How one of the world’s most popular sports apps applied predictive modeling to create more timely and relevant communications.
Onefootball is the world’s most comprehensive app for the world’s most popular sport. From the latest news and scores, to connecting with fans across the globe, the team at Onefootball has built a product that their users love -- and is living proof that they are truly crazy about football.
Onefootball’s data-driven culture of product and customer development permeates all levels of their organization. As Rafael Aviles, a Data Analyst at Onefootball, puts it:
“As we move towards our goal of personalization, we need to deeply understand the patterns that distinguish user groups so we can ensure they receive the most relevant and customized experiences.”
When we first spoke to the Onefootball team about predictive churn management, they lamented the fact that today they can only observe when a user churns after the fact - and at that point there is little they can do to convince those users to re-engage. If only they could predict which of their users were at-risk of churning before they actually left the app, they could proactively engage their highest risk users with incentives to continue their relationship with their app and product.
The first step towards tackling a complex problem like churn is defining what churn means, and establishing a baseline churn rate against which you can measure improvements. For the Onefootball team, a user is considered to have churned if they go 30 consecutive days without opening the app. This means that currently active users may potentially have ‘churned and returned’ - or experienced some 30-day consecutive period with no activity in the past, and then returned to becoming an active user - defined as having at least two sessions in the past 30 days and a first session that is older than 30 days.
To establish a baseline churn rate against which to measure improvements, Onefootball looked at the proportion of their churnable users from the past 180 days that they observed to have actually churned. For Onefootball, churnable users are those with at least two sessions in their lifetime. This means that they excluded ‘One-and-Done’ users -- who had only had one session in their lifetime --from their benchmarking and subsequent predictive modeling processes, so as to keep the focus on retaining already active and engaged users.
Once we defined what churn meant to Onefootball and established a baseline churn rate, we were able to identify the user behaviors and attributes that were related to churn. Using historical data, we extended behavioral patterns into predictions about which of Onefootball’s current active and churnable users would be the most (and least) likely to churn.
Our predictions gave the Onefootball team a view of the distribution of churn risk across their user base. Current active users were placed into one of three churn risk Likelihoods: Low Risk, Medium Risk, and High Risk.
A view of Onefootball’s users by predicted Churn Risk in their Localytics dashboard.
Alas, insight doesn’t mean anything without action. Using Localytics’ engagement tools, Onefootball was able to target a series of push messaging campaigns to their Medium and High Risk users.
Using Localytics, Onefootball was able to target a new push campaign to High Risk users.
The campaigns were timely in that they preemptively engaged at-risk users with personalized, relevant messages that ranged from promoting a delightful new feature in the app allowing users to rate their favorite player at halftime, to communicating just-in-time information about player line-ups and breaking news as early as one hour before the start of a match. Some example campaigns that ran include:
Onefootball took an experimental approach towards measuring the impact these predictive campaigns had on churn rate. First, they randomly sampled a proportion of current active churnable users to control and experimental groups. Again, it was important for them to only look at a sample of churnable users who had at least two prior sessions in their lifetime, and had been active in the past 30 days.
We then calculated the churn rate observed in each group, and were able to draw some positive and statistically significant results! The experimental group presented a 7.4% reduction in churn when compared to the control group.
Seven percent may not seem like a lot at first, but any movement for an app like Onefootball with millions of users makes a big difference, especially when compounded over multiple months in the case where that improvement can be repeated consistently.
Given that CPIs have risen above $1.30, if Onefootball were to scale this program and rescue 25,000 users the estimated savings earned in customer acquisition spend would represent nearly $32,500 in value alone.
As an app that monetizes by partnering with brand advertisers to reach highly engaged football fans, Onefootball can also realize a benefit from keeping users engaged and exposed to ad impressions. Using a conservative industry estimate of $10.00 eCPM and 60 impressions per user-month based on market values Localytics has observed for the sports app vertical, a similar scale in the program is estimated to represent nearly $15,000 in additional monthly ad revenue.
But perhaps the greatest dividend paid by Onefootball’s investment in predictive segmentation is the additional light it shed on their user base. Thies Gruening, Onefootball’s Mobile Marketing Manager, claims that:
“Before, we assumed things about why users churn, and based our targeting on those assumptions. With predictive churn segments rooted in our data, our campaigns are no longer based on blind assumptions. Now, our anti-churn campaigns are something that we can do in a repeatable and scalable way.”
With this success, Onefootball plans to continue using Localytics to keep users engaged with the app throughout the year.
Thanks for signing up. Look for your first email shortly!
We’ll reach out shortly to schedule a time to talk.