Introducing Lookout: AI-powered analytics and instrumentation for RudderStack

The gap between the people who produce data and the people who depend on it has always been a friction point. Lookout closes it.

You’re staring at a dashboard that looked fine last week. Conversions are down. A key metric has gone flat. You try some SQL, poke around in your BI tool, but the answer isn’t in the data you can see. It’s somewhere in the instrumentation code, in a tracking code change introduced by the iOS engineer last Tuesday or the DBT model that was updated by the analytics engineering team. So you do what everyone does: file a ticket, ping Slack, and wait.

Today we’re introducing Lookout, RudderStack’s AI-powered analytics and instrumentation platform. Lookout puts an AI agent at the center of your customer data infrastructure: one that understands, your instrumentation code, your tracking plans, your warehouse data and DBT transformations, your pipeline status and your team’s context, and can investigate issues, build dashboards, generate instrumentation PRs, and review code changes. All through natural conversation.

RudderStack has always been built around the belief that your data should work for you, not the other way around. Lookout extends that to the way teams understand and act on their data: less time assembling context across tools, more time making decisions with the answers.

Why Lookout works: Context we already have

Most AI analytics tools bolt a language model onto a query interface. You can ask questions, but the answers stop at the data you’ve already collected. When something looks wrong, the tool can tell you that a metric dropped. But it can’t tell you why, and it can’t fix it.

The reason Lookout is different comes down to one thing: context. RudderStack sits in the data journey from the moment an event is defined in source code to the moment it lands in your warehouse. Source code, tracking plan, pipeline health, transformation configs, it’s all there as a natural byproduct of how the platform works. That context doesn’t require separate maintenance, and it isn’t assembled on the side. We already have it.

That’s what lets Lookout’s agent go from a one-line business question to an accurate answer, and from a flagged problem to a real fix, with the same data and the same governance your team already trusts.

What Lookout does

Lookout is a workspace where you talk to an AI agent that has full context over your data infrastructure. The agent can query your warehouse, read your instrumentation code, reference your tracking plans, and connect what it finds across all of those sources to give you answers, not just query results.

Natural language analytics lets anyone on the team query their Snowflake warehouse without writing SQL. Ask what you actually want to know: “Why did checkout conversions drop last Tuesday?” or “Which events are we tracking on mobile but not web?” The agent handles schema discovery, query construction, and interpretation. Results come with the context assembled to find them, not just numbers on a screen.

Dashboard creation goes a step further. Describe what you want to track or monitor, and Lookout builds an interactive dashboard with charts, tables, funnels, and metrics, complete with a shareable URL and live queries on every load. No BI tool configuration. No SQL editor. Teams that previously waited days for a dashboard request to get prioritized can have one in a single conversation.

Agentic tracking and instrumentation addresses the other half of the gap. Lookout’s tracking agent starts from a high-level business goal: “Track the onboarding flow from the login screen to the product list.” It writes the instrumentation for you, ensures the tracking code is correct, conforms to your existing tracking plan, and follows the coding conventions your team already uses. Instead of pushing changes silently, it opens a pull request so engineers stay in the loop and in control. The same process that used to span multiple sprints can now be drafted, reviewed, and submitted without leaving the tool.

Automated PR review closes the loop. Connect your GitHub or GitLab repositories and Lookout reviews incoming pull requests for breaking data contract changes: renamed events, removed properties, type mismatches. Issues that used to surface weeks later as broken dashboards get caught at the point of code review.

Context Hub is a continuously updated knowledge base of your instrumentation, covering event schemas, code locations, tracking plan coverage, and the institutional knowledge the agent accumulates through investigations. It’s the documentation teams always meant to write but never had time for, generated as a byproduct of the work.

Agentic activation (coming soon) takes this further still. Instead of building segments through a drag-and-drop UI, you give Lookout a high-level goal and it creates the audience for you, layering in external requirements if needed. The goal is to let business users express what they want, not assemble the how.

For teams that work in Slack, Lookout integrates directly so data questions get pipeline-aware answers without switching to a separate tool. For teams building on AI tooling, Lookout exposes an MCP server that external clients like Claude Desktop and VS Code can connect to via OAuth 2.1. All of this runs on AWS Bedrock with zero data retention, meaning your data is never stored or used for model training.

What this means for your team

For business teams, Lookout means you stop waiting in the engineering queue to get value from your own data, from instrumenting an event to building a funnel to standing up an audience.

For engineering teams, it means fewer interrupt-driven requests and changes that arrive as reviewable pull requests instead of vague tickets. Same data, same conventions, far less friction.

Get started

See it in action in the video below:

Lookout is in private beta, but you can explore the sandbox today at https://lookout.rudderstack.com/w/launch-demo/join and see its power firsthand: Add missing instrumentation, build funnels to understand users, diagnose data quality issues, create and save dashboards, and more–-all with natural language prompts.

You can get started with two simple prompts: Try “Build me a simple sales dashboard” to build a dashboard, then start a new chat and try “How many users start typing a coupon code but never apply it?” to see how Lookout handles missing instrumentation.