Deploy AI

Overview

Once you have trained and tested your AI Agent, the final step is to deploy it so it can begin handling live customer conversations. Deployment is managed through the Flow Builder, giving you complete control over when and how your AI interacts with customers.

You can activate your AI in two ways: by using a pre-built template or by integrating it into a custom flow.

Method 1: Use the "AI Handles All New Chats" Template (Recommended)

This is the quickest and easiest way to get your AI up and running.

1. Navigate to Automations > Flow Builder.

2. Click + New flow.

3. From the template gallery, select AI Agent from the side menu and choose the "AI handles all new chats" template.

4. This will create a pre-configured flow designed to let the AI be the first point of contact for all new conversations.

5. Configure the flow and click Publish to activate it.

Method 2: Add the "Let AI Handle" Node to a Custom Flow

If you have existing flows or want more granular control, you can add the AI node to any flow.

1. In the Flow Builder, either open an existing flow or create a new one from scratch.

2. Add a new action node and select "Let AI handle".

3. Connect this node to your desired trigger, such as "Message received".

Key Configuration Settings

Whether you use a template or build a custom flow, there are two crucial settings to configure.

1. Set the Flow Trigger Correctly

For the AI to function as a frontline agent, you must configure the trigger settings correctly.

1. Click on the "Message received" trigger node at the start of your flow.

2. In the settings panel on the right, under Flow trigger settings, select "Trigger once every new open chat".

Why is this important? This setting ensures the AI gets one chance to handle the conversation per session. If the AI cannot resolve the issue and escalates the chat to a human agent, it will not try to jump back into the conversation, even if the customer sends more messages. The AI will only become active again after the chat has been closed and the customer starts a new one.

2. Configure the "Let AI Handle" Node Outcomes

The "Let AI handle" node has two main exit paths that you must configure to control what happens next.

• When AI agent cannot handle effectively: This is the escalation path. If the AI determines it cannot answer the question or is instructed to escalate by a Scenario, the flow will proceed down this path. You should connect this to an action like "Assign to agent" or "Add a label".

• If no customer response after...: This path is triggered if the AI has sent a message but the customer doesn't reply within a specified time (e.g., 1 hour). This is useful for automatically closing inactive conversations. You should connect this to an action like "Close chat".

Flexibility and Control

The power of using Flow Builder is its flexibility. You can use the "Let AI handle" node as a fallback mechanism. For example, your flow could first check if a customer's message contains specific keywords (like "order status"). If it does, it follows a standard flow. If no keywords are matched, you can then direct the flow to the "Let AI handle" node to manage the query.

Final Check: Before publishing your AI flow, make sure you don't have any other active flows with conflicting triggers. For example, having two active flows that both trigger on every new message can cause unpredictable behavior. Ensure your AI flow is the primary one for handling initial customer contact.