Twenty-seven years between the loan file and the regulatory return

Mandate

Establish the bank's first enterprise data office, give every critical data element an accountable owner, and prepare the bank's data and models for the mortgage renewal wave, OSFI's revised model-risk guideline and consumer-driven banking.

Outcomes

• Built a renewal register covering 51,200 mortgages worth C$10.8bn; 4,900 strained households contacted before maturity, and 90-day cohort arrears of 0.29% against a 0.55% forecast.
• Led, with the chief risk officer, a census of 274 candidate models; 163 classified as models against 88 previously listed, and 37 spreadsheets retired or rebuilt.
• Delayed a small-business credit model by five months to repair record matching; the approval gap at similar risk between language groups fell from 11 points to about 2.
• Raised confirmed language of service from 74% to 96% of active customers; wrong-language complaints fell from 1,930 a year to an annual rate of 410.
• Reduced unidentified payroll descriptions from about 6% to 2.4% in preparation for consumer-driven banking.

Decisions in context

The bank created the role of chief data officer in February 2021 and appointed me to it from within. My mandate was to turn a reporting function into an enterprise data office: named owners for the data elements that feed credit, capital and customer decisions, an analytics and model development team held to independent validation, and a governance process the board could see. Since 2024 the work has been dominated by three external pressures. Five-year mortgage terms signed at the lowest rates are renewing at much higher ones; OSFI's revised model-risk guideline has widened what counts as a model; and the federal consumer-driven banking framework will send our customers' data to other institutions at their request. I present to the risk committee each quarter on data and model risk, and to the audit committee twice a year on data quality behind regulatory returns. I hold no seat on the bank's board.

Reason for the move

Current appointment; completing the consumer-driven banking data work and the validation queue created by the model census.

Mandate

Respond to an internal audit finding that regulatory returns relied on more than a thousand manual adjustments a quarter, and build loan-level data the bank could report from directly.

Outcomes

• Reduced manual adjustments in quarterly regulatory returns from 1,140 to 190.
• Built a loan-level mortgage data mart of 212 defined fields, each with a named business owner and a reconciliation to the general ledger.
• Cut internal preparation time for the quarterly capital return from 24 business days to 15.
• Merged three customer master files into a single record for 1.2 million customers.

Decisions in context

I joined Aubreuil to fix a specific problem. Its returns were accurate, but only because a skilled team corrected them by hand every quarter, and internal audit had concluded that the bank could not show how many of its reported figures were produced. Reporting to the chief financial officer, I combined the regulatory reporting data team with the bank's small data management group and set out to report from loan-level records rather than summaries. The mortgage data mart took two years. Its most useful product was not the technology but the 212 field definitions, each agreed with a business owner who would answer for it. We also merged three customer master files inherited from earlier acquisitions. The work brought me into regular contact with the audit committee and with our supervisor's reviews, and it persuaded the bank that data needed an executive owner of its own.

Reason for the move

Appointed the bank's first chief data officer when the role was created in February 2021.

Mandate

Build renewal, prepayment and arrears analytics for a broker-originated book funded largely through securitisation.

Outcomes

• Reduced monthly reconciliation breaks between servicing and securitisation records from about 2,300 loans to 140.
• Built a renewal retention model; retention at maturity rose from 71% to 79% over three years.
• Built the prepayment model used in funding and hedging decisions, reviewed by an external validator in 2015.
• Gave the board risk committee loan-level stress results for the first time in 2014.

Decisions in context

Kestrelholm lent through brokers and funded much of its book by selling mortgages into securitisation programmes, so a loan's record lived in two places: the servicing system that collected payments and the files that described the pools investors held. When I arrived the two disagreed on about 2,300 loans a month. Our chief executive wanted the renewal model first, because retention at maturity drove the economics of a broker-originated book. I asked for six months to correct the records, and she gave them to me reluctantly. The renewal and prepayment models we then built were used in funding and hedging decisions, and from 2014 the board risk committee received loan-level stress results rather than portfolio averages. I learnt here how much of a mortgage lender's risk sits in the timing of renewals, a lesson that shaped my work at Aubreuil.

