Personalised offers should only ever undercut the shelf

Every loyalty programme personalises price, whether or not it uses the phrase. When Mirebrook sends one member twenty per cent off a moisturiser and another member nothing, the two pay different net prices for the same jar. I think that is acceptable. What I do not accept is the next step, which retail technology now makes easy: using what we know about a customer to estimate how much they would tolerate paying, and setting their offer or their price accordingly. My position is that in a pharmacy chain personalised pricing should be held to three rules. An offer may only ever take a price below the posted shelf price. It may not be sized on signals of need, urgency or vulnerability. And a defined list of essential health products should not be personalised at all.

The shelf price is a promise to everyone

The shelf price, and its online equivalent, is the price we are prepared to state to anyone who walks in. It is visible, comparable and open to challenge. A customer who thinks it too high can go elsewhere, complain or photograph the label. Personalised offers sit beneath it as discounts, and the worst a member can say is that somebody else received a better deal.

Electronic shelf labels and app-only pricing change that arrangement. Once a price exists only on a screen addressed to one person, the shelf stops being a promise and becomes an opening bid. In 2024 our pricing team proposed that app prices for some beauty lines should vary by member, upwards as well as downwards, within a band of eight per cent either way. The model behind the proposal was competent. It estimated each member's sensitivity to price from their purchase history and set prices to maximise margin, and on paper it added about forty basis points to front-store gross margin.

I recommended against the upward half of the proposal, and our chief executive agreed. A member who discovers that she paid more than her neighbour for the same serum because a model judged her unlikely to shop around has learnt that loyalty is penalised. That lesson travels faster than any offer. Competition and consumer-protection authorities in several countries, Canada among them, have begun asking how algorithmic pricing works and whom it disadvantages. I would rather explain a rule we already keep than defend a model we cannot easily describe.

Where personalisation turns into charging the ill

The second rule matters more than the first. Even offers that only ever reduce a price can be unfair if they are sized on the wrong signals. A model rewarded for margin will learn that a customer buying a cold remedy at eleven o'clock on a Sunday night is not comparing prices. It will learn that a parent buying the same infant formula every week is unlikely to switch brands. It will learn, if we allow it, that members in postal areas with fewer competing stores redeem fewer discounts and can be given smaller ones.

None of those inferences needs a protected characteristic. All of them produce systematically worse offers for people who are ill, exhausted, caring for someone or living with little choice. When we reviewed offer allocations in 2024, we found that members whose purchases suggested a chronic condition were receiving offers worth, on average, eleven per cent less than other members with similar spending. The model had learnt that they would buy anyway. Nobody designed that outcome. It emerged from an objective that measured incremental margin and nothing else, and it had been running for two years before we looked.

We removed late-night and weekend urgency signals, the same 38 health-inference categories we exclude from supplier advertising, and competitor density from the features used to size offers. We now test each quarter whether the value of offers differs by inferred need group, by province and by the language in which a member receives communications, and the results go to our executive committee. Offer-driven incremental margin fell by about C$6 million a year. I regard that as the cost of not charging ill people more.

Three rules we publish

The third rule is the bluntest. Some products should not be personalised at all. Our list of 212 essential health products, which includes children's fever medicine, oral rehydration salts, basic pain relief, glucose testing supplies and infant formula, carries one price for everyone and no personalised offers. Promotions on those products run through the flyer, open to every customer, member or not. Our loyalty team objected that this denies members value they have earned. I think it denies us the ability to make the cost of a sick child's medicine depend on a household's shopping history, which is a power a pharmacy should not want.

We publish all three rules in French and English on our website and in the Mirebrook Circle terms, in plain language: personalised offers only ever lower prices; we do not use signals of urgency or health to decide the size of an offer; essential health products are never personalised. Publishing them changes behaviour inside the company. An analyst who proposes an exception now has to explain why a public promise should change, rather than why one model feature is harmless.

There are serious objections, and I take them seriously. Downward-only offers still mean that some members pay more than others once discounts are counted, and a purist would say that is already unfair. I accept the point and answer it with visibility: members can see every offer they have received and the categories it was based on. Competitors who personalise upwards may earn more margin and use it to cut shelf prices for everyone, which would help customers who are not members. That is possible, and I would want to see the evidence before believing it. Finally, the essentials list has edges, and each product we add to it removes a promotional lever from a category manager.

I would not ask another retailer to copy our list or our rules word for word. I would ask its board three questions. Can any customer pay more than the posted price because of what the company knows about them? Do the offer models use signals that correlate with illness, urgency or a lack of alternatives? And is there a set of products whose price the company has decided should never depend on who is buying? If management cannot answer each one plainly, the pricing models are answering them instead.