Optimizing Your Supply Chain: A Model Approach

A wide array of technological tools are just the do-it-yourself kit companies can use to model, then optimize, their supply chains.


MORE TO THE STORY:


Applying the information you get by using the right supply chain modeling tool can help squeeze cost and inventory out of your supply chain while improving service to customers. Yet, while some companies do an excellent job of modeling their supply chains, others have barely tapped the potential for improvement.

“The high-tech companies—where capital equipment is very expensive—have spent significant money and effort in developing models within their organizations, and they use these tools extensively,” observes Kiron Shastry, a senior manager with consulting firm Accenture, Chicago. Similarly, companies in process industries, such as paper and beer manufacturing, “have traditionally used optimization routines to improve their process.

“But other industry verticals, such as the food and consumer product goods industries, are more demand-focused, and haven’t spent a lot of effort optimizing their supply chains,” he says.

Modeling tools that can help optimize supply chains have been around for years, notes Larry Lapide, vice president of supply chain management, AMR Research Inc., Boston. But they’re getting increased attention because of their lower cost and greater capability, coupled with companies’ quests for improved supply chain performance.

A variety of modeling tools are available today. Deciding which type is right for you begins with understanding the environment in which you work and how often you need to change your supply chain, Lapide says. Dynamic environments—which may experience numerous new product introductions, have products with short life cycles, or use segmented customer fulfillment with multiple service levels, lead times, and fulfillment methods—require more frequent supply chain reconfiguring.

“If you will use this type of tool several times a year, it makes sense to buy it, train your people, and create a group to do supply chain modeling,” Lapide says. Companies that model their supply chains once every three years or so may want to tap a third party’s modeling technology and expertise.

Saturn Service Parts Operations (SPO) has gotten excellent results leveraging third party modeling capabilities.

“Modeling is not our core competence,” explains Bryan Burkhardt, Saturn’s director for warehousing and distribution. “We don’t have the expertise internally to do sophisticated modeling, so we have always looked at working with someone else.”

For example, Saturn SPO wanted to validate that its distribution model, which uses a single point of distribution—a warehouse in Spring Hill, Tenn.—was still appropriate. The company turned to Wharton Business School several years ago to model its supply chain, taking into account variables such as inventory, facility, labor, and transportation costs.

“When we looked at the business model, it made sense to stay with a single point of distribution,” Burkhardt says. Saturn manages inventory for its 440 retailers, so the bulk of deliveries are replenishment shipments.

“We take care of emergency orders via overnight shipments, and sensitivity of response time for replenishment orders is not as critical,” he says.

When the automobile manufacturer revalidated its distribution strategy last year, the single DC continued to make sense. But the high cost of outbound transportation did not. So Saturn turned to Schneider Logistics, its third-party logistics provider, for modeling help.

“We looked at bidding out a national dedicated fleet system like we’d used previously, and at dividing the business among three or four carriers to see if we could leverage some regional strengths,” Burkhardt says. Saturn also looked at integrating its traffic with that of General Motors Service and Parts Operations (GMSPO), which also uses Schneider Logistics.

“When we got those results back, it was a no-brainer,” Burkhardt says. GMSPO and Saturn have customers in many of the same cities, and in some cases were running duplicate routes.

Today, Saturn is nearly two-thirds of the way to converting its transportation to the dedicated delivery system it shares with GMSPO. Close to 87 percent of Saturn’s parts volume is shipped through the dedicated delivery system.

Saturn retailers receive replacement orders from the Spring Hill DC twice a week. Orders for a group of retailers are shipped as a line haul to the carrier’s regional interline facility (located next to the 16 GM parts distribution centers). Then they’re broken down by route, shipped to the GM parts distribution centers, loaded with the GM freight, and delivered to the retailer.

“We’ve reduced transportation costs by 32 percent,” Burkhardt says. The shared delivery model has improved service to retailers—especially those in remote locations—and enhanced tote pickups and returns.

