Two things separate the Canadian businesses getting value from AI from the ones that aren't
BDC surveyed 1,500 Canadian business owners and found that 78 percent of SMEs using AI are satisfied with the return. That number climbs to 86 percent among firms that train their staff and falls to 53 percent among those that don't. The tool is not the project. Here is what the data actually says, and what it means if you're deciding whether to spend.
Most of the writing about small business and AI is either a sales pitch or a warning. There is not much in between, which makes it hard to answer the only question an owner actually has: if I spend money on this, do I get anything back?
The Business Development Bank of Canada published a study in June 2026 that gets closer to an answer than most. It surveyed 1,500 Canadian business owners and decision makers in February 2026, and rather than asking whether they were excited about AI, it asked what they had adopted, what they had gotten out of it, and how they had gone about it. The last part is where the useful finding is.
The headline number, and why it is the least useful part
The number that made the press coverage was $350 billion. BDC’s economic research team built a digital maturity score out of 100, covering technology use, corporate culture, data management, and how intensively a firm uses AI. It then modelled what would happen if the roughly 92 percent of Canadian SMEs below the top tier operated at the level of today’s top-performing 8 percent. The answer: productivity up nearly 38 percent, and Canadian GDP up close to 14 percent, or about $350 billion. AI on its own accounts for roughly 6 percent of GDP, around $150 billion, in BDC’s estimate.
Those are macroeconomic modelling numbers. They are interesting for policy and almost useless for a business owner in Scarborough deciding whether to spend $8,000 on something. A national GDP figure does not tell you whether your quoting process should be automated.
The adoption numbers are more grounded. Ninety-six percent of Canadian SMEs now use at least one digital technology, up from 91 percent in 2021, so the “get a computer” phase of this is over. But just under a quarter, 23 percent, score high or very high on digital maturity, made up of 15 percent high and 8 percent very high. And only 30 percent use generative AI at all.
So the gap is not awareness. Essentially everyone has bought software. The gap is between owning tools and running the business on them.
The finding worth acting on
Here is the part of the study that is genuinely useful, and that we have not seen quoted much.
Among Canadian SMEs that are using AI, 78 percent say they are satisfied with the return on investment. That is a high number on its own, and it is the one a vendor will quote at you. The interesting thing is what happens when you split that group:
| Group | Satisfied with AI return on investment |
|---|---|
| All SMEs using AI | 78% |
| Have a formal technology adoption plan that includes AI | 85% |
| Using AI with no plan | 66% |
| Train their staff on AI | 86% |
| Use AI without training staff | 53% |
Read the training row twice. Firms that trained their people were satisfied at 86 percent. Firms that bought the same category of tool and skipped the training were satisfied at 53 percent. That is a 33 point spread, and it is the widest gap anywhere in the data.
It is also the cheapest variable on the list. A formal adoption plan takes a few meetings. Training staff on a tool you already bought costs a fraction of the licence. Neither requires better technology, a bigger budget, or a data scientist. They are the two things a small business fully controls, and they appear to matter more than which product you picked.
The obvious caveat: this is a correlation, not a controlled experiment. Businesses organised enough to write an adoption plan and run training are probably better run in other ways too, and better-run businesses get more out of everything they buy. The study does not isolate cause. But the direction is consistent, the mechanism is plausible, and the intervention is cheap enough that the distinction is somewhat academic. Nobody was going to regret training their staff.
The same study reports that SMEs using generative AI show about 24 percent higher productivity than those that don’t, which carries the same caveat and should be read the same way.
Where the AI actually went, and where it didn’t
The study also found where Canadian SMEs are putting AI, and the pattern is telling. It clusters in sales, marketing, administrative work, management, finance, and HR. It shows up far less in production, distribution, logistics, inventory, and service delivery.
That is the reverse of where a lot of small business waste actually lives. Writing marketing copy faster is a real saving, but it is a saving on a task that was already cheap. The expensive problems in most of the businesses we work with are operational: a scheduling process that eats a day a week, inventory counts done from memory, quotes that take three days to go out because the information lives in four places. Those are less glamorous and considerably more valuable, and almost nobody is applying anything to them yet.
There is a structural reason for that, and the study names it: close to half of SMEs holding digital data have systems that are not integrated, or only partially integrated. AI on top of disconnected systems does not have much to work with. If your customer list is in one tool, your invoices in another, and your job history in a spreadsheet on somebody’s desktop, no assistant is going to answer “which of our customers are overdue and also profitable.” The plumbing has to exist first. That is unglamorous work, and it is usually the actual prerequisite. We have written before about why we consolidate onto one database rather than four services, and this is the business version of the same argument.
