FEATURE: Mining with Meijer
Maintaining a local product mix is not easy for a large chain, especially one that has grown as fast at Meijer. Over the past 10 years, the Grand Rapids, Mich.-based chain has expanded from 76 stores to the 156 locations it has today, with average SKUs numbering a quarter million—even more during holidays and seasonal periods.
The growth came at a price, however. "Many of the processes we used in our merchandising area were disparate between departments," says information technology and services manager Marlene Lieb, who presented at the NCR Teradata Partners Conference in October. "Foods had one way of doing things, merchandisers and buyers had another, fashion had their own processes, and hardliners again had a whole different set of tools. It was just an IT support nightmare, there was absolutely no consistency."
Lieb realized that single assortment products did not always sell the same in different stores. And the size of the chain made it difficult for them to evaluate the problem using their current systems. "We still wanted to have that local market feel, but to do that, we had to be able to take a look at our sales data by local assortment," says Lieb. "Out of all the tools we had, there was nothing that could do this."
With Wal-Mart Supercenters beginning to challenge Meijer in its home turf, Lieb needed a data mining solution. Fast.
After a rigorous evaluation process (see sidebar), the grocer selected the NCR Teradata Data Mining solution to help perform global market optimizations and develop local market assortments across the entire chain.
Art vs. science
There are two tools that Meijer analysts use in creating local market assortments, according to business analyst manager Lisa Rider. One is the Teradata data mining solution. Basically, what the data mining tool does is sort through huge amounts of detailed data stored in the data warehouse, using pattern recognition technologies and mathematical recognition techniques to discover hidden data patterns that may not be obvious to humans.
The other tool, she says, "is the one between our ears. We use our brains to interpret the data from the data mining solution and transform it into actionable business decisions."
This balance of "art versus science" is a crucial one to maintain when dealing with local market assortments. "Science can tell us about a lot of things," says Lieb. "It can tell us about our own data. It can tell us what is selling, and where. However, it can't tell you what new products are coming into the marketplace. It can't tell you what your competition is doing, or what they are selling. That is where our merchandisers and buyers come in. We wanted to be able to apply the art to the science using fact-based decision making."
To prioritize the products the analysts would focus on to develop the local market clusters, Rider created a scorecard based on two criteria. The category level of importance, which is determined by sales or margin contribution to the corporation, and the second—which came out of the Teradata Warehouse Miner—is the variability that is seen in sales patterns within that particular category.
What Rider found was that there were certain core items that needed to be in all the stores. These were products that sold similarly across local markets, and without much variability. "These core items have significant sales across all of the stores; these items are what the customers are going to be looking for," says Rider. "But when we looked at other products, we began to see the stores separate, and discovered different patterns in the data."
Case study:
water softener salts
The results of the local market clustering on Meijer's water-softening salts showed that, while there was a high-purchase rate of the branded product in one group of stores, in other store clusters, the in-house Meijer brand sold extremely well. "There was a distinct cluster of stores where consumers preferred our private-label brand, and another group of stores where the customers preferred a particular type of packaging —blocks of salt in large bags."
Armed with this data, Rider set out to determine the impact these local market assortments would have on sales. The planograms for the various clusters of stores were arranged to include the core set of items that sold well across all stores, and a space for the local assortments, which varied from cluster to cluster.
To gauge the performance between the different groups—and prove its ROI—she set up testing controls. In the control stores, planograms were not changed at all, while in the test stores, she implemented the local market assortments.
The test was a success. "We had a huge payback on this," says Rider. "Not only did we see sales increase and margins increase, but we saw reductions in inventory as well, because we weren't carrying the kinds of products that had low turns."
One of the lessons that came out of the initial local-market assortment tests was that it was crucial for the business side and technology side to work hand-in-hand in the process. "We can't just start out with the science component, run the cluster analysis and come up with the solution to hand off to the merchandisers," says Rider. "Instead, we meet with the business first."
Rider begins with a category that shows variability in its sales across stores, and meets with someone on the business side who handles the category, to determine if it has potential for a local market assortment. "They tell us what is happening in that particular category, what new products are coming out, what they are finding from our external sources, such as ACNeilsen," she says. "Once we take all these factors into consideration, then we decide whether to move forward or not."
