How Machine Learning can solve the Mobile App Uninstall Puzzle
March 3, 2017
Post Views: 357
The explosion of E-Commerce Mobile apps and their adoption by online businesses is a well-established phenomenon by now. Well on their way to becoming the toast of the digital world, 2016 is expected to see mobile ad spends dwarf their desktop counterparts by more than 200%. There are now well over 3 million apps across mobile app stores and they are expected to account for 4 billion users by 2020. That’s roughly 60% of the World’s population. Staggering!
Since 2014, driving app installs and enhancing Play Store ratings became key metrics or focus activities for E-commerce marketers. The app install spend stands testimony to the fact that the primary focus and attention of the marketer was towards driving installs, with more than 40% of marketing budgets and mind space on mobiles accounted for by the apps installed on them. In addition, tracking the installs gave rise to new businesses in Mobile Measurement Partners and we can see the pervasive presence across geographies indicating the importance of tracking Install campaigns.
Until recently, the general assumption was that gaining app installs was the ultimate criteria that would ensure business growth. However, that line of thinking was shown up to be misleading when app marketers eventually started witnessing 70-90% of installs being lost in a period of 1-3 months.
However, the fact that the top E-commerce apps successfully retain 30-40% of users on average suggests that this is not necessarily a bad starting point, albeit considering the next few steps are planned with an eye on preventing uninstalls from claiming a few victims . In simple terms, a higher retention curve is also possible provided sufficient steps are taken to focus on retaining app users.
Having seen that a more focused approach on user retention can do wonders to business growth, let us approach this challenge in 2 steps
1. Overhauling primitive strategies
Despite app user churn being a serious problem, the method initially adopted to tackle this was rather basic. Typically, marketers used to obtain the email/device IDs of the users who uninstall their app and reach out to them on social media and email channels. The flip side of this was, of course, that both are paid advertising channels.
This practice can also be quite inefficient as it requires manual intervention in setting up campaigns.
In other words, even though it helps in gaining lost users, this method incurs additional cost wherein a bit of intelligence from the systems could have, in hindsight, prevented channel isolation and saved thousands of dollars.
To put it succinctly, this campaign could have been way more efficient were marketers able to identify the users who are actually valuable to the business.
Be a Smart Marketer - Reduce Uninstalls with Machine Learning
Going back a few years, RTB (Real Time Bidding) became a rage as it ushered in greater efficiency in media buying when compared to the existing bulk media purchase. The bulk media purchase had long demonstrated obvious shortfalls as it did not qualify the user to whom an impression needed to be shown.
RTB, on the other hand, basically assigned a monetary value to the perceived value of the user to a business,a flexible quantity which could be further modified in real time and define a user’s perceived value to the business.
Similarly, uninstalls could be handled more effectively if marketers can assign a value highlighting the risk of a particular user who is uninstalling the app. This would help them identify what they are willing to spend to retain that user.
For instance, a high risk user who has purchased earlier with the business could be offered steep discount for the next purchase, while this sort of indulgence might not be afforded to a user who has installed the app that very day and is yet to make any kind of transaction on it.
We here at Vizury use proprietary algorithm for uninstall predictions which gives precisely this information to the marketer. An uninstall score of sorts is assigned to each user based on their engagement behaviour with the app, helping the marketer define a hierarchy of users categorized based on how likely they are to uninstall the app. This empowers marketers to prioritize their users by segmenting them thus,
Purchased User with Uninstall risk of >=70%
User with Products Added to Cart & Uninstall risk of >=80%
New User with Uninstall risk of >=90%
This also gives one more key advantage - Ability to engage with high risk users through the app push notifications channel without incurring any additional cost. Once the user uninstalls the app, that’s when email or display/social remarketing take over as the major channels to him/her. Since this consumes a pretty significant budget, it makes sense to keep this as a backup option.
So there you go. A smart way of approaching a problem is all you need to see your retention rates spike drastically. This may well prove to be the panacea for your business growth!
Got any questions for us? Get back to us at email@example.com with your thoughts and queries.
Nishant Kadian takes care of content marketing for mFaaS. He is passionate about sharing his learning on the ad technologies, mobile ad fraud preventions, and more. Drop him a 'Hello' on LinkedIn or Twitter to start a conversation with him.