The 32% Smart-Cart Number Is a Segmentation Signal, Not a Business Case

Researchers at Bayes Business School analysed 12,418 shopping sessions from a one-month smart-cart pilot at a German supermarket and reported that shoppers who used the cart screens spent 32% more on average, bought 25% more items, and stayed 23% longer than non-users ^[1][2][3]. That 32% is already travelling as a procurement argument for in-cart screens, and it cannot hold that load: the authors describe the work as a descriptive field study, not a causal experiment, and shoppers decided for themselves whether to use the screen ^[4][5].

Start with the shape of the sample. CNET reported that 9,422 of the 12,418 sessions involved active screen use ^[6], which leaves 2,996 sessions in the comparison group, about 24% of the total ^[7]. So the headline compares a large self-selected group against a smaller residual group defined by declining to engage. Lead author Sabrina Gottschalk told CNET the analysis cannot establish causation, and that customers who opt into a smart-cart interface may already differ in goals, habits, or expected trip length ^[8]. A shopper doing a full weekly stock-up has more reason to open a digital list than someone grabbing milk. Some unknown share of the 32% is that difference, not the screen.

The internal pattern makes the point harder. The study separated out "superusers," shoppers with more than 20 screen interactions, and found they bought more items and stayed longer but did not spend more money than the average shopper ^[9]. If the interface were driving spend through recommendations and promotions, heavier exposure should push spend further, and it did not. Co-author Yusuf Oc told FoodNavigator that some heavy users may have been comparing products and prices rather than responding to promotions ^[10]. That is a plausible mechanism in which more engagement means more price scrutiny.

Then the arithmetic that a business case actually runs on. A 32% spend lift on a 25% item lift implies roughly 5.6% more spend per item ^[11]. The same 32% against 23% longer visits implies roughly 7.3% more spend per minute of dwell ^[12]. Those are the intensity numbers, and they are an order of magnitude less impressive than the figure in the press coverage. Most of the association is simply that screen users ran bigger, longer trips. Anyone modelling incremental margin against cart hardware, mounting, charging infrastructure, and support should assume the causal component sits somewhere below 32% and does not yet have a floor.

None of this makes the dataset useless. It establishes screen engagement as a segmentation signal, and it shows that interaction volume does not map cleanly onto revenue, which matters for anyone tempted to optimise a cart interface on engagement counts ^[13]. The dataset also tracked recommendation, navigation, and digital-list interactions alongside purchase and dwell-time outcomes, which is the raw material for a better test ^[6].

What to watch is whether any retailer publishes a controlled version. The measurement that would settle this separates selection from interface effects, randomises or otherwise controls recommendation and promotion exposure, and tracks substitutions, basket composition, trip completion, and customer trust alongside revenue ^[14]. Watch also for whether the superuser finding replicates, because a flat spend curve at high engagement is the cheapest available evidence that the screen is a tool shoppers use rather than a lever operators pull. The observation window here was a single month, March 2025, at one chain, per FoodNavigator ^[15], so seasonality and novelty effects are unresolved.