The top barrier to AI in Canada isn't cost or privacy. It's relevance.

I published a piece about British Columbia's AI numbers on September 14. The next morning, at my desk in Langley, I did the thing I should have done first: instead of reading Statistics Canada's write-up of the barriers question, I downloaded the table behind it. All 8,442 rows of it, every province, every industry, every employment size band.

The write-up quotes you the national row. The table shows you the shape.

Here is the shape. Nationally, 40.0% of businesses say the thing limiting their use of AI is that it is not relevant to the business. Cybersecurity and privacy is 13.4%. Cost is 10.6%. And the smaller the company, the louder the first answer gets, while privacy and cost get quieter.

That last part matters more than the headline, because it means the objections my industry writes its marketing against belong mostly to the companies that have already adopted. Every "AI is more affordable than you think" campaign is aimed at a 10.6% problem. Every privacy reassurance is aimed at firms big enough to have a compliance officer. The company with four employees is not sitting there worried about cost. It is sitting there unconvinced that any of this has anything to do with what it sells, and almost nobody writes to that.

TL;DR. Statistics Canada put the question to 21,105 businesses in April and May 2026, and 9,251 answered: what limits your use of AI? The top answer was that AI is not relevant to the business, at 40.0% nationally and 40.6% in British Columbia. Cybersecurity and privacy came next at 13.4%, then cost at 10.6%.

The gradient: among businesses with one to four employees, 41.4% say AI is not relevant. Among those with 100 or more, 21.3%. Privacy runs the other way, from 11.6% to 30.0%. Across 16 industries, the "not relevant" share and the share actually using AI correlate at -0.81, which is my own calculation from tables 33-10-1169 and 33-10-1167 rather than a Statistics Canada figure.

The other number: a second question, asked a quarter later and only of the businesses with no plans to adopt, puts relevance at 79.1%. Narrower base, different sample, and both numbers are true. Below is what each one measures, and what to do if you are in the 40%.

What Statistics Canada asked, and who answered

Two questions, one survey, two different bases. Almost every argument about this data comes from mixing them up.

The figures come from the Canadian Survey on Business Conditions. For the second quarter of 2026, Statistics Canada invited 21,105 businesses and 9,251 responded, fieldwork ran from April 1 to May 6, and Vicky Do, Shivani Sood and Chris Johnston published the analysis on June 11.

The first question, table 33-10-1169, goes to every business: what limits your use of AI? The second, table 33-10-1208, goes a quarter later and only to businesses with no plans to adopt: why not? Both allow more than one answer, so neither totals 100%.

Two things before I start quoting numbers at you.

The questions word the option differently and I keep them apart throughout: 33-10-1169 says "not relevant to the business or organization," 33-10-1208 says "not relevant to the goods produced or services delivered." Either way it is the owner's own judgment, not anyone's assessment of them.

Every estimate carries a quality flag, set from the standard error as a percentage of the estimate: A under 2.5%, B under 5%, C under 7.5%, D under 10%. I name it whenever it is worse than B. The national 40.0% is A, the B.C. 40.6% B.

The full dataset is here, 597 rows, every figure I use below carrying the flag Statistics Canada gave it.

Relevance outranks cost and privacy three to one

Nationally, 40.0% of businesses name relevance. The next answer is less than a third of that.

Cybersecurity or privacy concerns, 13.4%. Cost, 10.6%. Uncertainty about the benefits, 10.0%. Lack of skilled workers and regulatory concerns, 7.0% each. Data limitations, 3.8%. All rated A.

British Columbia, where I work, gives almost the same answer: 40.6% relevance, 14.5% privacy, 10.4% cost, 8.6% regulatory, 7.6% uncertainty, 5.8% skills, 4.7% data.

I keep coming back to that B.C. line, because B.C. has not been slow about this. It leads the country on AI use, 25.1% against 19.2%, and on plans to adopt next year, 31.3% against 25.2%.

If "not relevant" were mostly a symptom of being behind, the province furthest ahead should report less of it. It reports slightly more.

