Where AI Saves a Small Business Time, and Where It Doesn’t

Where does AI actually save time?

The tasks below have three things in common: they come up often, a first draft is most of the work, and a person can check the result quickly. In each one the tool proposes and a person decides.

• Reading receipts and invoices. A tool can pull the date, the supplier, the total and the tax from a photo. The check: compare it with the bank or card transaction it belongs to, and fix what it misread.
• Drafting quotes from rough notes. Notes from a site visit become a tidy quote in your own format. The check: every price, quantity and exclusion is yours, read line by line before it goes out.
• Routine email replies. Booking confirmations, follow-ups, answers to the questions you get every week. The check: read it the way the customer will, and send it yourself.
• Sorting transactions. Suggesting the account each bank line belongs to, based on how similar ones were handled before. The check: someone who knows the business approves each one, and anything unusual gets a closer look.
• Summarizing documents. A long contract, a lease or a set of meeting notes, boiled down to the points that matter. The check: open the original for any clause you are going to rely on.

None of these removes the person. What they remove is the blank page and the retyping, which is where much of the time goes.

What should you not hand to AI?

• Final tax and accounting decisions. Whether a cost is deductible, how a payment should be treated, what to file. A tool can help you find the rule, but the decision and the responsibility for it stay with you, not with the tool.³
• Anything sent to a customer unchecked. A quote, an invoice or a reply is yours once it is sent, whoever wrote the first draft.
• Client or customer information pasted into a free consumer tool. Names, addresses, account numbers, health or financial details. Under PIPEDA, the federal private-sector privacy law, a business stays responsible for personal information it hands to another company to process.⁴

Why can a confident answer be wrong?

Generative AI produces what is statistically likely in response to a prompt, based on what it was trained on, which is why it can be fluent and wrong at the same time.¹ The Government of Canada’s guidance for its own staff says plainly that these tools can produce content that looks credible and reasonable but is inaccurate, and can point to sources that do not exist.¹

The Canadian Centre for Cyber Security adds that answers can be incorrect, can leave out factors that matter, and can be biased, and that content should be checked against credible sources before anyone acts on it.² The same federal guidance warns about the habit that makes this costly: some people accept a system’s output without question, even when it is wrong.¹

In practice that comes down to three habits. Check every figure against the document it came from. Open any source an answer cites. And be most careful where an answer sounds most certain, because certainty is not evidence.

Where does your data go?

Whatever you type or upload goes to the company that runs the tool. What happens next depends on its terms: whether your inputs can be used to train its models, how long they are kept, who can see them and where they are stored. The Cyber Centre’s advice is not to share private information with an AI tool unless you understand what it does with your data.²

The Office of the Privacy Commissioner of Canada asks organizations to use anonymized or de-identified information in prompts where they can, and to enter personal information only where they are authorized to.³ For a small business the simple version is to take the names and numbers out before you paste, or to use a tool whose terms you have read and your clients would be comfortable with.

How do we use it ourselves?

We use AI in our own bookkeeping, under the rule this guide describes. Our privacy policy puts it plainly: it proposes; a person decides. Nothing an AI suggests is booked to a client’s accounts without a bookkeeper reviewing it first.

The same policy names the one AI provider we use, says what reaches it, and says the provider may not use that information for its own purposes, including training its models. What the tool saves us is mostly reading and retyping. It does not save judgment, and we do not ask it to.

How can you start in an afternoon?

A first step does not need a project. A short checklist:

1. Pick one task you do every week that starts with a blank page or with retyping.
2. Read the tool’s terms on training, retention and storage before you put anything real into it.
3. Try it on a made-up or anonymized example first.
4. Decide who checks the result, and what they check, before you rely on it.
5. Run it beside your usual way for a couple of weeks, and keep a note of what it got wrong.
6. Keep it away from sending, paying or filing anything until you trust what it produces.

If you would like a second pair of eyes on where it fits your business, our AI Business Assessment looks at exactly that.

Frequently asked questions