AI agents in a small business — what to automate first

Every week another tool promises to "automate your business with AI". Most small businesses that try end up with a chatbot nobody uses, or a clever workflow that saves ten minutes a month. The problem is usually not the technology. It's the choice of the first job.

Here is how to pick one that pays off.

Look for work that is frequent, boring and checkable

The best first jobs for an agent have three things in common:

1. They happen often. Daily or weekly, not once a quarter. A job you do fifty times a month is worth automating. One you do twice is not.
2. They follow a pattern. A person could write down the steps, even if there are exceptions.
3. The result is easy to check. You can glance at what the agent produced and tell whether it's right.

Frequent, boring and checkable. If a job is missing one of the three, leave it for later.

Good first candidates

• Sorting the inbox. Reading incoming enquiries, labelling them (quote request, complaint, supplier, spam) and drafting a first reply for a person to approve.
• Chasing. Unpaid invoices, missing documents, unsigned quotes. Polite reminders on a schedule, stopping when the reply comes in.
• Getting information from one place to another. Enquiry form to CRM, email attachment to accounts software, booking to calendar. This is where most small-business admin time goes.
• Summaries. A morning digest of yesterday's enquiries, open quotes and overdue jobs.
• First-pass research. Pulling together what's publicly known about a prospect before a call.

Poor first candidates

• Anything where one mistake is expensive. Paying suppliers, sending contracts, changing prices on the website. Automate the preparation, not the final click.
• Emotionally delicate conversations. Complaints, bad news, anything a customer will remember how you handled.
• Work that's rare or different every time. The agent needs a pattern to follow.

Keep a person in the loop, at first

The safest pattern for a first agent is draft, then approve: the agent does the work, and a person clicks send. Two weeks of approving drafts tells you exactly how good the agent is. Then you can let it act on its own for the cases it always gets right, and keep approval for the rest.

Measure before and after

Before you start, write down roughly how long the job takes each week. A month later, write it down again. If you can't see the difference, pick a different job. That's not failure, it's the process.

The first agent is a lesson

What you learn from the first agent (where your data lives, which tools connect, what your team trusts) makes the second much faster to set up. Pick the first job to learn from, not to transform the business, and the transformation tends to follow.