Current tools around data collection, storage and processing have opened up the opportunity for real time and near-real time analytically driven personalization of user experiences on mobile devices. This personalization can be applied in a variety of ways for such purposes as boosting click-through by intelligently targeting users with contextually relevant advertisements. Personalization can also increase in-application purchases by determining who to make an offer to; this is a critical component given that in-app payments are more profitable than paid apps. The key to success in personalization on mobile is understanding the user as deeply as possible and predicting interests and engagement from that understanding. Rather than relying on blanket statements such as “women like lipstick, and I think this user is a woman because mostly women use this service, so I will show a lipstick ad to this user” predictive analytics is data driven by leveraging previous direct interactions with the user.

As the Data Strategist for Medio, when speaking to users about using event based logging for informing predictive analytics I often get the question “what should we be logging?” My answer is always the same, “everything”. One reason behind this approach is that data (often in the form of event logs) is used for many purposes, and if you fail to capture the specifics initially, they are likely to be lost forever.  Another reason is the context that thorough logging provides – if you only capture purchases but fail to include items viewed you can’t calculate a click through rate or determine the difference between items nobody looks at versus items nobody buys. If you only focus on monetization events, you have limited understanding of what leads up to a monetization event (or more importantly a failed monetization event). Yet another reason is that you never know what is predictive of the outcomes you are interested in (and by the way, the outcomes you care about today may change tomorrow) which is why we engage in data mining to begin with – to detect otherwise hidden patterns in data. This by no means is an exhaustive list.

Data is used for many things including basic reporting, tracking, monitoring, ad hoc analysis, user experience insights, satisfaction estimation, experimentation, feature testing, behavioral prediction, and forecasting. As such it is imperative that measurement be thorough – so thorough that you can recreate the user experience as faithfully as possible through the examination of the event logs. Employing a minimalist strategy to data collection is an approach fraught with future frustration and a disservice to your organization that relies on data for important insight into how your products and services are being used in the marketplace. Storage is cheap, and time is short. Take the time up front to properly instrument your products and services to capture as much fidelity as you can and you will never be stuck telling your boss “I’d love to answer your business question but we decided not to measure that.”