How to Use AI in Your Business: Twelve Jobs

Using AI in a business starts with a list of jobs, not a list of tools. Pick a job you already do and write down the steps. Then keep a person on the result. The tool is the last thing you choose, and for many jobs it is a small one.

Cited in this note

7%

Statistics Canada asked businesses that already use AI which tools they use and what they changed.¹ In its 2025 report, 35.7% of them used text analytics, which is software that reads and sorts written text. Another 23.1% used marketing automation. Also, 40.1% had built new ways of working, and 38.9% had trained their staff. In the same report, 12.2% of all businesses used AI to make goods or give a service, up from 6.1% a year before. These numbers show what firms said they did. They are not a shopping list, so there is no reason to open twelve tools on Monday.

The help note explains the gap between making goods with AI and using AI to write and sort. The automation note shows how to pick the first job. The agents note covers the time when the steps start to branch, and a fixed checklist no longer fits. We’d start with one job from the list below. Then we’d leave the rest alone until that job has a person on it and a week of real use behind it.

Twelve jobs you can write down

In each job below, AI does the first try and a person makes the call.

• AI drafts the customer reply, and a person sends it.
• AI sums up the meeting, and a person accepts the tasks.
• AI turns notes into a quote, and a person sets the price.
• AI sorts the inbox, and a person decides what gets an answer.
• AI flags the late bill, and a person decides how to chase it.
• AI describes the product, and a person approves the page.
• AI groups the help requests, and a person picks the answer.
• AI spots the stock problem, and a person places the order.
• AI rewrites the policy in plain words, and a person posts it.
• AI compares two supplier notes, and a person chooses.
• AI drafts the weekly report, and a person accepts the numbers.
• AI prepares the staff answer, and a person trains the staff.

Every line has the same shape. The tool gets the work ready, and a person decides. If you take the person out, you no longer have a job. You have a mailbox that hopes for the best. The reply can be drafted from the thread the customer wrote, but someone who knows that customer still sends it. The meeting summary can list the tasks, but someone who was in the room still accepts them. The tool was not there to hear the aside that mattered. The quote can be built from notes, but the price stays with the owner. We keep coming back to this shape because owners try to skip the second half, and the second half is the business.

The numbers above are about businesses that already use AI, except the 12.2% figure. That one covers all businesses in the Canadian survey. So please don’t read the text analytics share as a share of every firm. When we say 35.7%, we mean that share of the businesses already using AI, not the shop on the corner. Marketing automation at 23.1% is the same kind of share. The 40.1% and 38.9% for new ways of working and training are also among firms that already use AI. They tell you what those firms changed. They don’t say that a firm that has never tried this is behind, and they don’t say which of the twelve jobs to run first. The 12.2% figure stands apart, and we don’t want it mixed in with the others. Using a tool to draft a reply is a different claim from using AI to make the goods or give the service.

A worked example

Take a shop we keep using as our walkthrough. It is not a case study or a scoreboard. It is just a counter, a back room, and three people. The owner buys and sets prices, a clerk covers the floor and the inbox, and a bookkeeper chases what was not paid.

Yesterday, in this story, looked like any other day. A customer emailed about a size, and a supplier sent a new price sheet. Two bills sat past the date the bookkeeper cares about. A talk with the landlord left a page of scribbles and no tasks. The product page for a new item was still blank. None of that needs a new company, but someone does need to say which of those is the job for now.

Say they pick the customer reply, because it comes up every day. A wrong tone with a regular costs more than any chart can show. The clerk drops the thread into a draft. The draft pulls in the size the customer asked about and the line the owner uses for delays. The clerk reads it and changes one sentence, because this customer dislikes cheery notes. Then the clerk sends it. The tool did not know about the cheery notes, but the clerk did, and that is the whole design.

You can sort the inbox in the same pass if you like. A person still decides what gets answered today and what can wait. The late bill goes on a list the bookkeeper already trusts, and the bookkeeper decides how to chase it. A stiff note to a good account takes good sense. A reminder to a stranger who has been quiet for sixty days is closer to a rule. Even then, we like a person to take a quick look.

The quote job shows up the day the shop does custom work. Notes from a talk become a description, and the owner sets the price. We’ve watched owners let the draft set the price because it sounded sure of itself. But sounding sure is not the same as knowing the cost. The supplier comparison is the same trap in a different coat. There are two notes with two minimums and two lead times, all written in different ways. A draft can line them up. The owner chooses, because choosing is the job and lining up is only the chore.

