May 2000

BY SOREN KIRCHNER, PH.D., AND MIKE BRADWAY, NET PERCEPTIONS, INC.

The ultimate purpose of data mining for direct marketers is to solve business problems. In most cases, the ideal outcome of effective data mining is a newfound ability to target the right customer with the right product at the right time. Traditionally, "targeting" was the task of hitting enough targets with your product offer to move sufficient inventory. The more targets, the more chances for a hit.

The medium of choice for reaching targets was direct mail, due to its low cost in mass quantities. After "launching" the envelopes at the right target groups, direct marketers played the waiting game, hoping for a strong response rate -- anywhere from 1 in 10 to 1 in 20. Data mining's origins may be found in that kind of thinking by direct marketers. Historically, data mining in direct marketing was largely based on temporal "snap shots" of customers and prospects over time. By "launching" envelopes and collateral, we then looked at how well each effort did with our targets, time by time. We looked at factors that contributed to success and used data mining in staggered time to make conclusions as to whom we should "target" to drive up the response rate.

For a call center, where live conversations between operators and potential customers are taking place, the speed with which the operator gains relevant insight into an individual is key. New data mining technologies specially designed for call centers can now make this possible.

"Solutions that offer combined online analysis and offline analytics allow e-marketers to build stronger customer relationships," said Gene Alvarez, program director at META Group Inc. "This combination of analytics allows companies to identify what to market to whom and then provides the means for acting on that information at an individual level."

Many call centers now use some form of data mining, and some of the largest call center outsourcing agencies are attempting to differentiate themselves by offering database marketing and data mining to avoid being commoditized and merely offering "minutes for sale." The key for call centers and outsource providers is the effective leveraging of information about customers to create an experience in which the individual feels special because of the personalized service he or she receives.

Sophisticated Data Mining: An Opportunity For Your Call Center
Today your call center can become a profit center if you choose to make judicious use of data mining and data warehousing techniques.

Donovan F. Gow, Aberdeen Group senior analyst, predicts that, "Call centers that move aggressively to implement personalization technologies will gain a significant competitive advantage over those that do not."

Let's first outline the typical process steps that you will likely take in such an initiative and share some of the techniques that you will likely want to consider.

Bulk information for data mining is usually found in an operational/transactional system or data warehouse, and then extracted into a data mining database or "data mart." The primary advantages of extracting information from a data warehouse are that the underlying data are cleaner and more robust. (This is because warehouses typically are maintained via data integrity processes, with data derived from numerous sources.) These advantages, however, are offset by large resource investments necessary to build and maintain these warehouses. Because many companies cannot afford these warehouses, operational systems continue to survive as primary feeds for data mining.

Once the information is captured, it must be formatted in a way that is conducive to data mining. Unfortunately, most data warehouse schemes are built on various levels of aggregation, whereas data mining applications center on raw detail. It is not uncommon that databases for data mining require their own analytic engine, design and structure. Fortunately, dramatic decreases in hardware and software costs such as disk storage, CPUs and memory have enabled data mining databases to be economically feasible. In addition, improvements in parallel processing have enabled large-scale data mining applications to churn away on these massive data mining databases.

When the information is maintained in a data mining database or data mart, exploration of the data -- the mining part -- usually begins with the use of OLAP (online analytical processing) reporting. This reporting uses a multidimensional framework that enables a marketer to look at historical data in various ways as well as examine and measure business dynamics. OLAP tools typically access and report on business metric calculations including counts, sums and percentages of business metrics such as sales, orders, products and promotions.

The marketer can examine these metrics across different business criteria through a query process and report on these metrics through frequency distributions, cross tabs, columnar reports, etc. This combination of queries and reports enables the marketer to identify trends and monitor expectations.

For example, an OLAP report can compare sales across various regions. Suppose that average sales in your western region were down 20 percent last summer. The multidimensional capability of OLAP allows you to drill down into an area of interest and fine-tune your observation to understand the forces at play. You can take the lid off your hypothetical west region and look at sales by various geographic "slices," such as by state, city or store. By allowing you to explore an area of interest, OLAP empowers you to quickly uncover potential business issues and devise appropriate action strategies. Also, your understanding of the magnitude of each issue uncovered will be expanded, enhancing your ability to quantify/prioritize additional in-depth data mining analyses needs.

While OLAP reporting and analysis helps identify business issues, more in-depth data mining tools are needed to discover patterns and relationships in the underlying data to make valid predictions. A predictive model allows the marketer to select appropriate marketing strategies that address and optimize data-identified business issues.

The most common data mining applications include customer profiling, targeted marketing, market basket analysis and attrition management. Sophisticated data mining techniques can address these applications and optimize customer lifecycle stages (acquisition, retention and reactivation). These techniques include:

• Clustering: segments data into undefined groups.
• Association: identifies relationships within your data.
• Linear regressions: forecasting model based on historical data.
• Time series: forecasting model based on time-based historical data.
• Neural nets: forecasting model based on nonlinear, large-scale historical data.
• Decision trees: prediction model that produces rules-based tree diagram.
• K-nearest neighbors: classification model based on distances between neighboring data.
• Logistic regressions: forecasting model based on predicting binary variables.

Data mining products are packaged as broad-based tools or packaged analytic solutions. The broad-based tools consist of a data mining engine with embedded applications. With these tools, the marketer has flexibility in addressing custom business needs. One of the fastest growing trends in data mining is the presence of packaged analytical applications. These pre-built applications solve specific business needs (i.e., cross-sell, customer segmentation), providing the marketer with an easily deployable solution.

