Filtering Out Dirty Data

Computerworld - It all started with lots of data and very little information. Like many car companies, Mitsubishi Motor Sales of America Inc. in Cypress, Calif., had reams of files containing customer data but no easy way to access and use it.
In 1999, after a decade of flat sales, Mitsubishi was in the midst of a do-or-die marketing effort when management realized just how little the company knew about its customers. And the managers realized why: Mitsubishi had never really focused on customers. Systems were all about sales, service and financing -- there was no repository of customer information.
There was warranty information in the warranty database, service information on dealer systems throughout the country, retail delivery records in the sales database, financing and leasing information, and data from sales and service satisfaction surveys elsewhere. One customer might have owned several Mitsubishi cars during the years, financed them all through Mitsubishi, serviced them faithfully at dealerships and responded to many surveys, but there was no way for the company to know that.
The situation was grim but not unusual. Research firm Gartner Inc. reports that less than 10% of businesses have a single, integrated view of their customers. But that's what Mitsubishi needed in order to market itself effectively.
Although the road Mitsubishi would travel led from data mining and knowledge management to marketing and customer service, its origin was simple. "We were not CRM people or call center people when we got into it," says executive vice president and general manager Greg O'Neill. "We were people with a lot of customers who weren't managing them, and we were spending a lot of money ineffectively."
In 1999, Mitsubishi undertook a massive and painstaking effort to combine customer data from all departments into one database.
First, the company had to determine what data it needed and how to glean that information from the various databases. Warranty files, for example, hold lots of information, but most of it isn't about customers. "From my point of view, a lot of it is just noise, and I had to separate the noise from what I wanted," explains Carlos McEwan, manager of relationship marketing and market research.
Mitsubishi started with a series of data downloads. "It was basically a big dump," recalls Rebecca Caldera, project development manager. "We brought everything in and looked at what we had. We needed core demographics: anybody who had purchased a car, or a primary or co-borrower. We matched things up, and then it was a process of elimination."
But matching