Rudder Lookout Private Beta

Rudder Lookout is currently in Private Beta, where we work with early users and customers to test new capabilities and gather feedback before making it generally available.

Contact your Customer Success Manager if you are interested in enabling Lookout for your team.

Rudder Lookout is an AI-powered analytics and activation agent for marketing teams. You ask questions in plain language, and an AI agent answers them and acts on them against the customer data you already trust — without writing any SQL, raising a ticket, or waiting on an analyst.

Each result that Lookout shares is saved as a durable artifact you can revisit and share, rather than a transient chat reply.

Audiences and activation

Most of what teams do in Lookout starts with a sentence like “create an audience of my most valuable customers”. The agent reads your warehouse, works out which tables hold those people, and drafts the audience.

You then see what it built. Every rule can show you the data behind it — the spread of a numeric column, the values a text column actually contains, and how many people each choice keeps. Adjust a rule and the count moves with it.

When the audience is right, send it to a destination in your RudderStack workspace and it stays in sync from there.

See Audiences and Activation for more details.

What Lookout connects to

Lookout works with the systems you connect it to, for example:

• When you link your RudderStack workspace, it can activate audiences to your destinations and read your sources, transformations, and Tracking Plans
• When you link your warehouse, it answers questions on your data and builds audiences from it
• When you approve external AI agents or assistants, they can connect to Lookout through the Lookout MCP

The following integrations are currently supported:

What Lookout produces

Lookout’s output is a set of durable artifacts:

• Audiences: Saved definitions of who belongs in a segment, built from your warehouse data
• Activations: Live syncs that keep an audience up to date in a destination
• Answers: Charts and tables produced from a live warehouse query
• Dashboards: Saved, parameterized views that re-run their queries on every load
• Documentation: Generated descriptions of the data in your warehouse and repositories, stored in the Context Hub
• Memories: Learnings the agent captures from your conversations, stored in the Context Hub and recalled in future chats

How it works

Lookout reasons over the context a RudderStack team already has — the events flowing through your pipelines, the Tracking Plans that define them, the warehouse they land in, and the destinations they feed. Because of this, a single conversation can move from a warehouse metric, to the audience it suggests, to the ad account that audience is sent to.

Two rules govern how the agent operates:

• Least privilege, per task: The agent is given only the tools a task requires. With nothing connected, it can converse but has no external reach — connect a warehouse and it can query it, connect a repository and it can read it, and nothing more.
• A human approves consequential actions: The agent operates with a human-in-the-loop — it prepares the work and a person pulls the trigger. An audience is saved by you, sending one to a destination is an explicit action you take, and publishing a dashboard is always a human choice.

Next steps

• Set up your workspace with the Quickstart
• Build your first audience — see Audiences and Activation for details
• Review the Rudder Lookout Key Concepts
• Connect your tools — see Integrations for details
• Understand the data model in Security and Data Handling