Cross-Shopping Behavior: What Mixed Baskets Reveal About Retail Media Opportunity

A basket containing pasta, tomato sauce and parmesan looks like an obvious cross-shopping story. A basket containing laundry detergent and bananas does not. Both are co-purchases, but only one immediately suggests a useful occasion, placement or partnership.

That difference is the job of cross-shopping analysis. It asks which products, brands or categories share baskets, how directional the relationship is, whether the overlap is stronger than normal popularity would predict, and what decision the relationship can improve.

For retail media, the opportunity is not simply to find items that appear together. It is to identify a credible route from one shopping signal to another commercial action, then test whether media changed the outcome.

What cross-shopping behavior means

Cross-shopping behavior is the observed overlap between two defined products, brands or categories across eligible transactions or shoppers. The unit of analysis matters. Same-basket overlap captures products bought on one trip. Shopper-level overlap captures purchases made across several trips. Session-level overlap may include browsing, cart and purchase events.

These views answer different questions. Same-basket analysis is useful for adjacency, bundles and in-session media. Multi-trip analysis can support audience planning or category partnerships. Browsing overlap can reveal consideration, but it should not be reported as purchase behavior.

A clean definition names:

• the two entities being compared;
• the population of eligible baskets or shoppers;
• the period, stores and channels included;
• whether the relationship is same-basket, same-session or multi-trip;
• how returns, canceled orders, substitutions and missing identifiers are handled.

Google Analytics ecommerce documentation illustrates the event structure needed for this work: item arrays can carry product identifiers, brand, category, price and quantity through cart, checkout, purchase and refund events.[3] Retailers still need a consistent taxonomy if a product is to be compared across systems.

Direction matters more than a single overlap percentage

Suppose many salsa baskets also contain tortilla chips. That does not mean the reverse rate is equally high. Chips may appear across lunches, parties and standalone snacking, while salsa may be bought mainly for a narrower set of occasions.

Report both directions:

• A-to-B attachment: baskets containing A and B divided by baskets containing A.
• B-to-A attachment: baskets containing A and B divided by baskets containing B.
• Joint incidence: baskets containing both divided by all eligible baskets.
• Lift: P(A and B) / (P(A) x P(B)), where each probability uses the same eligible-basket population.

Directional rates help planners decide which item should provide the audience signal and which should receive the message. Lift helps prevent a popular product from looking strategically special merely because it appears in many baskets.

Lift above 1 indicates positive association, not causality. Before activation, require the joint-basket count as well as a stability check, such as a confidence interval, repeated-period result or predeclared minimum count. Low-frequency items can produce unstable rates, while promotions, store layout, availability and pack formats can create temporary patterns.

Complements, substitutes and coincidence

Cross-shopping relationships are not one type of relationship.

Complements help complete the same task, such as pasta and sauce. Substitutes satisfy a similar need, such as two brands in the same category. Portfolio companions may be bought together because one brand serves several parts of an occasion. Coincidental pairs share baskets because both are popular or because large stock-up trips contain many unrelated items.

Research on product relationships shows why labels need care. Work using basket and browsing data treats complementarity as something to infer rather than assume from one co-purchase count.[4] A recent commerce preprint from Allegro also warns that co-purchase data is noisy and may produce unexpected or inappropriate recommendations without category constraints and human review.[5]

The practical rule is simple: co-occurrence creates a candidate relationship. Taxonomy, business context, availability and review determine whether it is usable.

This boundary separates cross-shopping from shopping missions. A mission explains the trip or occasion. Cross-shopping shows which purchases cluster inside that context. The two become stronger when used together, but neither proves the other.

Turn relationships into media decisions

A useful cross-shopping finding should change a specific decision. Use the same compact screen each time: relationship type, directional rates, lift, joint count, availability and the campaign outcome to test.

Audience expansion

A brand can reach buyers of an adjacent category when the relationship is stable, relevant and large enough to activate. The brief should state why the adjacent category is a credible signal and whether current brand buyers are included or suppressed.

Contextual placement

An onsite or in-store unit can appear near the trigger category. The placement should respect the active context. Showing a dinner complement near recipe ingredients may help. Repeating the same message after the basket is complete may add clutter.

Brand or category partnerships

Two teams can coordinate creative, funding or measurement when the relationship is mutual. Directional analysis is essential because one category may contribute far more incremental demand than the other.

Assortment and availability protection

Media should not accelerate a relationship that stores cannot fulfill. Check whether both sides of the pair are ranged, in stock and correctly mapped in the targeted stores.

