Chief Data Officer / Canada
Decisions that changed what our customers could rely on
- Vessels whose voyage and fuel data my team governs
- 15 lakers
- Plus two bulk terminals and 210 road tractors, 2026
- Arrivals inside the published 90% window
- 88%
- 1,510 arrivals, 2025 navigation season
- Data function operating budget
- C$10.2m
- 2026; 48 colleagues; company revenue C$912m in 2025
Decisions & consequences
01Replacing the promised arrival time with a window and a record
In 2021 Morvaise sent terminals and shippers a single estimated arrival time for each of its 1,430 port arrivals. When I analysed the season, the median arrival came nine hours after the first estimate, 31% of arrivals missed it by more than twelve hours, and estimates were revised later far more often than earlier. The commercial desk set the first estimate, and a late one could cost a cargo nomination, so the errors leaned one way. Grain elevators on the St. Lawrence called in labour and ordered rail cars against our times; a Lake Erie steel mill planned its stockpiles around them. We paid C$2.9 million in waiting-time claims that year, and customers absorbed costs we never saw.
I proposed a likely arrival and a window designed to hold nine times in ten, computed from lock queue records, weather, berth performance and each vessel's speed history. Commercial staff could still change a forecast, but every override would be recorded with its reason. Bastien Corriveaux, our vice-president commercial, opposed the change in writing: customers wanted one number, a competitor offered firm times, and a window would make us look like a carrier that did not know where its ships were. Our fleet operations director wanted the windows kept internal. Rémi Vautrinel agreed to a pilot.
We ran the 2022 season with two grain elevators and the steel mill, sending the window and a monthly calibration report. We gave up the commercial desk's control of the first estimate, and the appearance of precision. In autumn gales the windows were sometimes forty hours wide, and we had to explain why.
By the 2025 season, 88% of 1,510 arrivals fell inside the published window, the median error of the likely time was 3.1 hours against 9.4 in 2021, and waiting-time claims had fallen to C$1.1 million. One feed-grain merchant moved two cargoes a year to the carrier quoting firm times.
The calibration report now goes monthly to the chief executive and to every shipper on our portal. What I would do differently is bring the terminals' own schedulers into the design from the start. In the second season they told us that a firm 'not before' time mattered more to them than the window's far edge, because it decided when to call labour. They were right, and we could have learnt it a year earlier.
Letting the ice set the last fixture of the season
Every December the same argument took place. The Seaway locks close for winter around the turn of the year, and a laker that misses closure is stranded on the wrong side of them until spring, away from its planned winter berth, shipyard slot or winter trade. Customers want one more cargo of ore, grain or salt before closure, and December margins are among the best of the year. In December 2022 ice in the upper St. Lawrence slowed two of our vessels on their final voyages. One needed icebreaker assistance it had not planned for, reached its winter berth nine days late and required C$1.3 million of hull repairs. Our salt customers ran short on fourteen depot-days that winter, because stockpiles had been positioned by habit.
In 2023 I proposed a season-close rule. For each proposed late fixture, my team would estimate the probability of the vessel completing its voyage and reaching its winter berth at least three days before the announced closure, using ice outlooks, cold-weather lock transit history, the vessel's power and hull, and berth sequences. Below 85%, the fixture would not be offered. Bastien Corriveaux argued that the rule would hand December business to competitors. Members of the founding family on our board asked whether a model should decide what our captains had always decided. Our masters supported the rule on one condition, which I accepted: the forecast governs what the company sells, never what a master does at sea.
We declined six late-season cargoes in 2023, about C$3.4 million of revenue and C$1.2 million of contribution. We also used depot drawdown forecasts to move 64,000 more tonnes of salt to our Ontario and Quebec depots before closure.
At the 2023, 2024 and 2025 closures no Morvaise vessel needed unplanned icebreaker assistance, and ice-damage repairs averaged C$0.2 million a year. Depot shortfalls fell to two depot-days in the 2024–25 winter. Milder Decembers helped, and I say so whenever these figures are shown.
