A data career built across research, supply and retail

Mandate

Turn a capable analytics department into an accountable data function: own governance and the loyalty data platform, set the rules for personalisation and supplier data use, and give the board a clear account of data risk.

Outcomes

• Separated consent for personalised offers and supplier advertising for all members; 61% and 44% agreed, and privacy complaints fell from 1,140 to 310 a year.
• Allocated scarce children's fever medicine by need in 2022–23; 590 of 648 stores kept stock at least five days in seven during the hardest weeks.
• Launched a supplier clean room excluding 38 health-inference categories; partner network revenue reached C$152m in 2025 after a C$23m shortfall against plan in 2024.
• Reduced front-store shrink from 2.3% to 1.8% of front-store sales without biometric identification.
• Raised peak-week availability of 212 essential health products from 91% to 97% between the 2023–24 and 2024–25 winters.

Decisions in context

The chief data officer role was created in January 2022, when our chief executive concluded that decisions about customer data were being made in too many places. I took over governance, the data side of Mirebrook Circle and the analytics behind the partner network, and moved from reporting to the chief financial officer to reporting to him. Within the first year we faced a national shortage of children's fever medicine and the arrival of stricter consent rules in Québec, and both forced decisions that I would otherwise have approached more slowly. Since then the work has been to make our refusals explicit: the consent questions, the health-inference register, the clean room rules, the limits on loss analytics and the published rules for personalised offers. I present to the audit and risk committee each quarter, with our chief privacy officer, and I am not a director of the company.

Reason for the move

Current appointment; completing the essentials forecasting programme and the board's review of the partner network.

Mandate

Bring three separate analytics teams in marketing, merchandising and supply chain onto one data platform with common definitions.

Outcomes

• Moved three analytics teams and 14 departmental data marts onto one cloud platform, retiring C$3.1m a year of duplicate licences.
• Built daily store-level demand reporting for online and curbside orders within five weeks of joining, during the first pandemic lockdowns.
• Agreed single definitions of an active member, a store visit and front-store margin across finance, marketing and merchandising.
• Started the register of sensitive inferences that later became the 38-category list approved by the board.

Decisions in context

I joined Mirebrook in April 2020, three weeks into the first lockdowns, and met most of my team by video. My mandate was consolidation: marketing, merchandising and supply chain each had their own analysts, their own data marts and, in several cases, their own definition of an active loyalty member. The pandemic made the work urgent. Online orders and curbside pickup grew faster than any of our systems were built for, and the first useful thing my team produced was a daily report of store-level demand for online orders. By the end of 2021 the three teams worked on one platform with shared definitions, and we had retired fourteen departmental data marts. In the same period I started keeping a written register of the conclusions our data could draw about customers' health, after a merchandising analyst asked, innocently, whether we could identify likely diabetics for a glucose meter promotion.

Reason for the move

Promoted when the chief data officer role was created with a reporting line to the chief executive.

Mandate

Rebuild how member data and insight were shared with partners after member complaints and a partner audit, without losing the partners who paid for insight.

Outcomes

• Replaced member-level extracts sent to about 120 partners with aggregated reports and a controlled analysis environment.
• Launched a bilingual preference centre used by 1.9m members in its first year; complaints about partner marketing fell 58%.
• Rewrote partner data agreements to forbid onward sharing and re-identification, with annual audit rights.
• Rebuilt the analytics team around fewer, larger partners after an anchor partner left in 2019.

Decisions in context

Vellacott's coalition let 8.7 million collectors earn points with about 120 partner brands, and its business depended on telling those partners something useful about their customers. When I arrived, much of that insight left the company as member-level files with names removed but postal codes, ages and transaction histories intact. A partner audit and a run of member complaints had made the practice untenable. Over four years we moved every partner onto aggregated reports and a controlled environment in which analysts could query data without taking it away, rewrote the agreements and built a preference centre in French and English. Two partners left partly because of the changes. I moved to Montréal for the role and learnt French properly, which has been useful ever since. In 2019 Vellacott lost one of its two anchor partners, and I helped restructure the analytics team before leaving in March 2020.