The implementation required that Meijer change its organization, according to Rider. "We need to become much more specialized than we had been in the past," she says. "The buyer is a buyer now, not also an analyst. We have a process in place that we know will lead to success. And we have the technology and tools in place to bring this process through."
The growth came at a price, however. "Many of the processes we used in our merchandising area were disparate between departments," says information technology and services manager Marlene Lieb, who presented at the NCR Teradata Partners Conference in October. "Foods had one way of doing things, merchandisers and buyers had another, fashion had their own processes, and hardliners again had a whole different set of tools. It was just an IT support nightmare, there was absolutely no consistency."
Lieb realized that single assortment products did not always sell the same in different stores. And the size of the chain made it difficult for them to evaluate the problem using their current systems. "We still wanted to have that local market feel, but to do that, we had to be able to take a look at our sales data by local assortment," says Lieb. "Out of all the tools we had, there was nothing that could do this."
With Wal-Mart Supercenters beginning to challenge Meijer in its home turf, Lieb needed a data mining solution. Fast.
After a rigorous evaluation process (see sidebar), the grocer selected the NCR Teradata Data Mining solution to help perform global market optimizations and develop local market assortments across the entire chain.
Art vs. science
There are two tools that Meijer analysts use in creating local market assortments, according to business analyst manager Lisa Rider. One is the Teradata data mining solution. Basically, what the data mining tool does is sort through huge amounts of detailed data stored in the data warehouse, using pattern recognition technologies and mathematical recognition techniques to discover hidden data patterns that may not be obvious to humans.
The other tool, she says, "is the one between our ears. We use our brains to interpret the data from the data mining solution and transform it into actionable business decisions."
This balance of "art versus science" is a crucial one to maintain when dealing with local market assortments. "Science can tell us about a lot of things," says Lieb. "It can tell us about our own data. It can tell us what is selling, and where. However, it can't tell you what new products are coming into the marketplace. It can't tell you what your competition is doing, or what they are selling. That is where our merchandisers and buyers come in. We wanted to be able to apply the art to the science using fact-based decision making."
To prioritize the products the analysts would focus on to develop the local market clusters, Rider created a scorecard based on two criteria. The category level of importance, which is determined by sales or margin contribution to the corporation, and the second—which came out of the Teradata Warehouse Miner—is the variability that is seen in sales patterns within that particular category.
What Rider found was that there were certain core items that needed to be in all the stores. These were products that sold similarly across local markets, and without much variability. "These core items have significant sales across all of the stores; these items are what the customers are going to be looking for," says Rider. "But when we looked at other products, we began to see the stores separate, and discovered different patterns in the data."
Case study:
water softener salts
The results of the local market clustering on Meijer's water-softening salts showed that, while there was a high-purchase rate of the branded product in one group of stores, in other store clusters, the in-house Meijer brand sold extremely well. "There was a distinct cluster of stores where consumers preferred our private-label brand, and another group of stores where the customers preferred a particular type of packaging —blocks of salt in large bags."
Armed with this data, Rider set out to determine the impact these local market assortments would have on sales. The planograms for the various clusters of stores were arranged to include the core set of items that sold well across all stores, and a space for the local assortments, which varied from cluster to cluster.
To gauge the performance between the different groups—and prove its ROI—she set up testing controls. In the control stores, planograms were not changed at all, while in the test stores, she implemented the local market assortments.
The test was a success. "We had a huge payback on this," says Rider. "Not only did we see sales increase and margins increase, but we saw reductions in inventory as well, because we weren't carrying the kinds of products that had low turns."
One of the lessons that came out of the initial local-market assortment tests was that it was crucial for the business side and technology side to work hand-in-hand in the process. "We can't just start out with the science component, run the cluster analysis and come up with the solution to hand off to the merchandisers," says Rider. "Instead, we meet with the business first."
Rider begins with a category that shows variability in its sales across stores, and meets with someone on the business side who handles the category, to determine if it has potential for a local market assortment. "They tell us what is happening in that particular category, what new products are coming out, what they are finding from our external sources, such as ACNeilsen," she says. "Once we take all these factors into consideration, then we decide whether to move forward or not."
The implementation required that Meijer change its organization, according to Rider. "We need to become much more specialized than we had been in the past," she says. "The buyer is a buyer now, not also an analyst. We have a process in place that we know will lead to success. And we have the technology and tools in place to bring this process through."