Dillard’s: The Big Fix


Jerry Blaker, a senior consultant in IBM’s Houston supply-chain practice, felt like he’d gotten off on the wrong floor when he took up his post at department store giant Dillard’s in February 2002.

Blaker was brought in to take on the role of outside project manager for a major supply chain and inventory replenishment project Dillard’s had launched a year earlier with i2 Technologies. At first, it looked like an easy assignment. In preparing for the project, he says he had never seen two parties so committed to each other’s success.

But within hours of getting off the elevator, Blaker realized that while relations between the customer and vendor were strong on the surface, the project was unraveling behind the scenes. In quiet moments between joint meetings, he was pulled aside to hear confidential concerns about the direction of the project and its prospects for success. “Both parties came to me and effectively said, ‘The other party’s not listening to us,'” recalls Blaker. “It was pretty clear that there were disconnects on several levels.”

The parties desperately wanted the project to get back on track. Dillard’s needed to turn over its inventory faster to keep up with rivals such as May Department Stores and Nordstrom. Inventory turnover is a key measure of a company’s ability to put the right products on the right shelves at the right time. i2, a provider of inventory, supply chain, and replenishment software, knew that failure at Dillard’s could cost it other retail customers at a time when it was already struggling for survival. Some of the problems would be relatively easy to fix, but others would not:

  • Goals of the project needed to be stated more clearly. The Little Rock, Ark.-based Dillard’s and i2 Technologies have different corporate cultures, says Blaker. Dillard’s has a culture based on long-term strategic thinking; goals are stated and executed sometimes without supporting documentation. i2 operates much more literally—goals have exact numbers and timetables attached.

  • Analysis of sales had to be sped up. I2’s software could slice and dice Dillard’s inventory and forecasting requirements in dozens of different ways but the results were not coming fast enough for Dillard’s. For example, i2 might be able to determine exactly how many bottles of Estée Lauder Intuition perfume should be stocked on a shelf in an individual store in Cincinnati, but if it could not make those precise calculations in time to get the delivery trucks out, the information was useless.

  • Transaction handling capacity required a big push after hitting a wall. By the time i2’s software was handling 2.9 million store items, or stock keeping units (SKUs), it was already taking too long to produce replenishment orders. Dillard’s needed the software to process up to 15 million SKUs, but the technology could not handle that volume within an acceptable timeframe.

    The face of Dillard’s changed dramatically in 1998 when family patriarch William Dillard put together a deal to acquire Mercantile Stores of Fairfield, Ohio, for $2.9 billion. The purchase expanded Dillard’s presence in the Midwest and solidified its position as the nation’s third largest department store chain, with 331 stores in 29 states, and annual sales of $8.2 billion.

    But the expansion exposed weaknesses in the company’s inventory control systems. In 1998, the company had an inventory turnover rate of just 3.7 compared with an industry average of 4.9. At the time, Dillard’s was using an 11-year-old homegrown inventory planning system called A.I.M.S. (Automated Inventory Management System) and a replenishment system called Trends. While the IBM-mainframe-based system was well suited to handle the 15 million product codes in the Dillard’s system, it was not great at precisely anticipating customer demand for an individual product, or for individual locations.

    Donna Walls, Dillard’s replenishment manager, says it was basically a “one-size fits all” system that did not take into account purchasing patterns based on geography, weather or other factors. For example, the company has five purchasing centers located in Little Rock, Ark., Phoenix, Ariz., St. Louis, Mo., Fort Worth, Texas, and south Florida. Phoenix handles buying not only for the company’s Arizona stores, but also for more northern locales such as Utah. In Phoenix stores, bathing suits are hot sellers right through the winter, but not so in northern states. Yet, Dillard’s inventory planning system could not fine-tune stock for individual stores without human buyers getting involved.

    Similarly, the system was good for setting broad parameters for product groupings such as cosmetics, perfume and shoes, but it was extremely labor-intensive to set replenishment parameters for an individual product. In the high-end retail business, where purchases often are emotional, that’s a definite disadvantage. Unlike selling boxes of Tide detergent, selling high-profit items like perfume and jewelry is a science. Walls says experience has shown, for example, that when women shop for expensive perfume like Christian Dior’s J’Adore, they are more likely to make the purchase when there are five or six bottles on the shelf. If there is only one bottle left, they’ll often take a pass.

    “I can’t explain it, I just know that it’s a fact,” she says. “It’s the same with products like towels—nobody buys just one towel, they’ll buy at least two. And they’ll likely buy hand towels at the same time. Having the right quantities of products is a key factor to success in this business.”

