Personalisation has an uncomfortable asymmetry: the retailer captures the benefit immediately, while the customer experiences the cost — in surveillance, in price opacity, in the nagging sense that the shop knows too much.
The retailers growing share with AI have closed that loop by making the trade visible. Explicit preference controls that genuinely change what is shown. Loyalty value that is legible in currency, not points. Clear statements of what data is used for, and a real option to withhold it without losing service quality.
Underneath, the operating change is assortment and supply. Personalisation at the shelf is only as good as the ability to have the right item in the right place; models that shape demand without touching availability simply move disappointment around.
The measurement discipline matters too. Most retail AI programmes report uplift against a holdout that was never truly randomised, which inflates the number and destroys trust when the finance team audits it. Running clean experiments is slower and it is the only way the benefit survives scrutiny.
Personalisation is not a technology problem any more. It is a permission problem, and permission is earned in the interface.
