Submit a ticket

Manychat AI Intention Recognition

Intention Recognition in Manychat is an advanced trigger that understands the intention behind a user's message, unlike traditional keyword triggers. This feature is ideal for handling complex language nuances and common typos. They offer more accurate and flexible responses, improving user experience and reducing manual intervention in chatbot conversations based on context.


Setting Up Intention Recognition

Step 1: Define the Intent

  • Identify User Queries: Think about common questions or requests related to your business, like "Where is your store located?".
  • Understand User Intent: Consider different ways users might express the same intent. For location queries, this could include "Address?", "How to find you?", or "Directions to your shop?" or even “I can’t find you on Google Maps”

Step 2: Create the Automation

  • Access Automation Section: Log into your Manychat account and navigate to 'Automation'.
  • Start a New Automation: Click to create a new automation. This is where you'll set up responses to user intents.

Step 3: Craft Your Response

  • Write a Clear Reply: Based on the identified intent, create the content box for the reply you will send when an intention is recognized and triggered.
  • Keep it Simple and Informative: Ensure your response is easy to understand and provides all necessary information.

Step 4: Set the Trigger

  • Choose 'New Trigger': Select this option and pick 'User sends a Message'.
  • Define the method of recognizing: instead of “detect specific words in a message”, select “Recognize intention of message”
  • Write the intention to recognize: a good formula is USER + verb + ABOUT + ITEM (e.g., user ASKS about DELIVERY).

Step 5: Activate Your Intention Recognition

  • Test Before Going Live: Ensure your setup works as intended by testing different phrasings.
  • Set Live: Enable your Trigger to start automating responses.

Writing Effective Intention Recognition Prompts

Simplicity is Key

  • Clarity Over Complexity: Use language that's easy to understand. Avoid jargon or overly technical terms unless they are commonly used by your audience.
  • Direct and to the Point: Your prompts should directly address the potential queries. For instance, if your keyword is about pricing, the prompt should be straightforward like, "Asking about product prices".
  • Use AI for Suggestions

    • Leverage AI Tools Like ChatGPT: These tools can help you brainstorm different ways people might phrase a question and how to prompt for them. For example, ask ChatGPT, "What is a prompt I can give an AI to recognize all incoming messages from contacts asking about my store’s location in a clear way"
    • Incorporate AI-generated Variations: Use the variations provided by AI tools to broaden the scope of your Intention Recognition. This ensures your chatbot can recognize and respond to a wider range of user inquiries. Examples: “asking about location”, “wants to visit us in person”.
    • Continuous Learning and Updating: Regularly update your prompts based on new variations you encounter in actual user interactions. AI tools can be used periodically to generate fresh variations.

Testing and Refining Prompts

  • Conduct Real-World Tests: Use test messages to see how well your Intention Recognition recognizes and responds to different phrasings. We highly recommend you use slang common to your region to test the accuracy of the reply and incorporate it into your prompt if it's not replying accurately.
  • Gather User Feedback: Encourage users to provide feedback on their interactions. This can offer valuable insights into how well your Intention Recognition is performing.
  • Iterative Improvement: Intention Recognition prompts are not set-and-forget. Regularly refine them based on performance data and user feedback.

Best Practices

One Topic, One Keyword

  • Avoid Overlap: Each intention should be unique to a specific topic or query. This prevents confusion and ensures more accurate responses.
  • Distinct Definitions: Make sure that each keyword is clearly and distinctly defined to avoid triggering multiple responses for similar queries.

Specificity Wins

  • Narrow Focus for Better Accuracy: The narrower the focus of your intention, the higher the accuracy of the response. For instance, differentiate between “Store hours in New York” and “Store hours in Los Angeles” if you have multiple locations.

Continuous Testing and Refinement

  • Regular Reviews: Continuously review and adjust your intentions based on their performance and user interactions.
  • Evolve with Your Audience: As your audience grows and changes, your intention prompts should evolve, too. Stay in tune with changes in language use, trends, and customer preferences.

Setting Conditions and Tags for Single Responses

  • Use Conditions Wisely: Implement conditions in your automation to ensure that a response to a particular AI Intent is sent only once to a user. This prevents repetitive messages that might frustrate users.
  • Tagging for Tracking: Once a response has been sent, use a tag like “LocationQueryAnswered” to mark that this user has already received a response to that specific query.
  • Automation Setup: In your automation flow, add a condition to check if the tag exists for the user. If the tag is present, the bot should not send the AI Intent response again.
  • User Experience: This approach enhances the user experience by providing necessary information without redundancy.

Monitor and Analyze Performance Data

  • Data-Driven Decisions: Regularly monitor the performance of your Intention Recognition. Look at metrics like response accuracy, user satisfaction, and engagement.
  • Adjust Based on Insights: Use these insights to make informed decisions about modifying and improving your intention prompts.

Examples for Inspiration

1. Formula: User is + (action) + about + (item)

  • Example 1: User is inquiring about product availability.
  • Example 2: User is asking about shipping options.
  • Example 3: User is interested in our return policy.
  • Example 4: User is curious about our services.
  • Example 5: User is concerned about privacy.

2. Formula: Person wants + (action) + information on + (item)

  • Example 1: Person wants detailed information on delivery times.
  • Example 2: Person wants information on our pricing.
  • Example 3: Person wants to know about our product warranty.
  • Example 4: Person wants information on our company history.
  • Example 5: Person wants guidance on troubleshooting.

3. Formula: User needs + (action) + assistance with + (item)

  • Example 1: User needs assistance with setting up their account.
  • Example 2: User needs assistance with tracking their order.
  • Example 3: User needs help with product installation.
  • Example 4: User needs assistance with payment methods.
  • Example 5: User needs help with a technical issue.

4. Formula: Lead is looking for + (action) + guidance on + (item)

  • Example 1: Someone is looking for guidance on product selection.
  • Example 2: Someone is looking for guidance on using our app.
  • Example 3: Someone is looking for guidance on returns.
  • Example 4: Someone is looking for guidance on our subscription plans.
  • Example 5: Someone is looking for guidance on troubleshooting tips.