What small business owners need to know about AI

01Where Canadian businesses are

Statistics Canada asks, and it asks businesses that have employees. In the second quarter of 2026, 19.2% of them reported using AI to produce goods or deliver services over the twelve months before the survey. Two years earlier that figure was 6.1%, so reported use has tripled in two years. Separately, in the third quarter, a quarter of businesses, 25.2%, reported plans to use AI over the next twelve months, up from 14.5% a year earlier.

Read those as two different measures. One counts what businesses said they had done. The other counts what they said they intend. Both have risen across the last three surveys, and that rise is a measured fact about the past rather than a prediction about the next three.

The survey covers businesses with employees, across every size class, so a sole operator is not in it at all and a six-person shop is averaged in with companies far larger than yours. Two in five businesses, 40.0%, said AI is not relevant to their business.

The number cannot tell you whether you are behind, because it says nothing about your business or about the people you compete with. What it does say is that in the middle of 2026 most businesses with employees were not reporting AI use in producing their goods or delivering their services, and that the share reporting it rose in each of the last two years. Neither of those settles anything for you.

02Before you accept that you are behind

Start with the problem you can name in your own business, not with what you imagine others have bought.

If you do want to look outward, ask what has changed in the way the businesses you compete with answer a phone, send a quote or follow up on a job. That is evidence about what a customer would notice. It is weaker evidence about the rest, because a better schedule, a lower cost or more capacity may not show from outside at all.

Ask where the pressure comes from. If it is coming from customers who want something you cannot give them, that is a business problem worth solving. If it is coming from vendors and from general noise, slow down.

03Four uses that matter to an owner-operated business

AI does more than the four things below. It generates images from a written description, among plenty else on offer. These four are the ones I would expect to touch a business with a phone, an inbox and a small team. Ask a supplier to demonstrate them on material like yours before you pay. The checks below are how I would assess the result; they are not a promise about what will work in your business.

Writing and rewriting text. Quotes, follow-up emails, job descriptions, the website page you have been meaning to fix for two years. Give it a job you have already done and compare what comes back against what you actually sent. It will still need someone who knows the business to correct it.

Listening and summarising. A phone call, a site visit recording, a meeting. What is offered is notes and a list of what was agreed. Part of that is transcription, and transcription accuracy depends on the recording: excessive background noise, echo and two people talking at once can each reduce it. Whether the summary and the list of agreements match what was said is a separate question again. So check the first several against the recording itself rather than against what you remember. Before you record a customer, check what you are allowed to record and who has to be told.

Sorting and routing. Reading an incoming message, deciding what kind it is and sending it where it should go. If your problem is that messages arrive in four places and get answered late, this is the part to look at first. It needs the accounts and permissions to reach each of those places, which is work in itself. It will also misread some messages, and confidently, so read a sample of what it routed as well as what it set aside, and sample again after anything changes in how work arrives.

Answering questions from material you give it. Your price list, your service area, your policies. Supplying your own documents points one of these systems at your business rather than at the world in general. It does not make the answers reliable. These systems produce confidently stated content that is wrong, including answers that contradict the material they were given, and they produce confident reasoning for wrong answers as well as for right ones. Treat any customer-facing use as needing a person to check anything consequential before it goes out, and a way for the customer to reach a human.

04Three ways this costs owners money

Buying a tool before naming a problem. A tool gets bought because it was impressive in a demo, and then somebody has to find work for it. Name the thing that costs you money or time, then ask whether any capability above touches it.

Paying for something nobody opens. Software you have had for years gets new features, and an add-on approved once becomes a line on the monthly bill nobody looks at again. Before you buy anything new, list what you are already being charged for and open each one.

Automating a process that is already broken. If your quotes go out late because nobody is sure who is responsible for them, automating the writing of the quote will produce late quotes faster. AI applied to a broken process makes the breakage more efficient.

05What is worth looking at first

The enquiries you already paid to get. Pull your call records, count the calls nobody answered and look at when they landed. Then work out what one real enquiry is worth to you, because a handful of missed calls from people ready to buy can be worth more than a long list of wrong numbers and suppliers. If those calls are worth having, an automatic text back to a missed call is a modest piece of technology to try. Judge it on work that came back, not on how many people replied, and leave it running long enough to see more than a handful of missed calls.

The repetitive typing in the day. Quotes, follow-ups, the same answer to the same question. This does not need a platform. It needs the work looked at once, honestly, to see what is genuinely repeated rather than what feels repetitive.

What only lives in someone's head. If your pricing, processes and customer history sit with one or two people, start by making that information searchable. Choose the tool after that: it cannot find information nobody has written down. Before you upload customer records or call recordings anywhere, check who can see them, how long they are kept and whether they are used to train the supplier's systems. Start with made-up examples until you have those answers.

06What to leave alone for now

Anything customer-facing where a wrong answer costs you the customer, until a person reviews consequential replies before they go out. "Someone is checking it" has to mean checking before the customer sees it, not after.

Anything that gives software discretion over money. Keep the invoicing and payment automation that is tested and working, along with the controls around it: someone who reviews what went out, a limit on what can go out unseen, and a way to reverse a mistake. A rule that repeats reliably is not by itself a safe rule, because it repeats a wrong amount as faithfully as a right one. A system deciding what to charge, what to refund or when to chase, on its own judgment, needs a person in front of it.

Replacing a person who is the reason customers stay. If people call because they like talking to Sharon, the technology question is how to give Sharon fewer interruptions, not how to remove her from the phone.

07Five questions to ask anyone selling you AI

1. Which specific thing in my business does this change, and how will I know in thirty days?
2. What does it cost in the second year, including what happens if my staff numbers change?
3. Where does my customer data go, who else can see it, and is it used to train anything?
4. If I stop paying, what do I keep?
5. What does this get wrong, and what happens when it does?

The fifth question separates the vendors worth talking to from the ones who are not. These systems can produce output that is wrong, because producing plausible output is what they are built to do and plausible is not the same as correct. A supplier who will not tell you how theirs fails has either not looked or is not saying.

08What I would tell you

I have run technology for eighteen years and built and sold a company, and the advice I give owners on this is duller than the marketing around it. Start with small repetitive work: the phone calls that come in after you close, the booking someone tries to make on your website at ten at night. Get one of those working and living in your business for a while. Then look at the next piece.

The important thing is not how far along you are. It is that you have looked once, honestly, rather than assumed the answer. Do that with me or do it yourself, but do it.

09Where to start

You do not need a strategy for this. Start with the work that gets stuck and the questions you want answered.

That is what the free hour is for. I come to you, we go through your priorities and where the gaps are, and you get a high-level direction. No obligation, and no quote on the spot, because a real quote needs a scope we have agreed together.