Mining Your Own Business
If there is one phenomena the computer industry is no stranger to, it is the concept of "explosive growth" touted by pundits. The concept has been applied to various technologies over the years, including pen computing, universal databases and artificial intelligence -- in other words, technologies that were largely stillborn.
Allow me to introduce a technology poised for explosive growth: data mining. The Meta Group (Stamford, Conn.) says that companies worldwide will spend upwards of $8.4 billion on data mining products and services by the turn of the century.
Another Roman candle that will rise brilliantly and briskly only to fizzle equally as quickly? Not a chance, not this technology, I guarantee it. The reasons are many, but the most important is that data mining is a technology that not only solves critical business problems, but also can act as a new business enabler.
That’s much different from some of the overpromised fizzlers I mentioned above. For the most part, failed technologies in this business are those that started out as technologies in search of a market.
By contrast, the market is absolutely ripe for a data mining (er, pardon the expression) explosion. Why? Corporations are awash in data -- not information, but data. It comes from a variety of sources, including, increasingly, the Internet. It is cheaply stored, and there it often rests, clogging disk drives like inert matter.
Technologies such as OLAP have been employed to analyze this data for trends and patterns as a means of making better predictive business decisions. But OLAP requires that the business analyst pretty much know in advance what patterns and trends he or she is looking for. For example, the analyst may know sales are down in a certain district and thus seeks to analyze data to find out why.
By contrast, data mining tools have the uncanny ability to ferret out trends and patterns in piles of corporate data that otherwise would have gone undetected. Once the analyst sees the relevance of something pop up through data mining applications, he or she can then go back to OLAP and perform more in-depth analysis.
That’s rule number one about data mining. It does not exist as an independent application, and unless you think about it as a potentially major part -- but just a part -- of an overall, integrated decision-support strategy, your data mining efforts will be doomed.
The second rule of data mining is that, as with data warehousing, you’d better build your users a system that can scale well, because the benefits of data mining will become very apparent very quickly to users throughout the organization. In addition, the actual data mining tools themselves are becoming much more user-friendly. They still are years away from becoming a tool for the computing novice. But a widening array of nontechnical users are becoming data miners thanks to the growing ease of use of the tools.
What are companies doing with data mining that is helping the enterprise with real business value? The best generic example was given by those beer and diapers ads that Tandem, a division of Compaq, used to run, featuring a pudgy young father wearing nothing but a diaper. The ad pointed out how, using data mining, retailers discovered that men who go out to buy diapers in the early evening also are inclined to purchase beer. That led convenience stores to begin placing the diapers next to the beer, increasing sales of both.
More to the point, consider Boston-based Fleet Bank. Like just about all big banks, Fleet uses many different tools to tear into customer data in warehouses and data marts. That’s rule number three, which is that the quality of your data mining analyses are only as good as the quality of the data being analyzed. Most of the IT department’s time spent in setting up the data mining application will be dedicated to ensuring that the data to be analyzed is valid in the first place.
Fleet did a lot of cross-selling -- that is, selling ancillary products to existing customers -- believing this to be profitable. But with some data mining analysis, Fleet discovered that is not always the case with all types of customers. Some customers, for example, do not want or need to be persuaded to move savings balances into CDs. So with data mining techniques, Fleet learned to better target customers, on the basis of profiles generated that showed which types of customers would be most amenable to such selling.
There are scores of great data mining stories developing. My advice is to go out and start making your company’s own great data mining success story. Bill Laberis is president of Bill Laberis Associates Inc. (Holliston, Mass.) and former editor-in-chief of Computerworld. Contact him at firstname.lastname@example.org.