Why Albertsons’ higher AI basket values do not prove ROI

Albertsons reports a result that should interest retail executives. Customers who use its standard conversational search have about 10% higher average order value (AOV), while customers using more comprehensive assistants for recipes, ingredients, and dietary preferences have about 26% higher AOV. The difference is commercially interesting because the richer experience addresses more of the shopping mission. But the observed association alone cannot establish that AI caused customers to spend more.

Bigger AI baskets are a commercial signal

Jill Pavlovich, Albertsons’ Senior Vice President of Digital Shopping Experiences, attributes the figures to customer use of the company’s conversational tools. She said, “We see anywhere from a 10% lift in average order value when they [customers] use standard conversational searching and about a 26% lift in average order value when they use more comprehensive assistants to find their recipes, find the ingredients that match their dietary preferences.” Albertsons has a commercial interest in showing that its AI investments improve customer and financial outcomes. Executives should treat the reported association as evidence to investigate rather than an independent causal finding.

The distinction matters for investment decisions. Customers who choose conversational tools could already be more engaged digital shoppers or could arrive with shopping missions that naturally produce bigger baskets. A customer planning several recipes, for example, has both a reason to use a comprehensive assistant and a reason to buy more products. Grouping customers by AI use cannot distinguish those explanations from spending caused by the AI experience.

The stronger business question is incremental value: spending or profit that would not otherwise have occurred. Usage shows whether customers engage with an experience, while AOV among users shows the value of orders placed by that group. Establishing incremental value requires a credible comparison with what similar customers or shopping missions would have produced without the experience. That turns an interesting association into a testable investment case.

Broader shopping missions may create more value

The difference between Albertsons’ two reported outcomes supports a useful hypothesis. Standard conversational search helps customers discover products, while the more comprehensive assistant also helps them work through recipes, ingredients, and dietary preferences. The richer experience participates in more decisions involved in building a grocery basket. Its larger reported AOV association is consistent with the possibility that handling more of the shopping mission creates more commercial value.

Consider two customer requests. Finding “pasta” is mainly an item-discovery task, while planning a meal around dietary preferences requires choosing several compatible products. The second task gives software more opportunities to affect which products customers consider and eventually buy. The breadth of the customer problem may therefore matter more than the conversational interface itself.

The same reasoning creates a confounder. Complex shopping missions can produce larger orders because they contain more requirements, and those requirements can also make a comprehensive assistant more useful. The customer’s underlying mission could therefore explain part of the association between richer assistance and higher AOV. The reported difference is a reason to test comprehensive assistants more carefully rather than count the entire observed uplift as AI-generated return.

For CEOs and CTOs, this becomes a practical product question. A conversational experience can reduce friction in finding products, but a system involved in building a basket must interpret customer intent and connect it to products that satisfy the request. A recipe constrained by dietary preferences provides a concrete test: the assistant must identify relevant ingredients and products while keeping the combined basket useful. As the system influences more decisions, rigorous economic measurement becomes more important.

Incrementality requires a counterfactual

A counterfactual is a credible estimate of what the same type of customer would have done without the AI experience. Controlled tests can create that comparison by exposing comparable customers or shopping missions to different experiences and measuring conversion, basket value, margin, and repeat behavior. For a recipe assistant, the comparison should account for mission complexity so that customers planning larger meals do not automatically make the AI group look more valuable. Channel and timing can matter for the same reason.

Profitability requires another layer of measurement. A larger basket can contain a different product mix, use different promotions, or create different fulfilment costs, so incremental AOV and incremental profit are separate outcomes. Executives evaluating an AI investment should define the economic measure before running the test. The measurement design should match the financial decision the company expects the system to improve.

Longer observation can test whether an apparent gain represents additional demand or a shift between sessions, channels, or purchasing periods. A customer might use an assistant to consolidate products into one order that would otherwise have been purchased separately. That change could still have operational value, but it should not automatically count as incremental revenue. Attribution needs to distinguish changed shopping behavior from genuinely additional economic value.

This discipline separates three questions that management can otherwise collapse into one. Adoption asks whether customers use the assistant. AOV among users describes the commercial value of orders placed by those customers. Incrementality asks what value deploying the assistant caused relative to a credible alternative.

Conversational commerce becomes an operating-model question

If a retailer wants an assistant to handle more of a grocery mission, the system must work with the business information that shapes the answer. A dietary request depends on accurate product attributes, while a purchasable recommendation also depends on current availability and fulfilment options. Making systems “agent ready” means giving an AI agent structured access to the business information needed to complete a customer task. In this setting, data quality directly limits which shopping decisions the assistant can handle reliably.

Albertsons identifies four areas of AI investment: digital customer experiences, merchandising intelligence, staff empowerment and supply-chain optimisation. Taken as Albertsons’ stated investment scope, these areas show how customer-facing assistance can intersect with decisions elsewhere in a retailer. An assistant can recommend a product, while merchandising and supply-chain systems influence whether that product is suitable, available, and practical to fulfil.

The connection changes the organizational requirement. A customer-facing team may measure assistant use and conversion, while merchandising teams manage product economics and supply-chain teams manage availability and inventory. When one AI-enabled shopping flow touches those functions, management needs a measurement model that traces gains and costs across them. Otherwise, a strong customer-interface metric can hide weaker economics elsewhere in the transaction.

The implementation burden also grows as the assistant takes responsibility for more of the basket. Product attributes must support the requests customers actually make, and availability information must prevent recommendations that cannot be purchased. Pricing and fulfilment information must remain consistent with the transaction the retailer can execute. A richer conversational experience therefore tests the retailer’s underlying commercial systems as much as its interface.

Main highlights

• Bigger AI baskets require proof of incrementality: Albertsons reports 10% higher AOV for conversational search users and 26% for users of more comprehensive assistants, but those associations do not establish causation. Leaders should use controlled comparisons before treating higher AOV as AI-generated return.
• Broader shopping missions may create more value: Assistants that support recipes, ingredients and dietary preferences can influence more purchasing decisions than basic search. Retailers should test whether this broader role produces incremental value while controlling for customers who already have larger, more complex shopping missions.
• Measure profit against a credible counterfactual: Adoption and AOV among AI users are not substitutes for incremental financial impact. Tests should measure conversion, margin, fulfilment costs and repeat behavior against comparable non-AI experiences.
• Conversational commerce requires operational readiness: Richer assistants depend on reliable product attributes, availability, pricing and fulfilment data across multiple business functions. Leaders should align customer experience, merchandising and supply-chain systems so gains at the interface translate into sustainable economics.