What Will Your Customers Buy Next?

For nearly 50 years, Cabela’s has been selling its hunting, fishing, and outdoor gear through mail-order catalogues; it now produces more than 100 different catalogue titles and sends out 130 million copies annually. But the company, which is based in Sydney, Nebraska, has been looking to cut down on mailings to unlikely prospects as paper and postage costs rise.

So over the past couple of years, Cabela’s has been relying heavily on predictive analytics software that draws on the mountain of data the company has generated from shoppers at its global website and its 30 stores across North America. Using tools from SAS Institute, a private company, Cabela’s built a model that ranks customers from those with the best buying history to those with the worst. Next, it adds more than 15 predictive variables, including a customer’s preferred product categories and zip code. Each customer is then assigned a score from 1 to 100—the higher the “star rating,” the greater the revenue projection. The score determines whether and when Cabela’s mails that customer each of its catalogues.

The goal is “to determine how much each customer is going to spend with us over the next 12 months,” says Corey Bergstrom, director of market research and analysis for the $2.6 billion company. Some high-value customers get special perks, such as more sophisticated telephone support. “If you deserve the white-glove type treatment,” Bergstrom adds, “we need to know who you are so that we can go the extra mile for you.”

Since introducing the model, Cabela’s has quadrupled the rate of responses to its catalogue. In other words, people who receive them are now four times as likely to buy something. It’s reached a point, says Bergstrom, where “we can we can predict what a customer is going to purchase next.” For instance, someone who purchased duck decoys might be in the market for a shotgun cleaning kit.

Never before have companies been able to gather such volumes of data about their customers. Yet predictive analytics software is a fairly small part of a $1.4 billion global market for business intelligence software, according to the market research firm IDC. That’s probably because implementing the software effectively requires a commitment from everyone from the CEO down to the customer service people, says Bergstrom: “It’s about building a culture of leveraging one of the biggest assets most companies have—data.”

Predictive analytics works by mining the troves of data captured by companies. First, sophisticated computer algorithms or mathematical equations are used to create a statistical model. Plug in variables such as age, gender, and credit report scores, and these algorithms can automatically calculate potentially predictive relationships—patterns that point to the future. Armed with this information, companies can better anticipate everything from inventory gaps and product returns to sales levels and loss of business—prophesies that can boost a company’s bottom line.

Using analytics software from IBM’s SPSS business unit, the company made a surprising discovery: many of its customers were experiencing billing errors that were prompting them to abandon the carrier as soon as their contracts expired. The model showed that customers were more likely to jump ship if they had received a mistaken account suspension notice. After the error was cleared up, they’d angrily wait out the remainder of their contract and then fail to renew it. Despite complaints to XO call centers about billing errors, the company simply never would have made the connection without predictive analytics, says Payne.

To rectify the situation, XO Communications introduced a “first bill” review, in which a customer support representative goes over the details of its billing practices and procedures with each new customer. According to Payne, the result was a 60 percent reduction in “customer churn,” which more than paid for the $750,000 investment in the new software within the first year of deployment.

“If we didn’t have predictive analytics, we’d probably still be struggling to outsell our churn,” says Payne. “It really has plugged a hole and allowed us to grow.”

Stories like this may actually help explain why so few companies take advantage of predictive analytics, though. Eric King, president of the Modeling Agency, a predictive modeling consultancy, says that not all companies are interested in an honest assessment of how their normal business processes are working. He refers to it as the Jack Nicholson effect—You can’t handle the truth!—and warns that “if we can’t carry out the advice that a predictive model provides, the project will fizzle.”

Predictive models also don’t help much if companies don’t tweak them to reflect changes in market or customer base. Cabela’s recently altered one of its models, reclassifying its potential catalogue recipients as “households” rather than “individuals.” In the past, the company delivered catalogues to only those customers who made more than a minimum purchase each year—a strategy that often alienated other household members who were loyal customers but fell short of the threshold. By combining customers into household rankings, Bergstrom says, the new model has increased response rates by 20 percent.

All this provides good reason to carefully guard confidential customer data: the competitive edge it gives companies like XO and Cabela’s. “Ultimately every organization is going to have to make predictive analytics a part of their standard business intelligence practice,” says King. “It’s going to be too expensive not to.”