Walmart Says Sparky Makes Customers Spend 35% More. Nobody Knows Why.

Walmart has a number it wants you to hear: customers who use its AI shopping agent Sparky spend 35% more, on average, than customers who do not. The figure comes from Walmart CEO John Furner on the company's Q1 FY2027 earnings call on May 21, and it is being reported as proof that the agent works.

Here is what we know. The 35% AOV figure is real. Weekly active users of Sparky grew more than 100% in the same quarter. Units purchased through Sparky more than quadrupled versus the prior fiscal period. The agent's intelligence and response quality improved 40% year-to-date. It now speaks Spanish and works in stores for automatic repeat reorders. These are specific, quantified claims traceable to Walmart's earnings call transcript, corroborated across multiple independent sources including Digital Commerce 360 and Sahm Capital.

Here is what we do not know: why those numbers are what they are.

Walmart presents the 35% AOV premium as evidence that Sparky helps customers find what they want. That is one hypothesis. Another is that Sparky's recommendations systematically favor higher-margin items — private-label brands, products with higher advertising fees, SKUs that optimize Walmart's basket economics. A third is that Sparky's engaged, higher-intent users would have spent more regardless of what the algorithm recommended. The earnings transcript does not distinguish between these explanations. It offers no breakdown of category mix, no control group analysis, no disclosure of whether the recommendation engine weights product margin in its ranking logic.

The honest answer is that the mechanism behind the 35% lift is unknowable from public data. Walmart has not published the analysis that would separate algorithmic effect from customer selection. Outside reporters cannot run that analysis without access to Walmart's transaction logs and recommendation engine logs. What we can observe is the architecture Walmart has built — and that architecture has both the incentive and the infrastructure to shape shopping outcomes in ways that favor the retailer.

The hypothesis that Sparky steers toward higher-margin goods is plausible — and it is also possible that engaged shoppers who use an AI tool simply spend more regardless of what they are recommended. Without a control group or a published breakdown of category mix, Walmart's earnings call does not distinguish between those two effects. The 35% figure is real. The mechanism is unknown.

What would answer the question is straightforward in theory: run the same shopping sessions through Sparky and through conventional search. Log every recommended product, every price point, every brand surfaced. Look for a consistent pattern of higher-priced alternatives, private-label items, or products with higher advertising placement fees. That analysis has not been published.

One piece of context makes the question more urgent. Walmart is not running one AI agent — it is running four. Sparky sits between consumers and product recommendations. Marty sits between commercial partners — brands, suppliers, marketplace sellers — and the algorithmic layer that determines what those consumers see. Brands cannot negotiate with Sparky directly. They go through Marty, an AI intermediary that manages their relationship with Walmart's recommendation infrastructure. Epinium, a retail analytics firm, documented that brands completing AI-assisted content enrichment on their top Walmart SKUs see catalogue quality scores rise by an average of 38 points — and those scores feed directly into shopping agent recommendation ranking. The intermediary is not neutral. The content signals that determine visibility are partly defined by Walmart.

Code3, a retail media agency, noted that Walmart is already testing a ad format called "Sponsored Prompt" inside Sparky — ads that appear inside the AI shopping experience and trigger an AI-generated answer plus a click-to-buy product. Brands that want visibility in Sparky's recommendations are already being offered a paid path to get it. That is consistent with an agent optimizing for Walmart's revenue, not only the customer's interest.

The 35% AOV figure will be repeated across earnings coverage this week. It is a real number from a real earnings call. Whether it reflects genuine customer value or algorithmic basket engineering is a question the transcript does not answer — and one that matters to every brand on Walmart's marketplace.

By Q4 FY2026, roughly half of all Walmart app users had interacted with Sparky, according to Constellation Research. At that penetration level, what Sparky recommends and why is no longer a product feature. It is infrastructure. And infrastructure that nobody has audited is infrastructure that serves its operator's interests first.