Creative framing

The message can name the occasion or missing benefit rather than advertise an isolated SKU. This is where category affinity can support audience relevance, while the cross-shopping relationship supplies transaction context.

A worked planning example

Consider a sparkling-water brand that frequently appears with premium prepared meals. The first temptation is to target every prepared-meal buyer.

A better process starts with the denominator. Is the overlap strong among all prepared-meal baskets, or only in urban stores and evening trips? Does the relationship remain after removing promoted weeks? Is it driven by one meal subtype? Are both products available in the intended stores?

The campaign might then use prepared-meal browsing or basket addition as a signal for a clearly labeled sponsored reminder. The outcome is not clicks alone. The plan should track whether sparkling-water incidence increased in eligible meal baskets, whether the total meal basket changed, whether another beverage was displaced, and whether the effect persisted outside promotional pricing.

This turns an association into a testable commercial hypothesis.

Measurement: association is not impact

The IAB/MRC retail media guidelines call for transparent definitions, data sources and attribution methods.[1] For cross-shopping campaigns, report the eligible population, exposure rule, product mapping, availability filters and treatment of repeat transactions.

Compare at least three layers:

1. Delivery: Did the shopper qualify because of the defined trigger, and was the media actually served?
2. Behavior: Did the target product enter the basket, and what happened to substitutes, complements and total basket composition?
3. Incrementality: What would have happened without the treatment?

IAB and IAB Europe distinguish attributed outcomes from causal impact and emphasize credible counterfactuals, bias control and separation of signal from noise.[2] A strong attachment rate among exposed shoppers may simply mean the campaign targeted people who already bought the pair.

Use randomized holdouts where feasible. If a matched method is used, match on pre-period incidence, store, mission mix, promotion, availability and other factors that influence the relationship. Report percentage-point change for incidence and the underlying basket counts, not only a relative lift headline.

Common mistakes

Using one denominator. A-to-B and B-to-A rates can tell different commercial stories.

Calling every pair complementary. Same-basket presence can reflect substitution, popularity or large-trip noise.

Ignoring taxonomy. Duplicate SKUs, pack changes and inconsistent category mapping can manufacture or hide relationships.

Activating tiny patterns. A high lift from a small base may not be stable or scalable.

Forgetting availability. A cross-sell audience is useless when the promoted item is not ranged or in stock.

Reporting association as incrementality. The campaign needs a counterfactual, not only an exposed-basket comparison.

Buyer checklist

Before approving a cross-shopping tactic, ask:

• What exactly are A and B?
• Is the unit a basket, session or shopper history?
• What are both directional attachment rates?
• Is the relationship stronger than base popularity would predict?
• Is it stable across stores, periods and promotional conditions?
• Is the relationship a complement, substitute or unresolved candidate?
• Can the retailer fulfill both sides of the relationship?
• What treatment and control design will test incremental impact?
• What decision follows a strong, weak or inconclusive result?

FAQ

Is cross-shopping the same as market basket analysis?

No. Market basket analysis is a broader family of methods. Cross-shopping is a specific commercial question about overlap between defined products, brands or categories.

Does high lift mean two products should be bundled?

Not automatically. Lift indicates unusual co-occurrence. Margin, mission fit, availability, shopper value and operational feasibility still need review.

Can cross-shopping use multi-trip behavior?

Yes, but label it clearly. Multi-trip shopper overlap supports audience and partnership planning, while same-basket overlap supports adjacency and in-session decisions.

What is the best campaign outcome?

The outcome depends on the job. It may be incremental attachment, new-to-brand purchase, category expansion or a profitable partnership effect. Define it before activation.

Bottom line

Mixed baskets reveal opportunity only after the relationship is defined, normalized and challenged. Use direction, lift, taxonomy and business context to identify credible candidates. Then test whether media changed the basket rather than merely finding shoppers who already bought the pair.

Sources

[1] Interactive Advertising Bureau and Media Rating Council, IAB/MRC Retail Media Measurement Guidelines, 2024.

[2] IAB and IAB Europe, Guidelines for Incremental Measurement in Commerce Media, 2025.

[3] Google Analytics, Measure ecommerce.

[4] arXiv, Inferring Complementary Products from Baskets and Browsing Sessions, 2018.

[5] Allegro research authors, Beyond Co-purchase Relation: Evolution of Complementary Recommendations at Allegro, September 2026 preprint.

Ready to see how this works in practice?

Footprints AI helps brands and retailers measure what matters. See our customer success stories or get in touch to discuss your retail media strategy.