"Data mining" is not yet within the grasp of most retailers, but those building data warehouses today are already reaching for it in hopes of taking decision support to higher levels of discovery.
By querying their data warehouses -- massive repositories of information previously stored on multiple, disparate systems -- retailers have extracted fact-based answers to simple questions such as "What were my sales in region A this week, and what is the increase or decrease compared to last week's sales?"
Data mining, by contrast, yields answers to questions no one thought to ask. The process is described as one of "discovery" because it uncovers unforeseen trends and relationships buried within the raw data.
In other words, "We'll know things we never knew we never knew," said Terry Ransom, chief information officer, Harry's Farmers Market, Roswell, Ga.
Among the key areas where data mining can produce new knowledge:
Segmenting the customer base according to demographics, buying patterns, geographics and other variables -- and mining that database can produce surprising new insights about customer preferences.
Tailoring localized product assortments tailored to each store become more possible as retailers understand the specific market dynamics, rather than rely on long-held assumptions about rural, urban, lower- and upper-income neighborhoods.
Sharpened buying practices can be developed as retailers replenish only what's needed, thereby reducing inventory levels.
Activity-based costing projects can reveal otherwise-hidden cost drivers and reverse long-held theories as retailers understand operational expenses on a more granular level.
"Data mining will get us to the point where we're going to find a lot of information and trends we never thought of," said Jerry Johnson, chief information officer and vice president of management information systems, Abco Foods, Phoenix.
He said data mining will be particularly beneficial for frequent-shopper programs and also can, for example, reveal a previously unknown relationship between the sales of diapers and a specific breakfast cereal.
Armed with the new knowledge of such product affinities, retailers might adjust shelf sets to position the two products together and maximize sales. Or, knowing that one item is bought in tandem with the other with high frequency, the retailer may choose to put one of the products -- not both -- on sale, he said.
Because this type of "free-form" data analysis is not guided by human instinct, it is therefore not limited by it, retailers said.
The most essential prerequisite for a data-mining project is a reliable data warehouse or data marts, which are subject-specific databases whose information may be replicated elsewhere.
Few retailers are currently engaged in full-scale data-warehousing projects today, but their numbers are growing. In SN's 1997 Technology State of the Industry Survey, for example, 12.7% of respondents ranked data warehousing as a top priority this year, up from 6.4% the year before.
Among those retailers poised to begin data mining is Hannaford Bros., Scarborough, Maine.
"It's one of the next things we'll be looking at as we complete our data-warehousing project," said Bill Homa, vice president and chief information officer.
Initially, Hannaford will employ data mining to conduct market basket analysis on a deeper level, but that is just the beginning, he said.
"Eventually it could be tied into demand-side initiatives in terms of computer-aided forecasting, ordering and replenishment," Homa added.
For Ron Glickman, chief information officer, Bristol Farms, Rolling Hills Estates, Calif., data mining represents a new opportunity to transform static historical data into dynamic and actionable information.
"When people talk about 'empowerment,' I think data mining is a great example of what that means -- in practice," he said.
He described how, after a store's grand opening in Long Beach, Calif., Bristol Farms' operations executives wanted to know the geographic breakdown of customers to determine the effectiveness of a special mailing.
Shoppers were asked to provide their residential ZIP code, and that information was recorded along with their transaction at the point of sale. Bristol Farms compared this information to the mailing list and learned that the mailing, for the most part, was effective --for the most part.
Sam Masterson, director of operations development and administration, performed the data-mining process and shared the findings with SN.
"We mailed our grand-opening piece to those areas surrounding the store location that we believed matched our demographic profile. However, after we did some data mining based on customer ZIP codes, we determined that we [inadvertently] avoided some areas that we should have mailed," he said.
The retailer then made corrections for the next mailing and adjusted advertising accordingly, Masterson added.
"That, to me, is data mining," said Glickman, "where you put information together so you can make a decision and do critical analysis. It's turning data into information."
At Harry's Farmers Market, which is renowned for its vast offering of perishables, the biggest opportunities to transform data into knowledge are in predicting product movement and sharpening buying activities.
"To do that effectively, we must be able to track sales by product, correlating that information with customer preferences, weather data, seasonal-buying variations, product availability and price points," said Ransom.
Sales and product movement data are already available for mining, he added, and the retailer is currently revising manufacturing systems to make accumulated data accessible, too.
"That is a critical step for us in embracing the future requirements in providing fresh-prepared food products to Progressive Food Concepts as well as Harry's in a Hurry stores," Ransom said.
Progressive Food Concepts, Golden Colo., is the affiliate of Boston Chicken that has partnered with Harry's to refine its Harry's in a Hurry prepared-meals store format.
Ransom and other retailers told SN that data-mining initiatives must be driven by sound retailing sensibility if they are to succeed.
"I think it is critically important to understand the process of food retailing before attempting a data-mining project, including the mechanics of product movement, how register data are captured and how buying orders are developed," he said.
For many retailers, the monumental task of collecting historical data from legacy systems, and cleansing it before loading it into a data warehouse or into data marts, is too intimidating a project when not guided by a clear business vision.
"Personally, I think data warehousing projects can be defined too broadly -- and that's what keeps most people from taking the first step," said Bristol Farm's Glickman.
Challenges related to data integrity appear to be a major obstacle for retailers who have long collected, but not managed properly, point-of-sale data.
"In the grocery industry, the competition is so keen that you [get so intent on building] better information systems and better POS systems that you don't always take care of the infrastructures to make something like data warehousing and data mining happen," said one East Coast retailer who requested anonymity.
"It's not good enough to be good anymore," added Terry Morgan, senior practitioner, Garr's Food Retail Technology Practice, a division of Deloitte & Touche Consulting Group, Marietta, Ga. Morgan, the former director of information technology at Food Lion, Salisbury, N.C., was heavily involved in Food Lion's data-warehousing project before he joined Garr earlier this year.
"The only way to get better is to use new information about sales, the marketplace and information about customers," he said.
In addition, using data mining to analyze financial information can bring ABC projects to new levels.
"So much of what people understand with financial information is an accountant's view of the world, not a good manager's view of the world," he said.
Through data mining, financial analysis can not only cut across multiple business areas, but also break free of the pre-established cost models that may be hindering ABC projects.
"Unless people have a good ability to analyze the data, which is what data mining provides, ABC can bring in its own inherent fallacies and misrepresentations because it is based on what people think is going on, and not necessarily what really is going on."