REAL WORLD EXPERIENCES WITH DATA MINING AND KNOWLEDGE DISCOVERY: PART III
by Jill Dyche & Evan Levy
After three months of using knowledge discovery tools to analyze existing data warehouse data at this company, Baseline was able to identify new and probable marketing opportunities for a variety of customers and products. With only a 1% probability rating, this resulted in an additional $8 million in new yearly revenue. Needless to say, the CIO had more than enough ammunition to move forward with knowledge discovery.
Big Bang at a Big Bank
The same process and algorithms described above apply to every industry. Baseline reproduced its basic process of analyzing data and mining it with a variety of knowledge discovery algorithms, this time at a major bank. Like the phone company, the bank was particularly interested in new product marketing strategies to existing customers, as well as on customer retention issues. We used a cluster analysis tool to process a selected set of customers (800,000) into unique clusters (segments). Once the cluster engine process was complete, the cluster definitions were loaded into a relational database. This allowed us to review clusters (and rules) associated with an individual product.
Right away, we found some startling information.
We reviewed the clusters associated with a specific type of consumer loan product (the Product). Cluster 1 contained 3.5% of the population (about 28,000 account holders) and a high percentage of these customers owned the Product (in fact, about 3 times the average). This particular result wasn't surprising because the bank already knew that the best "target customer" of the Product were likely to own (and use) numerous "sophisticated" financial products. However, the bank hadn't been able to quantify this information in the past, and they hadn't ever been able to identify this type of focused target audience in one week.
The other two examples were shocking. The second example cluster revealed that a large portion of business customers were purchasing a consumer-based debt product. (consumer-based debt products are typically less expensive than comparable business-based products.) This cluster clearly identified a situation where the bank was inadvertently losing profit; they were selling a product that cannibalized the revenue from a higher profit product. The third example cluster identified a very important and costly customer: the customer least likely to purchase the Product. In each example the bank was able to identify a distinct product-focused customer segment and respond with a simple, low-risk marketing strategy. They weren't guessing on business hypothesis, they were focusing on business action.
Other Case Studies
Baseline has been deploying knowledge discovery pilots at large companies for over two years now, using a wide range of different data mining technologies.
The choice of the correct data mining technology is matched to the business via comprehensive interviews and requirements gathering prior to the execution of the proof of concept, and is dependent on three main factors:
The sophistication and breadth of the company's existing data infrastructure
The company's ability to act on business information (are any of their current business processes supported by technology?)
The company's stated areas of interest (for example, customer retention, asset management, or behavior segmentation)
Some other interesting findings we have uncovered using Knowledge Discovery are:
At a major retailer, we developed a new product pricing strategy based upon purchase patterns relating to store location, customer demographics, and time of year.
At another major telecommunications company, we were able to identify customer-oriented call volume growth, and thus dictate where switching equipment upgrades should be installed.
At a credit card company, we were able to deduce risk for certain "prospect bands," and thus reduce the likelihood of fraud or default.
No matter how high the value of the new information, how promising the new revenue opportunities, and how quickly the business wants to act on the newfound knowledge, it must be remembered that data mining and knowledge discovery are not panaceas. Baseline has seen several companies with terabyte-range data warehouses use sophisticated technologies and specialized consultants to perform the work, only to discover later that they cannot use the results. Of critical importance here is that the business processes must be in place to leverage the results of data mining. Without an established infrastructure to deploy the results, even the most advanced mining algorithms are rendered just another failed pilot project.
The Bottom Line
As the Baseline Pyramid illustrates, and our data mining experience plays out, the data warehouse is not just technology, but a process. Data mining and knowledge discovery introduce a host of new capabilities into a customer's business, and can result in nothing short of increased competitive advantage and increased revenue.