What I would ask a board about its models and its data
My contribution to a board
I would bring a board a practitioner's understanding of how data and models make decisions about individual people, and of the places where those decisions quietly go wrong. My experience runs from a mortgage file on an underwriter's desk to a census of 274 candidate models across a bank, and most of it has been spent where data meets a supervisor: quarterly returns, capital data, deposit insurance records and now model risk. I have written many of the reports a risk committee receives, so I know which questions they tend not to answer.
Three contributions would be most useful. The first is oversight of models and automated decisions. Many organisations outside banking now use scores, rules and AI tools to decide things about customers, patients or policyholders without the discipline banks have been required to build. I can help a board ask what counts as a model, who validated it, what data it depends on and how it performs for different groups of people. The second is data quality as a governance question: whether the figures a board reads and a regulator receives can be traced to records with owners. The third is serving the public in two languages. I know how data that works well in English can fail silently in French, and how that failure reaches decisions.
I would not offer myself as the financial expert an audit committee designates. I am financially literate and have worked on regulatory and capital reporting for two decades, but I am not an accountant. Risk and technology committees are where I would contribute most, and I would serve a full year on either before chairing it.
Judgement in practice
01
Telling a board that its old scorecard had been unfair
When fairness testing delayed our small-business credit model in 2023, the easier presentation to the board would have been that the new model had a problem we were fixing. The truer one was that the scorecard we had relied on for nine years had the same matching flaw and had disadvantaged French-preferring applicants throughout. I presented the second version, with three years of reruns, and recommended that we write to applicants whose files had failed to match and offer a review. Two directors worried that the letters would invite complaints. The board approved them; we wrote to 312 applicants, and 87 were approved on review. As a director I would want management to bring backward-looking findings of this kind, and I would try to make doing so less costly for the executive who does.
02
Putting the list of non-models in front of a risk committee
When we presented the 2024 model census, the obvious report was the 163 models and their validation status. I insisted that the risk committee also receive the 111 items the panel had judged not to be models, each with its reason, because an inventory's weakness lies in what it excludes. One director read the list closely and challenged three items. Two were reclassified as models after a second review, including a rules engine that set credit card limits. The committee now receives the exclusions every year. On a board I would ask any management team for what it decided not to count, and for the reasons, before discussing what it did.
03
Declining free analytics at a literacy charity
In 2024, as chair of Clairbourne's audit and finance committee, I reviewed an offer from a technology company to provide free budgeting analytics for our programme participants in return for rights to use de-identified participant data. The executive director was keen, because the tool would have saved about C$40,000 a year. I recommended that we decline: our participants are often in financial difficulty, many had joined through social agencies, and none had been asked whether their data could support a commercial product. The board agreed, and we bought a smaller tool that took no data rights. As a non-executive I try to ask who would be surprised by a decision if they read about it later.
My first year on a board
I would treat a first year as an apprenticeship in the organisation's own risks rather than an opportunity to apply mine. Before the first meeting I would read two years of risk and audit committee papers, the most recent regulatory or supervisory correspondence the board has seen, and whatever inventory the organisation keeps of its models and automated decisions. If no inventory exists, that would be my first question rather than my first criticism.
During the first two quarters I would meet the chief risk officer, the head of internal audit and whoever is responsible for the organisation's data, separately and without slides. I would ask each to walk me through one automated decision about a customer or member, from the record to the outcome, and to tell me where they think it is weakest. I would also ask to see a complaint file in which the organisation got something wrong about a person.
For the rest of the year I would contribute mainly through questions in committee, holding back proposals until I had seen a full cycle of planning, reporting and year-end. At the anniversary I would ask the chair whether my questions had been useful or merely banking questions in another setting, and adjust accordingly.
Conflicts and their management
Three kinds of relationship would need examination before I accepted any seat. The first is competition: I would not serve on the board of a bank, credit union, mortgage lender, fintech lender or payments company, or of any business likely to receive Aubreuil customers' data under consumer-driven banking. The second is commerce: I would not join a company that borrows significantly from Aubreuil, supplies data, analytics or technology services to it, or competes for those contracts. The third is personal: my shareholding in Aubreuil, and Aubreuil's annual sponsorship of Clairbourne Financial Literacy Network, would be disclosed in writing.
Any appointment would also need the consent of our chief executive and a review by our general counsel, and I would keep a register of my interests that I updated each quarter and sent to the board secretary. If a matter touching Aubreuil came before the board, I would declare it, leave the room and ask for the minutes to say so. If that happened more than occasionally, I would resign rather than hold a seat I could only partly fill. Aubreuil's confidential information, including its model results and customer data, would never enter another board's discussion, however relevant it seemed.
My independence statement
I am a serving executive of a listed bank and I hold shares in it. Whether I would be independent of a particular company is for its board to decide under the rules that apply to it, after I have disclosed my employment, my holdings, my not-for-profit directorship and every dealing I know of between Aubreuil and that company or its management. Until that assessment is made, I describe myself only as a candidate, not as independent.
Board and committee exposure
Served as a director
Clairbourne Financial Literacy Network
Board of directors; audit and finance committee member from 2022, chair from 2024
September 2022–present
Director of an Ottawa not-for-profit that teaches budgeting and credit skills in English and French; chairs the committee that oversees its audit, budget and data-protection practices.
Presented to the Board or a committee
Aubreuil Bank
Board risk committee
February 2021–present
Presents a quarterly report on data and model risk, including the renewal register, the annual model census and fairness test results for credit models.
Reported to a committee
Aubreuil Bank
Board audit committee
March 2017–present
Reports twice a year on data quality behind regulatory and capital returns, including exceptions found after filing.
Presented to the Board or a committee
Aubreuil Bank
Full board
2023–2024
Presented the decision to delay the small-business credit model, the finding about the legacy scorecard and the results of the model census.
Presented to the Board or a committee
Kestrelholm Mortgage Corporation
Board risk committee
2014–2017
Presented loan-level stress results and the renewal and prepayment analytics used in funding decisions.