AI Center
Configure artificial intelligence agents to automatically qualify leads from WhatsApp.
In the AI Center you can create and edit artificial intelligence agents to integrate them into your WhatsApp conversational flows. These agents are designed to help you automatically qualify leads, collect important data, and answer frequent questions from your potential customers.
🧠 Create an AI Agent – Qualify Potential Customers
The AI Agent for qualifying potential customers is in Beta. We are attentive to any feedback about the agent to continue improving Treble for you.
This type of agent is optimized to detect sales opportunities and collect key information from prospects. Before starting, we recommend having a document with important information about your company that you can upload to the Knowledge Base.
You can assign a name and description to AI agents. They already have a pre-configured role by Treble, but you can add specific instructions in the Agent Role tab.
Although AI agents are powerful tools, they are not magical. They need specific context about your company, products, and processes to function correctly.
AI agents are not designed to make decisions outside their main flow. If you need conditional logic, connect it with a conditions block within the conversation creator.
General Configuration
Define whether communication is more oriented toward casual or formal conversation. This and the following configurations can be adjusted later.
Define the AI Agent Role
Include context about how the AI Agent should communicate or what its role is with the customer. The agent’s role can include special instructions about customer treatment; context about your company should go in the knowledge base section.
Interaction Logic
These are the questions the AI Agent will ask the customer. The agent will
follow the order of each step on this screen; you can reorder them. In this
step, you decide whether to use HubSpot, Salesforce, or Treble Variables.
Exit Rules
No configuration is necessary in this step; it only shows you a preview of the AI Agent.
Knowledge Base
Feed the AI Agent’s context with information about your company, products, and services.
Use the AI Agent in a Conversation
Once created and saved, you can go to your conversations and create or modify an existing conversation to integrate AI Agents into your flows.
📊 Integration with HubSpot and Salesforce
You can automatically save the information captured by the AI Agent to one of these destinations:
HubSpotSalesforceTreble Variables
When you select a destination, the captured variables automatically receive a prefix so you can use them in flows, webhooks, or conditions.
| Destination | Prefix | Example: firstname | Example: email |
|---|---|---|---|
| HubSpot | hubspot_ | hubspot_firstname | hubspot_email |
| Salesforce | salesforce_ | salesforce_firstname | salesforce_email |
| Treble | (none) | firstname | email |
The AI Agent also generates a JSON with conversation history and displays the extracted variables in a metrics table.
⚙️ Using the AI Agent within the Conversation Creator
The new block for AI Agents is called AI Lead Qualification within the conversation creator. When you click on it, you’ll see a new block that allows you to select which of your AI agents you want to use. You can use more than one AI Agent in the same conversation.
You can drag the AI Agent block and connect it with any flow within the creator. Use conditional blocks to control the experience and variable blocks to capture the obtained information.
When selecting an AI Agent, you’ll see a preview section that tells the agent what to do if it doesn’t receive a response from the user.
By default, it’s configured to ask again 2 times every 20 minutes without a response, meaning it will ask once at 20 minutes and then again 20 minutes later.
You can customize this according to what you think is convenient for your users. We know that users can get distracted or forget to respond to conversations, so configuring this will allow them to continue a conversation with the user automatically.
Exit Rules
You can configure different AI Agent outputs based on the responses that the AI considers it received.
Depending on your use case, you could connect all outputs to a human agent or take automatic action like sending a coupon to encourage returning to the conversation. This depends on whether the AI Agent collected all responses, some responses, or no responses.
The Critical Conversation output is designed by Treble to detect any user who is annoyed with the conversation. If we detect that the user is annoyed, the conversation will be interrupted, no more questions will be asked, and the user will be directed to this output.
We recommend the Critical Conversation output to a human agent, but the
final decision is yours.
📚 Knowledge Base
The knowledge base is fundamental for the correct functioning of the AI Agent.
What should it contain?
- Physical address, phones, business hours
- Product and service details (even if you don’t include prices)
- Delivery or coverage areas (and exceptions)
- Policies (shipping, changes, returns)
- Frequently asked questions
Allowed formats
The allowed file types are: pdf, doc, docx, ppt, pptx, txt.
You can upload up to 3 files, with a maximum size of 2MB each.
For a more detailed description and tips on how to feed the knowledge base for the AI, you can see it in the Knowledge Base section
🧪 Playground
The playground allows you to simulate how your AI agent will function within the flow.
If you detect that the AI is not collecting the appropriate information, remember that you can add instructions in each step such as the format of the response you expect, the responses that can be considered valid or invalid - all of this you can do with natural language.
How to use it?
- Execute each step as if you were a customer
- Review the extracted variables at the end
- If there are errors, adjust the step instructions
- Use the
Restart conversationbutton to start from scratch
The playground does not access the knowledge base. It only uses the logic defined in the conversation flow so you can test the AI Agent before using it in a real conversation.
The order of interaction rules matters. If you want to capture name, email, and postal code, they must be defined in that order in the flow logic.
⚠️ Limitations and considerations
- The AI needs clear context: the more detail, the better the results.
- Don’t rely on images. Use structured text.
- Treat it like a new collaborator: train it to understand how your company works.
❓ Frequently Asked Questions
Can the AI Agent have conditionals?
Can the AI Agent have conditionals?
Not directly. The AI follows a linear flow. You can use the conditions block in the conversation creator to handle decisions based on variables or schedules.
Can the AI Agent read my website?
Can the AI Agent read my website?
No. The AI Agent does not currently access websites. You must upload a document with the information you want it to learn.
Can the AI Agent interpret images?
Can the AI Agent interpret images?
Partially. The AI can read text within images, but we don’t recommend relying on images.
The ideal is to provide content in plain text and well organized.
Can the AI Agent listen to audio?
Can the AI Agent listen to audio?
Yes. If your user sends audio, the AI can listen to it and generate a text response. The AI will not send audio; it will always respond with text.
Can I configure schedules to redirect to a human agent?
Can I configure schedules to redirect to a human agent?
Yes. Use the time conditions block to:
- Send to a human agent during business hours
- Send to the AI agent outside business hours
Can it transfer a lead to a human agent?
Can it transfer a lead to a human agent?
Yes. You can configure the flow so that the AI, upon completing qualification, redirects the lead to a human or registers it in your CRM.
Can I see what the AI wrote?
Can I see what the AI wrote?
Yes. Each execution saves a history in JSON format that you can review.
You can also see the extracted variables in the results table.
Can human agents see the history?
Can human agents see the history?
Yes. If you connect the AI block to a human, the agent will see the complete history as if it were a previous conversation.
Additionally, if you’re using a connected CRM, the agent will see the captured properties in the user panel.
Can I have more than one AI Agent?
Can I have more than one AI Agent?
Yes. You can create multiple AI agents with different interaction rules, knowledge bases, and flows within each conversation.
Can I modify an AI Agent's behavior after creating it?
Can I modify an AI Agent's behavior after creating it?
Yes. We recommend always giving the AI as much context as possible, but if you detect that your customers are asking questions you didn’t expect, you can always update the agent’s context and behavior.
Can I use an AI Agent as inbound or as HSMs?
Can I use an AI Agent as inbound or as HSMs?
You can use the AI Agent directly as an inbound start.
To be able to send an AI Agent or start a conversation toward the customer, it’s necessary to add an HSM before the AI Agent in the conversation creator.
Can I use the same AI Agent in multiple conversations?
Can I use the same AI Agent in multiple conversations?
Yes, the AI Agent is designed to be used in as many conversations as you require. This will allow you to not have to feed a different knowledge base for each conversation.