Servigistics Service Parts Planning: More Science, Less Art
Cray Inc. takes the guesstimation out of managing its stock of spare parts.
When bad things happen to supercomputers, technicians rush into action. But they can get the machines back in service only as fast as they can get hold of the right replacement parts. High-end supercomputer manufacturer Cray Inc. used to stock as many parts as possible at or near customers' facilities.
"You didn't ever want to get caught short and have to tell a customer like Ford that its $20-million machine will be down for 24 hours because we have to ship a spare from Chippewa Falls," says Jim Tennessen, Cray's manager of logistics.
Large safety stocks worked well in the days when Cray enjoyed high profit margins. But in today's tight economic climate, a service organization, just like a manufacturer, has to keep inventory lean. And maintaining the right stock of spares is even trickier than maintaining the right flow of parts to a production line.In service, "the parts are planned and needed more in a just-in-case than in a just-in-time environment, because demand is variable and hard to predict," says Mike Landry, chief executive officer of Servigistics, an Atlanta-based vendor of supply chain software for service parts planning. "When you wake up in the morning, you don't know what demand will happen where, or the priority of importance of that demand.
"Planners must decide how many of each spare to keep in stock. They must also figure out where to store those parts so they can speed them to technicians without spending huge sums on expedited shipping.Since last year, Cray has been paring back its safety stock while still providing customers with excellent service. With help from Servigistics' web-based optimization and planning software, Cray has cut its spare parts inventory by 28 percent, saving $13.2 million. Customers in more than 30 countries use Cray's supercomputers for applications such as vehicle design, weather forecasting, and medical research.
Technicians employed by Cray, most of them based at customers' facilities, maintain those machines. The service organization keeps spare parts at customer sites, at eight parts depots located throughout the United States, at two parts banks in Tennessee and Germany, and at a central inventory site in Chippewa Falls, Wis. Cray's service business brings in about $70 million per year, Tennessen says.
Traditionally, when Cray designed a new product, planning specialists worked with designers to learn how soon each part was likely to fail. Planners would develop a list of spares for the computer, recommending how many of each one Cray should keep at customers' sites, at the depots, in the two parts banks and in central inventory.
That planning approach posed two major problems, according to Tennessen. First, when engineers estimated how long parts would last, the numbers were sometimes too high and sometimes too low. Also, because the service organization competes with manufacturing for Cray-built parts, it has to place orders six months in advance. To judge how many parts they would need in which geographical areas, planners used sales and marketing department projections. "Those projections typically would be in error as well," Tennessen notes.
Beyond that, as new products went into the field, no one updated the projections to reflect actual performance. "If you had astute planners who had plenty of time on their hands, they might say, 'Gee, we're not using as many of these $45,000 parts as we thought we would, so let's pull some back and return them to manufacturing. Use them in the build cycle and get full credit for them,'" Tennessen says. But as planners moved on to concentrate on new projects, they generally let their old estimates stand.When he took over the logistics operation in 1999, Tennessen found shelves full of spare parts that had never been used and probably never would be. Coming from an engineering background, he wasn't comfortable with imprecise estimates and projections."If you want to know how much power a machine takes, there is always a way to determine that to the tenth of an amp," he explains. "I didn't like spares planning because it was more of an art than a science."
Studying the Science Of Inventory Management
Seeking a more scientific approach, Cray's logistics organization evaluated Servigistics and signed a contract for the software in June 2001. It took about eight weeks to get the system up and running. Implementation included integrating Servigistics with Cray's inventory management application and configuring the new system to meet Cray's business needs.
"Generally, because this is all new enabling technology for our customers, a lot of business process analysis comes with it," Landry says. Users must decide, for example, how low to allow their stock of different parts, at different locations, to fall before the system sounds an alarm, or how many months a part should sit unused before asking planners to reevaluate its importance.
Past and Future
To create an inventory plan, Servigistics forecasts demand for every part at every location, Landry says. It bases the forecasts on both historical data—often captured from an inventory management or customer relationship management (CRM) system—and on projections, such as the estimated mean time between failures.
The system considers the service commitments written into contracts—for instance, repair within four hours—and users can set different priorities for different parts, Landry says. For example, for inexpensive but critical parts needed in a popular new product, the user might require a fill rate of 99.8 percent. But for parts costing more than $1,000 that are needed in an older product, a 70-percent fill rate might be acceptable.
Servigistics also considers how long it takes to obtain parts through different pipelines, including bad parts that are returned from the field and repaired. "How long does it take to order and receive that part?" Landry asks. "How long does it take to replenish it and move it out to the field? How long does it take to get the bad one back, then how long does it take to repair it?
"After creating a plan, the system monitors actual parts consumption through its link to the inventory management system or other execution software. Based on that data, it continually updates the plan."Every night, the system looks at what occurred during the day, and comes up with new numbers the next day," Tennessen says. The system also modifies engineers' estimates on parts failure."It says, 'We have 14 of these parts out here, and engineering said one would fail every year, but two are failing every year.' Then you can take appropriate action," he notes.
Don't Need, Don't Buy
Some of the $13.2 million Cray has saved so far came from pulling unused parts out of inventory. But, Tennessen explains, "the main thing I'm trying to do with this tool is prevent buying new parts—spares I'm never going to use.
"If planners find out they won't need certain parts, they can cancel their orders and release those parts to manufacturing. "It doesn't do a lot of good to find out eight years later that you never needed those power supplies," he says. "There's no secondary market for 95 percent of our spares."Along with reducing inventory, Servigistics improves efficiency and customer satisfaction, because technicians who have access to the parts they need get their work done faster, Landry says. "We also see planner productivity increase, because now they can be proactive rather than reactive expediters.
"Cray will get further advantages from Servigistics when it starts using a new module called Strategy. Comparing current inventory and activity against the plan, Strategy allows executives to monitor the performance of their parts logistics organizations.
"They track performance in terms of service level, fill rate, against the goals, as well as financial goals such as inventory levels against the optimal and expected goals," Landry says.Tennessen says Strategy will save him time by automating reports he now creates by hand. It also offers a credible way to show upper management how the logistics department is saving money.
"Three or four Excel spreadsheets I put together myself don't have quite the weight of a report my manager can run at any time he wants by clicking a button," he observes.In the future, Tennessen will also use Servigistics to help determine if Cray's spare parts banks and depots are in the best possible locations. "We do it manually now," he says. "We probably do a pretty good job. But I like having a scientific tool looking over our shoulders."