Basket Completion: How Retail Media Helps Shoppers Finish the Mission

A shopper adds pasta and tomato sauce to an online cart. A useful completion system may suggest parmesan, vegetables or a drink. A weak system may show another pasta, an unavailable product or a sponsored item that does not fit the meal.

Basket completion is the decision process behind that difference. It uses a partial basket and current context to identify what could make the shopping mission more complete. For retail media, it creates a high-intent placement near the transaction. It also creates a trust risk if commercial pressure overwhelms relevance.

The goal is not to make every basket bigger. It is to help the shopper finish the task while giving retailers and brands a measurable, clearly labeled way to influence an open decision.

What basket completion means

Basket completion starts with a non-empty basket during an active trip or session. The system infers one or more likely intentions, generates candidate additions, removes ineligible products, ranks the remaining options and presents a small set at an appropriate moment.

Research describes within-basket recommendation as completing a non-empty basket during the shopping session.[6] The same work notes that real baskets can contain several intentions and that shoppers can switch between them. A shopper buying breakfast food may also be buying cleaning supplies. Treating the whole basket as one mission can produce irrelevant recommendations.

Completed-basket analysis explains what transactions contained. Basket completion makes a different decision: what, if anything, to recommend while a non-empty basket is still open.

Start with the active mission, not the inventory

A completion system should ask which part of the basket is still unresolved.

Signals can include:

• products already added and their categories;
• sequence and timing of additions;
• search and browse context;
• store, channel and fulfillment method;
• known availability and substitution rules;
• recipe, occasion or list context supplied by the shopper;
• repeated purchase patterns when legitimately available.

The mission signal should remain revisable. A shopper can remove an item, change quantities or start a second task. Shopping-mission planning provides the context; basket completion handles the next-item or next-set decision.

Avoid sensitive or speculative inference. The fact that products co-occurred historically does not authorize a system to infer health, family or other private conditions. Use the minimum data needed for the defined shopping task.

Candidate generation: find plausible missing items

Candidate generation creates a broad set before ranking. Sources may include:

• frequently bought-together relationships;
• curated category or recipe rules;
• compatible accessory mappings;
• shopper replenishment history;
• products viewed in the current session;
• campaign-eligible sponsored products.

Google Cloud describes Frequently Bought Together as a shopping-cart expansion model and separates it from related-items and substitute models.[4] That distinction matters. A substitute may help when an item is unavailable, but it does not complete the basket in the same way as a complement.

Co-purchase is only a starting signal. Research using basket and browsing sessions shows that complement relationships can be learned from more than transaction pairs.[8] A current Allegro preprint argues that raw co-purchase is noisy and that category constraints and human review help prevent unexpected pairings.[9]

The candidate set should then pass hard eligibility rules:

• in stock and available for the chosen fulfillment method;
• compatible with the selected product or task;
• allowed in the shopper’s market and account context;
• not already present in an equivalent form;
• within campaign dates and budget rules;
• not suppressed by frequency, category or policy controls.

Media cannot rescue an ineligible recommendation.

One item or a coherent set

Some missions are completed by one product. Others require a set. A shopper buying ingredients for a meal may need several components. A device purchase may need a compatible case and power accessory.

Set-to-set basket-completion research frames this as recommending a coherent set rather than predicting one next item.[7] The practical implication is that the system should consider interactions among recommended items. Three individually plausible products can form a repetitive, expensive or contradictory set.

Use set-level constraints:

• avoid recommending multiple substitutes for the same role;
• cap the number and total value of additions;
• maintain compatibility across the set;
• follow explicit filters, declared preferences and constraints evidenced by the current basket or a shopper-provided list;
• diversify only when diversity improves the mission;
• explain the relationship in plain language.

Where retail media can enter

Retail media can support basket completion in several ways.

Sponsored complement

A paid product can enter the candidate set only after hard eligibility, safety, availability and policy gates. Shopper relevance and compatibility determine which eligible products deserve consideration. A commercial bid or sponsorship score may rank products only within that qualified set; it must not rescue an irrelevant or incompatible item. Label paid placement clearly and log the product’s organic rank so analysts can audit displacement and incrementality.

Sponsored bundle

A brand or partnership can fund a coherent set for a defined occasion. The retailer still needs to verify availability, price logic and whether each component is useful.

Contextual reminder

An onsite or in-store message can remind the shopper of a missing role without naming one product. This can be useful when several brands are eligible or when the final choice should remain open.

Offsite preparation

Offsite media can shape a planned list before the basket exists. Once the basket is active, onsite and in-store channels normally have stronger context.