Reason for the move

Recruited by Aubreuil Bank to lead enterprise data and regulatory reporting, which also brought me home to Ottawa.

Mandate

Build loan-level credit-risk data where only branch reports existed, and bring depositor records to the standard needed to identify insured balances account by account.

Outcomes

• Built a monthly loan-level arrears dataset through 2008 and 2009, replacing branch summaries.
• Reconstructed fourteen years of loss history from paper files to support loss estimates.
• Reduced duplicate depositor records from 11% to under 1%.
• Wrote the company's first credit-risk data dictionary, of 160 terms.

Decisions in context

Thornvale hired me to build credit-risk data at a trust company that had grown on the judgement of experienced lenders and had never needed loan-level history. I started in June 2007 with three analysts. Within months the financial crisis made the work urgent: the credit committee wanted to know, every month, which mortgages were slipping, and the only source was a set of branch reports in different formats. We built a loan-level arrears dataset and then went back through fourteen years of paper files to reconstruct what defaults had actually cost. In parallel we rebuilt depositor records so that the company could show which balances deposit insurance covered for each depositor, which meant resolving thousands of duplicate and inconsistent customer records. I completed a part-time master's degree in applied statistics during these years, and left to apply it to a larger mortgage portfolio.

Reason for the move

Moved to Kestrelholm Mortgage Corporation for portfolio analytics on a larger mortgage book.

Mandate

Prepare and reconcile the mortgage and retail lending schedules of quarterly regulatory returns and, from 2005, test retail exposure data for the Basel II capital framework.

Outcomes

• Documented 340 rules mapping product codes to reporting categories, and retired 12 that nobody could explain.
• Reduced unexplained differences between the mortgage schedule and the general ledger from about C$410m to under C$20m a quarter.
• Identified 23 gaps in retail exposure data during Basel II testing, all corrected before the bank's parallel reporting runs.

Decisions in context

I moved to Toronto to see how regulatory returns were produced at scale, and found that the principles were the same as at the caisse and the volumes were not. My work was the mortgage and retail lending schedules: taking balances from dozens of product systems, mapping each product code to a reporting category and reconciling the result to the general ledger. Many of the mapping rules had been written by people who had since left. Documenting all 340 of them was unglamorous, and it was the most useful thing I did there, because it exposed a dozen rules nobody could justify and a quarterly difference of several hundred million dollars nobody could fully explain. From 2005 I tested retail exposure data for the move to the Basel II capital framework, which introduced me to the idea that the quality of a bank's data could change its capital requirement.

Reason for the move

Recruited by Thornvale Trust Company to build its credit-risk data.

Mandate

Underwrite residential mortgages, then take over preparation of the credit union's quarterly regulatory returns.

Outcomes

• Underwrote or reviewed about 900 residential mortgages.
• Took over the quarterly regulatory returns in 2000 and wrote down how each figure was produced.
• Built the branch arrears tracker the lending team used until 2006.

Decisions in context

I joined the caisse two months after finishing my degree, as a mortgage underwriter serving members in French and English across eleven branches. Underwriting taught me to read a borrower's file closely and to notice when a document did not fit the rest of the story. In 2000 the manager who prepared our quarterly returns retired, and I volunteered to take them over because I had been asking how the figures were produced. I discovered that some depended on codes underwriters, myself included, entered inconsistently, and I wrote the first instructions explaining what each code meant for the return. I also built a simple arrears tracker for the branches that outlasted me by three years. The combination of credit files and regulatory figures at a small institution gave me the habit I have kept since: following a reported number back to the desk where it began.

Reason for the move

Moved to Lachford Bank in Toronto to work on regulatory returns at a much larger bank.