The Pitch Walmart Is Selling
On Walmart's August 25 earnings call, CEO John Furner offered one of the cleanest AI-ROI numbers any retailer has put on the record this year: customers who use Sparky, Walmart's conversational shopping assistant, spend 40 percent more per order than customers who do not. Usage of the assistant grew 70 percent year over year, and it now sits inside an e-commerce business that grew 23 percent for the quarter. Furner framed it plainly: "We believe AI will improve nearly every part of our business by making shopping better and our associates' work easier." It is a tidy story, and Wall Street wanted one.
Sparky launched roughly a year ago and works by matching conversational intent and purchase history to recommendations, letting a shopper describe a need in plain language rather than typing search terms and filtering results. That is a real product improvement over keyword search, and Walmart is right to invest in it. The question is what the 40 percent figure is actually measuring. Nothing in Walmart's disclosure indicates a controlled comparison between similar shoppers who did and did not use the assistant, which is the difference between a marketing statistic and an ROI number a board can act on.
Where the Causation Argument Breaks
Data scientist Jose Prabhu Michael Singarayan flagged the obvious hole publicly after the earnings call: the 40 percent gap is correlation, not demonstrated causation. Shoppers who bother to open a conversational AI tool inside a retail app are already a self-selected group, more likely to be frequent Walmart customers, more comfortable with the app generally, and more inclined to buy in volume regardless of which discovery tool they used. Comparing their basket size to the average shopper's basket size does not isolate what Sparky itself contributed.
A defensible version of this claim would come from a randomized holdout, some qualifying shoppers get Sparky prompts and some do not, with basket size compared across matched cohorts over time. Walmart has the transaction data to run that test and has not published results from one. Until it does, 40 percent is a useful adoption signal and a weak capital allocation input, and enterprise leaders evaluating similar assistants for their own stacks should ask vendors for the holdout data before the adoption number.
A Rocky Quarter Behind the AI Story
The AI narrative arrived at a convenient moment. Walmart posted $187.9 billion in quarterly revenue and extended a streak of 25 straight quarters of growth, yet the stock fell 9 percent on August 20 after results missed some analyst expectations on margin and comp guidance. Market capitalization held around $836 billion, but the reaction made clear that top-line growth alone is no longer enough to satisfy investors who are pricing in AI-driven efficiency gains across retail.
That backdrop matters for how to read the Sparky numbers. When core retail metrics wobble, a clean AI engagement statistic becomes a useful offset in the earnings narrative, something concrete to point to while margin questions get worked through on the call. That is not evidence the number is wrong. It is a reason to treat any AI metric that surfaces during a soft quarter with slightly more scrutiny than one volunteered during a strong one.
Sparky Isn't Alone in the Room
Walmart is not the only retailer running this playbook. Amazon disclosed that Alexa for Shopping reached more than 350 million customers over the past year, with interactions up roughly fivefold year over year, and cited the identical framing that engaged users spend 40 percent more per order than those who do not use it. The repetition of that exact figure across two competitors, using two different assistants, is itself worth noting.
Target and Albertsons are telling adjacent versions of the same story. Target says digital traffic arriving from external AI platforms is growing more than three and a half times faster than its industry average, with AI-assisted wish-list creation volume up 50 percent heading into back-to-school. Albertsons reports a 10 percent average order value increase tied to conversational search and 26 percent basket growth among shoppers using its fuller AI assistant. None of the four retailers has published a causal study behind these figures.
The Metric Retailers Are Really Chasing
These numbers are not really being aimed at shoppers. They are aimed at boards and investors who are asking hard questions about the return on years of AI infrastructure spending, and adoption or engagement statistics are the fastest available answer. A 70 percent jump in assistant usage is an easy line for an earnings deck. A rigorous incrementality study takes quarters to run and might return a less flattering number, which is exactly why so few retailers have published one.
That gap between what is easy to report and what is actually true is not unique to Walmart, and it is not necessarily dishonest. It reflects an industry still early enough in agentic commerce that the measurement discipline has not caught up with the deployment pace. Furner's aspiration that AI improves "nearly every part of our business" may well prove out. The current evidence just does not yet distinguish that outcome from a simpler story about who was already going to spend more.
What This Means for Enterprise Tech Leaders
If you are building the business case for an AI shopping assistant, customer service agent, or any conversational tool that touches revenue, borrow the scrutiny now being applied to Walmart and apply it to your own numbers first. Insist on a holdout test before the metric goes into a board deck, and be explicit with stakeholders about the difference between engagement lift and demonstrated incremental revenue. The gap between those two claims is exactly where AI capital allocation decisions go wrong.
Watch what Walmart, Amazon, Target, and Albertsons do next rather than what they said this quarter. If any of them publishes a controlled study behind these figures, that is the signal the measurement discipline is catching up to the hype and a genuine competitive differentiator is emerging. Until then, treat every AI-driven basket size claim coming out of this earnings season, including your own vendor's, as a hypothesis rather than a result.
