AI Personalised Pricing: What Shoppers Can and Cannot See
Dynamic and personalised pricing systems adjust offers per visitor, which raises a transparency problem stores need to manage openly. A plain guide to the practice and its trust cost.
Two shoppers open the same product and see different prices, different discounts, or different bundles. Sometimes that is a loyalty reward. Sometimes it is an experiment or a pricing model reacting to signals the shopper never knew were read.
Personalised pricing is spreading through AI tooling faster than the norms around it. This guide separates ordinary personalisation from price discrimination, and sets out the transparency practices that keep the first from feeling like the second.
Separate discounts from base price changes
Most shoppers accept personalised discounts: a loyalty price, a returning customer offer, a targeted promotion on top of a common base price. What corrodes trust is a personalised base price, where the headline figure itself moves by visitor with no visible reason.
Stores should know which one their tooling does. Pricing vendors sell both under the same personalisation label, and a store can drift from offers into base price variation without anyone choosing that deliberately.
Watch the signals being used
Pricing models can react to location, device, visit history, basket behaviour, and predicted willingness to pay. Each signal carries fairness and legal questions that vary by market, and some, like device type as a wealth proxy, read badly the moment a customer discovers them, which customers do, usually by comparing screenshots.
Ask vendors for the signal list in writing and review it like a policy document. If a signal would embarrass the store on a news page, it should not be in the model, whatever the measured lift says.
Disclose in plain language
A short, findable explanation, that prices and offers may vary by promotion, loyalty status, and timing, with the rules for the big programmes stated, covers most legitimate personalisation. Legal requirements differ by place, and counsel should review, but silence is not a strategy.
Support teams need the same explanation in usable form. When a shopper asks why their price differs from a friends screenshot, a vague answer converts a pricing question into a fairness complaint. A calm, factual one often ends the conversation.
Measure trust alongside revenue
Pricing tests usually report conversion and margin only. Add the trust metrics: price related contacts, fairness complaints, and social posts comparing prices. A model that lifts margin while teaching customers to screenshot check every price is training behaviour that taxes every future visit.
Set a walk away line before testing: the complaint rate or press risk at which the experiment stops regardless of lift. Lines drawn during a success are rarely drawn at all, so draw them at the start.
Related reading
If this topic is useful, these earlier guides go deeper on connected questions: AI Visual Search and the Photos Shoppers Bring, When AI Recommends Substitutes for Out of Stock Items, and How AI Search Summaries Describe Your Returns Policy.
Frequently asked questions
What is the difference between personalised discounts and personalised base prices?
Discounts layer offers on a common price, which shoppers generally accept. A personalised base price changes the headline figure itself per visitor, which is what triggers fairness concerns when discovered.
Which pricing signals are riskiest?
Signals that proxy for wealth or vulnerability, like device type or predicted willingness to pay, carry the highest trust and regulatory risk. Stores should hold the vendors signal list in writing and review it deliberately.
Should stores disclose personalised pricing?
In plain language, yes, at least the fact and the legitimate bases, promotions, loyalty, timing, with counsel reviewing local requirements. Silence guarantees the practice is explained first by an annoyed customer instead.
How should pricing AI be evaluated?
On trust metrics beside revenue: price related contacts, fairness complaints, and comparison sharing. A walk away line for those metrics should be set before the test, not negotiated after it succeeds.