advertisement
advertisement

Mixpanel Launches Predict, An Analytical ‘Magic 8-Ball’ For User Engagement

Mixpanel’s new predictive analytics product Predict claims to be able to get your business more engagement within 30 seconds.

Mixpanel Launches Predict, An Analytical ‘Magic 8-Ball’ For User Engagement
[Fortune Teller: Everett Collection via Shutterstock]

The small field of “predictive analytics” wants to turn all the data mined by businesses into a type of “Magic 8-ball” that can be used to predict and solve problems before they start. Today, analytics firm Mixpanel is releasing Predict, a predictive analytics product that claims to figure out how to get your business more engagement in 30 seconds.

advertisement

Mixpanel has been collecting business analytics since it launched in 2009 and it has claimed some high-profile clients in the past few years–names like Uber, Spotify, NBC, Healthcare.gov, OpenTable, and Fitbit, along with snagging $65 million in a funding round with an $865 million valuation last December. Mixpanel’s clients want to know how many actions a user needs to perform before they become engaged, like how many songs a user might listen to on Spotify before subscribing. Predict ingests a client company’s data and tells them how to get users to those tangible goals.

“The current state-of-the-art [in analytics] is modeling information. It’s actually just taking a taking a piece of data and graphing it,” says Mixpanel CEO and cofounder Suhail Doshi. “We take a look at all the people who reached that goal previously and figure out what it is about them that [gets them engaged]. If these users are like other users, then they have a higher likelihood of converting, and that’s the premise of most machine learning–to pattern most users.”

Based on a client’s previous user data, Predict aims to dissect which users will become engaged and which won’t. Of those that won’t, Predict will suggest actions the client can take to nudge that user into becoming engaged. This could be anything–an email newsletter, a push notification, an in-app alert–but Predict bundles all the actions in a client’s arsenal and lays out a path-to-engagement that aims to mimic exactly how similar users became engaged over time.

Predictive analytics have been around for a few years, but are unrefined. Predict’s machine learning guts are based on six years of business data Mixpanel has been collecting since it launched, and gives the platform a leg up on new competition. This data amounts to 50 billion user actions per month that Mixpanel tracks across all its clients, which could stack up to trillions of actions taken over the years, says Doshi.

But Predict itself is still a young product. Mixpanel has only released Predict for around 20 clients to try out over the last two weeks. We won’t be able to tell whether Predict can deliver on its claims to lay out a path-to-engagement in mere seconds until more usage data gives us a clearer picture of Predict’s accuracy. But like all machine learning that refines its algorithms with more data ingested, Predict should always be getting more and more accurate as it moves forward.

“The thing about machine learning is that it just takes time. Think about the first Google search. It was a great product, but not nearly as good as it was a couple of years later. Think of this as a first version of that, there’s a lot of work to be done” says Doshi.

Likewise, it will take a few months for Predict ramp up to that 30-second diagnosis, says Doshi. It will likely take under 30 minutes from ingesting company data to delivering a path-to-engagement, but getting to the instant diagnosis phase is a serious goal. In Predict’s case, the golden goose isn’t so much tons of users but tons of actions. This is why Doshi and the early Mixpanel team dispensed with pageviews and number of unique visitors back in 2009, the du jour metrics of the time, in favor of tracking actionable behavior like shares. This is how Predict is going to suggest paths-to-engagement for early-stage products that have few users, says Doshi.

advertisement

“What do we do when there’s not enough signal, and not enough users? There’s a point where the minimum is too small. Machine learning is nuanced and can find a lot of information from signals as long as there’s variance in the signal. It’s less to do with the number of users and more to do with how many things they are measuring and how much signal, how much engagement and behaviors,” says Doshi.

Video