The rule is now part of our winter operating plan. What I would change is where the threshold came from. I set 85% myself, and the board's audit and risk committee approved the rule without being asked what level of risk it wanted to accept. That should have been the committee's number, chosen with the forgone contribution in view.
Making border data a condition of dispatch
In 2008 Thibervaux Transport sent about 1,700 loads a week across the border, most of them southbound from Quebec and Ontario into the US north-east and Midwest. Electronic manifests had become mandatory before a truck reached a US crossing, and Canada was preparing its own equivalent. Our dispatchers built those manifests from paper bills of lading minutes before departure. That year 6.8% of southbound crossings were held or delayed for document-related reasons, for an average of two hours and forty minutes each. Every hour a driver waited came out of the driving time allowed under hours-of-service rules, and late deliveries followed.
The errors began long before the border: vague commodity descriptions, consignee addresses copied wrongly, weights estimated. I proposed that shipment data be captured when a shipper placed an order, through electronic data interchange or a bilingual web form, checked against each shipper's approved commodity descriptions, and that no cross-border load be dispatched without it. Laurent Thibervaux, who led sales and was the founder's son, opposed the proposal. About 7% of our cross-border volume was same-day freight booked by telephone, and he expected those shippers to go elsewhere rather than complete a form. Our vice-president of operations worried about trucks standing idle while data was chased.
The founder backed the proposal after a three-month trial on our two busiest lanes. We gave up some same-day business and accepted a slower booking process for small shippers. One furniture manufacturer, worth about C$6 million a year, left within six months rather than change how it booked freight.
By the end of 2010 document-related delays had fallen to 1.4% of southbound crossings. Drivers recovered roughly 7,000 hours a year that had been spent waiting, and on-time delivery on cross-border lanes rose from 84% to 92%. Our customs brokers reported fewer corrections on their side too.
Data quality at order intake became a line in the sales team's monthly review. If I did it again, I would share rejection reasons with the brokers every week from the beginning rather than after a year. Some of the errors we blamed on shippers were introduced in the hand-off between our systems and the brokers', and we could have found them within weeks.
Giving driver telematics a purpose and an expiry date
In 2014 Thibervaux decided to fit forward-facing and driver-facing cameras, with braking, speed and location telematics, across its 1,240 tractors, 310 of which belonged to owner-operators. Three parts of the business wanted the data for different reasons. The safety director wanted all video kept indefinitely to defend claims. Operations managers wanted location and idling data to rank drivers' productivity. Our insurer offered a 9% premium reduction in exchange for continuous access to event data and video. Driver turnover was already 58% a year, and a camera pointed at the driver is exactly the kind of change that sends experienced people to a competitor.
I proposed a purpose register before installation. Each data element would be tied to named uses: collision and claims defence, coaching, hours-of-service compliance and billing disputes. Video not attached to a triggered event would be overwritten after seventy-two hours; event video would be kept for two years or the life of a claim. Location data would be held at full resolution for thirteen months and then aggregated. Drivers would see their own events within a day and could add a note before any coaching conversation. The insurer would receive fleet-level safety measures, and video only for a specific claim.
The safety director disagreed about deletion, and the vice-president of operations about the productivity ranking, which the register excluded. Taking a 4% insurance reduction instead of 9% cost about C$560,000 a year. Our president approved the register over both objections.
Over the next two years preventable collisions per million kilometres fell from 0.84 to 0.57, and event video cleared our drivers in twenty-three disputed incidents. Driver turnover fell to 46% by 2016. I would not credit all of that to the policy, but exit interviews stopped mentioning the cameras, and no driver complaint reached a regulator.
The register became the template for every later data collection at Thibervaux, and I brought it with me to Morvaise. Two things I would change. Owner-operators should have been consulted separately, because those tractors were their businesses as well as their workplaces. And seventy-two hours was too short. A cargo damage claim reported on the fourth day found its video already overwritten, and we paid a claim we could probably have defended. We moved to seven days.