Virtual Agent

Virtual Agent is a technology that allows users to carry out their transactions by speaking or typing with their own words by using Natural Language Understanding (NLU). Each VA can be created only in one language. User can create multiple VA’s for several languages.

First let’s look at the how Virtual Agent works;
The first thing to do when creating a VA is to define intents. Scenario designer can create intents from scratch or use prebuilt ones. Creating intents is a way to map user’s input to specific action. When scenario designers define intent, they see utterance, entity and action pages.

• Intent
  • Utterance
  • Entity
  • Actions

Utterances: Writing utterances is an important step in training Virtual Agent. It is a part of Natural Language Understanding (NLU) trained with machine learning system; therefore, Virtual Agent understands the utterances that it is trained for and associates them with the intents accordingly. While writing the utterances, Generative AI can assist with writing additional sentences. Moreover, utterances and users’ inputs don’t have to be exact matches. Since, Virtual Agent is trained by machine learning, it can identify other sentences related to written utterances as well. This enables it to perform correct transactions.

For example, the “Account features” intent can be created inside the banking virtual agent. One of the utterances of this intention can be “What is premium account?”. By training the virtual agent with these utterances, when the user says, "I want to learn about premium account", it matches this expression with the “Account features” intent.

Entities: Secondly, entities are part of the intent branch. Various entities can be defined in an intent. They are key phrases that may be related to this intent. In other words, entities can be viewed as a kind of storage to catch and hold specific words, like e-mail addresses, zipcodes etc.

Action: Third one is designing specific action plan when a user input matches with that intent. Which questions should be asked, what to say, where to direct and etc. For example; in this design we are asking a question and according to the answer Agent replies with two different ways.

How to create a new Virtual Agent
When a user wants to build new Virtual Agent, s/he should click the "+New Virtual Agent" from Virtual Agent tab.

This will lead user to the page down below where Name, Language and Confidence Threshold are necessary fields. Language cannot be changed once it is created.

Confidence threshold is the threshold we set for our intents that we will create in this specific Virtual Agent. If the confidence value of the discourse with the intent is higher than the value specified here, it matches and the Agent perform specific action for the matched intent. If it is less than this value, it will not match.

We discussed creating intents from scratch at the beginning. Besides that, user can access prebuilt intents according to chosen language. For example, English_Banking virtual agent has 79 prebuilt intents.

After user clicked to the ‘Save’, Virtual Agent page will be launched with ‘welcome’ and ‘fallback’ intents.

The user can add new intents from right hand corner. Also, from the downward arrow, user can import/export the intents and use prebuilt ones.