The world of mobile service and apps has many moving parts. For the mobile operators there are issues of roaming and service quality. And app providers need to ensure their apps are working properly across every OS and geography, that they are profitable, users are engaged, and so on.
Yet with so many elements
- front end/ back end, advertising platforms, partners, and more -
there are many opportunities for something to break. If one element is working less than optimally, it can cause problems elsewhere, just as the allegorical butterfly wing caused a massive storm somewhere else.
Many companies use dashboards and alerts to track what is happening on their systems and apps. For example, important KPIs to track for apps could include:
-Number of user installations
-Cross promotion success rate
And for service providers:
-Minutes of Usage (MOU)
-Call Drop Rate (CDR)
-HOSR (Handover Success Rate)
-ARPU ( Average Revenue Per User)
-ARPM (Average Revenue Per Minute)
-International Roaming Revenue
Each of these KPIs has its own typical pattern (for example, installations may drop during evening hours and rise again during daytime hours; or drop overall throughout the weekend). The typical pattern may also have a trend associated with it, as ideally app usage is growing over time.
And yet each KPI can be broken down much further into more granular metrics. For example, if we look at just one KPI such as installations, it can be looked at for each geographic region, and for each type of device (e.g. iOS, Android), for each device version (Android Jelly Bean, Lollypop, Nougat, etc.) and more. If you consider every possible KPI multiplied by every possible permutation, you will quickly arrive at a massive number of metrics that you need to track. Is it better to track “installs” or “installs on Android in California”? The more granularity the better since it allows you to find the root cause more quickly if something goes wrong, or to quickly leverage something that is working well. Add to that the fact that you probably want to measure these at a minimum every minute, and you will quickly find that you are overwhelmed with dashboards and alerts.
This is where anomaly detection comes into play. Using machine learning, you can learn the normal behavior of each metric, or even combinations of metrics (e.g. revenue per install). Then identify in real time if a metric behaves abnormally. In this presentation, we will describe how an anomaly detection system needs to be built and what data to collect in order to keep your mobile service and apps and their revenue humming.
Ira Cohen is a cofounder of Anodot and its chief data scientist, where he is responsible for developing and inventing its real-time multivariate anomaly detection algorithms that work with millions of time series signals. He holds a PhD in machine learning from the University of Illinois at Urbana-Champaign and has over 12 years of industry experience.
Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?
Join the conversation here (requires login)
©2018, O’Reilly UK Ltd • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • firstname.lastname@example.org