My first chatbot
How I built it, what went wrong and what it says when it doesn't know the answer.
It's worth it when you get lots of repeated questions (opening hours, prices, order status, how to make a return) and you have that information in writing…
It's worth it when you get lots of repeated questions (opening hours, prices, order status, how to make a return) and you have that information in writing to give to the bot. If every enquiry is different, sensitive or requires a decision, the chatbot will save you little and may end up annoying your customers. Nobody wants to argue a complaint with a machine that answers with infinite politeness and zero solutions.
The key is what you feed it and how it reacts when it doesn't know. A good chatbot answers from your documents (your FAQs, your terms, your catalogue) and, when it can't find the answer, says so and hands the conversation over to a person. The problem with AIs is that they tend to make things up with confidence, what's called hallucinating, and in front of a customer that's expensive. It should also be clear that people are talking to an AI; the European AI law points that way, and misleading a customer never ends well.
How I built my first chatbot, what failed and what it does when it doesn't know the answer I explain in "My first chatbot". If you're considering one, do the paper test first: write down the questions you get for a couple of weeks. If most of them repeat, there's a case for a chatbot; if not, maybe what you need is a good FAQ page.
How I built it, what went wrong and what it says when it doesn't know the answer.
One email a week with what was verified, what was refuted and what is still on the bench. No news roundups.