Do you actually need an AI bot?

Probably not. That is the honest starting position, and I say it as someone who builds them.

Assistants are genuinely good now in a way they were not three years ago. They are also the thing every agency is selling, which means a lot of businesses are being sold one for problems it does not solve.

Here is how to tell which you are.

What an assistant is actually good at

One thing, mostly: answering questions whose answers already exist in your material, at a volume or an hour that makes a person expensive.

That is a narrow description and it is meant to be. Inside it, they work well:

• Repeat customer questions, especially out of hours, which is where most enquiries are lost
• Looking things up for staff who would otherwise interrupt someone who knows
• Taking an enquiry and turning it into a structured booking or lead
• Drafting the first version of a reply you send twenty times a week

The common thread is that the answer is already written down somewhere. The assistant is a faster way to find it, not a source of new knowledge.

Three tests

Volume. Are you answering the same handful of questions dozens of times a week? Under that, a decent FAQ page costs an afternoon and beats a bot. Over it, the maths changes.

Written knowledge. Does the answer exist somewhere, in a price list, a policy, a set of past replies? If it only exists in your head, an assistant cannot read your head. Part of the job becomes writing it down, and that work is worth doing whether or not you build anything.

Consequence. What happens if it gets one wrong? A wrong opening hour is embarrassing. A wrong dosage, price quote, or legal answer is not. The higher the consequence, the more it should hand over rather than answer.

Two out of three and it is probably worth talking about. One out of three and it is not.

What it costs

Roughly, in 2026:

• Off-the-shelf chat widgets: $50 to $500 a month depending on volume
• Something custom, built on your own material: commonly $1,000 to $5,000 to build simply, more when it has to connect to your systems
• Running costs: usually tens of dollars a month, not hundreds, at small business volumes

I build the custom kind at $5,000 to $10,000, because in practice the work is rarely just the assistant. It is your material, cleaned up and structured well enough for anything to answer from, and then the connections to wherever the enquiry needs to end up.

If your case is simple, buy the widget. I will tell you that on the call and you will have spent thirty minutes to save several thousand dollars.

The failure everyone worries about, and the one they should

The worry is that it invents answers. That is a real risk and a solvable one: you ground it in your documents, you make it cite where an answer came from, and you make refusal the default when it cannot find one. Done properly, "I do not know, here is a human" is a feature rather than an embarrassment.

The failure people do not expect is subtler. An assistant trained on material that is out of date confidently gives out last year's prices. It is not hallucinating, it is doing exactly what you asked with the wrong source. Which means the real ongoing cost of an assistant is keeping the material behind it true.

That is worth understanding before you buy one, because nobody selling a monthly subscription will mention it.

The honest recommendation

If you are losing enquiries out of hours and your answers are written down, an assistant is one of the highest return things you can build.

If you answer six questions a week, or your knowledge lives in one person's memory, spend the money on the FAQ page and the writing down. It is less exciting and it will serve you longer.