    In January 2001, after months of evaluating vendors, President Alex Dillard announced the company would deploy i2’s TradeMatrix Demand Planner and Replenishment Planner products to address the weaknesses in its homegrown applications. “This is the solution for which we have been searching for years,” he enthused at the announcement.

    Fast Forward

    By the time IBM’s Blaker came on board in early 2002, the initial enthusiasm was being replaced with concern. The system was successfully processing demand requirements and issuing replenishment orders for 2.9 million SKUs, and Walls loved the early results. “The forecasts out of this system are phenomenal,” she says. The problem was, it was taking up to 20 hours to complete the task.

    Dillard’s typically performs its stock replenishment processing over the weekend so that orders are ready for approval and can be sent out to suppliers by EDI (electronic data interchange) first thing Monday morning. This ensures products will be on trucks and delivered to stores later that same week. The 20-hour processing period was already cutting it too close for Walls’ comfort, and it was still only handling a fraction of the 15 million SKUs required. Both parties realized that unless the processing speed could be dramatically improved, the project was in jeopardy.

    “Sometimes it’s hard to say there’s something wrong with your baby,” says Walls. “I think we were so far along at that point … everyone had been working so hard, it was difficult to say something doesn’t work.”

    For its part, i2 was determined to overcome the roadblock. Pallab Chatterjee, i2’s solutions president, got on the phone to Alex Dillard and personally assured him i2 would do whatever was necessary to resolve the issue. Blaker also played a key role in getting to the next stage.

    For starters, he did what the parties admit should have probably been done from the start—clearly document strategic goals and desired metrics of the project, such as being able to produce replenishment orders for 15 million SKUs within a 12-hour window. By August 2002, the parties had developed what they referred to as a “Get Well Plan,” which included several minor changes and one major upgrade.

    On the minor side, there were parameters in i2’s software and IBM’s DB2 database software, which could be tweaked to improve the exchange of data between the two systems. The big change involved upgrading the applications from single thread processing, to multi-thread.

    Previously, Dillard’s used an IBM S85AIX server running DB2 EE 7.2, IBM’s database software. That version does not support database partitioning. Even though Dillard’s S85 server had six processors, only one was being utilized for the task of producing replenishment orders for the 2.9 million SKUs.

    By upgrading to DB2 EEE 7.2 (referred to as triple-E), i2’s replenishment software could run in a multi-partition mode and divide the workload:

  • The work of processing the desired 15 million SKUs could be divided among the six processors on the server.

  • By having each processor work on 2.5 million SKUs simultaneously, the task could theoretically be completed in 1/6th of the time.

    While the strategy looked great on paper, i2 had never attempted running the software in that type of environment, or at that large a scale. The software was shipped to IBM’s DB2 research center in Toronto, and technicians ran the applications through a battery of benchmarking tests. The results from those tests were positive; positive enough to convince i2 it had the answer. The tests showed that as the work was divided, the time required to complete the number crunching could be cut proportionately to the number of processors in use.

    Lessons learned in Toronto were then transported to i2’s headquarters in Dallas, and through the spring of this year i2 tested the software against sample data from Dillard’s. Meanwhile, Walls was anxiously waiting back in Little Rock for word on whether the upgrade to DB2 EEE would clear the roadblock. In early May, Vikas Goel and Ed Thompson, the project leaders from i2, walked into the office and gave her the news.

    “I knew it was going to be good, because they were grinning from ear to ear,” she says. In fact, based on its tests, i2 believes it will be able to process all 15 million SKUs and produce stock replenishment orders in a period of four to five hours. The actual processing time will take a little longer because Dillard’s asked for a few extra features. For example, it wants to break the processing up into tiers, so that its most important suppliers—the ones that provide products that bring in the most profits or are in greatest demand—will be processed first.

    “This was a critical challenge for us,” says Goel. “It was important for us in terms of our commitment to Dillard’s, but in the bigger picture, the lessons we learned here can be cross-leveraged across the retail industry.”

    The next few months will prove critical. The rollout onto the DB2 EEE architecture began in mid-June. By Thanksgiving, Dillard’s aims to be running all 15 million items through i2’s Demand Planner and Replenishment Planner. However, early gains from the limited use of i2’s software appear to be showing up in Dillard’s inventory turnover rate, which is now slightly better than the industry average. By early next year, Dillard’s and i2 should have a better sense of whether the big fix is generating big returns.