My contribution to a board
A board would gain from me a practitioner's reading of catastrophe risk from the data upward. I have rated hail by township, run earthquake models for reinsurance placements and rebuilt the exposure data behind an insurer's wildfire pricing. A company whose assets, customers or service territory sit in wildfire, flood or storm country can use a director who knows which modelled numbers deserve confidence, which are informed guesses and what the difference usually costs.
The second thing I would bring is experience of automated decisions about individuals: prices, fraud referrals and underwriting triage. I have removed a variable that worked but could not be explained, retired a fraud score I built myself, and extended a provincial privacy notice to customers who were not legally owed it. Boards are approving uses of machine learning faster than they are learning to question them. I can help a committee ask what a model does to the people it scores, and what evidence would show that it is drifting.
The third is data control behind regulatory reporting. I have reported to an audit committee on the lineage of the figures in our capital return, including an error that passed every control we then had. On a risk, technology or audit committee I would be the member who reads the data behind the paper and asks who owns each judgement in it.
I would not present myself as an audit committee financial expert, and I have not chaired a committee of a listed board. I would expect to serve a full year as a member before chairing anything.
My first year on a board
Before my first meeting I would ask the company secretary for two years of risk committee minutes and the papers behind the three decisions the board found hardest. In each paper I would look for the numbers that came from a model, who built each model and what the paper said about its limits. In most organisations the answer to that last question is short, and its shortness tells a new director something.
Over the first two quarters I would spend time with the people who produce the data, not only those who present it: whoever maintains the asset or customer records, the head of model validation if the company has one, and the internal auditor who last reviewed data controls. I would ask each of them what they would want a director to know and are never asked. Those conversations would be arranged through the chair and the chief executive, so that nobody mistook them for an investigation.
In the meetings themselves I would mainly listen. When I saw a question the papers did not answer, I would take it to the committee chair beforehand rather than raise it cold in the room, unless the decision could not wait. I would ask for one change of format at most in the first year, and only if it helped the whole committee.
At the end of the year I would ask the chair, and one executive I had challenged, whether my questions had improved decisions or merely lengthened meetings. I would act on the answer, including by asking fewer questions if that was what it took.
Conflicts and their management
Haverlock employs me and insures homes, vehicles and small businesses in five provinces, which creates some conflicts I would rule out in advance and others I would manage. I would not join the board of an insurer, a broker, a managing general agent, a reinsurer, or a company that sells catastrophe models, consumer data, telematics or claims technology to insurers. The overlap with my executive work would be constant, and recusal would leave me with little to contribute.
The harder case is a company Haverlock might one day pursue. An electricity utility whose equipment is alleged to have started a wildfire can face subrogation claims from the insurers that paid the resulting losses. If I joined a utility board, I would disclose at the outset any open or contemplated subrogation by Haverlock against it, absent myself from any discussion of wildfire litigation or settlement, and ask Haverlock's general counsel not to brief me on the matter. If a claim became material for either company, I would resign one of the two positions.
I would carry nothing from Haverlock into another boardroom: no customer data, no pricing, reserving or reinsurance information and no supplier terms. My chief executive's consent and a written conflict review would come before any acceptance, and I would refresh my disclosures every year and whenever my executive responsibilities changed.
My independence statement
I am a full-time executive of a listed insurer and hold Haverlock shares through its employee share plan. Whether I am independent of a given company is a judgement for that company's board under the securities rules and its own policies, made after it has seen my disclosures, not a claim I can make for myself. I would provide those disclosures in writing before any appointment: my employment, my shareholdings, my non-profit directorship, and any policy, claim, supplier or data relationship between Haverlock and the company.
Judgement in practice
01Bringing the risk committee a larger wildfire number than it expected
In early 2025, when we first ran the licensed wildfire model on our corrected exposure data, the modelled 1-in-200 wildfire loss for the home portfolio was about 18% higher than a preliminary run on the old file had suggested. Our reinsurance renewal was weeks away, and one colleague suggested presenting a range rather than the new figure. I took the corrected number to the risk committee with the limits note attached and said plainly that the earlier, lower figure had come from bad locations rather than lower risk. The committee asked management to review the retention on our catastrophe programme and chose a lower retention at a known cost. On a board, I would expect bad news about a model to arrive with its reasons and its limits, and I would try to make it safe for an executive to bring it early.
02Recommending resubmission when the error was immaterial
In 2022 I found that our capital returns had carried a line-of-business mapping error for eight quarters, moving our MCT ratio by about two points on a ratio above 200%. The amounts were immaterial, and our chief financial officer's preference to correct the mapping prospectively was reasonable. I argued that the supervisor should hear about it from us before the next filing, that the two most recent quarters should be resubmitted and that the audit committee should see the root cause, because the error had passed every existing control. The audit committee chair agreed, and the episode produced a lineage programme the committee still monitors. As a director, I would distinguish between an error that is small and an error that shows a control does not work, and I would treat the second kind seriously whatever its size.
03Holding back a dataset at a non-profit board
In 2023 the board of Harrowby Open Data Collaborative, where I chair the privacy and data ethics committee, considered publishing a municipal dataset of rental housing inspections at address level. Tenants' groups wanted it, and a funder had asked for it. My committee found that in small buildings an address combined with inspection dates and violation types could identify individual tenants and, in a few cases, accessibility modifications that revealed a disability. We recommended publishing at block level with a one-quarter delay. Two directors argued that we were weakening the dataset, and the funder reduced its grant by about C$40,000. The board accepted our recommendation. I learnt that a committee's advice carries more weight when it states plainly what the organisation will lose by following it.