Skip to main content
Sign in

Search

Internal Support Center

Emplifi Bot — AI Utterances_ Train Bot Intents with AI

  • Updated

ARTICLE: Emplifi Bot — AI Utterances: Train Bot Intents with AI

Source: https://docs.emplifi.io/platform/latest/home/ai-utterances

Last modified: 2024-04-10

PURPOSE

AI Utterances is a feature that uses AI to generate training examples (utterances) for your bot's intents — dramatically reducing the time and effort required to build a well-trained bot. Instead of manually writing 10–20 variations of how a customer might phrase a question, you write one example and let the AI generate the rest. This article explains the concepts of intents and utterances, and walks through the full procedure for using AI to train an intent.

KEY CONCEPTS

Intent

An intent is a recognized category of customer behavior or question — a profile built from training examples that the bot uses to match incoming messages. Examples: "track my order," "request a refund," "report a login issue."

Utterance (training phrase)

An utterance is a single example of how a customer might express a particular intent. For example:

  • "I'm having trouble logging in."
  • "My password isn't working."
  • "I can't access my account."

All three are utterances for the same intent ("login issue"). The bot uses these examples to recognize similar phrasings it has never seen before.

WHY TRAINING MATTERS

Without sufficient training utterances, a bot frequently fails to match customer messages to the correct intent — leading to escalations, poor containment rates, and frustrated customers. 10–20 high-quality utterances per intent is the recommended target.

PREREQUISITE

The AI intent example feature must be enabled on your account. Contact your Emplifi account manager to get access.

STEPS — CREATE AN INTENT AND TRAIN IT WITH AI

  1. In the Bot authoring menu, navigate to the Intents section.
  2. Click the + icon to create a new intent.
  3. A new intent is added to the list. Name the intent to reflect its purpose. Recommended naming convention: align the name with the customer goal — e.g., "Refunds," "Order Tracking," "Password Reset."
  4. Click into the new intent to open it.
  5. Click Add utterance. A text prompt field appears.

  6. Type your first example of how a customer would ask this question. Use a complete sentence, not just keywords. 

    Example: "I'm having trouble logging in."

    Important: You must add at least one utterance manually before you can use AI generation. Without this first example, the Generate with AI button will return an error.

     

  7. Click Generate with AI. 

    The AI generates five additional utterance examples based on your input, using similar language and sentence structure.

     

  8. Review the five generated examples:

     - Keep examples that accurately represent how a customer might phrase the request

    - Delete examples that are inaccurate, off-topic, or too similar to other intents

    - Click Generate with AI again to produce another five examples — each new batch uses all existing utterances as its prompt, so accuracy improves over iterations

     

  9. Repeat until you have between 10 and 20 utterances.Do not exceed 20 utterances per intent — too many can reduce accuracy by creating overlap with similar intents.
  10. Click Save.

The intent is saved with its training utterances. The bot's intent engine will now match customer messages to this intent.

CONFIGURE A RESPONSE

After creating the intent, configure how the bot responds when it matches this intent. Two options are available:

Option A — Routing rule

Create a routing rule that directs users to a specific conversation flow when the intent is matched.

Option B — Direct response

Add a response directly on the intent screen. Click Add Response to create a text response. You can change the response type (text, quick replies, cards, carousels) using the format icon to the right of the response field — supported types vary by messaging channel.

Note: Both options may require support from your Emplifi implementation team to configure the handler. Contact your Emplifi account manager.

BEST PRACTICES

  • Use complete sentences in utterances — not keyword lists
  • Aim for natural variation in phrasing — include formal and informal expressions
  • 10–20 utterances per intent is the sweet spot for accuracy
  • Review Utterance Matching analytics regularly to identify intents with low match rates that need more training examples

RELATED ARTICLES

- Emplifi Bot — Create and Manage Bots

- Emplifi Bot — Bot Analytics Dashboard

- Emplifi Bot — Unified Analytics for Bot

 
 


 

Powered by Zendesk