Most AI programmes fail for an unglamorous reason: they begin with the technology rather than with a decision that is currently being made badly. Pilots multiply, dashboards appear, and nothing in the operating rhythm of the business changes.
The retailers making real gains have been narrow and patient. They chose three decisions, what to buy, what to price, what to hold, and improved the quality and speed of those decisions with data the organisation already had. That is unfashionable work. It is also where the margin lives.
Slow AI has a second virtue: it protects trust. In luxury and premium, the customer relationship is the asset. Automating the parts of that relationship a human should own is the fastest way to spend goodwill you took decades to earn. Use the machine on the inventory, the forecast, the allocation. Keep the human on the client.
Governance matters more than models. Who owns the decision, who can override it, what happens when it is wrong, settle that before the first deployment, or the organisation will quietly ignore the output.
The measure of a good AI programme is not how many use cases are live. It is whether the leadership team now makes fewer, better, faster decisions than it did a year ago.