One more number, which I have not seen anyone quote. Asked what limits their use of AI, 23.0% of Canadian businesses said nothing does. No barriers at all, and more businesses gave that answer than named privacy or cost.

It is highest among the smallest firms, at 24.8% for one to four employees. It does not fall steadily with size, though: the low point is 16.5% at 20 to 99 employees, with the largest firms back up at 17.6%.

Whatever is happening between small Canadian businesses and AI, being blocked is not the common version of it.

The smaller the business, the more the answer is "not relevant"

At one to four employees, 41.4% say AI is not relevant. At 100 or more, 21.3%. Privacy moves in the opposite direction over the same range, from 11.6% to 30.0%.

Credit where it is owed: Autana, a Vancouver firm, put those two figures side by side in August, before I did. What I would add is the crossing point, somewhere between 20 and 99 employees. That swap is my industry's whole marketing problem in one image.

Cost never leads in any size band. It peaks in the middle, 15.1% at 20 to 99 employees against 9.8% at 5 to 19 and 11.3% at the largest. A mid-market worry, not a small-business one.

Look at who each barrier belongs to. Privacy, skilled workers and data limitations all rise with headcount, and those are the problems of an organisation that has decided to do this and hit the mechanics. Relevance falls with headcount, and it is not mechanical at all. It is the prior question of whether your work is the kind of work this technology does.

I pulled the B.C. cut too, which I do not think anyone has published: 41.4% at one to four employees, 20.5% at 100 or more, privacy climbing from 11.4% to 32.7%. Both 100-plus figures carry a D flag, so take that end of the B.C. gradient as indicative and the national one as solid.

If you run a five-person shop in Chilliwack, the barrier the survey finds for you is not one a vendor can sell you out of.

Cost and privacy are loudest where adoption is already highest

The sectors that complain about privacy, regulation and cost are the sectors that have already adopted. The sectors with the lowest adoption barely mention them, and say "not relevant" instead.

The five highest and five lowest of the sixteen industries Statistics Canada publishes, by relevance share. The six middle rows, from construction at 44.8% down to retail at 41.1%, are in the CSV. Barriers from 33-10-1169, adoption from 33-10-1167, both Canada, Q2 2026. Mining's relevance and the finance adoption figures carry C flags, the rest are A or B.

Across all sixteen, the "not relevant" share and the share that used AI give a Pearson correlation of -0.81. That is mine, calculated from the two tables rather than taken from Statistics Canada, and you can redo it from the CSV in a minute.

A correlation that strong invites a reading I want to resist. The unit is the industry, not the firm, so it says nothing about any one business. It does not tell me the relevance judgments are correct, nor which way the causation runs, since an industry that adopted AI has by definition found something relevant to do with it. What it shows is that this question sorts industries along the adoption line.

I ran the same calculation on every other barrier, and they do not all behave alike. Regulatory concerns come out at +0.89 against adoption, privacy at +0.85, cost at a looser +0.61. Lack of skilled workers is weakest at +0.37, and my own table shows why. Finance is the second highest adopter in the country and health care the fourth, and they report a skills barrier of 3.4% and 3.6%, while manufacturing reports 12.0% on an adoption rate of 13.1%.

It is the compliance and security barriers that track adoption, not the hiring one.

One row still argues with me. Wholesale trade reports 36.8% relevance on an adoption rate of just 7.9%, and I have no explanation for it.

Why 40% becomes 79% when you change the question

Ask only the businesses that have no plans to adopt AI, and 79.1% of them say the reason is that AI is not relevant to the goods they produce or the services they deliver. In B.C. it is 78.9%.

Same survey, narrower base. My 40.0% is the share of the whole economy naming relevance as a limit. The 79.1% is one subgroup of it, the businesses that have already decided against AI, answering a quarter later. Both are true, neither restates the other, and quoting one without its base is how I see this data misread.