Modeling 101

Models—which may take the form of a mathematical program or simulation—can be used at the strategic, tactical, or executional level, looking at such factors as demand, distribution networks, transportation routing, or warehouse operations.

“A model is a representation of the system,” explains Kevin R. Gue, associate professor of logistics, Graduate School of Business and Public Policy, Naval Postgraduate School, Monterey, Calif.

“In terms of improving the supply chain, the biggest bang for the buck can be obtained by getting a better picture of the demand signal,” says Shastry. “Many companies have grown distant from their end customers, so they make guesses about real demand. They don’t have a lot of good technology support, so they end up building buffers of inventory.”

Companies that are serious about cutting the fat in their supply chains will reduce those inventory buffers by reducing the uncertainty in the demand signal. “Once you have a good signal of what demand is, you can use this information to drive down safety stocks in your supply chain,” says Shastry.

Models are often used to achieve a single goal or a set of objectives, such as minimizing inventory costs or maximizing service, Gue says. Optimization tools can be used “to search through the set of possible solutions to try and achieve the best value of that objective, which might include finding the set of inventory policies that maximizes service within some cost constraint,” he says.

Mathematical tools tend to be more strategic, used to perform a study of capacity planning or develop a new network design or a plan for new product distribution, Lapide says. The cost of these may run into the hundreds of thousands of dollars.

Simulation tools using what-if concepts are more tactical, simulating plant or warehouse operations, for example. These are less expensive. PC-based simulation tools may cost in the tens of thousands.

A new type of tool is the inventory optimizer, which focuses on “where and how much you should stock products within an existing network,” Lapide notes. These tools are gaining popularity as companies look for new ways to minimize inventory.

As with inventory optimization, logisticians often use competing objectives, such as maximizing service levels while minimizing cost. These situations call for optimization tools that can incorporate a multi-criteria decision-making structure, Gue says. By considering the relative importance you’ve specified for service and cost, for example, the system identifies the answer that best satisfies the objectives, based on the tradeoffs.

A model may “search the hundreds of thousands of permutations of different possibilities,” Shastry says. The solution has to be feasible, and respect the constraints of your supply chain.

Deterministic vs. Stochastic

Models can be deterministic or stochastic, depending upon whether uncertainty is built into the model. “Say the demand for copier paper at an office supply store is 42 cases per month,” explains Gue. “In a stochastic model, demand has a mean of 42.” But demand isn’t static; it may be different every month. So the actual demand has some sort of distribution underlying the average of 42.

Incorporating probabilities such as these into models can be “mathematically and computationally difficult,” Gue says. “Finding optimal solutions to these kinds of models can be extraordinarily difficult, so people may use deterministic models,” which do not incorporate issues of uncertainty.

Another important concept in modeling is heuristics, which Gue describes as using “rules of thumb to help the model search all possible solutions to find what is hopefully the best one.”

“‘Heuristic’ literally means by trial and error,” says Mark Ridge, vice president professional services, Radical Limited, a provider of supply chain modeling software and consultancy based in Watford, U.K.

Heuristic approaches are common in everyday life. Imagine assembling a jigsaw puzzle. “We try alternative pieces in combination until we find pieces that fit together. We often shortcut the puzzle-building process by finding all the edge pieces first to build a frame for our solution. This is a heuristic solution,” Ridge says.

By its very nature, the heuristic approach does not always work. “But we accept its limitations, then take an appropriate course of action to deal with the shortfalls of our heuristic estimates as and when they arise,” Ridge notes.

Just as people try to fit pieces of a jigsaw puzzle together in various combinations until they find the one that completes the picture, “supply chain models are run over and over again against different network configurations to find the one that satisfies the scope and objective for any given project,” he says.

Mathematical approaches such as mixed integer programming optimization, or MIPO, use mathematical techniques “to make resource allocation decisions where there are competing demands upon resources, and overriding criteria to be satisfied,” Ridge says.

