We sell an AI support chatbot, so treat what follows with the scepticism it deserves. But a tool bought at the wrong moment gets blamed for problems it was never going to solve, and we would rather lose a sale than gain a customer who cancels in six weeks with a poor opinion of the whole category. Here are the situations where the honest answer to “are chatbots worth it” is not yet.

Your questions repeat twice a month, not twice a day

A support bot earns its keep on repetition. It answers the same delivery question for the four hundredth time without resenting it, which is the one thing it does better than any human.

If your inbox holds thirty enquiries a month and most are genuinely different, there is very little repetition for a bot to absorb. You would spend an afternoon connecting documents to remove perhaps twenty minutes of work a week. Write a good FAQ page instead and revisit this when the same question starts arriving daily.

Nothing is written down yet

A grounded bot answers from your documents. No documents, no answers — and the bot fills the silence with a handoff message every time, which is a slow way to discover you needed a help centre.

This one is worth sitting with, because it is the most common reason a chatbot trial disappoints. The work that makes a bot good is the work of writing down what you know, and that work pays off whether or not you ever buy anything. Customers can read it, new staff can read it, and if you do add a bot later it will be answering from something real.

The questions are emotional, or expensive to get wrong

Some enquiries should meet a person on the first line. A cancellation after a bereavement, a complaint that has already gone wrong once, a medical or legal question where a confident wrong answer is a liability rather than an inconvenience.

A bot can still sit in front of those and route them quickly. What it should not do is attempt them. If most of your volume is that kind of conversation, a chatbot improves your routing without changing your support model, and you should price the decision accordingly.

Almost nobody visits the page it would live on

The widget answers people who are already on your site. If your customers reach you by phone, WhatsApp, Instagram or email and rarely touch the website, a chat widget solves a problem you do not have. Serochat has no phone or social channel support, so on that traffic we are the wrong tool and there is no configuration that changes it.

What to do instead, for now

None of these situations need software. They need an hour or two.

  • Write the five answers you send most often and put them on one page a customer can find from the navigation.
  • Set up saved replies in whatever inbox you already use, so sending answer number four hundred takes a keystroke.
  • Add an autoresponder that states your delivery times and returns window, since that is what most out-of-hours emails are asking.
  • Keep a tally, for a month, of every question that arrives. That list is the specification for anything you buy later, and it is the evidence that tells you when the moment has arrived.

What changes the answer

Repetition, documentation, and out-of-hours volume. When you notice the same five questions arriving every day, when the answers exist in writing somewhere, and when a meaningful share of enquiries land at 11 pm or on a Sunday, a doc-grounded bot starts earning its fee — and you will have the tally to prove it rather than a hunch.

Our plans are a flat monthly price rather than a charge per answer, which means we have no way to profit from a bot that sits on a quiet site doing nothing. What we get from a customer in the wrong situation is a cancellation, so we would rather say this now: come back when the questions repeat.