The barrier nobody budgets for
One more figure from the study, because it lands close to what we spend a lot of time on: 45 percent of Canadian SMEs reported a cyberattack or attempted attack in the previous 12 months, up from 17 percent in 2021. Cybersecurity now ranks as the second largest barrier to digital adoption, behind only cost.
That is not a coincidence sitting next to the AI numbers. Every system you connect is another door, and AI tools in particular tend to want broad access to your email, files, and customer records in order to be useful. We wrote a longer piece on the specific design rule that keeps an AI agent from becoming a breach, and the short version is that the question to ask a vendor is not “is it secure” but “what exactly can it reach, and what can it send out.” Most of our security writing is downstream of the same idea.
There is money available, and it is a loan
Worth knowing if cost is the blocker: on April 24, 2026, BDC launched a program called LIFT aimed specifically at getting Canadian SMEs off the AI sidelines. It is $500 million, targeted at helping more than 1,000 businesses, and it combines advisory support with financing for technology adoption, covering digital tools, data infrastructure, cybersecurity, and equipment like automation and robotics. Loans run from $25,000 to $5 million, with the option to postpone principal payments for up to two years.
Read the terms carefully, because that is debt, not a grant. Borrowing $200,000 to buy software you have not scoped is a worse outcome than not borrowing. But if you have a specific operational problem, a costed plan, and a payback you can defend, the financing exists and the advisory piece is arguably worth more than the money.
BDC’s president and CEO, Isabelle Hudon, framed the launch this way: “SMEs are stretched thin and finding time for AI is a real challenge. But their competition isn’t waiting.” Both halves of that are true, and the first half is the one most coverage skips.
What we would actually do
If you are a small business owner reading this and wondering what the next step is, in order:
- Pick one expensive process, not one exciting tool. The question is “where does this business lose the most hours or the most money,” not “what can AI do.” If the honest answer is quoting, or scheduling, or chasing invoices, start there.
- Check whether the data exists and is connected. If the information the tool would need lives in three disconnected places, the integration work is the project and the AI is the last five percent. Better to know that before you sign anything.
- Write the plan down, even if it is one page. What the tool is for, who uses it, what success looks like in ninety days, and what it replaces. The study’s 85-versus-66 split says this is worth an afternoon.
- Budget for training, and treat it as non-optional. This is the 86-versus-53 number. If your quote for a tool has no line item for getting people to use it, the quote is incomplete.
- Ask what the tool can access before you connect it. Especially anything touching customer records, email, or financials.
- Measure the thing you said you would measure. Ninety days later, check whether the hours actually came back. If they did not, stop paying for it. That discipline is the difference between the businesses in the 86 percent and the ones in the 53.
One honest note on the research itself: BDC’s survey used a non-probability sample, so the results are not statistically projectable to all Canadian businesses and no margin of error can be assigned. That is BDC’s own disclosure, and it is the right way to read the numbers. Treat them as a strong directional signal from a large sample, not as census data.
The underlying point survives the caveat, and it is not really about AI. Businesses that decide what a tool is for, tell their people how to use it, and check afterward whether it worked will get more out of any purchase. That was true of accounting software and it is true of this. What has changed is the size of the gap between the firms that work that way and the firms that don’t.
Update, August 26, 2026: point 5 above now has a specific Canadian instance. The list says to decide deliberately what a tool is for before you buy it. Since we published, the most common AI purchase a Canadian small business will make has quietly stopped being a decision at all. On July 1 Microsoft made Business Standard with Copilot and Business Premium with Copilot permanent products, and microsoft.com’s main business plans comparison page now offers Standard and Premium only in those editions. The base versions are still sold at CAD $19.00 and CAD $29.80, but you have to arrive at a different Microsoft page to find them.
The gap is CAD $12.90 per user per month, which is CAD $154.80 per seat per year, so a twenty-five person firm renewing on the default path commits about CAD $3,870 a year to an AI tool nobody in the building has decided they need. Read against the numbers in this post, that is the 53 percent path: a tool bought without a plan, without training attached, and without a ninety-day check. It is not an argument against Copilot. Buy it for the people who will open it daily, and where you do, buy the bundled edition rather than the add-on, because the bundle is genuinely the cheaper route. The full price comparison and the renewal timing that decides when any of it hits your invoice are in the post on the July 2026 Microsoft 365 repricing.
If you have a process that is eating time and you want a straight answer on whether it is worth automating, send us a note. Sometimes the answer is that it is not, and that is a useful answer too.