How does a virtual agent work?

In this article we will focus on how the virtual agent (VA) itself works and how we build it. To understand this, we need to explain a few terms first: intent, prediction, training data and response.

An intent, short for "user intention", is a topic that the VA is able to recognise and respond to. A VA can have hundreds or even thousands of intents that we create in the admin panel. When a user sends a message through the interface, the VA has to decide what the user's intent is. We say that the VA is making a prediction – it is predicting what the user's intent is.

In order to make an intent prediction, the VA needs to know what kind of messages should go to what intent. That is what training data is – examples of user messages that the VA can learn from. In the admin panel, we write training data for each intent, and that is how the VA knows how to predict messages to the correct intents.

Finally, every intent also has a response connected to it. And once the VA has made its prediction, it will send a response back to the user.