The Information Store

CRM is thrusting data mining back to the future.

In the old days, neighborhood grocers knew their customers. When a family experienced a birth or a death, a merchant might have acknowledged the event with a gift. As income levels and lifestyles in the locality changed, the grocer identified emerging trends and made appropriate inventory adjustments.

That's CRM (customer-relationship management) from the seat of your pants," said Mark Brown, SAS Institute director of data-mining strategy.

Now, many wireless-service providers have millions of customers to become acquainted with, thousands of employees handling customer accounts and a flood of account information flowing into separate departments through various channels such as phones, e-mail and Website forms.

As a result, today's enterprise versions of CRM are being catapulted to the seat of a cockpit, along with Web-based data-mining applications, which have replaced the shop keepers' eyes and ears.

Data mining can impart knowledge about which customers are profitable, which aren't and which ones to target with specific marketing campaigns, Brown said. Unfortunately, extracting customer information from legacy systems, transferring it to a standardized database format and broadcasting it across departments in a timely manner can be challenging.

Technical Difficulties
The first step for companies preparing to integrate HNC Communications' data-mining software is data extraction. During this process, customer information from an existing database or databases must be transmitted to the software vendor.

Though HNC can work with any format, the company's implementation crew requires service providers to submit a detailed data dictionary, which describes each field of information and lists values associated with the fields, said Mike Balon, HNC director of implementation services. For example, a field entitled "income level" might contain a value of one to indicate an income of $10,000 to $25,000 a year; other values might indicate higher or lower annual earnings.

Service providers also must clean or uniformly format the data. Each record should contain the same number of fields, and field values must be defined.

The most common problem Balon has seen is missing tag data, information transmitted to HNC separately from active account records. Tag data identifies trouble accounts, those closed because of fraud or non-payment, for example.

The HNC team explores the tag information for common patterns. Any common patterns that are discovered are programmed into the neural network to enable the future detection of similar patterns in active accounts.

But often companies have no record of what went wrong with the problem accounts, Balon said. The customer service, billing or fraud staffers who serviced the accounts may not have created logs of interactions with the customer.

"(The companies) may have the data and not know they have it, or they may not have it at all," Balon said.

The complexity of a systems integration varies from provider to provider, Balon concluded.

"Some carriers have all of the information, are highly motivated and have good technical resources to put on these projects," he said. "Others haven't really thought about what they need, or they've been focusing on customer acquisition rather than focusing on what the (customer) port folio they've acquired is doing. When they start to address that, they realize they have not collected the necessary data to analyze and maximize that portfolio for value or churn."

Moving Forward
Rob Epstein, BellSouth database marketing consultant, admits that preparing to analyze customer data is hardwork.

"The hard part is getting the data, getting it cleaned and getting it ready to run through your office," he said. "ButIen courage people who are interested in it to take the effort, because it's very gratifying once you've done it."

BellSouth began using SAS's Enterprise Miner software suite about a year ago.

"The product will run statistical models and determine which one gives you the most predictive results," Epstein said.

He conjures the example of an executive who wants to build a direct-marketing program to target people who like voice mail. The predictive software conducts what's known as a regression analysis on the company's existing voice-mail customers to isolate distinguishing characteristics and identify the fields within the data base that best describe the typical voice-mail customer.

Using these findings, the executive can search the database for people who have the characteristics of a voice-mail customer but don't yet subscribe to the service.

"So now when you do a direct-marketing campaign, you're doing it based on knowledge and science rather than intuition, which is how we had to operate four or five years ago," Epstein said.

BellSouth also uses the software applications in its retention programs.

"You can use these models to predict churn or you can build lifetime-value models to look at your customers based on their total value instead of revenue," Epstein noted. "We have a limited number of dollars we can spend on retention. To be able to identify the best customers to spend those on is great."

Identifying its high-value customers enables BellSouth to offer incentives tailored to them.

"An example would be priority call forwarding," Epstein said. "When somebody calls customer service, the customers that are our highest value customers can get routed more quickly and to our most skilled reps.

The company's marketing department transmits some of the customer-value information to CSRs' account records. Marketing also receives information from the customer-service department, such as the number of calls received. All of the information lands in the predictive models.

"I see us reengineering the business based on knowledge and using that science," Epstein boasted. "Lifetime value is a great example. Before we had lifetime value, we had to assume that our highest-revenue customers were our most profitable customers. So, in the past, if somebody called customer service 20 times and somebody didn't call at all, we wouldn't be able to account for that. Now we can."

Refining Customer Evaluations
Denise Nieves also has seen changes in the way her company learns about its customers' needs. Alltel's vice president of marketing communications said the company takes a more microscopic approach to analyzing customers than it once did. Taking the new approach has resulted in more focused marketing campaigns, she said.

"Marketing programs used to be done in bulks of 200,000 and 400,000 people at a time, with the same offer to everyone," Nieves said. "But it's not unusual for us now, with a customer base of nine million, to do campaigns that are targeted to 24,000 people. We have gotten much more granular in our approach to the customer by using data mining."

Alltel achieves this granular view of its customers by combining third-party market research information with internal account information. According to Nieves, this approach allows the company to sculpt its correspondence to particular customer preferences.