Compare this with recommendation AI, which covers the broader recommendation system. Basket completion is the narrower partial-basket use case.

A worked example

A shopper adds burger patties and buns. Historical data links those items with cheese, sauces, salad and drinks.

A weak system ranks by global co-purchase and shows the most popular products. A stronger system separates meal components from unrelated stock-up items, checks what is already present and removes unavailable products. It may follow a vegetarian path only when the shopper selected vegetarian products, filters or substitutions in the current session; it should not infer diet from unrelated history.

A sponsored cheese can be eligible if it fits the basket, is available and is labeled. The measurement plan should compare eligible sessions with and without the sponsored treatment. It should track cheese attachment, total meal completion, substitution from other cheese, basket value, abandonment and returns.

If cheese attachment rises but abandonment also rises because the module interrupts checkout, the tactic is not automatically successful. Shopper utility belongs in the outcome set.

Measure useful completion, not only clicks

Google Cloud recommends using purchase-complete events with revenue and quantity fields for revenue and order metrics, and its attribution tokens identify interactions served by the recommendation system.[5] Google Analytics also supports cart, checkout, purchase and refund events at item level.[3]

Predeclare one primary outcome before launch. A useful default is incremental attachment of the eligible missing role per eligible session, with the role set and eligibility rule fixed in advance. Then report:

• eligible baskets and served modules;
• product or role attachment;
• completion of the predefined mission set;
• incremental units, categories and value;
• substitution or cannibalization;
• checkout progression, abandonment and removal;
• availability failures and suppressed recommendations;
• repeat satisfaction signals where appropriate.

Treat basket value, abandonment, substitution, removal and returns as secondary outcomes or guardrails unless one is explicitly chosen as the primary estimand. Do not assemble a mission-completion composite after seeing which components improved.

The IAB/MRC guidelines require transparent outcome definitions and attribution methods.[1] Clearly separate organic recommendations, sponsored recommendations and other campaign exposures.

Attribution still does not prove impact. IAB and IAB Europe call for a credible counterfactual when estimating incrementality.[2] Randomize recommendation eligibility or ranking treatment where feasible. Avoid comparing people who clicked a recommendation with people who ignored it, because the groups differ in intent.

Failure modes and safeguards

Mission collapse. The system treats every basket item as one task. Use clusters, sequence and shopper-provided context.

Substitute shown as complement. Maintain relationship types and compatibility rules.

Popularity bias. Global bestsellers crowd out more relevant or local products. Normalize for base demand and include assortment constraints.

Commercial override. Sponsorship pushes an item that fails the shopper’s requirements. Relevance and eligibility must be hard gates.

Out-of-stock recommendation. Use fresh availability and fulfillment data.

Repetition. The same prompt follows the shopper through cart and checkout. Apply stage-aware frequency controls.

Basket inflation as the only objective. Track abandonment, returns, removals and shopper value as well as revenue.

Causal overclaim. Use holdouts and report the eligible denominator.

Buyer checklist

Before approving a completion placement, ask:

• What partial-basket state triggers the system?
• How are multiple missions separated?
• Which relationship types generate candidates?
• Which eligibility rules cannot be overridden by sponsorship?
• How fresh are inventory and price inputs?
• Is the recommendation one item or a coherent set?
• How is paid influence disclosed?
• What is the primary completion outcome?
• What shopper-friction and substitution metrics are monitored?
• What counterfactual supports the incremental claim?

FAQ

Is basket completion the same as frequently bought together?

No. Frequently bought together is one candidate source. Basket completion also needs mission inference, eligibility, compatibility, ranking, placement and measurement.

Should every basket receive a recommendation?

No. Some baskets are complete, some intentions are unclear and some contexts make another prompt intrusive.

Can sponsored products appear?

Yes, if paid influence is clearly labeled and the product passes the same shopper-relevance, compatibility and availability gates.

What is the best metric?

Use the outcome tied to the job, such as incremental attachment of a missing role, plus basket value, substitution and friction measures. Click-through alone is weak evidence.

Bottom line

Basket completion is a decision system, not a list of popular pairs. Infer the active mission, generate plausible candidates, enforce eligibility, distinguish complements from substitutes and measure useful incremental additions. The best completion placement helps the shopper first and monetizes that help transparently.

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] Google Cloud, *About recommendation models*.

[5] Google Cloud, *View analytics*.

[6] arXiv, *Within-basket Recommendation via Neural Pattern Associator*, 2024.

[7] arXiv, *S2SRec2: Set-to-Set Recommendation for Basket Completion with Recipe*, 2025.

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

[9] 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.