26.7%; Confidence B → A: 33.3%. The figures are illustrative: enter your own to see your result.

Why and how to reason

If items are bought together more often than by chance, it is the basis for a bundle or adjacent placement. Support shows the frequency of a pair, confidence the direction, lift the strength of the link. In the example 10,000 receipts, A in 1,500, B in 1,200, both in 400: support 4.0%, confidence A → B 26.7%, lift 2.22, leverage 0.022.

When to use it

Apply when looking for bundles and planning placement and cross-selling.

How to read the result

Rank by lift among pairs with enough support; a lift above one means a link but not a cause.

Common mistakes

They take pairs with little support and confuse lift with confidence.

What data is needed?

  • The number of receipts in the period.
  • For pairs: receipts with A, with B and with both.
  • The minimum support for conclusions and your lift threshold.
  • An example: 10,000 receipts, A in 1,500, B in 1,200, both in 400: support 4.0%, confidence A → B 26.7%, lift 2.22, leverage 0.022 (illustrative numbers).

Inputs

FieldUnitValue in the exampleMeaning
Total receipts in the periodreceipts10,000All receipts and not only those with the analyzed items. The example value is illustrative.
Minimum support of a pair for conclusions%0.5Pairs rarer than this share are not taken into the ranking: there is little data on them. The example value is illustrative.
Lift from which to suggest a bundle×1.2Your own decision threshold; there are no common norms. The example value is illustrative.

How it is calculated

StepFormula
Supportreceipts with A and B ÷ all receipts
Confidencesupport of the pair ÷ support of A
Liftsupport of the pair ÷ (support of A × support of B)
Leveragesupport of the pair − support of A × support of B

Worked numbers

Default values from the app; in the interactive version you enter your own.

Best pair by lift
Dress + Belt
Support of the best pair
4 %Receipts with both ÷ all receipts.
Confidence A → B
26.7 %What share of those who bought A also took B.
Confidence B → A
33.3 %
Lift of the best pair
2.22×How many times more often than by chance.
Leverage of the best pair
0.022
Pairs with enough support
5
Pairs with lift at or above the threshold
4
Lift of pairs
Bar chart: Dress + Belt 2.22, Sneakers + Socks 1.92, Jacket + Hat 1.9, Jeans + T-shirt 1.22, T-shirt + Bag 0.7200.511.522.5Dress + BeltDress + BeltLift: 2.222.22Sneakers + SocksSneakers + SocksLift: 1.921.92Jacket + HatJacket + HatLift: 1.91.9Jeans + T-shirtJeans + T-shirtLift: 1.221.22T-shirt + BagT-shirt + BagLift: 0.720.72
Data as a table
MeasureLift
Dress + Belt2.22
Sneakers + Socks1.92
Jacket + Hat1.9
Jeans + T-shirt1.22
T-shirt + Bag0.72

Pairs by lift

PairSupportA → BB → ALiftWhat to do
Dress + Belt4 %26.7 %33.3 %2.22bundle or adjacent placement
Sneakers + Socks4.5 %34.6 %25 %1.92bundle or adjacent placement
Jacket + Hat1.2 %13.3 %17.1 %1.9bundle or adjacent placement
Jeans + T-shirt7 %31.8 %26.9 %1.22bundle or adjacent placement
T-shirt + Bag1.5 %5.8 %18.8 %0.72bought together less often than by chance: do not combine

1 pair(s) with support below 0.5 % are not ranked: on rare pairs the lift is large by chance.

A lift above one means the pair occurs more often than with an independent choice but does not prove that one item causes the purchase of the other. Do not confuse lift and confidence: confidence depends on the popularity of B.

Receipt data becomes personal when tied to a customer: use aggregates.

What to do with the result: test the bundle or adjacent placement in part of the shops and compare with a control; do not combine pairs with a lift below one.

A typical mistake: taking pairs at support below 0.5% and confusing lift with confidence.

Nuances and limits

  • Taking pairs at support below 0.5%.
  • Confusing lift and confidence.
  • Taking lift for a cause: it shows co-occurrence and not influence.

Terms used

Calculate on your data →