AI and jobs in SMEs: how to automate without cutting positions

Will AI eliminate jobs in your SME? Canadian data says that, so far, very few. In the second quarter of 2025, among businesses that had used AI, 89.4% reported no change in their employment levels, according to Statistics Canada. What changes is the tasks: AI takes on data entry, follow-ups and compiling, and frees up hours.

So the useful question is not “how many positions will AI cut?” but “what will we do with the hours it frees up?” That decision belongs to management, and it is made before the project, not after.

Statistics Canada has been asking businesses about the effect of AI since 2024. Here are the results published for the second quarter of 2025, among businesses that had used AI in the previous 12 months.

The same survey measures the effect on tasks previously done by employees. In 2025, 47.2% of businesses using AI reported a small reduction, 32.4% a moderate reduction and 5.3% a large one, while 15.1% saw no reduction. In Quebec, the Institut de la statistique du Québec reports that 37.7% of businesses say AI has reduced those tasks to a moderate or large extent.

These numbers have limits. They cover the 12 months before the survey, at a time when most businesses are only at their first uses. They also rest on what businesses report, not on payroll records. Nobody can say for sure that the trend will hold once agents take on whole processes. What can be said is that the outcome is not automatic: a business that automates chooses what it does with the freed hours.

Put differently, AI nibbles at repetitive work without making positions disappear. That matches what we see on the ground: an agent that enters invoices removes the keying, not the person who understands suppliers, jobs and exceptions.

Once tasks are automated, management has three options. They are not mutually exclusive.

Les Industries Romy, a structural steel fabricator in Granby, faced falling orders caused by US tariffs. It needed to prospect, but its three project managers spent their weeks on invoices, follow-ups and coordination.

According to management’s estimate, AI agents now free up 152 hours a week, including 109.2 hours for the project managers, who went from 120 to 10.8 hours of administrative work. No position was cut. The project managers reinvest those hours in client relationships, and the owner says he sends roughly two to three times as many quotes. The drafter, freed from producing the drawings he now reviews, spends 80% of his time on project management.

These figures are management’s estimates, not yet measured over a full year. The details are in the Romy case study and in our article on the hours freed.

In a well-run project, the human share is not a leftover: it is designed. A price, a deadline, a commitment to a client or a payment above a threshold always goes through a person. An email to a client is left as a draft. The drafter approves every drawing before it goes to the shop.

That split moves work toward what calls for judgment: understanding what the client really wants, catching the exception the rules did not foresee, negotiating. Those are also the tasks employees find most interesting. An AI project that promises zero human hours is really promising that nobody will be left to check.

How you run the project matters as much as the technology. Five practices make the difference.

An AI acceptable use policy puts in writing what AI may do, what it never does alone and which data it may read. On the Law 25 side, our compliance support covers the privacy impact assessment. And when the agents run as sovereign AI, on a server on your premises, employee and client data stays inside the company.

To go further, our guide on how to integrate AI in your business lays out the plan, and our pages for manufacturers, law firms, healthcare and real estate show uses by industry. If you run Business Central or Acumatica, the agents connect directly. Read about our approach.