Walmart Sparky: What the Spend and Basket Size Numbers Actually Say About In App AI Agents

Most retail AI coverage this year has been long on capability demos and short on actual numbers. Walmart has gone the other direction with Sparky, its in app shopping assistant, with executives naming specific percentages in public remarks across three separate quarters. That specificity is unusual and worth taking seriously. It is also company reported, attributed to named executives in earnings commentary rather than an independently audited study, and treating those two facts, real specificity and no independent audit, as both true at once is the right way to actually use these numbers.

What Sparky does today, and how it got here

Sparky launched in June 2025 as a feature focused on summarizing and answering questions about customer reviews, offering product recommendations, and helping shoppers plan purchases, built on retail specific large language models with task agents handling comparisons and personalization according to Walmart's own description. By 2026 the stated scope had expanded toward proactive behavior, EVP Daniel Danker has described a version of Sparky that can, on autopilot, automatically restock a pantry or fridge based on consumption patterns, and toward cross platform presence, with Danker noting in March that shopping can now start in a ChatGPT conversation as much as inside Walmart's own app. The trajectory matters for reading the numbers correctly: the spend and basket figures below describe an assistant that has been getting more capable and more proactive over the measurement period, not a static feature.

The numbers that matter, and exactly who said them

CEO John Furner said in April that Walmart app users who interact with Sparky spend about 35 percent more than those who do not, and that roughly half of all Walmart app users have now interacted with it. US CEO David Guggina, speaking June 9, gave more granular figures for the first fiscal quarter, ended April 30: Sparky attributed value of goods sold rose 150 percent year over year, users build baskets 35 percent larger than non-users, units sold through Sparky rose roughly fourfold year over year driven partly by consumables and food replenishment, and weekly active users climbed more than 110 percent quarter over quarter. Guggina framed the strategic logic plainly: "We connect AI-driven shopping directly to our omni-channel fulfillment network," tying the assistant's growth directly to Walmart's existing delivery and pickup infrastructure rather than treating it as a standalone chat feature.

What these figures actually support, and what they do not

A 35 percent spend lift among Sparky users compared to non-users is a real, large, repeatedly cited number, but it is a correlation between two groups, not a controlled experiment isolating Sparky's causal effect. Shoppers who choose to engage with an in app AI assistant are plausibly already more engaged with the Walmart app generally, more frequent shoppers, more likely to use other app features, which would produce some amount of spend difference independent of Sparky itself. Walmart has not published methodology showing it controlled for that selection effect, and no outside party has audited the underlying data. None of that means the numbers are wrong. It means a 35 percent lift attributed entirely to Sparky's influence is a stronger claim than the public data actually supports, and the more defensible read is that Sparky usage correlates strongly with higher value shopping behavior, which is still a meaningful signal for anyone evaluating whether an in app assistant is worth building, just a narrower one than the headline number implies on its own.

Why the fulfillment network framing is the detail worth copying

The part of Walmart's approach most transferable to a mid-market retailer is not the specific percentage lifts, it is Guggina's framing of Sparky as connected directly to omni-channel fulfillment rather than existing as an isolated chat feature layered on top of an unchanged backend. An AI shopping assistant that can see real-time inventory, delivery windows, and pickup availability can make commitments a purely conversational layer cannot, confirm a restock will actually arrive before the pantry runs out, which is closer to what Danker describes as the "autopilot" version of Sparky. We have covered the broader shift toward agent to agent commerce protocols that Walmart is also participating in, and Sparky's fulfillment integration is the same underlying principle applied to a first party assistant rather than a third party agent: the assistant's usefulness scales with how much real operational data it can actually act on, not with how naturally it can converse.

What a mid-market retailer should actually take from this

Walmart's scale, fulfillment network, and existing first party shopping data are not reproducible by a mid-market retailer building an equivalent assistant from scratch, and treating Sparky's specific percentage lifts as a template to replicate misreads what made them possible. The transferable lesson is narrower and more useful: an AI shopping assistant's commercial value appears concentrated in tying recommendations to real inventory and fulfillment commitments, not in conversational quality alone, and measuring any pilot program against basket size and repeat purchase rate for engaged users, the same metrics Walmart is citing, rather than against engagement or session count, which are easier numbers to move but weaker evidence of actual commercial impact.

FAQ

Are Walmart's Sparky numbers independently verified? No. All the figures cited here come from named Walmart executives in public remarks and earnings commentary, not from an independent audit or academic study, which is a meaningful caveat even though the figures are specific and consistently repeated across quarters.

Does the 35 percent spend lift mean Sparky caused that spending increase? Not definitively. The comparison is between app users who engage with Sparky and those who do not, which could partly reflect that more engaged shoppers are both more likely to use Sparky and more likely to spend more, independent of Sparky's specific influence.

What underlying technology does Sparky run on? Walmart has described it as built on retail specific large language models with task agents handling comparisons, personalization, and completing shopping journeys, without naming a specific foundation model provider in its public statements.

Is Sparky available outside the Walmart app? Yes, Walmart has described shopping starting inside third party AI assistants like ChatGPT in addition to its own app, part of the broader move toward agent to agent commerce rather than keeping AI shopping confined to Walmart's own interface.

When this applies to your stack

If you are evaluating whether an AI shopping assistant is worth building for your own retail or DTC platform, treat Walmart's percentage lifts as a directional signal rather than a benchmark to hit, and focus any pilot's measurement on basket size and repeat purchase rate for engaged users rather than raw engagement metrics, since that is the evidence actually worth collecting. If your platform already has real-time inventory and fulfillment data, prioritize connecting an assistant to that data before investing further in conversational polish, since Walmart's own framing suggests the fulfillment integration, not the chat quality, is doing the commercial work. Contra Collective helps retail and DTC teams scope AI shopping assistant pilots around metrics that actually predict commercial impact rather than ones that are simply easy to report.