An AI chat on your website answers visitors' questions from your own material, such as the website and a knowledge base. It helps when the answer is written down but hard to find. When the answer is missing, it should say so and hand over to a person or a call instead of guessing.
What can the AI chat answer?
A typical example: an installation company has its services and terms on its website. A visitor writes: "Do you service air to water heat pumps in Borås?"
The chat replies that the company services air to water heat pumps in and around Borås and can link to the service page where that information is.
The visitor goes on: "What does it cost to replace the compressor on my pump?" That price is not written anywhere. The chat says it cannot answer that and offers a call. The visitor leaves their number and is called by someone who can assess the job.
The example shows the line. If the answer is in the material, the chat can give it. If the question is about a specific job or price, it goes on to a person.
The difference from an ordinary chatbot
A rule based chatbot follows a decision tree. Someone has guessed which questions will come and written an answer to each. If a different question comes, or the same question phrased another way, the bot often gets stuck and shows a menu option that does not fit.
An AI chat has no tree. It reads the question and looks for the answer in your material. You do not need to predict the question, but the answer has to be in the material the chat has access to.
How the website content and knowledge base are used
The chat gets its answers from two places: the pages on your website and a knowledge base with things that are not there, such as service area, delivery terms or how a quote works. That assumes both the material and the chat's sources are up to date. If an old delivery time is still on the site, that is what the chat will answer with.
Questions the chat often cannot answer show where the material has gaps. It is often the quickest way to see which text your website is missing.
How contact details are collected and passed on
When the visitor wants help from a person, the chat should ask for what is needed to take the request further, usually name, phone number and what it is about. Do not ask for more than you need.
What has already been said should come along. Whoever calls back should be able to read the conversation and see what the visitor asked about and which answers were given, so the customer does not have to explain everything again.
When is a call or a person needed?
This is where good and bad solutions differ most. A chat that always answers sounds helpful until one day it promises a price or a time that does not exist. A made up answer costs more than no answer.
So hand over when the answer is missing from the material, when the question is about the price or assessment of a specific job and when the visitor asks for it. If you are closed, the visitor should be able to leave a message instead.
How do you follow the quality of answers and leads?
Read a sample of the conversations every week. Did the chat answer correctly, and did it speak up when the information was missing? Also track how many conversations become leads and how many of them lead to a call or a deal. Before you put the chat live, test it with five kinds of questions:
- one clearly answered in the material
- one phrased in an unusual way
- one whose answer is not written anywhere
- one in another language
- one where the visitor wants to talk to a person
If the chat guesses on the third, it is not ready to go live.
The AI chat in Leadcaller
Leadcaller's AI chat answers from your website and the knowledge base you add to in the admin panel. It is built to say when the information is missing and then offer a call back, or the contact option you have switched on when you are closed. The chat sits in the same widget as callback, so the visitor can be called without switching tools, and it supports Swedish, Norwegian and English. Conversations are collected in your inbox, and those who receive notifications get them by email with a summary.
Key takeaways
- An AI chat answers from your own material, which has to be kept up to date.
- If the answer is missing, the chat should say so and offer a call or a message.
- Follow up both the answers and the leads the chat passes on.