Go Behind the Scenes with Inuvo!

Today we are kicking off a brand new segment of the Inuvo Blog: “Behind the Scenes.”  This series will give you an exclusive look at what goes into making the Inuvo products you use day-to-day.  In this first blog entry of the series, David Hess, Data Analyst, provides an overview of how Inuvo uses analytics to improve our products and control potential fraud.

A regular Joe from Minneapolis who uses his credit card for day-to-day expenses like groceries and gas manages to save up enough extra money to take a trip to Rio de Janeiro.  The day he arrives, he tries to use this same credit card to get a cash advance from an ATM next to his hotel.  The transaction triggers a call from his bank to his cell phone to verify that the card has not been stolen.

A young couple in Tampa who wants to buy their first house makes a visit to the bank to talk about financing their purchase.  Their loan officer tells them that they’re eligible for a home loan with a rate of 5.8%.  Their close friends could only get a rate of 6.5%, the difference being that the young couple had always paid their bills on time.

A 28-year-old data analyst living in New York finds a life expectancy calculator, decides to give it a whirl, and is told he should live to the not-so-ripe age of 74.  He makes a vow to eat more vegetables.

What do these three stories have in common? Predictive analytics!  Predictive analytics is, at its most basic level, combining current and historical data to make predictions about the future.  The process starts by selecting a value that needs to be predicted, and then historical data that might be relevant is collected.  That data is analyzed to determine which characteristics are useful in predicting the target value, and a mathematical model that relates those characteristics to the target value is constructed and tested.  You can see this process embodied in the above examples.  In each case, known factors (past spending habits, credit history, and health/family/demographics info) are used to predict an unknown value (probability that the credit card has been stolen, probability of the couple defaulting on their loan, and age of death) based on how those factors have related to the value in the past.

Here at Inuvo, the best example of our current use of predictive analytics is the FeedPatrol click fraud detection process, employed in our cost-per-click (CPC) search system.  We collect dozens of characteristics related to an individual click, and using historical data, we determine how each of these characteristics relates to the chance of the click being fraudulent. If the click’s properties indicate that it’s likely fraudulent, then it’s thrown out, and the advertiser doesn’t pay the publisher for that click.

Currently, one of the analytics team’s top priorities is to add predictive analytics to the Inuvo Platform in the form of affiliate quality ratings. Based on input from current and past clients, we’ve selected several different metrics as targets for prediction, and will combine those predictions into an overall quality score, so an advertiser can see at a glance whether they should consider working with an affiliate.  One of the target metrics is chargeback rate (CBR); we’ll be using factors such as an affiliate’s historical CBR, tenure in our system, rate of program expulsion to predict future CBR.  We’re also working to add predictive analytics to the Inuvo Platform in other areas, including transaction-level cost-per-action (CPA) fraud detection, expected earnings-per-click (EPC) projections, and a recommendation engine for advertiser-affiliate relationships.

We’re pursuing these projects not because we like predictive analytics – though we do – but because we think they will add value for the advertisers and affiliates using the Inuvo Platform.  Ultimately, a feature’s value is best judged by its (future) users, so please let us know what you think!  Give us your opinion on our current plans, and tell us what you’d like to see us predict in the future.  Leave a comment here on the blog, submit feedback via our Feedback Forum or email us at analytics [at] inuvo [dot] com.