I'll tell you what's worth building.
Then I build it.

Probably not yet, and this is the single most common thing I talk people out of. A chatbot solves a volume problem: too many people asking the same question for your staff to keep up with. Most businesses around Hamilton don't have that problem. They have a repetition problem, which is a different thing. If your site gets forty visits a week, a chatbot is answering nobody while costing you a subscription and a fortnight of setup. The reminder calls and the same five email replies your staff retype every day, that's where the time actually goes. Fix that first. If you reach the point where real people are genuinely waiting on answers, then a chatbot earns its place.

Not the free consumer version, no. The consumer tier's terms allow your inputs to be used to improve the model. The business and enterprise tiers have different data terms, and for a law office, a clinic, or an accountant that distinction is the whole ballgame. It's the difference between a normal tool and a confidentiality problem. The practical rule: assume anything typed into a free AI tool has left your building. Before anyone on your team uses one for real work, decide which tier you're on and write down what is and isn't allowed to go into it. That takes about twenty minutes and it's the cheapest risk you will ever remove.

Whatever your staff retype most. Nine times out of ten it's appointment reminders or the same five email replies. Boring, small, and it pays back every single week. The reason to start there isn't that it's the most impressive use, it's that it's the one where being wrong costs you nothing. A reminder that reads slightly oddly is a shrug. A quote with the wrong number on it is a real problem. Start where the mistakes are cheap, let your team build a feel for what the thing is actually like, then decide about the bigger stuff. Most people do this backwards and start with whatever is hardest to check.

The subscription is the small number, and it's the only one anybody quotes you. Per-seat tools generally run somewhere between twenty and a hundred dollars a month per person. That is rarely what the thing costs. The real cost is the three months your team spends half-using it, the workarounds they build because it doesn't quite fit how you work, and the one person who quietly becomes its unofficial owner. Then there's the afternoon everyone goes back to the old way and nobody says anything, because it was the boss's idea. When I map costs for a business I write down four lines: the licence, the setup time, the training time, and who owns it in six months. If that last line is blank, the tool is already dead.

Not in a business your size, and anyone telling you otherwise is selling something. What it replaces is tasks, not people, and specifically the typing-shaped ones. The places I see get real value out of this aren't cutting headcount, they're getting the people they already have out of retyping the same things all day. That said, I'd be lying if I said nothing changes. Roles that are mostly transcription and copying between systems do get thinner over time. If that describes a job in your business, the honest move is to start shifting that person toward the parts of the work that need judgment, rather than waiting to find out.

An agent is software that takes several steps on its own instead of answering one question. Ask a question, get an answer. Ask an agent to do something and it goes away, makes a series of decisions, and comes back when it's finished. That's genuinely useful, and it's also where the expensive mistakes live, because nobody is checking the middle steps. If it gets step two wrong, steps three through seven are all confidently built on top of that mistake. For most small businesses the honest answer right now is no, not yet. Get value out of the simple things first. An agent is what you graduate to once you know what this technology is actually like when it's wrong.

Three questions, none of them technical. Where do the answers come from? If the salesperson can't tell you, the tool is generating them rather than looking them up, and it will invent details about your business. What happens when it's wrong? If the answer is a shrug, you are the one holding that risk, not them. And who on my team owns this in six months? If nobody does, it's already dead. The tool that gets abandoned usually isn't the bad one, it's the one nobody was responsible for. You don't need to understand how any of it works to ask those three, and they sort most of the pitches you'll get this year.

It's worth it when the alternative is people quietly using these tools badly, which is what's happening in most businesses right now whether or not anyone has approved it. Your staff are already pasting things into ChatGPT. The question isn't whether AI is in your business, it's whether anybody has told them where the lines are. A session covering what the tools do, what they get wrong, and what's not allowed to go into them costs you a morning and removes most of the risk. The other reason is ownership. A tool somebody was taught properly gets used. A tool that was simply installed gets abandoned. That difference is most of the return.

ChatGPT is a general engine. It knows a great deal about the world and nothing about you. A tool built for your business is usually that same kind of engine with your own documents wired in, so it answers from your prices, your policies, and your history. That wiring is the part people call RAG. The practical difference is that a general tool will confidently invent details about your business, and a properly connected one won't, because it's reading rather than guessing. Whether you need the second kind depends entirely on whether the answers have to be specifically right. For drafting a supplier email, no. For anything a customer relies on, yes.