Uncategorized

Demystifying Predictive Analytics for Mobile and Web Apps

Predictive analytics is unfathomable to many of us. Terms like artificial intelligence, neural networks and deep learning just deepen the mystery. Yet predictive tools are already ubiquitous. Not in self-driving cars, or robots that look like humans. Rather, on your phone.

Spotify looks at your past listening habits, and that of people who like similar music, and recommends new tracks to listen to. Youtube does the same with videos. Your email service predicts whether a mail message sent to you is spam by analyzing past emails sent to you and others. Siri and Google Now use predictive technology to understand your voice, and offer answers to your questions.

Predictive Analytics Tools on Mobile

Predictive analytics is the practice of using advanced statistics and historical data to predict future outcomes. For the most part, the use of predictive analytics in business is still shrouded in mystery and the domain of data scientists. This is changing. A few analytics vendors now deliver products that provide more than just insight into past trends, but also predict the future. These tools don’t require a data science background either, and are ideal for business users.

What if you could predict which of your users were most likely to remain engaged? Or churn? And which users were most likely to convert (and drive revenue)?

What does this mean for product managers and marketers? Trailing indicators don’t help you with understanding what will happen in the future. These indicators include what features are used, how many users stay engaged, or churn, and how many convert or monetize. They’re also prone to misinterpretation. As humans, our brains are wired to see causality where there may only be correlation. We have difficulty determining what is meaningful in a torrent of data.

What if you could predict which of your users were most likely to remain engaged? Or churn? And which users were most likely to convert (and drive revenue)?

Predictive app analytics tools provide these insights, and allow product leaders to build tools that take action where necessary. For example, intervening when a user has a high likelihood of churning.

But how do these tools work?

A company running an accommodation marketplace might collect data on app usage. From installs, to which users log in and how many times they do so, how many different properties are viewed by a user, whether users provide their phone number and connect to Facebook, whether they book accommodation (the conversion goal), and many other events that occur that in the customer’s journey.

It can be difficult to determine which of the many events captured are influential in a user booking accommodation. What do users who convert do in the app before booking accommodation? And what can be done to increase the conversion rate?

Predictive app analytics tools take the guesswork out of understanding user behavior.

Predictive analytics tools look at historical data (called training data), and use mathematical algorithms (often called machine learning algorithms) to identify patterns in the data that have a high probability of predicting an outcome. These patterns are used to create a “model”, and the model is utilized to predict an outcome from new data that becomes available.

To identify these patterns, the machine learning algorithms determine which “features” (in this example, which app events) are relevant to predicting an outcome. They essentially filter out the noise from the data.

The machine learning algorithm may determine that users who book accommodation first view at least 3 properties. This pattern is employed to classify which users are most likely to convert, or not. Knowing this, we could automatically send an email and/or push notification to users whose behavior indicates they are unlikely to book accommodation, prompting them to view several exciting properties in exotic locations.

Predictive app analytics tools take the guesswork out of understanding user behavior. The tools reveal to product managers and marketers which users are most likely to convert or not, and allows action to be taken before it’s too late.

Get the latest from Mixpanel
This field is required.