Statistics Canada published its analysis of AI use by Canadian businesses for the second quarter of 2026. The headline number is striking: 19.2% of businesses report having used AI in the preceding twelve months, up from 12.2% a year earlier and 6.1% in 2024. A tripling in two years.

That is not the interesting number.

The interesting number sits lower in the table

Among businesses that use AI, 44.4% changed their training or hiring practices. In detail: 32.0% trained existing employees, 21.6% trained managers. More than one AI-using firm in two (55.6%) is therefore deploying AI without having changed either its training or its hiring.

And the gap widens with company size:

100+ employees1 to 4 employees
Trained existing employees68.1%24.0%
Trained existing managers51.7%15.6%
Hired AI skills32.8%8.2%
Used external consultants30.2%10.7%

Put those proportions next to the adoption gap, from the same survey: 27.8% of businesses with 100 or more employees use AI, against 19.9% of businesses with 1 to 4 employees. A ratio of 1.4 to 1. In Quebec, the Institut de la statistique du Québec measured 26.1% against 12.2% in the second quarter of 2025; Quebec figures for 2026 were not published at the time of writing.

The training gap is 2.8 to 1 for employees and 3.3 to 1 for managers. Twice as wide as the adoption gap. Small firms adopt almost as much as large ones; they support it far less.

What the data says, and what it does not

Let us be precise, because this kind of number is easy to bend. Statistics Canada measures reported practices. The survey establishes no link between training and results: it does not say that firms which train get better returns. Anyone presenting these figures as proof of ROI is inventing half the argument.

What they show is simpler, and enough: in large organisations, deploying an AI tool comes with a training effort by default. In small ones, it does not. The difference is not technological. It is a difference in deployment method.

Why manager training is the real weak point

Look again at the “trained existing managers” row: 51.7% against 15.6%. It is the widest gap in the table.

That is not incidental. Managers decide which processes are candidates, what budget follows, and what risk is acceptable. A trained team under untrained leadership produces orphan pilots: it works, nobody knows what to do with it, and it dies at the first reorganisation. That is the scenario we see most often.

The order that works runs against intuition: leadership first, in half a day, so they can arbitrate. Teams next, on the tools they will actually use.

The number one obstacle is addressed by training

In the same survey, businesses cite cybersecurity and privacy (13.4%) as the leading obstacles, followed by cost (10.6%).

The first obstacle is not a tooling problem. A team that knows what it can paste into a prompt, what it cannot, and why, reduces that risk without buying anything. A team that does not know will do it anyway, with free tools, out of your sight. Our post on Law 25 when your employees use AI covers what that means in practice.

Put differently: the spend that unblocks the most frequently cited obstacle costs less than most licences.

Three things to do before buying another tool

Train leadership before the teams. They do not need to know how to write a prompt. They need to recognise a viable use case, estimate what it costs, and say no to the rest.

Train on your processes, not on this season’s tools. Generic ChatGPT training expires in six months. Training built on your three heaviest processes produces habits that survive a change of tool.

Budget training inside the project, not beside it. Large firms do not train because they are virtuous. They train because the line exists in the budget from the start. It is the one thing in that table an SMB can copy without extra means.

To see where your teams stand: our AI tools literacy check takes a few minutes and collects nothing. And if you would rather start at the top, our training programs are built in that order.