How to set up and use the Social Engagement Agent (Multi-Channel Engagement Agent with Approval)

The Multi-Channel Engagement Agent with Approval helps businesses manage incoming social interactions across channels and locations by generating brand-compliant, contextual responses for customer messages. The agent analyzes each interaction, determines whether a response is needed, and routes generated responses through an approval workflow before they are sent.

Birdeye also provides other pre-configured engagement agents for different response approaches, including Multi-Channel Engagement Agent with Auto Responses, Multi-Channel Engagement Agent with Triage, and Multi-Channel Engagement Agent with Suggested Responses. This article focuses on configuring and using the Multi-Channel Engagement Agent with Approval.

Click Multi-Channel Engagement Agent with Approval to open the agent configuration panel.

From the right panel, you can:

The Goals and Outcomes are already configured as part of the pre-configured agent, so you typically do not need to modify them. The primary configuration at this stage is selecting the appropriate locations for the agent.

The New social engagement is received trigger determines when the Social Engagement Agent starts processing a social interaction.

By default, the trigger runs when any of the following conditions are met:

The Message updated and Marked as not spam conditions work together. This allows the agent to process an existing message when it has been identified as legitimate rather than spam.

The pre-configured agent already includes these trigger conditions, so you generally do not need to modify them. You can use Add condition to further customize the trigger logic based on your requirements.

The Summarize message context action creates a concise summary of the incoming social interaction and the information surrounding it. This gives the next steps in the workflow the context they need to accurately classify the message and determine whether a response is required.

The action uses information from the incoming interaction, including the message text, parent post text, message type, and message media.

You can review the following configuration in the right panel:

The pre-configured agent already includes the required prompts, context, and input fields. Review these settings and modify them only when you need the action to behave differently for your specific requirements.

The Classify message action analyzes the incoming social interaction and produces a structured classification that determines how the agent should handle the message.

It classifies the interaction across five areas:

The action uses information such as the message text, channel, message type, parent post, and conversation thread to make the classification. It also uses dedicated Social intent classifier, Social sentiment classifier, and Social urgency classifier configurations for their respective classifications.

The pre-configured agent already includes these settings, so you generally do not need to modify them. You can review the System prompt, User prompt, Context, and Input fields in the right panel to understand how the action is configured.

After the Classify message action determines whether an interaction is spam, the agent uses the Based on condition branch to route the message through the appropriate path.

The workflow has two paths:

When the message is classified as Spam, the agent follows the Spam branch and runs the Mark as spam action. The message is marked as spam in Birdeye, and the workflow ends without generating or sending a response.

This prevents the agent from creating unnecessary responses to spam or irrelevant interactions.

Messages that are not identified as spam follow the Fallback branch. The agent then runs the Update classification action to update the message's intent, sentiment, and urgency based on the triage rules.

The Run triage actions tool receives the Message ID and updates the classification using:

After the message is updated with its intent, sentiment, and urgency, the agent uses the Based on condition branch to determine whether the business needs to respond.

The branch has two paths:

When the classification determines that a reply is required, the agent follows the Reply required path and runs the Generate reply action.

The action generates a contextual response based on the message context and the classification results.

The pre-configured action includes rules for:

The Generate reply action is designed to produce a natural, human-like response while keeping it aligned with the business's brand voice.

When the classification determines that a response is not required, the interaction follows the Fallback branch and the workflow ends without generating a response.

This allows the agent to avoid unnecessary replies for interactions that do not require a business response.

Once the agent generates a reply, the Approval workflow action routes the response for review before it is published. This allows you to control who reviews the generated response, when reminders are sent, and what happens if the response is not approved.

Click the pencil icon next to Send posts for approval to configure the approval workflow.

In Send posts for approval, select the approval workflow you want to use for the generated response.

Use the Select approval workflow to choose the appropriate workflow. You can also configure Remind if not approved, to determine when approvers receive reminders.

The available options shown are:

You can also specify who should be notified based on the outcome of the approval process.

Under Who should be notified if the post is rejected or auto-published?, search for and select the users who should receive notifications.

You can also enable Send automatically if not approved. When enabled, the approval will be skipped if no action is taken within 7 days, and the post will be sent automatically.

Once configured, click Save.

After the response passes through the Approval workflow, the Response scheduler determines how and when the generated reply is handled.

Click the pencil icon for the Response scheduler action to configure how the generated response should be handled.

The scheduler requires the Message ID, Response text, Response reasoning, and Approval Info from the previous steps. Under How do you want the reply to be handled?

You can choose:

For the Multi-Channel Engagement Agent with Approval, the generated response is routed through the approval process before it is sent.

When Send reply is selected, you can choose when the approved response should be sent:

Select the appropriate handling and scheduling option, then click Save.

After configuring the Social Engagement Agent, click Run test in the top-right corner to test the agent before publishing it.

After clicking Run test, the Preview settings window opens with a list of social posts/messages.

Select one of the available posts you want to use for testing, then click Run test to execute the agent against that interaction.

The selected message is used as the test input so you can review how the agent processes the interaction through the configured workflow.

After you run the test, the Test details panel displays the result of each step in the Social Engagement Agent workflow. A green check mark indicates that the step executed successfully.

You can review the results from the trigger through the final response-handling actions to verify that the agent is following the configured workflow as expected.

Each completed step can be expanded to inspect the information generated or used at that stage.

Click View inputs under a step to expand its details and review the corresponding inputs and outputs used during execution.

For example, you can review the inputs used by the Generate reply action, including the parent post summary, message summary, intent, urgency, message text, spam status, and sentiment.

Reviewing these details helps you verify that the agent is receiving the correct information at each stage and that the generated response is based on the relevant message context and classification results.

After configuring and testing the agent, you can choose how to save it.

Use Save as draft when you want to return and make additional changes before making the agent available.