What AI agents can actually do for a small business
AI agents are software that takes on one defined job and acts on it. What that looks like in a small business, where it works, and where it doesn't.
Search for “AI agents for business” and page one is mostly software companies ranking their own products, plus a few reports written for large enterprises. Useful if you are choosing a platform. Less useful if the question on your mind is simpler: which of the jobs in your business could an agent actually take on, and what happens when it gets one wrong?
This article answers that question, using two real agents that are running in small businesses today.
What an AI agent is
An AI agent is software that takes on a job and acts on it, rather than only answering questions in a chat window. It works through steps towards a goal and uses tools to get there: a phone line, an inbox, a CRM, a job-management system.
The difference from a chatbot is that an agent does something. A chatbot tells a caller your opening hours. An agent answers the call, takes the caller’s details and puts them where your team will act on them.
What agents are doing in small businesses today
The clearest examples are also the least glamorous. Both of these started from the same problem: a trade business losing enquiries because the phone rang while everyone was busy.
- A solar installer in Queensland. When the team is already on a call or out on site, an agent picks up instead of letting the call ring out. It asks for the caller’s name, number and what they are after, then emails those details to the team for a callback. Read the case study.
- A window tinting company in Victoria. Same starting point, one step further: the agent adds each caller’s details into ServiceM8, the job-management system the business already runs on, exactly as a website enquiry would arrive. Read the case study.
Neither agent replaces anyone. Each takes one job that was falling through a gap, and hands the result to a person.
The same pattern fits other jobs that arrive often and look alike: taking down enquiry details and sorting them, or updating the systems your team already works in so nobody re-types what a customer already said.
Where agents work well
An agent is a good fit when the job is:
- Narrow. One clear task, not “help with admin”.
- Frequent. Something that happens every day, not twice a month.
- Predictable in what it needs. The same few pieces of information, collected the same way each time.
- Costly when missed. A missed call that becomes a lost customer is worth solving. A slightly slower internal report may not be.
- Easy to hand over. There is an obvious point where a person takes over.
Where they don’t
Agents are a poor fit for judgement calls, for anything where a mistake is expensive and hard to spot, and for jobs that are rare or different every time. They are also a poor fit for a process that is already broken: automating a broken process only makes it break faster. Sometimes the better answer is a simple rule, a tidier form, or leaving the job with a person, and part of the work is saying so before anyone spends money.
Which AI agent is best for a business?
It depends on the job, and on what has to happen afterwards. An off-the-shelf app suits a standard job that ends when the task is done, like taking a message. If the result needs to land inside your own systems, a CRM, a job system, a shared inbox, an agent built into those systems saves someone copying it across by hand.
Whichever route you take, ask three questions before you commit:
- Where is your data processed? If nobody can tell you, that is your answer.
- What happens when it gets something wrong? There should be a clear point where a person takes over.
- What does it cost to run each month, not just to set up? Voice agents, for example, are typically charged by the minute.
How to start
Start with one job, not a transformation programme.
- Write down the job: who does it now, how often, and what it costs when it is missed.
- List exactly what the agent needs to collect, and where that information should end up.
- Decide when a person takes over.
- Run it where the risk is lowest first, such as overflow calls only, and read what it actually does.
- Adjust, then decide whether the next job is worth an agent at all.
If you want help finding which of your jobs an agent should take on, and which it shouldn’t, that is where every AI project starts. For phones specifically, see the AI receptionist.