The rest of the holdouts' list is worth reading for how short it is: privacy 10.8%, not knowing what AI can do 9.9%, not mature enough 9.2%, too expensive 4.4%, bias 4.4%, skills 2.8%, data 2.6%, laws 2.0%. And "previous or current use did not meet expectations" sits at 2.3%. Almost none of them are there because of a bad experience.

I wanted a second Canadian source before trusting any of it. The Bank of Canada's staff analytical note 2026-22, by Chawla and Arnburg in June, put AI questions to the December 2025 Business Leaders' Pulse, and of the firms not adopting and with no plans to, the Bank writes, "a majority cite a lack of relevance to their operations as the reason." I read that twice before using it: the Bank says plainly that the Pulse is built to read aggregate conditions rather than to produce population-representative estimates, and it rests on 314 responses. I cite it because it agrees, not because it settles anything.

Then set it beside the American numbers that fill the search results. NEXT Insurance surveyed 1,500 US small businesses in April 2025 and found 62% citing a lack of understanding, 55% citing cost. A real sample, answering a question put to owners already weighing AI. Put that question to a whole economy instead and the top answer changes.

Some of the four in ten are right

Relevance is a judgment about the work, and the owner is better placed to make it than I am.

Two people with nothing to sell me arrive at the same place.

The Information Technology and Innovation Foundation argued in August that Canada aims its adoption support at the firms least able to convert it into productivity, citing a BDC survey of 1,247 firms in which 25% of micro-businesses using AI reported reduced costs against 41% of firms with 100 or more employees. Lawrence Zhang's argument is about policy. What carries over is that identical adoption pays differently at different sizes.

Avi Goldfarb, speaking to The Hub in June, described the mechanism. Early adoption is mostly point solutions, a tool dropped into a process that does not change around it, and "the productivity benefits of point solutions are necessarily incremental." The real gains arrive with what he calls system change, redesigning the workflow itself.

A twelve-person company has fewer workflows to redesign, and less slack to redesign them with. Incremental, at that size, can round to nothing.

None of this proves any particular business is right to sit it out. Plenty of owners say "not relevant" when they mean "I have not looked," and from the outside those are hard to tell apart. From the inside they are not. That is what the five-question test in the previous article is for: run the five questions against one task, not against the business.

What to do if AI genuinely doesn't apply

Nothing with the word AI in it, this year. Here is what I would do instead.

Get the work into something you can query. Data limitations sit at only 3.8%, which I read as most holdouts not having got far enough to hit the problem rather than their records being in order. If the job lives on paper, in phone calls and in one person's memory, that is the real project, and it pays on its own merits whether or not anything automated follows.

Write down one process. The cheapest item here, and it pays whether the answer is yes or no.

Pick a number and record it today. Not one you would start tracking to justify a purchase. One you already have.

Look again in a year. B.C. moved from 11.5% to 25.1% adoption in twelve months, and whatever moved it has not finished. A no about 2026 is not a no about 2028.

I have one thing to sell and would rather be straight about it. If you run the test and cannot tell how you scored, say so and I will tell you what I would do. If you score a clear no, you do not need me, and I would rather you learned that from the table than from a sales call.

I spent a morning on a spreadsheet to learn that four in ten Canadian businesses looked at AI and decided it was not for them, and that the smaller they are, the more often they say so. My industry has spent two years insisting those owners are confused about cost or nervous about privacy. Cost ranks third among the barriers, and the people worrying about privacy are mostly the ones already using it.

I do not know how many of the four in ten are right. Neither does anyone selling to them. The difference is that they can find out, one task at a time.

Data appendix

Cite this as: Raf Apocalypse, "The top barrier to AI in Canada isn't cost or privacy. It's relevance.", Pixel2 Consulting, September 15, 2026.

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

Figures come from Statistics Canada tables 33-10-1167, 33-10-1169, 33-10-1207 and 33-10-1208, and from the Canadian Survey on Business Conditions analysis of June 11, 2026, all extracted September 15, 2026. The full 597-row dataset if you want to check any of it.