Use of data mining techniques is essential for any database marketer. It is important to note that data mining will not explain why something happened, but it will identify patterns in the data so you can deploy prediction techniques to anticipate future occurrences and optimize current business dynamics.

Real-Time Personalization Applications Within Call Centers
Real-time personalization allows marketers to automatically interact with customers on a personal level at the actual moment of contact. This level of personalization fosters the creation of added value for an individual with every transaction. The marketer can identify the products or services that would be of most value to the customer and instantly offer them, creating a sense of added customer service as opposed to "canned" sales promotion suggestions that are more likely to be perceived as blatant requests for additional sales and do not address the customer's particular wants, needs or interests.

"In a one-to-one relationship, a firm makes loyalty more convenient than disloyalty for a customer, partner or any other stakeholder. It does this by retaining information for the stakeholder and using that information to provide customized services and products," said Bruce Kasanoff of Accelerating 1 to 1.

Advances in personalization techniques now allow businesses to have customized one-to-one levels of interaction with their customers. The open architecture of the more advanced systems allows them to tap into multiple formats or sources of information, such as customer data, product data and transactional information, and work across a company's systems to provide a single integrated view of the customer. For example, if a customer purchases something at a company's store and then contacts the call center, the business is aware that it's the same individual and can integrate both pieces of information and respond to the customer's needs from a more informed position.

Collaborative filtering is one of the newest and most popular forms of personalization. It is based on the premise that taste is a shared preference. If you have a friend who likes the same authors as you and who recommends a book that she really enjoyed, you're likely to like it, too. Collaborative filtering works from an individual customer's indicated preferences, as determined by past purchase behavior and real-time indications, and locates a group of people within a company's database that match those preferences. It then identifies what other product tastes the group has in common and offers them back as additional product recommendations.

One of the strongest advantages of collaborative filtering is its ability to respond immediately to customers' unique, real-time preferences. It also continually learns and adapts, since preferences change with time. Each new piece of data that is collected slightly affects which other customers' tastes are the best match; therefore, recommendations change accordingly.

Another advantage of collaborative filtering is also one of its challenges. A collaborative filtering application can reliably pinpoint what a customer wants to buy; however, that item might not be something the retailer particularly wants to sell at the moment. Therefore, it is important to include an outbound application with inbound applications so the "personalization-centric" enterprise can fully serve its own needs in addition to those of the customer.

Rules-Based Applications Of Data Mining In Call Centers
What is rules-based customer relationship management (CRM)?
It's quite simple: Rules should be the "what do do" expressions of the output from data mining. Since you are mining data to solve a business problem (e.g., what product or service to offer a customer and when and how), the output is a rule to operate with and act on that derived truth.

The output of data mining should imply workable action items for selling and customer care. If, for example, a particular demographic grouping tends to purchase a particular product, and your data mining singles them out as likely candidates for upselling or outbound calls, rules help.

Who is the rule-maker or rule-keeper?
Rules can be dangerous, especially if they don't square well with personalization and what the customer will actually prefer, want and buy. So be careful with rules because if not judiciously applied, rules can wreck your CRM.

Your call center can operate using hundreds and thousands of rules, and each rule must be updated and maintained to assure that it is correct and makes business sense. Rules offer the marketer the ability to interject his or her own will into the data mining process. Sometimes an overstocked item needs to be moved at all costs and the ability to manually intervene and make a rule that supersedes the output of data mining is necessary in the minds of some marketers. We see this as an essential undoing of data mining that exacts what customers actually want and will buy.

Rules require stewardship and maintenance. They can be an expression of customer desire, but too often they are an expression of political backbiting that sometimes occurs in some companies to vie for control over the marketing process. Using rules outside of customer desire and need leads to problems that can later diminish customer loyalty and the effectiveness of CRM in your call center.

New Directions For Rules-Based CRM Driven By Data Mining
Data mining, data warehousing and customer relationship management seem to be branching into three new directions, as discussed below.

Data mining CRM will be real-time.
As stated above, CRM and the supporting data mining must be real and be in real-time. The call center and e-commerce both require real-time CRM. Since the movement in CRM is clearly real-time, the supporting data mining must be in real-time or must be very adaptable to real-time CRM.

Personalization is and shall remain key!
Bruce Kasanoff, CEO of Accelerat-ing1to1, observed that, "Companies are running hard to enable firms to identify at an individual level what to market to whom and then provide the means for acting on that information. The pot of gold at the end of the rainbow is a closed feedback loop that is both responsive to individual customers and more profitable for the enterprise."

Enterprisewide interactive capability will be critical.
Your customers interact with you typically at many touch points, such as in your store, your call center and on your Web site. Whether your customer touch point is walk (kiosk and POS), click (e-commerce) or talk (call center), your CRM strategy must be personalized and seamless to your customer. Imagine customers receiving e-mail based on their call-in activity and then beginning to come to your site and buy there, saving you overhead! Using an enterprise-based CRM strategy can indeed result in migrating your customer from your call center to your site or vice versa. You can also increase your store traffic as a result of call center and e-commerce activity. The strategy is up to you and your company's needs.

If you cannot make recommendations for products and services to your customers across the enterprise (all touch points) and in real-time, you may as well close up shop. This is why it is important to function as an empowered whole -- any nonenterprisewide solution is amelioration at best and foolhardy at worst.

Soren Kirchner, Ph.D., is the call center product manager and Mike Bradway is the analytics program manager at Net Perceptions, a provider of data mining and real-time personalization to call centers.