Predictive analysis has a long and profitable history with direct mail. Millions of dollars have been saved by focusing on those customers most likely to buy—and not mailing to those who are unlikely to be interested.
Email marketing is a different story. Emails are so cheap to send that the attitude has been "Mail 'em all. It costs peanuts to blast the entire list." Accordingly, predictive-modeling analysts have been shut out of most email-marketing operations. That is beginning to change.
The reason? Inbox clutter. Every corporation in America is now sending emails to the same US online households. Billions are sent daily. Consumers cannot handle it all. Open rates are falling—average rates are less than 15% today. That means 85% of legitimate commercial emails never get opened or read by anybody. Millions of consumers are unsubscribing from emails that they had recently signed up for.
If marketers want to keep their subscribers, they had better make sure that what they send is content that their subscribers are really interested in. Send too many irrelevant emails, and subscribers will unsubscribe. You will have lost them forever. Many marketers are losing more than one million or more subscribers per year. That is a significant loss, considering that a typical email subscriber is worth $20 or more in long-term profits.
How can predictive analytics help? In two ways: First, segment your subscribers, and develop emails designed specifically for each segment; second, use analytics to predict which subscribers are most likely to unsubscribe, and—if those subscribers are valuable—find a way to keep them.
In our talk at the Predictive Analytics World conference last February in San Francisco, Anna Lu—director of research and analytics at e-Dialog—and I explained the email world to a room full of card-carrying predictive modelers. When we began, the room was half full. At the end of the hour, all seats were filled and lots of analysts were standing around the sides of the room. What was the excitement?
We began with a simple example using Next Best Product (NBP) analysis. In this example, a marketer with 2.5 million subscribers selected 273,334 among them who would be most likely to respond to the offer of a particular product. Those subscribers were sent an email featuring that product. To be sure that the NBP modeling had been done correctly, a control group of 20,000 was randomly selected to receive the same mailing. Here is what the example demonstrated: