Q. In your work with predictive analytics, what behavior do your models predict (e.g., attrition, response, fraud, etc.)?
A. My models predict customer lifetime value as well as churn in customer engagement.
Q. How does predictive analytics deliver value at your organization? What is one specific way in which it actively drives decisions?
A. At AOL, predictive analytics helps us decide which partnerships we should invest more in and which ones we should abandon. Our models show which partners will provide a positive ROI and looking at how engagement and lifetime value are trending in our model, we are not only better equipped to make these decisions, we can make them much quicker than before (incorporating a predictive analytics model).
Q. Can you describe a successful result, such as the predictive lift of your model or the ROI of an analytics initiative?
A. Originally, our model predicted lifetime value of traffic generated from the various partners on a monthly basis – looking at different cohorts every 30 days. With one partner, in the first half of a 30 day cohort, we noticed that engagement and ROI was much lower than usual, so we sprung into action and adjusted our model to look at the data on a weekly basis as well and predict engagement trends on a week to week basis, as well as work with the partner as to how to get their engagement up to a level that was acceptable and ROI positive. Now, we look at this for every partner to anticipate issues like this before they arise again.
Q. What surprising discovery have you unearthed in your data?
A. One surprising discovery that I have unearthed in my data is that month 1 engagement (not monetization) is the leading predictor of customer lifetime value. The higher engagement of customers in the first month, regardless of monetization, will drive a higher LTV than a similar increase in monetization (in our model, monetization is CPC – cost per click). For example, a 10% higher CPC will have a smaller impact on LTV than a 10% higher month 1 engagement.
Q. Sneak preview: Please tell us a take-away that you will provide during your talk at Predictive Analytics World.
A. By taking a step back from the complex data intensive models, you will see how we created a 3 pronged model looking at audience, engagement, and monetization to predict the lifetime value of users coming from different properties or partners. This model is proven empirically, when looking at the expected (predicted) vs. actual data. You will also see how this model is adaptable to other areas and by changing the individual metrics that are measured, you can look at the lifetime value of customers in nearly any industry – not only what is the value of someone who came to a website; but how much is someone who enters a store worth or how much is someone who signs up for a mailing list worth, amongst others.
Q. What has been some feedback from stakeholders?
A. From the team that is in charge of the partnerships that my predictive models help assess: “this is an awesome tool that really helps us make smart decisions” and “the ability to get accurate LTVs timely has been helpful in making some quick allocation decisions.”