What belongs together — and what comes next?
01 / THE APPROACH
Understand support, confidence, and lift
Support is the share of eligible orders containing both products. Confidence from A to B is the share of orders containing A that also contain B; the reverse direction can be different. Lift divides that confidence by the overall share of orders containing B. Lift above one indicates a positive association in the observed order population, not a proven causal relationship.
02 / THE APPROACH
Check the size behind the relationship
A very high lift based on two shared orders can be less commercially useful than a moderate lift supported by hundreds. Inspect shared-order counts, each product’s total order count, and the chosen date range. Product popularity, existing bundles, promotions, and catalogue structure can explain an association. Volume thresholds are screening rules, not statistical certainty.
03 / THE APPROACH
Separate same-basket and next-purchase opportunities
For sequential cross-sell, connect qualifying orders for identified customers and exclude items purchased in the same order. Define a forward window, such as 30 or 90 days. Customers need enough observation time to complete that window. Report unique customers converting to the destination product and the delay to purchase, rather than treating every repeated pair as a separate converting customer.
Before you act on the analysis
- Count each product once per basket for pair presence.
- Report both directional confidences and shared-order volume.
- Keep same-order affinity separate from sequential purchases.
- Test recommendations against a baseline and measure the result.
WHAT WE’RE BUILDING
From the question to a useful next step.
Retail Science Co is developing same-order product pairs and sequential purchase relationships. The planned report surfaces relationships with sufficient volume and a clear commercial question to test.
Join the early-access waitlist