Reason for the move

The coalition contracted after losing an anchor partner, and I wanted to hold customer data at a retailer that faces its customers directly.

Mandate

Rebuild demand planning after forecasting failed during the H1N1 pandemic, and give the commercial team an honest view of which promotions paid.

Outcomes

• Reduced weighted forecast error from 38% to 24% across the portfolio over four years.
• Showed that 31% of trade promotions lost money after accounting for forward buying; C$19m was moved to better-performing events.
• Built shopper analysis from retailer loyalty data for the six largest Canadian customers.
• Introduced a seasonal model for hand hygiene and cold-season products that held service levels above 95% in the 2014–15 winter.

Decisions in context

I joined Hedgecombe in the middle of the H1N1 pandemic, when orders for hand sanitiser ran at many times their usual level and our forecasts had stopped meaning anything. The first year was spent rebuilding demand planning for hygiene and cold-season products, and the next five extending the same discipline to the whole portfolio of about 1,400 products. The work I found most valuable was post-event analysis of trade promotions. Once forward buying by retailers was accounted for, almost a third of our promotions lost money, and the president moved C$19 million of spending to events that worked. I also bought insight from retailers' loyalty programmes, some of it as member-level files with complete baskets. That experience, seeing how much a supplier could learn about another company's customers, shaped every data-sharing rule I have written since.

Reason for the move

Recruited by Vellacott to lead analytics and member data at a coalition loyalty business.

Mandate

Modernise panel data collection and keep the panel representative as response rates fell.

Outcomes

• Moved panellists from paper diaries to home barcode scanners between 2005 and 2006; twelve-month panel retention rose from 71% to 82%.
• Rebuilt weighting against census benchmarks, reducing bias in estimates for households under 35.
• Delivered the firm's first linkage of panel data with a retailer's loyalty data in 2007.
• Wrote the panellist privacy commitments that the firm used with clients until it was sold.

Decisions in context

As director of panel methods I was responsible for the statistics behind everything Brimscombe sold: how the panel was recruited, how it was weighted and how far its estimates could be trusted. Response rates for paper diaries were falling, and the households that stayed were older and more settled than the country. Moving panellists to home barcode scanners improved retention and accuracy, and rebuilding the weighting against census benchmarks reduced the bias in estimates for younger households. In 2007 we linked the panel with a retailer's loyalty data for the first time, which showed both how much richer linked data could be and how loosely the contracts protected the people in it. I wrote the panellist privacy commitments partly because of the call I had taken from a withdrawing panellist years earlier. Marjorie Ennals, who founded the firm, gave me unusual freedom to tell clients what our data could not show.

Reason for the move

Wanted to apply demand and shopper analysis inside a manufacturer rather than sell measurement to one.

Mandate

Maintain the panel's sampling and weighting and produce estimates for client studies.

Outcomes

• Rebuilt the panel's regional weighting scheme, cutting the variance of provincial estimates by about a fifth.
• Introduced a written notice to panellists explaining how their data could and could not be used, after handling a withdrawal in 1998.
• Completed a part-time master's degree in statistics in 2001 while working full time.
• Promoted to senior statistician in 2000.

Decisions in context

My first job after university was at a research firm that most shoppers had never heard of, measuring what they bought. I spent my days on sampling and weighting, the unglamorous work of making 14,000 households stand in for a country. It taught me to ask who was missing from a data set before trusting what it said, a question I still put to every model my team builds. In 1998 I took a call from a panellist who was leaving because she believed her shopping diary had been sold. It had not been, but the conversation convinced me that people judge data use by its consequences rather than its mechanics, and I drafted the firm's first written notice to panellists. I completed a part-time master's degree in statistics in 2001 and was promoted twice before becoming director of panel methods in 2003.

Reason for the move

Promoted to director of panel methods.