Spark: Bringing generative AI to Mixpanel

The rules of SaaS are changing. For so long, using the apps and services we need to be productive has required technical formulas or exhausting interfaces. Generative AI is unframing all of that.

Have some scratch notes you’d like expanded into a new product requirements document (PRD)? You can now click a button to get AI to write and organize it for you. New efficiencies like these, and the wow factor they bring, are things we’ve never seen before in software.

You can add all of us at Mixpanel to the list of folks inspired by this technology. And when we started exploring the possibilities with generative AI months ago, we saw pretty quickly how it would help us accelerate our vision to bring analytics to everyone.

Introducing Spark

Today, we’re giving an early glimpse of that vision with the introduction of Mixpanel’s first step into generative AI. It’s called Spark, and the concept is simple. If you have a question about your data in Mixpanel, just ask your question—in plain language:

“What does our signup trend look like in the past year broken down by geo?”

“How many people visited our website yesterday?”

“What’s our ad spend this month?”

Spark will build the right report to get you the right answer, complete with the corresponding chart.

Spark will be rolling out in a closed beta soon (sign up here to join the waitlist), and we intend to continue working to make it even more robust in the coming months.

Since this is only the beginning of our plans for AI in Mixpanel, I thought it would be important to lay out the values underpinning all of our work in this area as we venture on. We want to ensure we’re creating as much wow factor as we are laying building blocks to make powerful analytics a part of how people everywhere work; here’s how we’re going to do it. 

We’re building transparent AI processes, not black boxes

First, we’re not pretending our AI is magic, so we’re happy to let you behind the curtain.

As a center principle, any generative AI feature we deploy in Mixpanel will be able to “show its work,” which means you’ll always be able to check for yourself exactly how analysis or other content is being generated.

When Spark builds a report, it’ll be viewable and editable like any other report, meaning you can go into its query builder view and see details like what events are being used. From there, you can even add your own edits to the report to make modifications or improvements.

This transparency is how we’re making generative AI in Mixpanel a powerful and instructional aid to your workflow rather than a feature that can only be trusted to take you so far.

We’re not trading off analysis power 

Speaking of power… It would be much easier to build a generative AI function in Mixpanel that serves basic analysis needs and leaves the advanced stuff to manual report-building. But that’s not our goal.

We want to make Spark capable enough to build reports that are just as complex as what your data science team can produce. All you’ll have to do is ask Spark the same way you would them.

We’re starting off strong already, with our early beta testers having access to advanced features like:

• Aggregate Properties: What’s the average number of minutes per video watched?
• Custom Buckets: How many videos were watched for the age groups 18-24, 25-29, and 30-34?
• Behavioral Breakdowns: For users who liked five videos in the past week, how often did they comment in that week?
  

And since getting to deep, insightful reports and visualizations can take some digging and discovery, Spark doesn’t stop listening after a single question. You can ask follow-ups to zoom in on areas of any report that’s generated, letting you further your analysis with plain language. This matches a natural step-by-step analysis process.

We’re getting everyone answers faster

Finally, Spark’s natural language querying is just the beginning of how we plan to tear down the friction between ideas and answers. We believe generative AI has the potential to create massive productivity gains for users of all levels who are doing analysis generation as well as analysis explanation, sharing, and other collaboration.

By building AI into Mixpanel as an implementation method and not a separate product, we’re taking away steps to get things done, not adding them. For example, to build a Mixpanel report with Spark, users go to the same place they would to build a report manually.

As AI becomes more capable and commonplace in software, user experience decisions will be more important than ever. Our design approach will continue to emphasize simplicity without compromising on power as we weave in AI to speed up workflows. You can continue to count on Mixpanel to get you—and your team—to the right data answers efficiently every time while putting you in control with inputs you can understand and validate.

Built with thoughtfulness and heart

The generative AI moment we’re in right now is exciting, and it’s easy to feel that at Mixpanel as much as anywhere else. For example, though we’d been developing AI use cases at Mixpanel for some time, Spark is a new method born out of sheer passion during our most recent hack week.

As we build Spark and similar features throughout the product, we won’t rely on a separate AI team at Mixpanel to develop it all. Instead, we have AI experts and enthusiasts working on every team, exploring possibilities of how this technology can be used to improve Mixpanel for our users today and the ones we hope to get tomorrow.

Let’s build 🤖

Spark is a new way of getting answers from your data. With natural language, you can ask for the advanced product, marketing, and revenue reports and visualizations you need to spark your next big idea. Where will your next Spark take you? Sign up for the waitlist here.