How Small Businesses Are Really Using Generative AI

Layla LevesqueOctober 7, 20267 Mins read3

How Small Businesses Are Really Using Generative AI

Photo: Free laptop coffee shop photo via Rawpixel, CC0

In the spring of 2024, Statistics Canada asked thousands of businesses a simple question: had they used artificial intelligence to produce goods or deliver services in the previous year? Just 6.1 per cent said yes. When the agency asked again in April and May of 2026, the figure had climbed to 19.2 per cent. That’s a tripling in two years, and it happened mostly on the back of chatbots and writing assistants that cost less per month than a decent lunch.

Look at the same survey from another angle, though, and a different picture appears. Four in ten Canadian businesses told StatCan that AI simply isn’t relevant to what they do. Restaurants, farms, trucking firms and builders remain far behind law offices, software shops and insurers. And the businesses that have adopted AI mostly use it for a handful of narrow jobs rather than anything you’d call a transformation.

So what are small and mid-sized firms actually doing with generative AI, what does it cost them, and where does it fall flat? We went through the most recent government and industry data to find out.

The numbers: faster growth than expected, but uneven

Statistics Canada tracks AI through its quarterly Canadian Survey on Business Conditions. The second-quarter 2026 round, conducted from April 1 to May 6 with 9,251 responding businesses, is the most detailed snapshot available. Here’s how adoption has moved:

Survey period	Businesses using AI to produce goods or deliver services
Q2 2024	6.1%
Q2 2025	12.2%
Q2 2026	19.2%

Size matters, but perhaps less than you’d guess. According to StatCan’s 2026 analysis, 27.8 per cent of businesses with 100 or more employees used AI, while 19.9 per cent of businesses with just one to four employees did. The bigger divide is attitude. More than 41 per cent of the smallest firms, and 41 per cent of those with five to 19 employees, said AI wasn’t relevant to them, compared with 21.3 per cent of large firms.

Industry matters even more. Information and cultural industries (42.3 per cent) and finance and insurance (40.4 per cent) led the pack, with professional, scientific and technical services close behind. Construction sat at 9.2 per cent and agriculture, forestry and fishing at 4.5 per cent.

International surveys tell a similar story. An OECD survey of more than 5,000 SMEs in seven countries, including Canada, found in 2024 that 31 per cent were using generative AI tools. The Business Development Bank of Canada landed on almost the same number: in a February 2026 survey of 1,500 owners, 30 per cent of SMEs said they used generative AI. (StatCan’s lower figure reflects a narrower question about using AI to produce goods or services, not just dabbling with ChatGPT.)

What small firms actually use it for

Forget the robot-run company. The most common uses are modest, and that’s largely the point. Among Canadian businesses using AI in 2026, StatCan found the top applications were:

Data analytics (36.6 per cent): summarizing sales figures, spotting trends in spreadsheets, building simple forecasts.
Text analytics (34.5 per cent): sorting customer emails, pulling key terms out of contracts, digesting reviews.
Virtual agents and chatbots (28.2 per cent): answering routine customer questions on a website or messaging app.
Natural language processing and large language models (27.0 and 24.8 per cent): the general-purpose writing and question-answering tools most people mean when they say “AI”.

Qualitative research from the U.S. fills in the texture. When the Federal Reserve Bank of San Francisco analyzed open-ended answers from its 2024 Small Business Credit Survey, owners described using AI for productivity tasks, social media and search optimization, written communications, visual design, customer service and some coding. Nearly 40 per cent of respondents were using it or planning to.

In practice, that looks like a bookkeeper drafting routine client letters, a small agency producing first drafts of ad copy, or a contractor turning rough notes into a cleaner quote. None of it is glamorous, but it adds up.

Where the evidence for gains is strongest

The best-documented productivity result still comes from customer support. A widely cited study by economists Erik Brynjolfsson, Danielle Li and Lindsey Raymond, published through the National Bureau of Economic Research, followed 5,179 support agents using a generative AI assistant. Issues resolved per hour rose 14 per cent on average, and 34 per cent for novice and lower-skilled workers. The most experienced agents saw minimal gains.

That pattern is useful for small business owners. AI tends to help most where a newer employee needs to perform like a seasoned one: answering common questions, drafting standard documents, following a playbook. It helps least where someone already knows the job cold.

BDC’s own data points the same direction. Its survey found AI-using SMEs generated, on average, 24 per cent higher sales per employee than non-users after controlling for industry and location, as The Logic reported. That’s a correlation, not proof that AI caused the gap. Better-run firms may simply be quicker to try new tools.

What it costs

The sticker price of generative AI is low, which is one reason adoption has climbed so fast. Microsoft’s Copilot Business add-on, which puts AI into Word, Excel, Outlook and Teams, was listed at US$18 per user per month on an annual plan when we checked, or more on month-to-month billing.

