Automatic Insights at Mixpanel: past, present and future - Mixpanel
Inside Mixpanel

Automatic Insights at Mixpanel: past, present and future

Jenny Booth

Two years ago, we realized that the number of new Mixpanel users who were sending us events to track had dropped 10%. We did some digging and saw that the open rate of our welcome email had also gone down around the same time. Eventually, we realized that a minor edit we had pushed to our email campaign was cutting off key instructions on how to integrate Mixpanel.

It was a small mistake that had a significant impact on our KPIs, and we figured that our customers had probably experienced similar things. That’s what spurred us to build Anomaly Detection.

Since we first started building our Automatic Insights product in 2014, we’ve shipped five different models. In additional to Anomaly Detection, we built Predict to show which users are most or least likely to convert, Automatic Segmentation, which highlights interesting segments of users in saved reports, Anomaly Explanations, which looks for those users that are causing anomalies, and Signal, our correlation report.

Speaking of Signal, we’ve recently given the report a makeover, and bolstered it with new features. With new functionality to add multiple correlation events, filter by property, and break down results by property, it is even clearer which actions correlate with higher retention, conversion, engagement—whatever the goal may be.

Automatic Insights helps large enterprise companies such as Ticketmaster and Vente-Privee, as well as up-and-coming startups save time and work more productively.

Nikhil Karnik, the Head of Data Science at Yoshi, the ExxonMobil and GM-backed on-demand car maintenance startup, uses Mixpanel’s Automatic Insights suite extensively:

“Anomaly detection has been helpful in flagging issues on our servers, so I can alert the engineering team right away. It’s also helped us plan for growth, by highlighting spikes or trends in the number of people scheduling. Signal has also surfaced interesting correlations in behavior, of those who cancel orders but actually do end up converting at a later date.”

Olga Kostyshena, a Digital Marketer at award-winning mobile payment provider PM Connect, uses Mixpanel’s machine learning tools to spot unusual user behaviour trends or subtle changes in their data, that she might otherwise have missed:

“A good example of Mixpanel’s value is when it alerted us to the fact that subscriptions were lower than expected, caused by a technical issue with one of the carriers. Thanks to this, we were able to react immediately, reporting back to the technical team.”

For those Mixpanel customers looking to save time and boost productivity, simply save any Insights, Retention, or Funnel report to a dashboard, and we’ll take it from there. Any anomalies or interesting user segments we detect will appear in the Notifications tab, underneath the bell in the top right hand corner.

So what’s coming next? We recently hired Adam Kinney, an expert in machine learning and data science to lead the team going forward. He’s a seasoned veteran in building out and managing data and analytics teams at data-driven companies like Twitter, Google, and Schibsted, and already has big plans for machine learning at Mixpanel.

On the product side, we’re working on the ability to customize Notifications, so that users have more control over what reports our algorithms monitor, what triggers a notification, who gets notified, and more. If you’re interested in sharing your feedback and being the first to try out the new functionality, please apply to the beta program today.

Get the latest from Mixpanel
This field is required.