Looker and Generative AI: Get Automated Insights From Your Looker Dashboards

Starting With a Challenge

At Devoteam, we’re passionate about helping clients embrace data-driven decision making. Tools like Looker have revolutionised how businesses access and explore their data, providing real-time dashboards and powerful data exploration tools. Even with amazing dashboards, a critical challenge remains: information overload. Imagine having to monitor dozens of dashboards daily. How do you efficiently pinpoint the insights that truly matter?

To tackle this, we’ve developed a novel solution that combines Looker’s analytical power with the intelligence of Generative AI in Google Cloud’s Vertex AI. Our goal? To automate insight delivery and free users from the burden of manual analysis. Now, they can receive scheduled summaries of key findings, making data-driven decision making more efficient and impactful than ever before.

Current Landscape

Together with the issues mentioned before, we found that deploying these tools was not a trivial task at all. Most data analysts will know how to work with SQL and Python, but now having to troubleshoot Vue and Javascript (which Looker makes use of for their extensions) will put analysts away from trying to touch these tools. If something breaks or needs to be customised according to your business needs, then it will require a lot of research beforehand if you do not have anyone on your team with the needed knowledge.

Before building our solution, we explored existing tools for automating Looker dashboard summaries. Two options stood out but ultimately fell short of our needs:

• The Dashboard Summarisation Tool: an open-source project, it adds a button to Looker dashboards for generating on-demand summaries. This is helpful for interactive exploration and ad-hoc analysis, but doesn’t address the need for automated, proactive insights. While it can export to channels like Slack or Google Chat, it still requires manual intervention.
• Spectacles TLDD: a Looker extension, offers scheduled email summaries. However, Spectacles converts the dashboard into a PDF which is then used by Gemini in the summary instead of directly leveraging the much richer Looker’s API. Furthermore, its complex configuration and reliance on front-end technologies like Vue and Javascript can be challenging for data analysts accustomed to SQL and Python.

Ultimately, both tools presented limitations in terms of automation, ease of use, and deployment.  This, coupled with the desire for a more streamlined and analyst-friendly solution, motivated us to develop our own.

Solution

Architecture

Our solution builds upon Looker’s existing Actions and Scheduled Plans framework, ensuring seamless integration and a user-friendly experience. We leverage the power of serverless computing in Google Cloud to further simplify deployment and maintenance.

Here’s a closer look at the technical architecture:

• Looker Integration: A Looker Action triggers a custom Scheduled Plan, sending dashboard data to our serverless application running on Cloud Run.
• Cloud Run Application: This application retrieves dashboard metadata from Looker’s API and passes it to Vertex AI for summarisation.
• Vertex AI Summarisation: Vertex AI’s advanced generative AI models create concise and insightful summaries of the dashboard data.
• Flexible Delivery: The application delivers the generated summaries to your chosen channels, providing flexibility and customisation options.

This architecture ensures a robust, scalable, and easily maintainable solution. The deployment process is streamlined and automated, requiring minimal manual intervention. Once deployed, the solution is fully integrated with Looker, allowing users to schedule and manage summaries with ease.

Demo

Below you will find a short demo of our solutions. Your automated insights from several Looker Dashboards will appear in your specified channels like Slack or Google Chat.

Conclusion

We understand that this is just one piece of the puzzle when it comes to addressing the challenges faced by data teams.  However, by automating the delivery of concise and insightful Looker summaries, we hope to alleviate some of the burden and empower businesses to make faster, more informed decisions. In the future, we plan to add more functionality, like follow-up questions and prompt refinements directly from the chats themselves. This will make it easier for users to work with the end result in whatever way they want.

Further reading

Want to know more about Looker? We got you! Here are more resources from our experts:

• How to become data-driven with Looker?
• Looker’s 4 unknown capabilities
• Unleashing the Power of Looker’s Machine Learning Accelerator