IBM'S MARK DAMICO TALKS ABOUT DM CHALLENGES
by Alan Beck, editor in chief
D S * : What primary advice would you give CIOs who are considering whether or not to invest in data mining technologies?
DAMICO: I think there isn't even cause for an issue or debate. They will have to embrace these technologies to at least achieve parity with their competitors, let alone exceed the marketing ability of their competitors. However, when considering the options available, they'll find that there are basically only two types: The traditional "roll-your-own" version of warehouse where all the different components are developed specifically for them and their own unique requirements, and the newer generation of verticalized warehouse systems that have a number of preintegrated components.
To decide between these, I'd recommend a careful evaluation of overall business objectives. If the objective is to address issues such as churn, customer retention or acquisition, then many of the verticalized warehouse systems offer substantial benefits in the form of lower price, comparatively short implementation time, and overall reduced risks. If, however, the objectives are very broad and diverse then I'd recommend the roll-your-own version that can be built to address the diverse nature of issues facing the company.
D S * : If such diverse issues are pending, what are the requirements for an appropriate system?
DAMICO: For the roll-your-own type, the warehouse components consist of the server platforms, a data model, tools to allow end-users access to information, and a service offering that integrates the whole package into the operational environment. On the personnel side, there is a need for DBA skills, hardware installation, software engineers who are able to use commonly available tools or develop new ones to extract, transform, and load the data, business analysts to identify appropriate access tools, and systems administrators who can maintain the system on an ongoing basis.
D S * : What kind of capital investment is required?
DAMICO: You're looking at a minimum entry price of $1.5 million, and it goes up from there depending upon demands to be placed upon the system. If the system is limited exclusively to one departmental function such as marketing then the entry-level costs will be at the lower end of the spectrum. But if there's an attempt to address issues that span multiple departments then the cost will increase proportionally with the complexity of the system.
D S * : How should results be benchmarked, and what kind of timeframes should be allotted for significant results?
DAMICO: The timeframe for a data mart system addressing a single department's informational needs is about six months. For a more sophisticated system addressing multiple departmental needs and requiring warehousing of large amounts of data, you're looking at 12-18 months.
Value is quantified in terms of return on investment. When purchasing this type of system, there is an assumption that you have specific problems you're trying to solve, such as customer defection, low acquisition rates, various retention issues, etc. The system is justified by evaluating how this information-management system has helped deal with a specific problem.
Case-in-point: Churn is a very significant problem in the telco industry. Depending upon whom you talk to, churn rates among wireless clients can range from 20 to 40 percent. Because of the competitive nature of that segment, a large amount of money is spent on promotional campaigns to attract new clients. It has been estimated that it costs the telephone companies upwards of $400 per person to acquire a new client. So if you can reduce your churn rate, you can maintain your present base of installed customers and minimize that high pay-out just to achieve parity with respect to lost customers. Given such a situation, it is not difficult to judge what kind of return on investment you've obtained.
There's a 1996 IDC report that detailed return-on-investment figures realized by a number of different companies in different industries. Several saw 300-400 percent return with this type of technology.
D S * : Don't such figures represent an extreme?
DAMICO: No! In fact the median figure was 400 percent. Some firms experienced a 1000 percent increase, while others saw only 100 percent.
D S * : As all firms in a given industry adopt these technologies, what key factors will ultimately provide the competitive edge?
DAMICO: First, there must be an efficient and easy way for end-users to get the information. In many cases, companies spend a great deal on just technology: very fast processors and immense data storage. But if you can't leverage the resultant information by providing meaningful knowledge to endusers, it's of no value. So you must have the tools and applications in place that allow end-users to access your information.
Second, what will become increasingly important is not the technology but the underlying information that's stored. Take a look at database systems in general. The technology offers the capability to ask questions and get answers, and if you have a high-performance system, you might get the answer a little faster -- but it's going to be the same answer. But if you are able to actually extend the underlying information -- client profiles, for example -- that will truly improve the answer you get. With more baseline information about clients, markets, usage rates, preferences, etc. you can develop more accurate models. This, in turn, leads to better results.
D S * : Are any revolutionary technological developments on the horizon?
DAMICO: I expect advancement in the basic hardware technologies: primarily processing power. This will facilitate ever larger systems, where more information can be stored and accessed. A few years ago, a one-terabyte warehouse was considered very large. Today, we have many warehouses of 7-8 terabytes. This growth will continue and will be fueled by the technology.
D S * : What additional points would you like to emphasize?
DAMICO: There are a number of different data warehousing technologies available...One of the big problems you have with customized systems is that you don't know what you have until you're finished with it. Prebuilt, preintegrated systems designed to solve specific problems can often prove quite valuable, because they allow you to try before you buy. If necessary, such systems can be further customized to accommodate specfific information requirements.
For more information, see http://www.ibm.com/bi
Alan Beck is editor in chief of D S * and vice president of publications for Tabor Griffin Communications. Comments are always welcome and should be emailed to: alan@tgc.com