Adding a chatbot is the most requested AI feature and the one most likely to be regretted. Not because the technology is bad, but because it is usually deployed against the wrong problem.
The question nobody asks first
What are people actually contacting you about? Pull the last hundred support emails and sort them. On most sites the answer is dominated by three or four questions: where is my order, what is your returns policy, does this fit, and can I change my delivery address.
Those are not conversation problems. Three of the four are answered by a page, and the fourth by an account area. A chatbot that answers them is an expensive way to compensate for information the site should already show.
When it genuinely earns its place
- Your catalogue is large enough that people cannot find things by browsing. A conversational filter beats twelve dropdowns.
- The product needs explaining before it can be chosen. Insurance, technical parts, anything with compatibility rules.
- You have real documentation and a support history worth searching. Grounded answers need a source.
- Support volume is high enough that deflecting 30% is worth real money. Below a few hundred tickets a month it rarely is.
When it costs you
A bot that cannot answer becomes an obstacle between a customer and a person. Everyone recognises the pattern: three rounds of unhelpful replies, then hunting for a way to reach a human. Customers who arrive already frustrated leave more frustrated.
The other failure is confident invention. A model with no grounding will happily state a returns window, a delivery date or a compatibility claim that is wrong. That is a support problem and, depending on what it said, a legal one.
If you build one, build it this way
- 01Ground it in your own content. Retrieval over your product data, policies and docs, so answers come from your material with a link to the source.
- 02Give it a hard scope. It answers questions about your business and declines everything else. An assistant that will write a poem is an assistant that will invent a refund policy.
- 03Put a human escape hatch on every screen, not behind a menu.
- 04Log what it gets asked. Within a month it will have told you exactly which pages to write, which is often worth more than the deflection.
- 05Measure the resolution rate, not the conversation count. Volume is easy and meaningless.
The cheaper thing to do first
Take your top ten support questions and answer them properly on the site. Order tracking on the order page, sizing on the product page, returns in plain language rather than a legal paragraph.
Do that, measure again, and then decide whether the remaining volume justifies a bot. Plenty of clients get to that point and find the answer is no. That is a good outcome and it costs a fraction of the alternative.
Working on something like this?
We build websites, stores and custom applications, and we will tell you honestly if the thing you are describing does not need one.