How a forecaster's log became a way of running data

In February 1998 I told a ship's master that his vessel would clear the ice in the Cabot Strait in nineteen hours. It took forty-one. I was eight months into a job as an ice and weather routing analyst at Brumelle Marine Routing in Quebec City, and my estimate had been the midpoint of a calculation whose spread I had not thought worth mentioning. The master was courteous about it. My chief forecaster, Aurèle Penvenan, was less so, and asked me to keep a log of every estimate I issued beside the time the passage actually took. I kept it for three and a half winters. It showed that my estimates were reasonably centred but far too confident, and that the people relying on them would rather have been given the range.

I had studied applied mathematics at Université Ferrandine, finishing in 1997, and had imagined a career in meteorology. Brumelle's clients ran ore and grain carriers through the Gulf of St. Lawrence in winter and on the Great Lakes in the shoulder seasons. One of them was Morvaise Navigation. In 2001 its fleet operations director, who had read our advisories for years, offered me a job planning voyages for what were then seventeen lakers. I accepted partly because I wanted to see what happened to a forecast after it had been sent.

At Morvaise I learnt how a schedule is actually assembled: from berth availability, lock queues on the Welland Canal and the St. Lawrence, draught limits that change with water levels, and the commercial desk's view of which customer could least afford a delay. Arrival times were promised by the commercial desk and revised by the vessels, and the two rarely agreed. I began recording lock transit times from our own vessels' logs, a habit that years later became the backbone of our arrival forecasting. I completed a part-time master's degree in operations research at Tourmelle Institute of Technology in 2005, studying in the evenings and through the winter lay-up, when the fleet is idle and planners finally have time.

In 2006 I left the water for the road. Thibervaux Transport, a family-owned truckload carrier on Montreal's South Shore, sent about 1,700 loads a week across the border and was struggling with the electronic manifests that customs authorities required before a truck arrived. I joined as director of network planning and border systems. The problem turned out to be where the data came from. Dispatchers were typing commodity descriptions and consignee addresses from paper bills of lading minutes before a driver left, and the border was simply where the mistakes surfaced. We moved the capture of that data to the moment a shipper placed an order and refused to dispatch a cross-border load without it. Delays fell sharply. We also lost a large customer that would not change its habits, and I learnt that a data standard is a commercial decision whether or not anyone calls it one.

From 2011 I led operations analytics at Thibervaux, which grew to 1,240 tractors, a quarter of them driven by owner-operators. When the company installed cameras and telematics in 2014, three groups wanted the data for different ends: the safety director for claims defence, operations managers for productivity rankings and our insurer for a premium discount. I wrote a purpose register that said what each data element was for and how long we would keep it, and I gave drivers access to their own events. Some of those choices cost money. They also made the system acceptable to the people driving the trucks, which was the condition for its working at all.

In December 2016 the Thibervaux family sold a majority stake to a US carrier that intended to run analytics from its own headquarters. Morvaise's chief executive asked me to come back and assemble fleet data that still lived in noon reports, logbooks and terminal spreadsheets. As vice-president of fleet and network data from 2017, I brought fuel flow-meter readings, engine data and transponder positions for every vessel into one platform and reconstructed eleven seasons of lock transit times. When Rémi Vautrinel created the role of chief data officer in 2021, he appointed me to it.

Since then my work has been to treat the data we send outside the company with the care usually reserved for financial statements. We publish arrival windows together with a record of how often they held. We decline late-season cargoes when the data says a vessel may not clear the locks with margin, while the master keeps the final word on every voyage. We give each shipper its own voyage data and nothing of anyone else's, and we send customers fuel-based emissions figures that include their share of the voyage home.

I lead forty-eight people, half of whom came from operations rather than analytics, and I prefer it that way. I ask for the miss rate before the accuracy figure, the method before the number, and the name of whoever will be inconvenienced by a new rule before the rule is adopted. Colleagues would say I am slow to publish a forecast and quick to publish its record. I speak French at home and English at most of our ports, and I still keep a log of the estimates I sign.

Open to Board appointments · Advisory roles · Speaking invitations · Executive opportunities

Languages / French · English