Predicting Debt

Tell us about your organization. DTE Energy is an energy company that provides electric and gas service to more than 3 million residential, business, and industrial customers in Michigan. In 2006, we had sales of $4.7 billion in electric and $1.9 billion in gas. What problems were you facing? The economic conditions in our market are difficult, to say the least. Michigan has the highest rates of unemployment, bankruptcy, and foreclosure in the nation. Due to the local economic conditions, we're consistently faced with a large percentage of customer accounts going into collection, and we were looking for a better way to understand those portfolios. We needed to identify ways we could work more efficiently with customers, without sacrificing customer satisfaction or raising rates. In the past, we had segmented accounts using factors such as balance, age of debt, and account history. We needed to take a more scientific approach in predicting customer behavior. We were looking for a predictive analytic solution that would give us the ability to segment accounts and predict a customer's behavior and propensity to go into debt or collections. How did you select a vendor? We evaluated a number of products on the market and choose Intelligent Results' Predigy platform. Flexibility was the key: Our selection process was driven by the ability to develop and customize predictive models. Predigy offered us the most flexibility to create models based on our own data, such as the "champion/challenger" method, to statistically compare new strategies for our customers. We began the project with a proof-of-concept in the beginning of 2006, and within six months we had deployed Predigy in most of our operational departments. What have been the main rewards? We've experienced a 700 percent increase in net savings and have reduced our collection efforts by more than 15 percent. In addition, the reporting makes it easier for DTE to understand the distribution of an economic factor within a certain population. For example, we wanted to maintain the amount collected while reducing operating costs, and to document any lessons learned for future use. We analyzed accounts and were able to prioritize efforts based on a charge-off model score. Using the "champion/challenger" strategies for account groups in various stages of collection, we determined that most early-stage accounts provided better results when contacted less frequently: People were more likely to pay if DTE contacted them less. Another example is "service disconnect," or when we have to shut off a customer's gas or electricity due to late payments. Service disconnects are expensive for DTE because we need to dispatch a field collector to the location, and disruptive for customers who often pay their bills and end up having the service reconnected within days, which results in a fee. Using Predigy, we analyzed two areas: whether the customer is likely to be at the location for the disconnect and whether the customer is likely to reconnect within seven days. We found that those customers who are most likely to be there are not likely to reconnect within seven days, and that these accounts are the ones most likely to be written off. As a result, we're in the process of changing our service operations to reflect those findings. That's going to result in reduced costs for DTE, and that means better rates for our customers and improved customer service. What future plans do you have? We're planning on developing more predictive models, which would allow us to predict whether a customer would make a payment or a promise to pay if DTE called. We're also having initial discussions about using Predigy in our power stations, so engineers and maintenance employees can predict equipment failures. When equipment does fail, it can damage other equipment upstream or downstream. Being able to predict likely locations of failure could help us avoid a multitude of obvious problems. 5 Fast Facts

• AGE OF THE INITIATIVE? Two years
• WHO WAS INVOLVED? Myself, all of our operation managers, our collections director, and our executive vice president
• BEST IDEA? Developing the "champion/challenger" predictive model to statistically compare customers
• BIGGEST SURPRISE? The accuracy of the predictive models. They've exceeded our wildest expectations
• BIGGEST CRM MISTAKE MADE? The inability to understand and correctly leverage the data an analytics solution provides

[UPDATE: The print edition of this article and earlier versions of this online version inaccurately characterized Mr. Tuzcu's title; he is risk manager at DTE Energy. CRM magazine regrets the error.]