"For instance, if somebody falls into a high-tech category, they're considered a techy," Nieves said. "Historically, we would have just sent them a direct mail like we do everyone else in this category. Now, we send those that are considered high tech an e-mail. Then we might take the next segment, which is not considered high tech but we know they're computer users, and send them a letter with a URL address so that they can go to the Web site if they feel like being interactive with us."

Because making sense of the data has become a priority, Alltel uses many types of data-mining software, including regression modeling and neural net systems. The company uses Business Objects' software as its primary front-end mining tool.

The company's use of customer data transcends marketing. Alltel's pricing group analyzes call volume and call-detail data to determine the peak calling hours and the typical duration of calls. This information also helps the company forecast future network requirements.

Alltel's financial groups, accountants, business-planning groups and customer-service department scan the data for negative and positive trends. The idea is to detect and resolve problems quickly and to identify and promote strengths.

"We all have access to the data," Nieves said. "It has been posted on an internal Internet site, and basically anyone has access to running the data in any way they can. We have also put out many canned reports so that people can watch trends."

Gaining the ability to view and analyze customer information on the Internet has accelerated Alltel's information gathering process significantly, Nieves said.

"A report that took four people and two weeks to produce five years ago can now be done in a matter of seconds at the lowest-level user," she said. "As competitive as the marketplace is, we don't have time to wait even a week or two for information that used to take months."

Gathering information that can help with customer retention is the goal of most of Alltel's data-mining efforts.

"Seventy-five percent of our efforts day in and day out is doing nothing but looking at customer satisfaction, watching to see if a negative trend appears and trying to cut it off before (customers churn)," Nieves said. "Before, we'd wake up six months later and find out that we were losing customers off a rate plan. Now, we really get very pro-active."

Nieves predicts the next five years, like the last five, will bring many changes in how Alltel learns about its customers. She foresees company executives categorizing customers into micro subsegments based on criteria such as the products customers have purchased, where they live and their hobbies. That information would be accessible by CSRs, which would allow customer service to become quite personal.

Data Mining 2000 & Beyond
Like companies' information-gathering methods, data-mining technology also is evolving.

"What we're seeing is the evolution of what we're referring to as collaborative data mining, said Joe Bergera, Micro strategy director of industry solutions marketing. Collaborative data mining enables service providers to go beyond simply identifying patterns in data to discovering the causes of those patterns, Bergera said.

"A lot of the data-mining technology is based on artificial intelligence. So you're basically cranking through large data sets and trying to identify patterns," he said, launching into a hypothetical scenario.

"What you'll get back is an answer that there seems to be a relationship between people who are between the ages of 20 to 25 and the purchase of a particular product," he said. "But unfortunately data-mining technology doesn't allow you to understand why those people are buying that product. That could be unfortunate if you were to launch a marketing campaign based on the assumption that that demographic group is inclined to buy the product when, in fact, some other variable was the underlying driver."

This is where decision-support software comes in. Decision-support technology makes it possible to query data using various hypotheses and ultimately, to get an answer about the causes of a trend, Bergera said. He foresees data-mining software becoming more decision-support oriented.

Bergera also has seen the size of customers' data files balloon as businesses adapt a customer-centric approach. In an effort to gain a deeper understanding of customers, service providers have added many fields to customer files to monitor a plethora of customer attributes.

BellSouth's database, for example, contains hundreds of fields for each of its more than five million customers, according to Epstein.

"We made our database more robust by adding demographic data, as well as billing-system data," he said.

The expansion of corporate databases and the proliferation of information entering companies through the Internet have created a need for data-mining software that can handle large data volumes.

"A lot of telcos are now able to record all of their customer interactions and track the history of their customer relationships," Bergera said. "Because of that, they are able to export information from all the conversations they have had with their customers into single customer-data warehouses, which five years ago you weren't able to do because everyone had all of these different legacy systems."

Bergera foresees more changes in the future of data mining, driven by the advent of the ability to track the location of wireless devices.

"Wireless providers are going to have a whole new set of information available to them," he said. "They're going to be able to track the transactions that I make, the interactions that I have with them and, potentially, with third parties over their networks."

But there's a downside - the growing public concern about privacy. However, Bergera remains hopeful.

"I think (service providers) are going to be able to come up with extremely interesting pricing models and very interesting product bundles," he said. "You'll really see people differentiating themselves on the ability to figure out how to leverage this information set."

---

Data Warehousing Gotchas

1. You are going to spend a lot of time extracting, cleaning and loading data. The usual figure quoted is that 80% of the time to build a data warehouse will be spent on this type of work. The amount of time to be spent on these tasks is often grossly underestimated.

2. You are going to find problems with systems feeding the data warehouse. Problems that have gone undetected for years will pop up. You will have to decide whether to fix the problem in what you thought was the read-only data warehouse or fix the transaction-processing system.

3. You will need to store data not being captured by any existing system. In this case, the data-warehouse developer faces the possibility of modifying the transaction-processing system or building a system dedicated to capturing the missing information.

4. Your warehouse users will develop conflicting business rules. Many warehouse tools allow users to perform calculations. The tools will allow users to perform the same calculation differently.

5. You are going to have a problem with security, especially if you make your data warehouse Web-accessible. Restricting people to information on a need-to-know basis does not cut it in the organization of the 2000s. But exposing the information to theft from anyplace on the globe is not too great either.

6. You are building a high-maintenance system. Reorganizations, product introductions, new pricing schemes, new customers, etc., are going to affect the warehouse.

Source: LGI Systems