I judge a bank's data by how early it lets us notice a customer's circumstances changing, and by whether we can explain what we saw to that customer in the language they chose.
I am chief data officer of Aubreuil Bank, a federally regulated bank headquartered in Ottawa with C$52.4 billion of assets, 138 branches in Ontario, Québec and New Brunswick and about 1.3 million customers, 43% of whom prefer to be served in French. The bank appointed me its first CDO in February 2021, after four years leading its enterprise data and regulatory reporting. I lead 162 people, with an operating budget of C$38 million in fiscal 2025. I report to our president and chief executive, Danielle Ferland-Moss, and to the board's risk committee each quarter.
I started in 1999 as a mortgage underwriter at Caisse populaire Marécourt, a francophone credit union in Ottawa's east end, where within a year I was also preparing the quarterly returns to our provincial regulator. I spent four years on regulatory returns at Lachford Bank in Toronto, five building credit-risk and depositor data at Thornvale Trust Company through the financial crisis, and four and a half running mortgage portfolio analytics at Kestrelholm Mortgage Corporation before coming home to Ottawa and to Aubreuil in 2017.
The questions I work on are ones a mid-size Canadian lender cannot hand to a vendor. Which of our mortgage holders will struggle when five-year terms signed in 2020 and 2021 renew at much higher rates, and how will we know before they miss a payment? Which of our spreadsheets, scores and AI tools are models in the sense our supervisor means, and who has checked them? Does a credit model treat a borrower named Thériault as it treats one named Thompson, and if not, is the fault in the model or in the data it was given? My answers have delayed two launches, raised our loan-loss allowance and cost our marketing team a practice it valued.
A renewal register before the arrears
For 51,200 mortgages worth C$10.8 billion renewing in fiscal 2025 and 2026, we built a register of payment strain from loan and deposit data. Advisers called 4,900 households months before maturity. Ninety-day arrears in the cohort were 0.29% in June 2026, against a forecast of 0.55%.
A model inventory built by census
Our census for OSFI's revised model-risk guideline examined 274 spreadsheets, vendor scores, rules engines and AI tools. It classified 163 as models, against 88 previously listed, and we retired or rebuilt 37 spreadsheets that informed lending or treasury decisions.
Matching names before testing fairness
Testing before launch showed a small-business credit model approving French-preferring applicants 11 points less often at similar risk, because accented and compound names failed to match bureau files. We delayed launch by five months; the gap is now about 2 points.
Language recorded, not guessed
We stopped inferring customers' language of service from their surnames and asked them instead. Confirmed language rose from 74% to 96% of active customers, and complaints about communications in the wrong language fell from 1,930 a year to an annual rate of 410.
Current focus
I am preparing our customer and account data for the federal consumer-driven banking framework, beginning with the fields that will travel to other institutions when customers ask. I am also working through the validation queue our model census created, so that every high-materiality model has a named data owner and a tested lineage to its sources before OSFI's revised guideline takes effect.
Conversations I welcome
I am not seeking another executive role. I would consider one non-executive seat on a risk, audit or technology committee at an organisation outside lending that neither competes with nor borrows from Aubreuil, with our chief executive's consent. I accept a few advisory and speaking conversations each year, in English or in French.