AI automation for a small business means handing the repetitive computer work your team does every week to software: quotes, invoices, follow-up emails, data entry, reports. In most of the automations I build, the AI part is small. Plain code does most of the work, and an AI model only steps in where something needs judgment.
I've been building software for clients since January 2020. One of them, a landscaping business called Mastersealer, saves more than 10 hours a week on quotes and invoicing because of an app I built for them, and that app doesn't use any AI at all. I'll get to why later on.
What AI automation is, and what to expect from it
If it's a task someone does on a computer, you can probably automate it. Copying data from one tool to another, writing the same kind of email over and over, turning a form into a PDF, pulling numbers into a weekly report. A machine does all of that faster than a person, and it makes fewer mistakes.
You have two kinds of tools for the job. Plain code handles the steps that should run the same way every time; it never makes anything up, and it costs almost nothing to run. An AI model (an LLM, like the ones behind ChatGPT or Claude) handles the steps where it has to read messy text or make a judgment call.
Most business processes are almost entirely the first kind. The quotation app I built for Mastersealer is plain code from start to finish. The process was simple enough for it, and I didn't want an AI model anywhere near the numbers on a customer's quote.
A lot of business owners come to me expecting AI to take over someone's job completely. With today's models, that doesn't happen. AI can do about 90% of a person's work on a task, and a human still has to review the last 10%. Models keep getting better, and I expect that number to move one day. If you're automating your business now, though, plan for the technology we have now.
Break jobs down into tasks
The biggest change in thinking is to stop looking at roles and start looking at tasks.
If I asked you to automate your video editor, you wouldn't know where to start. List what the editor actually does, though, and it gets much easier: they clean up the audio, cut the silent parts, add captions and background music, and upload the finished video. Each of those is a small task you can automate on its own, and once enough of them are automated, most of the job is too.
I did exactly this for my own YouTube channel, and my videos are edited and posted automatically now. I didn't build it in one go. I started with a few scripts to clean up the audio, then added silence trimming, then captions, then uploading, one step at a time. Between that pipeline and the other ways I use AI, I get at least 20 to 30 hours a week back. It still takes real work to keep running, and I don't want to pretend otherwise.
Any business works the same way. Mastersealer didn't need to "automate the office". Their 10+ hours a week went into a handful of separate tasks: customer communications, quotes, invoicing and follow-ups. Each one could be looked at, and automated, on its own.
You also don't have to automate a task 100% for it to be worth doing. If someone spends 10 hours a week on something and you automate 80% of it, they get 8 hours back, every week. Waiting for a perfect, fully automatic version usually means getting nothing.
I go through this way of thinking in more detail in my video on how to build a business that runs itself.
How to choose what to automate first
Don't brainstorm. When people sit down and ask themselves what they should automate, their gut gives them bad answers: something that would be fun to build, or something they do twice a year.
Track everything you do for a week instead, in as much detail as you can stand. A week is long enough to see where your time really goes, which is rarely where you think it goes. When I do this with a client, I give their team a way to log their work. Depending on the business, that might be a spreadsheet, a small app that records what each person is doing, or an automatic time tracker.
Then look for tasks that eat the most time and that come back over and over, following the same steps each time. The more manual and boring a task is, the easier it is to automate.
I also tell clients to leave some tasks alone. If something only happens once a month, building the automation can take longer than doing it by hand for years. If it needs complex reasoning or real creativity, people still do it better. And if a mistake would hurt a customer or the business, a person should be doing it, or at least signing off on it.
Before you build anything, write the steps down. If you can't do the task by hand and explain every step of it, you can't automate it reliably. I still haven't automated my YouTube thumbnails for exactly this reason: I don't have a repeatable process for them yet.
I cover the whole method, including the warning signs, in what to automate first.
Do you need an AI agent? Usually not
Everyone wants an AI agent right now. Almost nobody needs one.
An AI agent is an AI model that gets a goal and a set of tools, and decides by itself which steps to take. I use AI every day and agents can be really useful, but they're also the most expensive and least predictable way to automate something. A lot of the business owners I talk to have bought into the hype without knowing that.
I'd build an agent in two situations. The first is when people need to ask open-ended questions in plain language and still get work done, like a chat assistant that answers questions about your products and can act on them too. The second is when the work has many moving parts and a model has to pick the right tool for each one, because the steps change every time and a fixed workflow can't follow them.
