When an owner starts thinking about AI, the first question is usually which tool to buy. Somebody at a chamber breakfast mentioned an app, or a nephew set up ChatGPT on the office computer. Start somewhere else. Start with the work your business does the same way every week, and find out what it costs you.
Put a dollar figure on the boring stuff
Pick one repetitive task and time it for a week. Copying web enquiries into a spreadsheet. Texting crews who forgot to clock out. Building the month-end report the bank wants. Writing the fourth "yes, we deliver to Beaumont" email of the day.
Then do the math. Say your office manager spends three hours a week on quote requests and earns $35 an hour. That's about $5,500 a year, and it doesn't count the jobs that went to someone else because your reply went out on Thursday instead of Monday night.
Now you have a number to hold any tool, or any quote, against. If a fix costs less than a year or two of that, it's probably worth doing. If it costs more, keep doing it by hand and move to the next task.
What to automate first
The best first targets happen often and follow the same steps every time. In most Edmonton service businesses, these come up first:
- •Web enquiries that someone retypes into a spreadsheet or a CRM
- •Quotes that need a follow-up when the customer goes quiet
- •Timesheets, missed clock-outs and the scramble before payroll
- •Review requests after a finished job
- •Monthly reports built by hand from three different dashboards
- •The same handful of customer questions, answered by email over and over
If you can write the steps on a sticky note, a system can probably do them. If the steps start with "well, it depends," leave that one with a person for now.
Where AI helps, and where it doesn't
AI is good at reading and writing. It can sort an inbox, turn a rambling voicemail into a tidy job ticket, draft a reply that sounds like you, or boil a week of timesheets down to the few lines the owner should look at.
It's bad at being exactly right every single time. Alberta pays overtime on hours past 8 in a day or 44 in a week, whichever adds up to more. Ask a chatbot to apply that to a crew's week and it will usually get it right. "Usually" is a problem on a pay stub.
So the split is simple. AI drafts and code calculates. Before anything goes to a customer or touches money, a person approves it.
The version you can try this week
You don't need to hire anyone to start. A ChatGPT or Claude subscription costs less than a tank of gas a month. Paste in a customer email and ask for a reply in your tone. Paste in your price list and your delivery area, and ask it to draft answers to the questions you get most.
Your email program already has saved replies for the messages you send ten times a week. Tools like Zapier can move a form submission into a Google Sheet without a line of code, and they have free plans to test on.
Do that for a month. You'll learn which tasks AI handles well in your business, and you'll have real hours saved to point at.
When it's worth paying for a system
The do-it-yourself tools hit a wall in predictable places: when a task runs across programs that don't talk to each other, when a mistake costs real money, when a dozen people depend on it, or when nobody has time to babysit it.
We built one of these for an industrial shop south of Edmonton. Crews punch in on tablets at each bay, the system works out Alberta overtime, and anyone who forgot to clock out gets asked when they actually left before payroll runs. It keeps punching when the shop Wi-Fi drops. There's no AI in the payroll math at all. That part is plain code, because it has to be right.
The price of a system like that only makes sense next to the hours it replaces, which is why the first step is always timing the task.
Keep a person on the send button
The fastest way to lose faith in automation is letting it email a customer something wrong. A good system lets a person look before anything goes out, at least for the first few months. Our own blog drafts arrive as one link to approve, and nothing publishes until someone clicks it.
Trust builds from there. Once you've approved forty drafts without changing a word, you'll know which ones can go out on their own.
This week, keep a list of every task you or your staff did twice the same way. Time the three longest. If one of them costs you more than a few thousand dollars a year, that's your first automation, and we can tell you what it would take to build.