Summary
AI intent matching in Wati Rules lets you automate replies and actions based on what a customer means, rather than relying only on exact keywords. Describe the customer's intent in one sentence, set a confidence level, and choose what Wati should do when a message matches. AI intent matching supports messages in different languages and uses 1 Wati AI credit each time the rule runs its action, without using your Monthly Automation Triggers quota.
AI Intent limits by plan: The number of AI intents you can create depends on your Wati plan:
Growth: Up to 5 AI intents
Pro: Up to 10 AI intents
Business: Up to 30 AI intents
Note: The AI intent matching filter is being rolled out gradually to all Wati customers. If you don't see it in your account yet, it should be available soon.
Instructions
Use AI intent matching when you want to identify customer requests even when they use different words or languages to express the same need. For example, you can create a rule that detects refund requests and automatically sends a relevant reply.
How AI intent matching works
AI intent matching checks the meaning of an incoming text message against the intent description you provide.
For example:
Customer message: "I want my money back."
Intent description: "The customer is asking for a refund."
Action: Wati identifies the matching intent and runs the action configured in your rule, such as sending a reply.
The same intent can match messages that express the same request in different words or languages.
Create a rule using AI intent matching
Go to Automation > Rules.
Click Create Rules and select a trigger, such as New WhatsApp message is received.
Enter a name for your rule.
Select Add a Filter, open Filter by and select AI intent matches.
In Intent description, enter a sentence describing what the customer wants. The maximum length is 100 characters, including spaces.
Set the confidence level. The rule runs when the AI's confidence in the match meets the configured threshold. The default is 75%.
Optional: Enable Run top intent only if you want the rule to run only when this intent is the best match for the message. This helps when multiple intents match.
Add the action you want Wati to perform, such as sending a reply, starting a chatbot, or assigning the conversation to a team.
Turn on the rule and save your changes.
Set the confidence level
The confidence level determines how closely a message must match your intent before the rule runs.
Start at 75%: Use the default setting as a starting point.
Increase the confidence level: Do this if the rule runs for messages that do not match your intended request.
Decrease the confidence level: Do this if the rule misses messages that should match.
Test your rule with real customer messages and adjust the confidence level based on the results.
Use Run top intent only
The Run top intent only setting controls whether your rule must be the best match for a message.
Off: The rule can run when the message matches the intent above the configured confidence threshold.
On: The rule runs only when its intent is the best match for the message.
Enable this setting when you have multiple rules that could match the same message and want to prioritise the strongest match.
Write an effective intent description
An intent description tells Wati what the customer wants. Keep it to one sentence and within the 100-character limit, including spaces.
Describe the customer's request clearly and include relevant context when needed. If certain requests should not match, you can also describe what to exclude.
Example 1: Identify one clear request
Intent description: The customer is asking where their order is
Use this when: You want to identify order-tracking questions.
Example messages that should match:
"Where is my parcel?"
"Has my order shipped?"
"Can you share my tracking number?"
Example 2: Include relevant context
Intent description: The customer wants a refund for an order they already paid for
Use this when: You want to identify refund requests for paid orders, rather than general questions about refunds.
Example messages that should match:
"I want my money back for my order."
"Please refund my payment."
Example message that should not match:
"Do you offer refunds?" asked before making a purchase.
Example 3: Distinguish between similar requests
Intent description: The customer wants money back for a delivered order, not a cancellation or a policy question
Use this when: You need to distinguish refund requests from cancellations and questions about refund policies.
Example messages that should match:
"The item arrived broken, and I want my money back."
"Please refund me. This isn't what I ordered."
Example messages that should be handled by other rules:
"Cancel my order before it ships." This is a cancellation request.
"What is your refund policy?" This is a policy question.
"Where is my refund?" This is a refund-status question.
Because similar requests can overlap, test your intent descriptions with real messages to confirm that Wati distinguishes them as expected.
Tips for writing intent descriptions
Describe the customer's goal, not a list of keywords. For example, "The customer is asking for a refund" describes an intent. If you only need to match exact words, use a keyword filter instead.
Include context that changes the meaning. Details such as whether an order has been paid for or delivered can help distinguish similar requests.
Keep exclusions short. If needed, state what the intent should exclude in the same sentence. Avoid adding too many conditions.
Use one intent per customer need. Create separate rules for different requests, such as refunds and cancellations.
Avoid vague descriptions. "The customer is unhappy" is too broad, while "refund" does not explain what the customer wants.
Test with real messages. Try approximately 10 messages from your inbox and adjust the intent description and confidence level based on the results.
Understand AI intent matching costs
AI intent matching uses Wati AI credits rather than your Monthly Automation Triggers quota.
1 Wati AI credit is used each time a rule runs its action.
Messages that do not match the intent do not use a Wati AI credit.
Rules that are turned off do not use a Wati AI credit.
Failed AI checks, timeouts, and failed actions do not use a Wati AI credit.
You can review your credit usage on the Wati AI usage page.
Limits of AI intent matching
Intent description length: Each intent can contain up to 100 characters, including spaces.
Message type: AI intent matching checks text messages. Images, audio, and documents are not checked.
Available intents: The filter displays how many AI intents you can still add.
Best practices for AI intent matching
Start with your 3–5 most common customer requests, such as refunds, order status, pricing, cancellations, and requests to speak to a person.
Use AI intent matching to identify meaning across different wording and languages.
Use keyword filters for exact values, such as order numbers or promotional codes.
Review the Executed count in the Rules list after a week to understand how often your rules run.
Adjust your intent descriptions and confidence levels based on the results.
Pair each intent with an action that helps your team respond faster or reduce manual work.
Frequently Asked Questions (FAQs)
AI intent matching and credits
1. Does AI intent matching use my Monthly Automation Triggers quota?
No. AI intent matching uses Wati AI credits instead of your Monthly Automation Triggers quota. Each time an AI intent matching rule runs its action, it uses 1 Wati AI credit.
2. Am I charged Wati AI credits if a message does not match the intent or a rule fails?
No. You are not charged a Wati AI credit if the message does not match the intent, the rule is turned off, the AI check fails or times out, or the action fails.
AI intent matching filters and compatibility
3. Can I combine AI intent matching with keyword filters?
Yes. You can combine AI intent matching with keyword filters to identify the customer's request and check for specific words or values.
4. Does AI intent matching affect my existing keyword rules?
No. AI intent matching is a separate filter and does not change how your existing keyword rules work. Your existing keyword rules continue to work as configured.
5. Which channels support AI intent matching?
AI intent matching is supported on all channels, as long as the incoming message is text-based.