FIVE POINTS ABOUT USING DATA MINING TO YOUR COMPETITIVE ADVANTAGE 11.11.97 by Robert Groth D S *
Data mining is a remarkable industry. In what other field will you find big business outsourcing much, if not most, of the very activities it views as too big a competitive advantage to talk about. Using data mining as a competitive advantage within your own organization can be extremely profitable, but, given the odd nature of the industry, there are some caveats to be aware of.
The natural questions to ask when trying to justify data mining within your own organization is: how is data mining is being used, and, what is the return on investment? What shouldn't be too hard to figure out is that if data mining is viewed as an "emerging" market, yet is expected by Meta Group to be a 4.7 billion dollar industry in 1997, then something is clearly being left unsaid.
Why are there so few vocal examples of companies who have gained the enormous return on investment promised by this technology? Without fail, every company I have worked with wants a nondisclosure on their data mining projects. Perhaps the industry silence is not for lack of customers, but rather because too many people view data mining as a competitive advantage. Companies are often willing to talk about their success to others in non-competing industries, but they seldom seem up for much press. Executives can stand to learn much by gathering contacts from people within the industry, but it does involve effort.
2. Preparing Data Isn't Easy
The topic of a recent talk I did at a DCI Data Mining Forum (which was prepared for me) was "The experts agree that 60-80% of the costs of a data mining application are related to data preparation and data cleansing". Data preparation is a huge issue, especially when gathering data at the customer level, where customers are prone to move, marry, age, and commit many other such crimes against the internal order of data management systems.
The bulk of the data mining industry is outsourced largely due to the complexity of data management and preparation. While Meta Group's projections are that the data mining market will be growing from 3.2 billion in 1996 to 4.7 billion in 1997, it turns out that roughly 3.7 billion of this number is earned by integrators, service bureaus, and applications. Seruvice bureaus like Fair Isaacs, Harte-Hankes, Acxiom, and Equifax are companies in business because of the complexity of managing information at the customer level and keeping this data up to date and clean.
Corporations are only now trying to bring data mining in-house as demonstrated by Meta Group's projection that the applied data mining market (no outsourcing involved) will grow from 270 million to 640 million this year. At the heart of the movement to bring data mining in-house is the maturation of the data warehousing market. A great deal of thought is involved in preparing and cleansing data, which is only the first step to bringing data mining in house.
3. Keep a Goal oriented approach
There are data mining companies who will mine everything you give them and take six months to a year to analyze it; however, for companies meaning to take ownership of data mining themselves a simple rule of thumb is invaluable.
Strive for a goal oriented approach to data mining
The enormous amount of work involved in collecting and preparing data can hardly be understated. Established telecommunications players routinely deal with millions of customers, billions of transactional records for customer billing and Terabytes of data. Is it cost effective to mine this data?
The only way to evaluate a return on investment for enterprise-wide data mining is to look at a compelling business case to solve. For example, cellular providers can pay as much as $400 for each new subscriber and have had an average of 30% attrition a year on their customer base. This translates into 12 million a year to maintain 100,000 installed client base! Clearly if data mining can identify those likely to churn and reduce the number, there is a huge cost benefit.
It is interesting to note that some of the most successful data mining companies out there, many are choosing offering goal oriented solutions versus generic tools.
HNC is a prime example. The San Diego based company is a 70 million dollar company with a stock price/earnings ratio of 70. The company's flagship product, Falcon, is aimed at reducing credit card fraud. Almost every card swipped in America is analyzed by HNC's. HNC claims a detection system in place to monitor more than 160 million payment cards this year. HNC is conspicuously absent from many of the generic mining forums, although they were pioneers in neural Networks.
NeuralWare, now a subsidiary of Aspen Technology Inc. as of August 27, 1997 is another vendor concentrating on a solution oriented approach. NeuralWare is strong at providing solutions in process manufacturing. "This is a technology acquistion of enourmous strategic value to AspenTech," said Larry Evans, chairman and CEO of Aspen Technology. "It provides us with state-of-the-art neural network technology that we will incorporate into our solutions for advanced control, process modeling and information management."
DataMind Corporation is yet another example of taking a solution-oriented approach. DataMind announced that it is now delivering a data mining-enabled, marketing automation solution for the telecommunications industry. The new DataMind LifeCycleV suite focuses on telecom providers' customer life cycle planning demands -- helping them to attract, retain and build their business with subscribers. "In working with our telecommunications customers, we identified significant opportunities for these companies to compete more effectively in the crowded telecom industry by using data mining for their customer life cycle planning," said Eric Archambeau, president and CEO of DataMind.
The trend could be enforced by many other recent announcements in this industry. The point is, having specific solutions in mind for data mining is making the industry an easier sell. For executives selling data mining internally, the point is doubly true.
4. Central to Data Mining is the process of building a decision support system
There are many steps to building a decision support system and data mining is just a piece of the puzzle. Below the process, and vendors involved in those processes, are discussed briefly in order to emphasize the elements involved in decision support. Data warehousing is central to a good decision support system because in order to build a warehouse, you need to go through a process of checking data integrity. Without good data, the decisions you will get from data mining are circumspect.
Below are a few of the vendors involved in the steps of building a decision support system:
Cleansing Data
The first step to building a system that allows for good decision support is to cleanse data. This involves, in the case of customer systems, ensuring that names are correct, that addresses are correct and up to date. There are many companies who make a living doing this and outsource their services to large companies. These companies include:
There are many vendors in the space of data wharehouse and data mart implementation. The obvious vendors being:
Often the biggest problem in the process of creating a decision support system is gathering disparate data sources to populate the database. Again there are many vendors who play is this area including:
Query Tools and DSS
Once a data warehouse has been created and populated, there are a number of tools to query the database, including:
Looking at data in a multi-dimensional way is complementary to a data mining process. Multi-dimensional analysis is yet another piece of a decision support system. Some leading vendors in this space include:
Data Mining
Data Mining is the next logical step in building a decision support system. Data mining is actually also very useful in helping discover what is important to put in a decision support system. There are many vendors in this space. Some of the leading vendors include:
5. Make your results actionable
Regardless of what the intention of data mining, if the results of a data mining study can not be acted on, the effort put into data mining may be in vain. Many managers lose site of this point. For example, it is fine that a cellular provider can rank the customers most likely to churn, but what are you going to do about it? Within large corporations the role of analyzing problems and the power to act on the results of that analysis are often divided.
---
For more information, see http://www.datamindcorp.com