Chief Data Officer / Canada
Decisions that cost revenue first and earned trust afterwards
- Loyalty members active in the past year
- 10.6m
- Mirebrook Circle, June 2026
- Mirebrook revenue, fiscal 2025
- C$8.3bn
- C$3.6bn from dispensaries and C$4.7bn from the front store
- Front-store shrink, 2025
- 1.8%
- Down from 2.3% in 2023, without biometric identification
- Colleagues in the data function
- 186
- C$54m operating and C$19m capital budget, 2026
Decisions & consequences
01Asking 9.8 million members to choose again
When I became chief data officer in January 2022, Mirebrook Circle's terms gathered points, personalised offers and the use of purchase data for partners' marketing into one consent given at sign-up. The programme then had 9.8 million active members. Québec's amended private-sector privacy law was bringing stricter consent requirements for sensitive information into force by September 2023, and the federal commissioner's guidance on meaningful consent had long asked for clearer choices. Our external advice was that the bundled consent could probably be defended outside Québec with better notices. Personalised offers produced an estimated C$310 million of incremental front-store sales a year, and the partner network's plan assumed it could use the whole membership.
I proposed asking every member in every province two separate questions, one about personalised offers and one about supplier advertising, with a refusal carrying no penalty in points or member prices. Delphine Corriveau, our president of loyalty and marketing, argued for changing Québec only, where the law was clearest. Callum Pocklington, who leads the partner network, expected to lose half his audience. Our chief financial officer, Geneviève Belleau, asked for a revenue floor before agreeing. We estimated C$21 million of lost incremental sales in the first year, and the chief executive accepted that estimate as the price of one national rule.
We tested the wording in both languages with 4,000 members, set every choice to no until the member answered, limited reminders to two and made sure no store colleague was paid on agreements. By December 2023, 61% had agreed to personalised offers and 44% to supplier advertising. Offer redemption among consenting members rose from 6.1% to 9.4%. Incremental sales from offers fell by C$17 million in 2024 and were above the 2022 level by 2025. Privacy complaints fell from 1,140 a year to 310.
Every proposed data use now carries a consent mapping before development starts. What I would change is the supplier question. It did not name the kinds of brand involved, and refusals were highest among members who told us they could not picture what they were agreeing to. I also left pharmacy assistants without a script for six weeks while they fielded members' questions at the counter.
Rationing children's fever medicine without ranking our members
In the autumn of 2022 children's liquid fever and pain medicine was scarce across Canada. Our stores were receiving roughly 40% of normal supply. Parents were driving between stores, photographs of empty shelves were circulating, and several pharmacist-owners in Québec asked to control allocation for their own customers. Stock was also being bought in quantity for resale online.
Our loyalty team proposed alerting high-tier members with a history of children's purchases when deliveries arrived, and reserving part of each shipment for app orders. I opposed both. Alerts would have rationed scarce medicine by past spending, and identifying the households to alert meant inferring which members had young children from their baskets. I proposed allocating stock to stores in proportion to each store's demand in the three years before the shortage, adjusted weekly for provincial respiratory illness reporting, with a limit of two units per purchase and the product kept behind the dispensary counter so that a pharmacist could advise. Delphine Corriveau disagreed, arguing that our most loyal customers deserved first notice. Two franchise owners wanted to keep stock for families they knew. The chief executive backed the allocation rule. We gave up app sales, some goodwill among our best customers and the comfort of concentrating stock where it sold fastest.
My team ran the allocation daily for sixteen weeks and flagged any store without stock for three consecutive days. In the eight hardest weeks, 590 of 648 stores had the medicine on the shelf for at least five days in seven, against an estimated 430 had we allocated by sales value. Resale listings traced to our stock fell sharply after the counter rule began.
The episode produced our list of 212 essential health products and an allocation rule the executive committee approved in advance of the next shortage. What I would do differently: the two-unit limit was too blunt for families with several children, and pharmacists should have been able to vary it with a recorded reason from the first week rather than the fifth. We should also have published the rule to customers, who saw the limits without the reasoning.
Building the supplier clean room on rules the brands disliked
By 2023 the Mirebrook Partner Network, which sells sponsored placements in our app, on our website, on in-store screens and in the flyer, was earning C$96 million a year. Cosmetics houses and consumer health manufacturers were asking for two things: to match their own customer lists with our members, and to reach audiences defined by health purchases such as pregnancy tests, smoking cessation products or incontinence care. The 2024 plan assumed C$141 million of revenue, and competing platforms were offering both.
I designed a clean room with five rules. Only members who had agreed to supplier advertising would be included. Results would be aggregates of at least 500 households. No row-level data would leave. No audience could be built on any of 38 health-inference categories, including by the brand that makes the product. The privacy office would review query logs monthly. Callum Pocklington argued for a threshold of 50 households, closer to what competitors offered, and warned that excluding health categories would send consumer health budgets elsewhere. He put the argument in writing and I attached it to my paper for the executive committee. We gave up the most profitable audiences in the store.
Two over-the-counter medicine makers withdrew planned campaigns worth about C$9 million. The network earned C$118 million in 2024 against the C$141 million plan. It then grew to C$152 million in 2025, as suppliers found that campaign measurement on consenting members was more reliable than the broader audiences they had bought elsewhere, and supplier renewal reached 87%. We have had no privacy incident involving the clean room.
The board's audit and risk committee now approves the health-inference list annually and sees every proposal to remove a category. What I would do differently is the threshold. I set it at 500 without testing its effect on small brands, and an independent Québec skincare maker could obtain no usable results at all. We introduced a pooled campaign product for small suppliers in 2025; it should have been there at launch. I would also bring two or three suppliers into the design work instead of presenting them with finished rules.
Theft analytics that follow transactions, not faces
In 2023 front-store shrink at Mirebrook was 2.3% of front-store sales, about C$105 million, and 4.1% in cosmetics and fragrance. Organised theft was concentrated in downtown stores in Vancouver, Toronto and Montréal, and incidents in which colleagues felt threatened had risen 38% in two years. Dermot Mabbutt, our vice-president of asset protection, proposed a twenty-store pilot of facial recognition linked to a shared list of people suspected of repeated theft. Several franchise owners supported it, and so did store colleagues who had been confronted.
With our chief privacy officer, Yasmin Thorvaldsen, I recommended against the pilot. It would capture biometric data from every customer entering those stores, most of them buying cough syrup or shampoo. Error rates on low-quality store footage were uncertain, a mistaken match could see an innocent customer confronted, and Canadian privacy commissioners have been sharply critical of facial recognition in retail settings. I proposed instead exception analytics on refunds without receipts, voids, price overrides, pickup claims and gift card activity, together with product-level loss maps to decide which lines went into locked fixtures. Dermot's memo arguing for the pilot went to the board with mine. The chief executive chose our approach, subject to a twelve-month review. We gave up a tool that might have identified repeat offenders, and accepted that locked fixtures would cost some sales.
Front-store shrink fell to 1.8% of front-store sales in 2025, about C$23 million a year less loss, and cosmetics shrink fell to 3.0%. Refund and pickup fraud losses fell by C$6.4 million, and the exception reports identified internal theft that the stores had not suspected. Sales of fragrance lines moved into locked fixtures fell about 6%.
Any proposal to deploy surveillance technology in our stores now requires a privacy impact assessment and goes to the audit and risk committee. What I would do differently: our analytics addressed loss, not fear. Colleagues in the thirty worst-affected stores needed guards and redesigned entrances sooner than my programme delivered them. We also locked fourteen product categories at first and unlocked five within a year, after learning that they were losing more in sales than in theft.