Unlike heuristic programs, mathematical programs are not iterative. “They’re run once to deliver a solution,” says Ridge. “In the context of supply chain modeling, the mixed integer programming optimizer allows a mathematically optimal network configuration to be found given a user-defined set of candidate locations subject to any constraints or parameter settings.”

Users can use a combination of the two approaches. “If the heuristic is run before the optimizer, the user can benchmark the heuristic ‘preferred’ solution to the optimum solution in a quantifiable way,” Ridge says. “If the optimizer is run first, the optimum solution can be modified according to the practical requirements for implementation.”

Daily Planning Drives Breakthrough Model

When advanced integrated circuit solution provider Agere Systems spun off from Lucent Technologies last year, the company had already started moving from a spreadsheet-based weekly planning system to a system-based daily planning system. This advanced planning system, a key component in the company’s customer-centered supply chain program, enables Agere to improve its performance significantly, reports Andy Micallef, vice president of supply chain management for the Allentown, Pa.-based company.

Despite a down market, inventory turns have increased from six to eight a year. Shipping performance—against original acknowledgement date—has jumped from mid-60 percent to mid-90 percent, Micallef says.

That’s just the beginning, as Agere begins to leverage visibility and real-time data to implement a breakthrough supply chain model.

Agere’s move to daily planning was driven by its desire to provide customers with better service and the ability to carry less inventory. “To the extent that we can be flexible, when a new order comes in, we’ll make every effort to start the order process the next day,” Micallef says. This can slash the order cycle by as much as one week.

Two information systems enable the change to daily planning:

1) Agere’s proprietary enterprise system solution, Total Order Management, which ties into the company’s factory schedule system and Oracle ERP, and

2) a proprietary system built on an Access database with Excel spreadsheet tools, co-developed by Agere and Lehigh University’s Department of Industrial and Systems Engineering.

This system “considers demand, then models inventory buffer levels so we can improve shipping performance,” Micallef says. “It takes a demand profile from a specific customer on a specific product, and looks at shipment history against the order book. It also looks at the forecast.” With this information, the system models the positioning of inventory based on demand volatility, enabling Agere to improve the way it meets customer requests.

“We look at the plot of demand volatility against volume,” which enables Agere to make better inventory decisions, Micallef says. Take the customer whose demand is shown in Figure 1. Even though demand looks tightly distributed, “the frequency of upside is 46 percent, which means that half the time they require more than what they ordered,” he says. The weighted average change is 17.7 percent.

“If we want to have shipping confidence of 95 percent, modeling that against the normal distribution of what could happen, we’d need to have 30 percent buffer in place to always cover them.”

The ability to profile demand “allows us to realize unforecasted demand, which is a tremendous service to our customers,” Micallef says. Agere currently performs this analysis quarterly for its top strategic customers, then builds to demand and buffer on a daily basis.

Agere’s staff of supply chain planners in Allentown, Pa., begin their mornings reviewing system-generated daily move plans for individual products and specific technologies. “The system isn’t perfect; there are planning factors and operational issues to consider,” Micallef explains.

Changes might be caused by such factors as data integrity issues, capacity constraints, or new products. Planners have several hours to make changes in the move plan, then all the data is gathered into an integrated execution plan for Agere’s wafer fab facilities and packaging/test factories in Asia.

The supply chain staff reviews each day’s actions. “We want to constantly improve our system,” Micallef observes. “If we’re overriding what the system recommended, we want to know why.”

Agere is using the information generated by the tool to develop “a joint, full-stream, constraint-based buffer policy so that we can minimize inventory throughout the supply chain,” Micallef says, pointing out that the new approach removes data latency, “one of the biggest challenges in planning a supply chain.”

With the traditional supply chain model, demand data may take 60 days or more to travel from the end customer up the supply chain to Agere’s suppliers (its foundry partners).

“If we can introduce a system that shares data real-time, then everybody gets the same information at the same time,” he notes.

