JAN 11, 2006 6:37pm ET

Web Seminars

Why Data Virtualization Can Save the Data Warehouse
September 17, 2014
Essential Guide to Using Data Virtualization for Big Data Analytics
September 24, 2014

Perfect Pitch

Print
Reprints
Email

The famous British analytic philosopher Bertrand Russell once declared, "The trouble with the world is that the stupid are cocksure and the intelligent are full of doubt." Russell, a Nobel laureate and social commentator who lived nearly a century until his death in 1970, believed mathematics - as surely as poetry - could define greatness in man, giving us better insight to ourselves and our surroundings.

Much of what Bertrand Russell believed is being tested today in the business fields of sales and marketing, where numbers define winning and losing in enticing measures like sales, growth and market share. The metrics are simultaneously real and abstract - old and new, deep and superficial, believable and not, understood and misread. This makes the application of predictive analytics to business problems more than mathematical modeling of future behavior. In practice it is science and art - rigor and exactitude combined with an open mind and industry sense. Change takes place over time, though certain minds are inclined to such pursuits.

"I started out as an economist trying to understand how data behaves," says Seymour Douglas, director of CRM and database marketing at Cox Communications. "That led me to start learning how consumers behave." As the price of computing has fallen and companies have become inundated with data, a good part of Douglas's mandate today is to make productive information out of that data. Douglas joined Cox Communications in 2002 to develop capabilities in database marketing. That is when his superiors at the Fortune-500 company became industry innovators and laid down a straightforward objective in a simple phrase, underlined many times since: "Get the right product in front of the right customer at the right time."

Models and Findings

The challenge was thrown down as Cox was diversifying geographically and into new products for broadband DSL and voice communications, while facing new satellite competition in its traditional cable TV niche. "It was no longer a market in which a good customer was someone who took HBO in addition to digital cable," says Douglas. "We wanted to get some understanding around it as fast as we could." Today, Cox supports more than six million broadband customers alone. In 2002 however there was no analytics-centered database to analyze to determine how customers were behaving and which campaigns were successful. "There was little or no skill set to support what we wanted to do and in fact, the tools to help us didn't even exist," says Douglas. As the company set strategy and built a marketing database to support a wider view of the customer. (See sidebar: Predictive Marketing Building Blocks.) Douglas was building models to help identify customer lifetime value as well as propensity to buy and propensity to churn.

Traditional database marketing campaigns might offer a certain product to people with annual income above a certain dollar amount. Such campaigns do not attempt to address multiple variables that include income, recent activity, lifestyle or payment history. At Cox, the buy/churn dichotomy became a central part of the analytic strategy that would translate into selling strategy. For example, Douglas found that for products like high-speed broadband and VoIP (Voice over Internet Protocol, used for Internet telephone service), lifestyle is a stronger indicator than income when it comes to making purchases. "Part of the vision is understanding that churn is a cost of doing business," he says. "When I sell you VoIP I have to install a $500 network interface on the side of your house so you can use the service. If the churn is 6 percent, it increases my cost to recondition those boxes, and increases the risk that I might lose some of those boxes." So, Cox rolled out a product recommendation engine that ranks each customer by their likelihood to buy a product, discounted by their likelihood to cancel the service at some point in the future. "Now you can start asking yourself if this is the best product to offer to a given customer," says Douglas. Though this exercise is straightforward, it's the same thinking that is applied more complex predictive models.

In terms of customer lifetime value, Douglas found that the "triple-threat" customer who takes voice, data and digital products is about six times as valuable as the single-product buyer, based on revenue and the fact that the number of products purchased reduces the likelihood of churn. "We have found that by increasing the value of the customer we are seeing that our long term cost of serving the customer has gone down as our product portfolio has gotten more complex."

Analysts at Work

For the third time in his career, Douglas had applied a tool from a vendor he was familiar with, KXEN and its Analytic Framework, a suite of predictive and descriptive modeling engines that in this case sits on an Oracle database pointed to SAS data tables. Technicalities aside, Douglas believes the ease of using intuitive click-through menus lets analysts focus on creating models without extensive data preparation. "It's a tool but its complexity is hidden," he says. "My customer-facing employee really doesn't really need to understand how we came up with a recommendation around products. The critical thing is the rate at which customers accept the offer that we make, and whether we are making the offer in the most cost-effective way."

This model has allowed Douglas to apply a tiered approach, with four senior analysts reporting directly to him, and another analyst in each of Cox's 26 regional markets. "Not all of my markets need someone at the Ph.D. level. We can recruit at the regular analyst level and have them self-sufficient in creating and supporting analytic models," the most successful of which are returned to the senior analysts for further development. This minimizes cost at the corporate level while allowing customer-facing analysts to support and generate models based on their own observations.

Get access to this article and thousands more...

All Information Management articles are archived after 7 days. REGISTER NOW for unlimited access to all recently archived articles, as well as thousands of searchable stories. Registered Members also gain access to:

  • Full access to information-management.com including all searchable archived content
  • Exclusive E-Newsletters delivering the latest headlines to your inbox
  • Access to White Papers, Web Seminars, and Blog Discussions
  • Discounts to upcoming conferences & events
  • Uninterrupted access to all sponsored content, and MORE!

Already Registered?

Advertisement

Twitter
Facebook
LinkedIn
Login  |  My Account  |  White Papers  |  Web Seminars  |  Events |  Newsletters |  eBooks
FOLLOW US
Please note you must now log in with your email address and password.