The AI question every ASX retail board will ask

American retailers spent their second-quarter earnings calls talking about artificial intelligence. Retail Dive's roundup on 9 September 2026 has Ulta's chief executive describing the company as "in the early stages" of applying AI, Kohl's reporting stronger conversion and higher revenue per visit from its AI shopping assistant, and Dollar General's chief executive talking about agentic operating systems for enterprise workflows. Bain's global head of retail, Aaron Cheris, notes that most are overhauling one part of the business rather than transforming the whole of it.

Then read what those same companies actually filed. Dollar General's second-quarter release of 27 August 2026 reports net sales of $11.3 billion, up 5.2 per cent, same-store sales up 3.5 per cent and diluted EPS of $2.48. Its capital expenditure disclosure for the 26 weeks shows $31 million on information systems upgrades and technology-related projects against $168 million on distribution and transportation projects. The only mention of AI in the document is a forward-looking risk about competitors using it effectively. Gap's release of the same date guides to approximately $650 million of capital expenditure for fiscal 2026 covering stores, distribution centres and technology together, on net sales of $3.7 billion, down 2 per cent. Its AI mention also sits in the forward-looking statements.

Why it matters here

ASX annual general meeting season starts next month, and Australian retail boards are about to face the same question American ones did. Wesfarmers has already shown what the answer looks like when a group takes it seriously: its 2026 full-year results name three AI assistants, Buddy, Joy and Ollie, and a shared customer data asset of approximately 12 million records, with digital sales penetration of 7.6 per cent at Bunnings and 10.5 per cent at Kmart Group. The loyalty numbers are filed with figures attached. The assistants are named without any.

The take

My view is that the gap between the call and the filing is the whole story, and the discipline worth adopting is simple: say about AI only what you would be willing to put in a results release with a number next to it.

That is not a counsel of silence. It is a counsel of specificity. The reason the earnings-call language is so vague is that it is doing a different job from the filing. A call answers an analyst who wants to know the company has a plan; a release carries legal weight and gets read by anyone who cares to. So the AI story lives where nobody has to prove anything, while the capital expenditure line, where they do, shows $31 million into information systems and five times that into trucks and warehouses. That is not dishonesty. It is a company being precise where precision is required and expansive where it is not, and the expansive half is the half that gets quoted.

The risk for Australian boards is copying the expansive half. An AI narrative assembled for an AGM, with no number behind it, is a liability twelve months later when someone asks what it delivered. Worse, it distorts internal priorities: teams start building the thing that sounds good in the chairman's address rather than the thing that moves a metric the company already reports. If a retailer cannot name the metric an AI project is supposed to move and state its baseline, the project is not ready to be announced.

The better answer to the board is narrower and more defensible. Here is the one process we have changed, here is the measure it affects, here is where it stood before and where it stands now, and here is what we have not attempted yet and why. That answer is harder to write and much harder to argue with.

What to watch

Watch which Australian retailers put an AI figure in a filed document this AGM season rather than in a speech, because that is the group that has actually measured something. And if you are preparing board material now, run the test on your own draft: strike every AI sentence that has no number in it, and see what is left. Whatever survives is the part worth saying.