Retailers have every reason to be interested in AI. A shopping assistant that answers questions at any hour, compares products instantly and offers personalised recommendations sounds like a win for customers and businesses alike.
But the technology can be impressive while the experience is disappointing.
Consider a customer who asks an AI assistant whether a laptop is suitable for university. The tool responds confidently and immediately, but overlooks a crucial software requirement. Or a supermarket assistant proposes substitutions that ignore a dietary preference. These are hypothetical examples, but they illustrate how an AI interaction can appear efficient while failing the customer's actual needs.
In Australia, Bunnings' Buddy and Woolworths' Olive show how conversational shopping is entering mainstream retail. Amazon's Alexa for Shopping , introduced in beta to selected Australian customers, can help users research products and compare options. Walmart's Sparky offers similar assistance in the United States. These are examples of what retailers say their systems are designed to do, not proof that every shopper has a good experience.
The stakes are higher as AI shifts from answering questions to taking action. A mistaken recommendation is inconvenient; a mistaken purchase or an unwanted disclosure of personal information could be more serious.
That gap between technological capability and customer confidence motivated our research, published in the Journal of Retailing. Across eight studies, we developed and tested a new way of assessing how customers accept AI-enabled retail services.
We found that acceptance has eight distinct dimensions: personalisation, accuracy, safety, speed, convenience, humanisation, friendliness and emotionality. In everyday terms, customers care whether AI provides a useful, reliable and manageable experience, and whether the interaction feels appropriate rather than awkward or impersonal.
An AI tool might perform brilliantly on speed but poorly on accuracy. It might sound cheerful yet leave users questioning what happens to their personal data. And a very humanlike interaction is not necessarily the right solution for every customer or task.
Our analyses found that higher overall acceptance was associated with greater word of mouth, share of usage, willingness to pay and likelihood of adopting AI services. The scale performed better than several established alternatives in the samples tested. This matters because retailers need to know more than whether a new tool is being clicked. They need to understand why customers do or do not accept it.
There is a cautionary example from the broader service industry. In 2024, McDonald's ended a US trial of AI-powered drive-through ordering after reporting mixed results, amid customer reports of incorrect orders. Those reports do not establish the sole reason the pilot ended, nor was McDonald's examined in our study. But they illustrate why reliability deserves attention alongside innovation.
Retail managers can act on this now. Before rolling out a shopping assistant widely, they can ask customers to rate its accuracy, safety, speed and other qualities. They can compare an existing version with a redesigned one. They can investigate whether an apparently minor problem, such as confusing replies about returns, is undermining an otherwise useful service.
Our research offers a 32-item measure for detailed diagnosis and an eight-item condensed version for quicker assessments. The point is not to burden shoppers with questionnaires. It is to give businesses evidence about where improvement is needed.
Of course, the eight dimensions do not establish a universal recipe. Our research tests acceptance and its predictive value; it does not show that fixing any single design flaw will cause a particular rise in sales. Priorities will also vary between a grocery assistant and an AI tool helping with expensive purchases.
A retailer also needs to recognise that customers notice different failures at different moments. A misleading product detail during browsing, a poorly handled complaint and a privacy concern during checkout are not the same problem. Measuring those experiences separately gives managers a better chance of finding the underlying weakness before it becomes part of the brand's reputation.
The shopping revolution will not succeed simply because retailers adopt AI quickly. It will succeed to the extent that customers find these services accurate, safe and worth returning to. Moving too fast without measuring that experience is where the backfire risk begins.
