An Intent represents what the user wants to know or do through the bot.

Intents are entry points that trigger a Use Case. There are two types of intents, depending on where users are in the flow:

• Start intents (in the Start User Action)
• Intents (in a User Action)

Each Use Case has one or more Start Intents.
Each User Action has one or more Intents.

Create an Intent

Intent Name

Name each intent clearly to represent the user's intention.

👍

Good practice in naming intents

Name intents consistently to help teammates easily understand and differentiate them from similar intents

Intent Description and Labeling Hints

The intent description helps chatbot builders understand the user intent.

👍

Good practice for description and labeling

Specify the use cases covered by the intent, other intents it might get confused with, and the variables the bot detects in user phrases

Intent Types

Intent types describe the kind of element the intent is supposed to detect:

• Text: Detects a user phrase sent to the chatbot.
• Misunderstanding: Directly triggers the "OUT" class of the model. Only one exists per bot.
• Image: Detects an image sent to the chatbot.
• Video: Detects a video sent to the chatbot.
• Audio: Detects a voice note sent to the chatbot.
• Location: Detects a location sent to the chatbot.

📌

Note

The bot detects the format of images, videos, or audio files but cannot understand their content. You can set a specific response to inform the user that the file was received but not understood.

Hints

Hints are examples of phrases that help chatbot builders and labelers understand the intent.

Priority

Priority is used to resolve overlapping User Actions that cause confusion when predicting the correct intent.

Bot Action Triggered

Select the Bot Action triggered when the intent is detected. Start Intents of the Use Case are systematically connected to the Start Bot Action of the Use Case.

Erase

Specify variables to erase from the context when the intent is triggered.

📌

How to indicate variables to erase

• ""` (empty string): Context remains unchanged.
• "key1, key2, key3" (comma-separated keys): Erases the listed variables.
• "erase_all": Clears the entire context.
• "erase_all_except key1, key2, key3": Clears all variables except the listed keys.

📌

Spaces are ignored when listing keys.

Add Entities to Your Intent

Entities and associated wrappers can be added to your Intent.

Entity Name

Enter a name for the entity to be used in the context.

👍

Good practice

Use lowercase names without special characters to avoid typos or errors.

Wrapper

Select a wrapper to apply to your entity.

Entity Type

Select an entity type for your intent.

Required

When selected:

• The intent cannot be triggered if the entity is missing from the user message or Quick Reply/Button payload.

📌

Trace Inspection

In the Trace, check if entities are missing under:
Raw > NLP Controller > Instance Classifier: `"missing_entities": "True"

Optional

When selected:

• The intent can be triggered even if the entity is missing from the user message or payload.

Add Training Phrases to Your Intent

Rule-Based

Rule-based intents are triggered by detecting specific keywords. Define sets of words to trigger the intent when found in user messages. Refer to Rule-Based Syntax for more details.

📌

Note

Rule-based intents are checked only after the model has checked all AI-based intents and none were triggered.

AI-Based

AI-based intents are triggered by analyzing complete sentences (Training Phrases). Provide a variety of representative phrases for training the model. Changes to training phrases in a published use case are applied immediately after retraining.

📌

Good practice

Provide at least 10 training phrases for AI-based intents. Intents with fewer phrases will not be included in model training.

NLU Quality Score

Each training phrase receives a quality score, indicating how well it matches the intent:

• Green 🟢: Quality score ≥ 0.5, and no overlapping intent has a score > 0.2.
• Orange 🟠: Quality score < 0.5, or an overlapping intent has a score > 0.2.
• Red 🟥: Another intent’s quality score exceeds the current phrase.
• White ⚪: Quality score not computed or phrase duplicated. Retrain to update scores.

🖋️

Tip:

An orange score doesn’t mean the intent won’t be triggered. Use it as a warning to refine your training phrases

Resolve NLU Quality Score related issues by:

• Adding more phrase variations.
• Removing irrelevant phrases.
• Adjusting phrases in overlapping intents.

Upload/Download Training Phrases as CSV

Upload a CSV file to add training phrases or download existing ones for editing.

CSV Format:

message,use_as,smart_label,label
"What's the weather today in Paris?",example,false,false
"It is raining?",example,false,false

📌

Note

Use double quotes for phrases with commas to avoid conflicts.

⚠️

To delete training phrases, use the trash icon 🗑.