The Small Business Owner's Guide to AI agents

What even is an AI agent?

Fair question, because the word "AI" gets attached to everything now and most of it is marketing.

The plain version: an AI agent is a program that answers questions in normal language, based on information you gave it.

The best way to picture it is an assistant who has read every document about your business and never forgets any of it. It's not a person, though, and that's the important part. It can't use judgement, it can't read the room, and it doesn't know anything you didn't tell it. But it has read your entire price list, your FAQ, your policy documents and your website, and it will answer questions about any of it instantly at any hour of the day.

So when someone asks it "do you deliver to Croydon", it looks up your delivery area in the content you gave it and tells them the truth. It doesn't need to understand the question the way you'd understand it. It only needs to find the right bit of your content.

That's the whole idea.

Is it worth it?

You're right to be sceptical. Most software for small businesses is a waste of money, and an AI agent is no exception. So let's be specific.

It's probably worth it if:

• You or your staff answer the same handful of questions constantly
• You get enquiries when you're closed, and they bother you
• You don't have anyone whose job is support
• You already have a website with content on it, because that's what the bot reads

It's probably not worth it if:

• Someone already handles support properly and has time to spare
• Your questions are all one-off and complicated
• You don't have ten minutes to set it up
• You're hoping it will replace understanding your customers

That last one deserves expanding, because it's the most common disappointment. An AI agent doesn't know your customers. It knows your documents. It will handle "I'm wondering if you'd be able to do something different on the invoice" about as well as a form letter, which is to say: not well. If your business runs on those conversations, you want a person.

For what it's worth, we build ChatterEase, so we've tried to lean sceptical in the list above.

What can an AI agent do for your business?

Here it is by industry, because what "good support" looks like differs.

A restaurant. "What time do you open?" "Are you open Sunday?" "Do you do gluten-free?" "Can I book for eight on Friday?" It answers from your menu and your opening hours, and captures the booking request so it arrives with the details attached rather than as a vague email. Our restaurant guide has more on the specifics.

A salon. "What do you charge for a cut?" "How far ahead do I need to book?" "Do you do colour on Sundays?" Price list and availability, answered at 8pm on a Tuesday when the salon is shut.

A clinic or dental practice. "Do you take new patients?" "What are your opening hours?" "How do I reschedule?" The admin questions get answered instantly. It's bad at anything clinical, which is exactly where you want a human. More in the clinics guide.

An online shop. "When will this arrive?" "Can I return this?" "Do you have this in a size 10?" The three questions that are most of your email.

The pattern: the questions you already answer, that a customer can look up if they can find them, and that never need your judgement.

What it cannot do

This is the section most vendor pages skip, and it's the one that matters most.

• It can't handle a difficult conversation. Someone who is angry and wants to discuss a damaged order deserves a person. No AI agent should be negotiating a refund.
• It can't process a refund, change an order, or authorise anything. Those need access to systems and accountability behind them. An AI agent that claims it can is lying, and it will lie to a customer on your behalf.
• It can't make a judgement call. "I'll waive the fee" is a decision. Even where an AI agent could technically trigger it, you don't want a tool making that call unsupervised.
• It won't know anything you haven't told it. Not your new opening hours from last week. Not the stock level that changed this morning. It knows what you trained it on, which means keeping that content current is your job.
• It won't replace being known. Regulars who like you will still prefer you. This is for people who've never heard of you and have a question at midnight.

Read that list and you'll notice it describes the boring bulk of support. That's the point. The rest is the part you want to keep doing yourself.

How much does it cost?

For a small business, somewhere between $15 and $50 a month is the real range, and anything well above that is aimed at larger operations.

ChatterEase is free to start, then $14.99 a month for one AI agent with 500 conversations, $29.99 for three AI agents and 2,000 conversations, and $49.99 for the higher limits. Nothing asks for a card to begin. The pricing page has the current detail.

Free options exist, mostly the free tiers of bigger platforms. The trade is usually message limits and less control over what the bot is allowed to say. For a business answering under a few hundred questions a month, a paid plan at fifteen dollars is usually the cheaper option once you count your own time.

What changes the price:

• Number of AI agents. One per business is one AI agent. More if you run several sites.
• Conversations per month. A "conversation" is a chat session, not a question, so a visitor asking six things in a row is still one. High-traffic sites hit limits faster than people expect.
• How much content you load. More documents means more work for it to search, which is why plans have limits.
• Automatic retraining from a URL. Pointing the bot at your site and having it keep itself updated is a useful feature and it's on the higher plans.

To sanity-check the cost: a part-time support person is roughly $15–20 an hour before employer costs. If an AI agent removes two hours of repetitive email a week, it pays for itself in the first month. That's the calculation worth doing, not the sticker price.

How to set one up (short version)

The full walkthrough is here, but the short version is four steps:

1. Create an account. Free to start, no card.
2. Make an AI agent. Give it a name. That's it.
3. Feed it your content. Either upload the documents you already have, or give it your website's address and let it read your pages. This is the step that matters. The bot is only as useful as what you gave it.
4. Paste one line of code into your website. There's a full guide for WordPress, Shopify and Wix if your platform isn't obvious.

Then leave it for a fortnight and read the logs. Not before. Nobody gets this right first time, and the first week's conversations mostly teach you what people really ask, which is rarely quite what you guessed.

What happens if it gives a wrong answer?

Two things, and both are under your control.

It says it doesn't know. Set a fallback message for when the bot has no answer. Make it honest and give it an exit: offer to pass the question to your team, or point somewhere useful. A bot that says "I don't have that, here's how to reach us" looks competent. One that invents an answer looks broken, and worse, looks like something you said.

It answers, but wrongly. This happens, and it's usually one of two causes. Either the content you gave it is ambiguous (a price list with two prices and no dates, say) or you gave it something it shouldn't have had, like an outdated discount.

The fix is quick. You write the correct answer as a Q&A override, which is your own answer that always wins, whatever the bot would have said. It's useful even when the answer is right most of the time, because it pins the exact phrasing for anything you care about being said a particular way.

There's also a version of this worth doing ahead of time: read your real questions in the conversation log and write overrides for the top ten. That's most of an afternoon and it makes the bot noticeably better.

The safety net is escalation. When someone asks something it shouldn't handle, it should pass them to you rather than improvise. Make sure you know where those go and who reads them. Escalation you don't check is worse than no escalation at all, because the customer waits either way.

My recommendation

If you own a small business and you answer the same questions every day, start here: write down your ten most common questions, and get the answers right on a page. That work is worth doing whether or not you ever add an AI agent, and a good FAQ page solves a surprising amount of the problem on its own.

Once that's done, the AI agent is a small addition rather than a big project, because there's already something for it to read.

Two rules I'd hold to. One: don't let it answer anything you'd be uncomfortable putting your name to. Two: check in on it. Ten minutes a week reading what people asked will improve it more than any setting you can toggle.

If you want a straight comparison of what to use and when, AI agent vs live chat covers that, including where automation is the wrong call. And if your inbox is the bigger problem right now, 7 ways to reduce support emails starts with changes that need no software at all.

Try ChatterEase free — no credit card, and a trial ends on its own if you start one and do nothing.