Scandit survey finds retail AI is built on unreliable stock data
Retail stock information matches what store associates find on shelves only 46% of the time, according to global research released by Scandit. The resulting gap is pushing workers to rely on paper notes, chats, photos and other informal tools, leaving AI systems with an incomplete view of store operations.
- Stock records matched shelves 46% of the time.
- Only 8% of associates completely trusted store stock data.
- Associates estimated they could recover more than 4.4 hours each week.
- Inaccurate data could cost U.S. retail as much as $22 billion annually.
- Pilot participants reported a 74% rate of store tool use.
Unreliable records create a costly shadow store
Scandit commissioned Sapio Research to survey 2,200 store associates online in September 2026. The respondents worked in physical stores for retailers with more than 1,000 employees across the U.S., Germany, UK, France, Italy, Canada and Poland.
Only 8% of associates said they completely trust their store's stock information, while just one in four said it is accurate more than 70% of the time. When records are wrong, 49% of associates correct them themselves and 44% raise a stock adjustment. Among those who take action, 42% bypass the official system entirely, while 68% use at least one informal method.
Scandit describes the result as a “shadow store”, where important information exists outside the records used by head office, dashboards and AI systems. The informal methods identified in the survey include paper notes, team chats, personal phone photos, apps and memory.
Labor absorbs the cost of inaccurate inventory data
Associates estimated that they would each recover more than 4.4 hours a week if they had data they could trust. Scandit says a retailer with 10,000 store associates faces more than $41 million a year in labor costs tied to compensating for inaccurate information. That estimate is based on an average self reported $18 hourly wage.
Extrapolating the survey findings across the entire pool of retail associates in the U.S., inaccurate data could cost the industry as much as $22 billion annually, according to Scandit. Associates also reported spending an average of 4.1 hours each week handling customer questions that store technology should answer.
They were asked to physically check whether an item was in stock 5.85 times a week on average. More than one in five said they performed those checks more than ten times a week.
“Retailers do not have an AI adoption problem as much as a store truth problem,” Scandit CEO and co founder Samuel Mueller said.
Worker involvement was linked to higher tool use and trust
The research also examined how associates were involved in technology rollouts. Those who participated in a pilot reported using most store tools at a rate of 74%, compared with 61% among associates who were consulted before rollout and those who were never asked.
Half of pilot participants reported high trust in their stock data, compared with 25% of those who were never asked for input. Andrew Smith, Managing Partner and co founder of ThinkUncommon, said trust rises sharply only after accuracy passes 70%.
The findings suggest that deploying AI and other store technology depends on accurate shelf data and input from the workers who use the systems. For retailers, improving the underlying record could determine whether automated tools support store operations or add to the work required to verify what is actually available.