A clinic can run the same list without acting like a shop. The practitioner accepts the tasks from the meeting summary, and a person trains the staff on the answer the tool prepared. A person also posts the plain-word policy. That step matters because a policy that sounds friendly but says the wrong thing about care is worse than a stiff one that is right. A trades desk lives in the quote and the stock problem. The lead tech’s note becomes a description, and the owner sets the price. Then a person checks the problem and places the order. We are not saying every firm should run all twelve. We are saying that each of the twelve is a job with an edge, and the edge is a person.

How we’d run it

First, write the twelve down where the owner can see them, in the firm’s own words. “Flag the late bill” might mean “the Tuesday list the bookkeeper already prints.” So use that name for it. A job you can’t say in their words is not a job you can hand over.

Second, pick one with a simple test. It should happen often, the result should be easy to check, and a mistake should show up before a customer gets hurt. The customer reply and the late bill flag usually pass that test. The weekly report often fails it, because a wrong number becomes the record if nobody is named to accept it. So if you can’t name the person who accepts the numbers, don’t start there.

Third, write the steps of that one job the way it happened yesterday, not the way a brochure tells it. Say who gets the message and who is allowed to send. Say what done looks like and where the draft waits. If the draft doesn’t wait, you skipped the design.

Fourth, run it on live work with the old way still allowed. The clerk can still type a reply from scratch. You are watching to see if the draft gets good enough that they stop wanting to. If they keep typing from scratch, then either the draft is wrong or the job was the wrong one. Fix that before you add a second job.

Fifth, name the person who checks, in a sentence another worker could read. It might say, “The clerk sends customer replies, and the owner sends anything with a price or an apology.” That sentence will save you more pain than a new tool. Group the help requests the same way, so the tool groups them and a person picks the answer. Describe the product the same way, so the tool drafts and a person approves the page.

Sixth, only then look at a second job. The pattern in the survey that we trust is not the tool share. It is the pair of follow-up moves, which are new ways of working and staff who were truly trained. The 40.1% and 38.9% are shares among firms already using AI, and even there they are not everyone. A second job with no new steps and no trained person is how the first job dies. So leave the rest alone until the first job feels dull.

Common mistakes

Starting from a list of tools is the mistake the title is about. Someone asks which chatbot to buy before they can say which reply they want drafted. The tools will pile up, but the jobs won’t get any clearer. We’ve sat at tables where three logins were already paid for. Nobody could point to a sent reply that had passed through any of them. The money was small, but the mess was not.

Treating the survey as a ranking of your firm is the next one. A text analytics share among AI users is not a grade. The 12.2% for making goods and giving services is a different question, and it is smaller for a reason. Most of the help open to a small firm is still a draft, a sort, or a summary. That is fine, but calling it “AI in the product” when it isn’t will send you chasing a build you don’t need. The help note walks through that gap. Use it before you scare yourself with the small figure or flatter yourself with the large one.

Skipping the person on the result is the mistake that reaches the customer. A drafted reply that goes out alone will sooner or later be warm to the wrong person. Or it will be sure about a price the owner never set. A weekly report that nobody accepts will be quoted in a meeting as if it were the books. A policy that is rewritten and posted unread can change what staff think the rule is. The second half of every line in the list is there because we have seen the first half travel alone.

Running all twelve at once is impatience dressed up as ambition. A shop that tries replies, bills, stock, the product page, and the staff answer in the same month will do all of them badly. Then it will decide the tools don’t work. But the tools were never given a job, only a pile. So pick one and let it get boring, because boring is what success looks like here.

Forgetting training is how the survey figure and the real room come apart. A new way of working that lives only in the owner’s head is the old business with an extra tab. Show the clerk on a real thread and the bookkeeper on a real late bill. If the only person who can run it is the person who set it up, you don’t have a job. You have a hobby.

What we’d do

We’d pick the customer reply or the late bill and turn down every other job for a while. That is our view, because twelve examples are still only a list, and a list doesn’t run the counter. One checked job does. We’d write the steps in the clerk’s words and put their name on the send. Then we’d sit with them once on real mail before we called it done.

We wouldn’t buy a second tool to cheer the first week. We wouldn’t hand the quote price to a draft because the sentences looked finished. We wouldn’t post a rewritten policy just because it sounded friendly. And we wouldn’t read 35.7% as a target for our shop, since it is a share of a different group, in a Canadian survey, among firms that already use AI. It works as a map of what those firms touch, but not as a quota.

If you do nothing else after reading this, circle one line from the twelve and write the name of the person who finishes it. If you can’t circle one, you don’t have a use yet. You have curiosity, which is a fine place to start and a poor thing to automate. The question about goods and services can wait until a job you already do has a person on the result and a path someone else could follow.