The real costs are elsewhere:

Time to learn. Staff need to figure out which tasks AI handles well and how to check its work. StatCan found that 68.1 per cent of large AI-using firms trained existing employees, but only 24.0 per cent of the smallest ones did.
Review overhead. Every AI-drafted email, quote or report still needs a human to read it. If that review takes nearly as long as writing from scratch, the savings vanish.
Integration. Connecting a chatbot to your booking system, inventory or CRM usually means paying a developer or consultant, and that bill can dwarf the subscription fees.
Security. BDC found 45 per cent of SMEs had experienced a cyberattack in the previous 12 months, up from 17 per cent in 2021. New tools that touch customer data add to the attack surface.

BDC’s survey also offered a telling detail about planning. Owners with a formal AI plan reported 85 per cent satisfaction with their investments, against 66 per cent for those without one. Those who trained employees reported 86 per cent satisfaction, versus 53 per cent for those who didn’t. In our view, that gap is the single most practical finding in the recent research.

The risks small businesses tend to underestimate

You own what your chatbot says

Canada already has a landmark case on this. In February 2024, British Columbia’s Civil Resolution Tribunal ruled in Moffatt v. Air Canada that the airline was liable for its website chatbot, which had wrongly told a grieving customer he could apply for a bereavement fare discount after travelling. Air Canada argued the chatbot was effectively responsible for its own actions. The tribunal rejected that and awarded the passenger roughly $650 plus interest and fees, according to a summary by law firm Torkin Manes.

The dollar amount was small. The principle wasn’t. If a small business puts an AI assistant on its website, the business is answerable for what it tells customers about prices, refunds or warranties.

Privacy rules still apply

Canada’s federal, provincial and territorial privacy commissioners issued joint principles for generative AI in December 2023. Among other things, they advise organizations to use anonymized or de-identified information in prompts wherever possible. Pasting a client’s full file, including their health details or SIN, into a public chatbot is exactly the behaviour those principles warn against. Whether a specific use complies with PIPEDA or provincial law depends on the details, so it’s worth checking with a privacy lawyer before building AI into anything that handles personal information.

Hallucinations and confident errors

Large language models can invent facts, citations and numbers in fluent, plausible prose. For a marketing tagline, that’s a minor annoyance. For a tax summary, an employment contract or a safety procedure, it can be costly. The rule of thumb many firms settle on: AI drafts, a qualified person decides.

Where it doesn’t pay off

Not every experiment succeeds, and some high-profile research suggests most don’t. MIT’s NANDA initiative reported in August 2025 that about 95 per cent of enterprise generative AI pilots it studied were not producing measurable profit-and-loss impact, as Fortune reported. The study, based on interviews, employee surveys and public deployments, found that buying tools from specialized vendors worked far more often than building them in-house, and that back-office automation often delivered better returns than flashier sales and marketing projects.

McKinsey’s 2025 State of AI survey found that 88 per cent of organizations used AI in at least one business function, yet only 39 per cent reported any effect on earnings before interest and taxes. Smaller companies were also slower to scale: 29 per cent of firms with under US$100 million in revenue had moved beyond pilots, versus 47 per cent of those above US$5 billion.

And sometimes the tool makes people slower. In a 2025 randomized trial by research group METR, 16 experienced software developers took 19 per cent longer to finish tasks when allowed to use AI tools, even though they believed they had been sped up. METR has since said newer models may change the result, but the perception gap is worth remembering.

Based on that evidence, generative AI tends to disappoint when:

The work is already done by an expert who rarely gets stuck.
Errors are expensive and checking takes as long as doing.
The business has no clean, digital data for the tool to work with.
The core of the product is human contact, as in many trades, hospitality and personal services.

A sensible way to start

The firms that report the best results tend to follow a similar, unexciting path. Pick one repetitive task that eats hours every week, such as answering the same customer emails, writing listings or summarizing meeting notes. Try a mainstream tool for a month. Measure the time saved honestly, including the time spent checking. Write a one-page policy on what data staff can and can’t paste into AI tools. Train people, even informally.

If the numbers work, expand. If they don’t, stop. For anything touching contracts, taxes or regulated data, get advice from an accountant or lawyer before leaning on AI output.

The real takeaway

The story of small business AI in Canada isn’t one of wholesale transformation. It’s one of steady, practical uptake in knowledge-heavy industries, a stubborn gap in hands-on sectors, and a large group of owners who haven’t yet seen why it matters to them. As The Hub noted in June 2026, Canadian firms now roughly match U.S. adoption rates, but they lean on point solutions rather than redesigning how work gets done. That’s where the productivity gains are likely to come from next. The tools are cheap. The planning, training and judgment around them are the part worth investing in.