Does your small business need AI? A five-question test

I went looking for a number last week that I was sure I already knew.

Everyone in my industry repeats that small companies are falling behind. It gets said from stages, in pitch decks, and across most of the marketing aimed at owners, almost always with no source attached. So I pulled the Statistics Canada tables expecting to find my province trailing by a few points.

It's the other way around. B.C. businesses used AI more last year than those in any other province or territory, and the sharpest increase came from the small ones.

The harder number came out of the same survey and points the opposite way. Among B.C. businesses with no plans to use AI at all, 78.9% say the reason is that AI doesn't apply to what they make or sell. Not cost. Not privacy. Relevance. Both are true at once, and almost everything written about AI for small business tells you the first and pretends the second isn't there.

TL;DR. B.C. businesses used AI at 25.1% over the last 12 months against 19.2% nationally, yet 78.9% of the province's holdouts say AI doesn't apply to their work. Some of them have looked hard and concluded there's nothing there yet. Some haven't looked. From the outside those are hard to tell apart. From the inside they aren't, and the test below is how you tell.

What actually happened in British Columbia this year

B.C. went from slightly behind the national average to 5.9 points ahead of it in 12 months.

In 2024, 5.1% of B.C. businesses reported using AI to produce goods or deliver services in the previous 12 months, against 6.1% nationally. In 2025, 11.5% against 12.2%. Close both times, and on the wrong side of the line both times. Then 2026: 25.1% in B.C., against 19.2% for Canada.

Quebec came in at 20.1%, Alberta 18.4%, Ontario 18.1%. The closest is Yukon at 20.6%, flagged for caution because the sample is small, and that cuts both ways: the true figure could sit above 20.6 as easily as below. B.C.'s lead over Quebec is the one I'd call settled.

What this measures matters later. It's self-reported and the bar sits very low: one person using ChatGPT to draft supplier emails counts. It tells you how many businesses have touched AI, not how many get anything out of it. The figures come from the Canadian Survey on Business Conditions, 9,251 responses, and I've published the full dataset with Statistics Canada's quality flag on every figure. Check my work.

The jump came from the smallest companies

B.C. firms with 5 to 19 employees used AI at 24.9% last year against a national 14.9% for the same band. That 10 point gap is, as far as I can tell, the widest any size band holds in the country.

That band sat at 5.1% in 2024 and 8.0% in 2025, behind the national figure both years. Then 24.9%, tripling in a year while the national figure for companies the same size grew by about 60%.

Since I'm being pedantic about other people's sources, here are my own. The two the headline rests on, 25.1% and 19.2%, are rated A. The size-band figures behind the chart above, 25.1% for 1-to-4 and 24.9% for 5-to-19, are rated B, as is the 78.9% below. The one band where B.C. trails, 20 to 99 employees at 24.3% against 25.8%, is rated C, the same caution I applied to Yukon, so that dip may not exist. The 37.0% for firms of 100 or more carries a D, and it's the number I'd trust least on this page.

So the "small businesses are being left behind" line isn't holding up in this province. If you run a 12 person company in Surrey and you've spent a year feeling late, the data says you're probably not.

Which is worth knowing and isn't an answer. Being on time tells you nothing about whether any particular job inside your company is worth automating, and that's the question that costs real money to get wrong. The rest of this piece is about that one.

The numbers only look like they disagree

Go looking for how many Canadian businesses use AI and you'll find answers from 8% to 71%. That isn't six sources disagreeing about one quantity. It's six different questions.

Sort them by what was asked and they fall into three groups. Ask whether anyone has used AI at all, 45% to 71%. Ask whether AI is in the work itself, or whether anyone paid for it, 17.7% to 19.8%. Ask whether the use is significant, 8%. That last figure is the Bank of Canada's top band, not a competing estimate.

The middle group is the one I'd trust: three different questions with blind spots pointing opposite ways, one counting bank transactions rather than asking anybody, landing within a couple of points of each other. I've since pulled that apart properly, because the weighting matters more than the wording: why published adoption rates run from 17.7% to 78%.

None of these numbers, mine included, tells you whether AI is doing real work inside a business.

Four in five holdouts say it doesn't apply to their work

Statistics Canada asks businesses with no plans to adopt AI why not, one quarter after the adoption figures above. In British Columbia, 78.9% say AI is not relevant to the goods they produce or the services they deliver. Businesses can pick more than one reason, so these don't total 100. Everything else is a rounding error beside it.

A different question in table 33-10-1169, put to all B.C. businesses a quarter earlier rather than only the holdouts, gives the same shape: not relevant 40.6% (rated B), privacy 14.5%, cost 10.4%. Breaking that out by size and industry is a separate piece.

