Name the Hour

The province that taught the world’s machines to learn has not yet taught its own shops what to ask them. We built, in Montreal, one of the densest clusters of machine-learning research on the continent. And by one survey from our own employers’ council¹, close to three quarters of Quebec businesses cannot say which problem artificial intelligence would solve for them. That is not a technology gap; we have the technology in embarrassing abundance, much of it invented here. It is a naming gap. No tool, however clever, can be pointed at a problem nobody has named.

A naming gap can be closed on a Tuesday afternoon with a notepad. Naming is cheap, and the help for what comes after is already funded and under-used. The firms that named their hour report the gains the rest of us are still reading about; the firms that never did bought a subscription and got a subscription.

The barrier that is not a barrier

Our statistics institute reported late last year² that AI use among Quebec firms rose from 9.4 to 12.7 percent in a single year, a slower climb than Ontario’s. Firms of a hundred or more sit at 26 percent; firms of one to four people, at 12. The rest, asked why not, named the cost of implementation, an uncertain return, and a lack of specialized knowledge.

Those are not three reasons but one condition described three ways. The cost of what? The return on what? Knowledge for what? An unnamed problem has an infinite price and a return of nothing, and no course teaches its solution. What we lack is the sentence that begins “the hour we lose every day is spent on…”

Picture a machine shop of a dozen people in Saint-Hyacinthe, quoting the way the founder’s father did. Every request arrives as a PDF drawing; the estimator reads the tolerances, counts the operations, and types it all into the quoting sheet, the better part of an afternoon, twice a week. Nobody there calls this a problem, because a problem is something that might not be there, and this has always been there. Yet that afternoon is the entire AI strategy of the shop, in plain sight, unnamed.

Or a family food distributor in Laval, where orders arrive by phone, text and email, in French and English and sometimes both in one sentence, and someone whose real job is the route schedule spends the first two hours of each morning re-keying them. The owner would say, honestly, that AI is not relevant to distribution; four in ten small firms across Canada³ said the same this summer. They mean that AI has been introduced to the industry but not to the two hours.

The machine is a mirror

One finding governs the whole conversation. In April, Statistics Canada found⁴ firms that had adopted AI about 17 percent more productive than those that had not. Account for how productive those firms were before they adopted, and the gap falls to about 10 percent. Account for what else they had (trained people, usable data, defined processes), and it falls to about 5 percent, a figure their own tests could not tell apart from nothing. The agency’s own words: adoption in isolation is “likely insufficient to deliver transformative productivity gains.” A study of small entrepreneurs in Kenya⁵ found the same shape from the other side: no gain on average: stronger operators gained while weaker ones lost ground.

We read that as bad news. It is the best news in the file. The machine is not a substitute for a well-run shop; it is a mirror of one. A shop that already knows where its hours go will see that knowledge multiplied. A shop that has not named its problem will buy a very expensive reflection of its own confusion.

Where the task is named, the evidence stops being timid. Customer-support agents with an assistant⁶ resolved about 15 percent more issues an hour, the least experienced about 30 percent more. Physicians in British Columbia and Ontario⁷ using an AI scribe report three to four hours a week returned; ask any clinic on the South Shore what three hours a week of a doctor is worth. Singapore’s evaluation of its small-business digital program found tools built for a specific trade delivered roughly twice the gain of generic ones: a general tool answers general questions, and no firm was ever troubled by one. None of these is a company that “adopted AI.” Each named an hour and removed it.

Our own firms say the same. In a survey run with Investissement Québec⁸ late last year, Quebec SMEs that had automated named productivity and the labour shortage as their motives and reported a median gain of about 17 percent. Self-reported, and about automation broadly rather than AI in particular; but reported by the neighbour who did it, not the consultant who sells it.

The help is already paid for

The same survey turned up a detail that should embarrass us gently. Eight in ten of those firms paid for their automation from their own pockets, and fewer than one in five used any public accompagnement. The head of Investissement Québec, Bicha Ngo, said in February⁹ that our SMEs are “encore trop peu nombreuses à tirer avantage des leviers de financement et d’accompagnement qui sont à leur disposition”: too few take advantage of the financing and support on offer. The province funds AI training for some three thousand small firms and has budgeted over half a billion dollars more for SMEs. BDC has announced a 500 million dollar initiative¹⁰ supporting technology adoption, including AI; eligible borrowers can postpone principal payments for up to two years, subject to conditions. Nearly nine in ten¹¹ small-business financing requests are approved.

Money is rarely the constraint, and neither is the will: BDC’s chief economist said in August 2025¹² that in thirteen years at the bank there had never been more entrepreneurs asking about productivity, and that “it’s often about cutting waste.” BDC’s own impact study¹³ found that financing-only clients had productivity growth 3.8 percentage points higher than comparable non-clients between 2020 and 2021, while advisory-only clients showed no statistically significant difference in productivity growth, though their employment growth and survival were higher. The study does not explain the difference, and it does not test whether anyone had named a problem; the practical lesson stands on its own. The help is real, and the help is shaped like a tool. The missing piece is shaped like a sentence.

One more number belongs to the shop alone. Across Canada, about a quarter of the smallest adopters³ trained their staff, against more than two thirds of large firms. If a previous essay argued that we should rent the frontier and own the workhorse, then for a shop the workhorse is not the software. It is the named problem, the two trained pairs of hands, and the people who now know what “done” means for that task.

An inventory of one hour

We have some standing to be proud. By one analysis, Quebec’s hourly productivity gap with Ontario narrowed from nearly five dollars in 2020 to under a dollar in 2024, whatever the causes. Pride is only dangerous when it sits down.

So the first step is not a strategy, a vendor, or a grant application. It is an inventory of one hour. Name the hour we lose every day, in a sentence a new hire would understand. Pick the narrowest tool that removes that hour and nothing else, and prefer the one built for our trade over the one built for everyone. Train the two people who touch it, because those two are the whole project. Then use the accompaniment that is already paid for, and measure the only thing that counts on a shop floor: hours returned. The machines learned from us. It is our turn to ask them a proper question.

---

Disclosure: I work at BDC. This article draws on publicly available sources and is written in my personal capacity. Comments or opinions expressed here do not necessarily represent the views of BDC or its management.