All partners in the supply chain will work toward the same set of intentions simultaneously. Instead of each supply chain partner maintaining an inventory buffer, the buffer will be statistically sized and located. The new model will thus minimize volatility, lead time, and inventory, enabling Agere to deliver improved service while lowering costs.

Agere is just beginning to implement this new breakthrough supply chain model. “As we move and involve more of the supply chain, partner with our customers, and bring our suppliers and their suppliers into the mix, we’ll have more of an optimized supply chain than we do today,” Micallef concludes. “It takes the right people, right attitude, and right systems to reduce this data latency and improve processes.” Agere is well on its way to making it happen.

Worth the Effort?

With all the complexities and costs of supply chain modeling, is it worth it? Accenture’s Kiron Shastry thinks so. “There is definitely a lot of benefit—cost savings as well as improved customer service—that you can get from using the right supply chain modeling tool to improve your logistics,” he says.

“But don’t underestimate the amount of time and organizational change it takes to implement these tools,” he warns. “It takes a fair amount of time to understand the tools, and how best to apply them.” That’s why, for many logisticians, the time to start those modeling lessons is now.


7 Myths of Modeling

The concepts underlying supply chain modeling are very complex, but logistics managers don’t need to go back to school for an advanced degree in math, says Kevin R. Gue, associate professor of logistics, Graduate School of Business and Public Policy, Naval Postgraduate School, Monterey, Calif.

They do, however, need to know enough “to be smart consumers of modeling technologies, and be able to ask smart questions and understand the terminology,” he says.

The first step is understanding that “a model is, by definition, not the reality,” he says. “So you have to understand the assumptions that have been made to represent the system, whether mathematically or in a simulation. Acknowledging and recognizing the assumptions up front help ensure that there are no surprises at the back end.”

You also need to understand how uncertainty and variability were used to build the model. “In many mathematical programming and optimization models, uncertainty may not be represented at all,” Gue says. Simulation models, on the other hand, can represent uncertainty with underlying distributions.

Being a smart consumer of modeling technologies also means having an accurate understanding of what modeling is, and what it can do. Consider these supply chain modeling myths identified and dispelled in a white paper by Schneider Logistics Inc.’s Jim Jeray:

Myth #1. There is one modeling or software package to solve the whole range of your problems. Wrong. “Different kinds of questions require different kinds of answers, different ‘objective functions,’ and, often, different software applications,” Jeray says. Don’t expect a single modeling tool to be able to handle all your needs.

Myth #2. Once the model is built, it will work forever. Changes in variables such as demand patterns, carriers, product lines, and customer and supplier locations can make models obsolete. How often you need to model depends upon your supply chain environment.

Myth #3. Supply chain modeling is purely a mathematical exercise. Not so, Jeray says. “Although the underlying mathematical setup is important, it is usually constrained by customer requirements and current business practices.” In addition, the modeler must understand the questions that are being asked, as well as the required degree of precision and the timeframe involved.

Myth #4. Model results are easily implemented. The analysis isn’t over once the model is run. “Occasionally, models provide solutions that look good on paper but are not operationally practical,” says Jeray. That’s why, no matter how sophisticated tools become, the models will still require analysis and validation by supply chain pros.

Myth #5. Gathering the data to support the modeling effort is quick and painless. “Modeling efforts are highly dependent on the quantity and the quality of data that is available,” Jeray says. While getting enough of the right data is not an easy task for many organizations, the model will be no better than the data it is built upon.

Myth #6. All models can be created quickly and provide good solutions in less than one week. “Some models can be created and run in a matter of hours, while others can take months just to develop,” Jeray notes. “The time to develop, validate, run, and analyze the results of a model depend on the complexity of the inputs and the desired level of accuracy.”

Myth #7. Models are 100-percent accurate. “In fact, models are not 100-percent accurate,” Jeray says. “However, they are still better than other alternatives, such as ‘gut feel’ or strings on a map.”