On September 14, 2026 I logged the first page of organic results for five searches, 50 in total: does my small business need AI, AI for small business BC, should I use AI in my business, is AI worth it for small business, and AI adoption small business Canada. Every result treated "not relevant" as a mistake it needed to correct. Run them yourself and see if it still holds. The framing is always that you don't find AI relevant because you don't understand it yet, and here are 11 use cases to fix that. The non-adopter is never once allowed to be right about their own business. I'm not claiming all of them are right. Plenty say "not relevant" when they mean "I haven't looked," and from the outside the two look identical. That's why my industry treats them as the same thing.

One case of my own data correcting me. I assumed a slice of the gap between the high and low answers would be businesses that tried AI and stopped. Statistics Canada asks exactly that: 1.5% in B.C., 2.3% nationally. Almost nobody is a holdout because of a bad experience.

A five-question test for whether AI fits a task in your business

This is for owners deciding whether AI is worth starting at all. It measures whether one specific task is the kind of work automation can take over. Run it one task at a time, not against the whole business.

1. Does this task happen at least weekly, and in the same shape every time?

No if either half is a no. Weekly is workable; monthly rarely repays the maintenance. On shape, I don't mean the same content: every invoice is different, but processing one is the same job every time.

2. Could you explain this task to a new hire in under five minutes?

If yes, the rules are explicit and a machine can follow them. If the honest answer is "you'd have to watch Maria do it for a week," they aren't written down anywhere. That's a documentation problem, and software bought before you solve it moves the mess somewhere more expensive.

3. Does the task use text, numbers or images already sitting in a system?

Email, spreadsheets, your accounting package, a CRM, photographs, PDFs are all workable. Work that happens on a phone call, on paper or on a job site is not, unless you first build the system that captures it. That's a bigger project than you think.

4. If the output were wrong, would you find out within a day, and would checking be cheap?

No if either half is a no, and this is the one I'd keep if I could keep only one. If a bad output can sit undetected for a month, the first half fails. If verifying takes as long as doing the work, the second fails. Either way you've moved the labour, not removed it.

5. Is there a number you already track that would visibly move?

Already track. Not a number you'd start tracking to justify the project. If nothing on your dashboard would move, either the task doesn't matter or you can't tell. Both are reasons to wait.

These five come out of engagements I've scoped, including the ones I turned down.

Scoring isn't complicated. Four or five yeses, start with that task. Three is borderline; the decider is usually question 4. Two or fewer, the answer is no for now, which is a real answer, not a failure.

One warning: the test measures fit, not whether anyone will pay you. I built a diabetes platform, Glicemias Online, that reached about 16,000 patients in Brazil, never found a paying customer, and shut down in 2020. It passes four of these five cleanly, failing question 3: the readings lived on paper, and ten months went into building the system that captured them. The test still wasn't what killed it. What I never checked was whether the people with the problem were the ones with a budget. Ask that one on its own.

What to do with your answer

If you passed, write down the number before you buy anything. Take the metric from question 5 and record where it sits and how you measured it. Without it you can't tell in six months whether it worked. You'll have opinions, not evidence. Then automate a single workflow, not five. Every company that size I've worked with got there the same way: one small thing that worked, then the next.

If you failed, most of the honest list isn't AI. If question 3 was a no because the work lives on paper and in phone calls, that's your actual project: most of what people attribute to AI comes from being able to search your operation. If question 2 was a no, a process lives in one person's head, a risk whether or not you automate. And if the task happens twice a month and nobody is complaining, you've found a part of your business that works. Look at one that doesn't.

One more thing the test can't do. It scores a task you already suspected, one at a time. It won't walk your operation and tell you which of the 40 jobs in it should be at the top of the list, or which three are load-bearing enough that getting them wrong would hurt. That ranking is a different piece of work, and it's the one I get hired for. If you'd rather do it yourself, run the five questions across every recurring task you can name and sort by question 4. That's most of the method.

Then revisit when you've done the work the failed questions pointed at, not on a calendar reminder. Something changed in B.C. between 2025 and 2026 and I don't expect it to sit still. A no today is only a no about today.

I went looking for a number I was sure I knew, and it came back 5.9 points the other way. The only reason I found out is that I checked. Almost everyone telling you you're behind is selling something downstream of you agreeing with them. I'd rather you ran the test. If you're unsure how you scored, book a call and I'll tell you what I'd do.

Raf Apocalypse is the founder of Pixel2 Consulting in Langley, B.C. He has spent 20 years building software, including two diabetes platforms that together reached more than 16,000 patients, and has shipped systems for Nike Brazil, MercadoLivre and Hospital Albert Einstein.

The B.C. figures come from Statistics Canada tables 33-10-1167 and 33-10-1208, extracted September 14, 2026. The full 597-row dataset.