You can use Conversational Analytics to generate a scheduled agentic workflow with an Explore data agent, dashboard agents, or within a Conversational Analytics conversation with a standalone Explore.
A scheduled workflow delivers data reports and natural language insights to designated recipients at specified intervals without requiring a threshold trigger condition. Workflows can deliver notifications to Slack and email in accordance with Looker email domain allowlists.
Before you begin
Before you can use Conversational Analytics to generate a scheduled agentic workflow, you must fulfill the following requirements.
Required permissions
To create a scheduled agentic workflow, a Looker admin must grant you the chat_with_agent permission for at least one model that underlies the dashboard, Explore data agent, or Explore that is used in the conversation.
For more information about Conversational Analytics permissions, see the Set up Conversational Analytics in Looker documentation page.
Scheduling frequency
When configuring a scheduled workflow, you can set the delivery schedule using standard Looker evaluation frequencies, including the following:
- Minutes: Run every specified number of minutes. Minute-level configurations are limited to a maximum frequency of 15 minutes. If you specify an interval shorter than 15 minutes, Conversational Analytics adjusts the schedule to 15 minutes and notes the adjustment in the workflow plan summary table.
- Hourly: Run at specified intervals throughout the day.
- Daily: Run once each day at a designated time.
- Weekly: Run on specified days of the week.
- Monthly: Run on a specific day of the month.
- Custom days: Run on specific, selected days of the week or month.
Delivery methods and message content
You can configure scheduled workflows to deliver reports to Email or Slack:
- Header disclaimer: For insights delivered over email and Slack messages, the header to reports, emails, and messages is "Powered by Gemini. Verify as AI can make mistakes."
- Natural language narrative summary: The body of the email or the text block of the Slack message renders an AI-generated natural language summary generated directly from the execution of your prompt and instructions.
- Data tables: Deliveries include a clean, readable data table presenting the underlying query results. Multiple data blocks are rendered in separate blocks.
- Embedded visual charts: Visual charts and graphs generated by the agent can be rendered and embedded directly into the email body or Slack message blocks.
Slack delivery limits
When delivering scheduled workflows to Slack channels or direct messages, the following limits apply:
- Narrative character limit: The AI narrative summary is delivered in a single text block capped at 3,000 characters. If the narrative exceeds 3,000 characters, Conversational Analytics truncates the text.
- Table capping: Native data tables in Slack are limited to a maximum of 100 rows and 10 columns. If the dataset exceeds either limit, the system renders only the top 100 rows or top 10 columns.
- Truncation notice: If a message or data table is truncated due to Slack limits, the message displays the following notice: > This insight has been truncated due to Slack limits. Click Investigate below to view the full analysis and dataset
PDF report generation
When you select the PDF option for a scheduled workflow, Conversational Analytics compiles and generates a high-fidelity PDF report artifact:
- Combined content: The PDF combines the agent's natural language insights summaries with the corresponding visualizations (charts and graphs) according to the presentation formatting defined in the workflow instructions.
- Delivery: The generated PDF is attached directly to the delivery email or uploaded as a file attachment within the designated Slack channel or direct message.
Investigate workflow notifications
When you receive a workflow delivery by email or in Slack, you can select the Investigate link in the notification message to ask follow-up questions and explore the data in Conversational Analytics.
When you investigate a workflow delivery, Conversational Analytics initiates an investigation session with the following behaviors:
- Private and isolated conversations: Each recipient who selects Investigate receives their own dedicated, private conversation thread with the primary data agent where the workflow originated. Other recipients, including members of email distribution lists or shared Slack channels, can't see your questions or analysis.
- Persistent conversation history: Your investigation conversation is persistent. If you leave the conversation and select the Investigate link in the notification again later, Conversational Analytics returns you to your existing conversation history for that specific workflow run.
- Primary agent routing: The investigation conversation automatically opens on the primary agent that generated the workflow. If the agent is deleted and the associated Explores can't be identified, an error appears.
- Access control and data permissions: You must have access to the underlying Explores associated with the conversation. If you don't have access to any of the associated Explores, an error message appears informing you that you lack the required data permissions to investigate the insight.
Related resources
- Agentic workflows overview
- Create a triggered agentic workflow
- Admin settings – Workflows
- Admin settings – Workflow Executions
- Manage your agentic workflows