For everything else I'd build a workflow: a fixed set of steps that runs the same way every time, with an AI model called in only for the one or two steps that need judgment.
AI models are never right 100% of the time, and that adds up when you chain them. Say each agent in a chain gets it right 90% of the time. Two in a row get it right 81% of the time, and by the seventh you're under 50%, which makes your automation a coin flip.
So in most of my builds, I use AI to write the code once, and then that code runs thousands of times, the same way every time, at almost no cost. In my experience about 80% of any workflow can be plain code, and only 10 to 20% needs an AI model to think.
The full comparison is in AI agents vs workflows, and I explain my reasoning in the video on why AI agents are overrated.
Automations break, so plan for it
Every automation breaks eventually. The tools it connects to get updated, or an AI model returns something nobody expected. That part is normal. The real damage happens when nobody finds out.
I've seen businesses with lead-generation automations, customer follow-ups, invoicing systems and more that quietly stopped working as the software under them changed. Nobody noticed for weeks, because nothing was set up to tell anyone, and for all those weeks the work those automations were supposed to do just didn't happen.
So every automation I build keeps a log of each step it takes, which means that when something goes wrong I can see exactly what happened. If it fails, the client and I both get an alert straight away (clients on a retainer can leave the alerts to me). And anything sensitive waits for a person: if the output goes to a customer or makes a permanent change to the business's data, someone reviews it first.
At Mastersealer, the app builds and sends the quote, but Johnny, who runs the business, still checks every quote before it goes out. The software does the work and a person makes the final call, so a mistake never reaches a customer.
If someone has to babysit an automation all day, all you've done is swap one kind of work for another. More on keeping them running in how to make AI automations reliable, and in my short videos on logging and alerts and requiring final approval.
What AI automation costs
I don't have a price list, because almost everything I build is custom software for one business. I quote each project on its full scope after we've talked it through, and we agree on the price and the timeline before I start.
A project gets bigger when it has to connect to a lot of other tools, or when it needs an AI agent that juggles many tools or answers from a knowledge base. Screens add to it too: if your team needs a dashboard or an internal app to work with, that's more to build. Most of my projects take a couple of weeks, and bigger builds take longer.
The Mastersealer quotation app is a good example of a small one. It took me somewhere between 10 and 20 hours to build, and since then it has needed a minor fix about once every six months.
After launch there are a few running costs to expect. If the automation uses AI, you pay for model usage, though for most small-business tasks cheap open-source models are good enough and several providers offer free tiers of them. A small company with a few automations may never leave the free tier. You also pay for hosting. Mastersealer's internal tools don't get much traffic, so I moved them to a free hosting tier, which saved them $500 a year. Add any third-party subscriptions the automation relies on, and a retainer if you want me to keep things running and add features later. The retainer is optional, and most clients skip it.
I try to keep running costs as low as I can for every client. There's a full breakdown in what AI automation costs.
A real example: Mastersealer
Mastersealer is a landscaping business in Vaughan, Canada. When they came to me, Johnny was writing every quotation by hand and emailing each one as a PDF himself, and a lot of those emails were landing in spam.
I built them a quotation app: Johnny fills in a short form, and the app turns it into a branded PDF and emails it to the customer. It's plain code with no AI in it. I also moved their email to Google Workspace and cleaned up a long list of errors in their domain settings. Their internal app had been running on a server that was switched on all year, even though the business is seasonal, so I moved it to a free tier.
The results were 10+ hours a week saved on quotes and invoicing, email deliverability up 50.61%, $430,000 CAD in quotations sent in three months that would otherwise have landed in spam, and $500 a year saved on hosting.
The big wins came from fixing problems nobody at the company could see, and none of them needed fancy AI. You can read the full Mastersealer case study, or watch me walk through it in how I generated $430k CAD for my client.
How to get started
If you want to do it yourself, block out dedicated time for it, like an evening or a weekend. Pick the one part of the business that looked worst in your week of tracking, automate just that, and make sure it works before you move on to the next one. Trying to automate everything at once is how you end up with nothing finished.
If you'd rather have it done for you, that's what I do. We talk about your business, I quote the full project with a timeline, I build it and keep you updated, and then I deploy it or hand it over. If you're not sure what to automate yet, I can help you work that out first. Here's how I work and what I build.
