Predictive Analytics Remains Huge Challenge

What happened to Web sites that personalize the content on a page based on the visitor's browsing history? Not too long ago I heard about this type of technology all the time. Not so much anymore. What happened to analyzing my real-time actions to determine the type of content or ad to serve me? And then there are the ecommerce platforms that think they can count inventory, so shirts, pants and skirts don't get oversold. Frankly, I'm tired of retailers showing me what I need and then getting the "this product is backordered" email from online retailers like Anthropologie after purchasing something on sale.

Predictive algorithms are about matching physical attributes to needs and behavior and finding patterns in data that suggest the way people will behave. It's easy to show me recently viewed items, but what about those items I may need most today or in the near future? Most companies seem to have access to the data, but they don't seem to get the required results. I'm not convinced marketers understand how to parse the data to find the gems.

Personalization and fulfilling orders by predicting behavior still seem to be a huge challenge. The Harvard Business Review provides a great example of predictive algorithms in its most recent post, "Why Websites Still Can't Predict Exactly What You Want." It looks back to 2006, when Netflix offered $1 million to anyone who improved its home-grown algorithm to predict movie ratings. It took three years before a combined team of seven researchers managed to improve predictions by 10%. HBR explains the winning entry contained more than one-hundred algorithms.

There's no shortage of U.S. Patent and Trademark office filing describing predictive algorithms. Facebook describes a method for predictive ad targeting based on social information stored in user profiles within the network.

"Users of the social networking system are mapped to a specific income bracket based on statistical correlations derived from the predictive model," per the U.S. patent filing. "Advertisements are targeted to users based on income bracket. The system may use a machine learning algorithm to analyze conversion rates of targeted advertising to retrain the predictive model."

Microsoft's U.S. patent filing for predictive price targeting based on historical data describes how to use historic pricing to determine future pricing. In some instances, predictive pricing information is used to assist customers when purchasing airline tickets.
Recommend (1)