Most app marketing campaigns are focused on driving a large volume of app installs at the lowest possible cost. While this may be a good initial strategy, it must give way to more downstream focused strategies as the campaign matures. More savvy app advertisers are increasingly recognizing the folly of an install-focused approach and are instead exploring mechanisms for optimization of post-install events.
Common post-install metrics of interest include:
- Cost Per Retained User at Day-n (CPRU-n) - cost of acquiring a new user who is retained through the end of a certain period represented by the number of days (n)
- Return On Investment at Day-n (ROI-n) - return on advertising investment resulting from in-app purchases and other user monetization over a certain length of time after app install represented by the number of days (n)
Use of deep tracking in combination with machine learning based campaign optimization is the most robust way of implementing an effective post-install optimization strategy. In order to be effective, a good post-install optimization strategy requires two things: robust and deep data collection, and a powerful optimization algorithm.
Big Data Analytics
Aarki’s big data analytics capability is built upon a streamlined architecture that covers data storage, real-time retrieval, robust redundancy, real-time analysis, insights generation, and behavioral forecasting. This architecture enables the company to query and analyze historical user behavior and develop a dynamic user model for post-install events within seconds.
With the Aarki Encore platform, advertisers have the ability to specify a pre or post-install goal of their choice while setting up the campaign or at any time during the campaign lifecycle. The platform has an easy to use tool that allows mappings between arbitrary keywords and human-readable event names. When a goal is defined, the machine learning algorithm picks it up and automatically drives the campaign towards optimal performance against that goal.
Because of this infinitely flexible data architecture, campaigns can be optimized on thousands of potential metrics such as cost per retained user at day-n, return on investment at day-n, number of events of interest of different types, registrations, listings, app usage, and other metrics. Since advertiser goals tend to be unique, especially in the post-install scenario, this architecture enables greater alignment with the actual business objectives. For example, a classifieds app advertiser may be interested in optimizing on the number of listings users put up within the first 24 hours after installing the app. A social networking app, on the other hand, would want to maximize the number of users installing the app and registering for an account.
Our programmatic advertising platform - Aarki Encore - uses proprietary machine learning algorithms to expand the target universe of users (i.e., lookalike targeting) and to better inform real-time bidding decisions by estimating post-install metrics.
Lookalike targeting in Aarki Encore allows us to leverage historical data received from the client as well as learning gathered from prior campaigns to predict user interests, purchase behavior, and key post-install action drivers. Other user information such as demographics, interactions with other apps, engagement with different types of ads, device ownership, and geographical factors are also used in developing these predictions. Results from this predictive analysis are then used to mine billions of user profiles to identify other users who are most likely to engage in similar post-install actions.
The underlying machine learning methodology for lookalike targeting consists of a combination of unsupervised learning for feature selection and predictive modeling for profile identification.
In a previous article, we discussed some unique aspects of the machine learning algorithm in Aarki Encore. The platform uses a proprietary Bayesian Logistic Regression algorithm to predict the probability of pre and post-install events in a real-time bidding (RTB) situation. This predicted probability is used to determine which ad impressions to bid on and the bid amount. Hence, the accuracy of prediction not only determines the placement of the ad but also the CPRU-n and ROI-n.
Use of the Bayesian algorithm ensures that the platform is able to capture both prior knowledge as well as random changes to the system in a robust manner. As a result, the campaign team can make campaign decisions that are statistically sound and practically trackable.
Recent Campaign Example
The chart below show performance data from a recent campaign that was optimized for a post-install metric. In this case, the optimization metric was Day-1 Cost Per Retained User. The pink line represents the actual campaign CPRU-1 and the dotted line is a negative exponential fit.
We can see that the Aarki Encore algorithm converges very swiftly to the optimal solution with a half-life of 3 days and overall optimization period of 2 weeks. Once the optimal solution is accomplished, the algorithm continues to maintain this over the remainder of the campaign duration (even while volumes are scaled up).
In this article we highlight the unique big data architecture and machine learning features that make Aarki Encore especially suited for post-install optimization of app marketing campaigns. The platform not only delivers superior performance, it also has the flexibility to cater to an infinite variety of post-install objectives in a robust manner.
For more information or to schedule a complimentary 15-minute consultation with one of our analysts, please email email@example.com.