An intent represents the wish of a user: request for an invoice, request for a quote, follow up a request, see the supporting documents ...
There are two editing modes. The simple mode for an FAQ chatbot and the expert mode for very advanced conversational agents (dictionary, context ...)
Basic mode : The simple mode proposes all the elements that are necessary to create Questions/answers with no memory, no entities, no contexts ...
- It will be displayed in the knowledge base, in the tests and in the usage statistics to get insights from your automation use. So choose it carefully.
The disambiguation phrase¶
- The disambiguation phrase is used to clarify intents and guide the user. Disambiguation helps to resolve the ambiguity between two intents during a user request.
- It will also automatically generate an FAQ page from the Chatbot knowledge base.
Prisme.ai uses Supervised Learning to understand automatically the intents of your users composed in real life of various and varied sentences. It's not magic. In fact, to help the machine understand, it needs training sentences. Let's draw a parallel with image recognition, to identify a cat, you need variations of cat pictures annotated "Cat" . In natural language, to understand a "I'd like to pay my 20€ bill", it needs variants of "I'd like to pay my 20€ bill"...annotated 'bill.pay' for example.
Training phrases are used to train users' natural language intent. Identify examples of sentences, rephrases, that users will be able to express in natural language, and enhance this function to the maximum to improve accuracy. (Go to Settings-> Developer to set up the Machine Confidence Score [0,1] )
- IMPORTANT: To use Unsupervised Learning, go to AI Tools -> Supervised Learning. Prisme.ai will use Unsupervised Learning algorithms to automatically propose intents but it will always be up to you to accept or reject them.
A workflow is made up of one or more steps. A step maybe:
- Send a message (text, file, carousel, ...)
- Perform an action (Call an API, Search in a document, Read from a database, Write to a database, climb to a Live Chat ...)
Prisme.ai offers the possibility of customizing the workflow according to the chosen channel. Indeed, a text type experience will be totally different from a voice experience. It is for example not possible to display links, images or cards orally. You therefore have the option of creating a personalized response according to the distribution channel. The same intent for several types of responses or different actions.
Advanced mode :
The advanced mode offers additional features such as the notion of context, the management of parameters, events or the Handover which consists in passing the hand to an API or another chatbot.
- This very powerful feature allows the chatbot to be equipped with a memory. This is like creating links between intents to get as close as possible to a human conversation. This allows to structure all the conversations, in a personalized and non-linear way, and to bring a relevant and contextual answer to the user request.
- Events are elements that trigger a specific intent. Events allow you to invoke an intent based on a non-verbal signal, such as a click on a button or the start of a new conversation with a bot, or a TV that turns on, a lift that stops ...
- The list of events specific to the platform.
It is therefore possible by managing these events to launch a specific workflow or a personalized reaction.
** Events common to all channels: **
- WELCOME: used to trigger the Welcome Intent
** Events specific to the telephony channel: **
- TELEPHONY_PLAY_AUDIO: launch an audio for the end user
- TELEPHONY_TRANSFER_CALL: transfer the call transparently for the end user
- TELEPHONY_HANGUP: hang up the call
- TELEPHONY_HOLD: put the call on hold
- TELEPHONY_TRANSFER_CALL_INDIRECT: transfer the call provided the recipient accepts
- TELEPHONY_NO_INPUT: The end user does not speak.
- TELEPHONY_LEAVE_MESSAGE: Leave a message in the voice mailbox of a recipient who has not responded to a bot (in case the bot is calling)
** Specific events for the management of the handover to experts: **
- HANDOVER_HUMAN: start handover
- HANDOVER_HUMAN_TRIGGERED: handover was automatically triggered (according to rules configurable in the console)
- HANDOVER_HUMAN_FINISHED: disable handover
- This feature allows you to manage what is called Slot Filling. It allows you to build a personalized conversation. Several types of parameters (See entity management) exist and can be completed according to your context.
- If you mark a parameter as required, the chatbot triggers the mechanism to extract the parameter from the sentence. It will also automatically ask the user if the parameter is not specified by asking the right question. Questions are customizable without coding.
List of conditions : - exists / does not exist = match if the value is present in the entity's dictionary - equals / not equals = match only with a particular value in the entity dictionary - lower than / superior than = comparison between integers - match / does not match = match if a string is present in the sentence
- In this panel you will be able to configure different behaviours in the conversation linked to an intent.
Some of these behaviours are available only when deploying through Prisme's connectors, such as webchat, portal and mobile app.
Example: disable the user text input when this intent is triggered with a custom placeholder. Here is a video showcase.