Executive Briefings

MIT Professor Sees Promise In Supply-Chain Modeling

A conversation with Jeremy Shapiro, professor of operations research and management emeritus in the Sloan School of Management at the Massachusetts Institute for Technology.

As professor emeritus, Shapiro continues to teach courses at MIT in supply-chain management and related topics. For nine years, he served as co-director of MIT's Operations Research Center and has lectured and consulted extensively in Europe, Asia, Australia and South America. Prior to his appointment at MIT, Shapiro was employed by Procter & Gamble, Hughes Aircraft Co., and the Port of New York Authority. Shapiro received his BME and MIE degrees from Cornell and a Ph.D. in operations research from Stanford. He has published more than 50 papers and is the author of two books: Mathematical Programming: Structures and Algorithms, published by John Wiley and Sons in 1979; and, Modeling the Supply Chain, published by Duxbury Press in 2001.

Q: What are some of the key supply-chain issues you see companies grappling with today?

Shapiro: One issue that I am particularly sensitive to is this: IT developments have enabled companies to collect, transmit and communicate transactional data about what is going on extremely well, but there is more to running a supply chain than making sure the orders get shipped out today. Companies need to be planning ahead for next week, next month and next year. I think many are beginning to realize this and are looking for the right tools and processes. This becomes more difficult, by the way, if a company relies heavily on outsourced manufacturing in places like the Far East, because they don't have as much control over contracted operations. As companies increasingly perceive the need to do a better job of planning ahead, they are running into issues of understanding what kind of capacity they have at contract manufacturers, what their future costs will be and perhaps how they might renegotiate contracts. Without this information, they have a lot of exposure.
And that touches on some of the ongoing and classic issues of risk management that have to do with longer-term planning - things like currency exchange rates. Exchange rates could now be very favorable to a company manufacturing in a Far East plant, but if the exchange rates vary significantly next year, that arrangement might not turn out to be such a great deal. So it comes down to what your contract looks like and how you plan around that.

There also is the issue of cycle time when goods are coming from so far away. A company might be able to procure a product really cheaply in the Far East, but the lead time to get it to distribution centers and stores in North America might be very long and less predictable. So companies have to ask what the tradeoff is of cost vs. lead time. I know in the retail business, they are starting to ask those questions.

Q: What kind of answers are they coming up with?

Shapiro: Well, for example, I worked with one large Canadian retailer that is considering leasing or building one or more DCs in China in order to do inbound consolidation. That would give them better control over how long it would take goods to get to ports of entry and then on to North American DCs and on to stores, so this retailer is facing up to it. These are complicated issues because some of the classical analytic approaches for looking at cycle time and lead time vs. inventory really look at items one at a time and don't consider the complication that items travel together. If you do inbound consolidation you can control that. Of course, this also depends on the quantities you buy. If you are buying container loads or truckloads it's less of an issue, but smaller quantities may be kept waiting for up to 20 days, until there are enough orders for a full container.

Another interesting example is Obermeyer, the ski apparel manufacturer. Obermeyer makes a lot of products in the Far East-this was written up in the Harvard Business Review a while back. They focus on getting early forecasts of which products are going to be really popular. Instead of just using a no-information forecast, they get some information early in the season and use that to fine-tune their orders. Of course, they had to structure their contracts with clothing manufacturers to allow them to do this. It turned out to be very successful for them.

So going back to my original theme, you want to manage the execution of the business well, but you also need to develop longer-term plans because, while things may be going great guns right now, they can easily get off track.

Q: Is visibility a big part of that?

Shapiro: Visibility helps a lot, but I would say that more important are modeling tools that allow companies to run scenarios of the future. This lets them see how sensitive and cost effective their plans are in different circumstances. These tools allow you to change what the future looks like and compare different ways of responding. For example, you can evaluate issues such as having redundancy in product or component production, so that if you have a surge in demand or if you have a plant outage for a week, you can shift production from one place to another and not lose a lot of business. A modeling tool also can help you understand what kind of inventories you should carry and where you should carry them. Or, if you want to look at using airfreight as a backup, you can easily see the tradeoff of occasional airfreight vs. other contingency plans that might seem to be less expensive but really aren't. So you can build these holistic models that look at sourcing, manufacturing and distribution together. Then, as a result of (hopefully) having accurate and comprehensive cost and capacity information about various parts of the supply chain, you optimize for next year's plans, or maybe optimize to 10 versions of next year's plan. Then you use management judgment to say how you shall actually run the supply-chain next year. So in that sense, you are exploring the options for running the company next year as opposed to just empirically extrapolating numbers from this year to next year and waiting to see how the metrics go. But it takes more work to develop these exploratory models. Given the payoffs, it is well worth it, but I would say that that type of modeling is new and companies are just getting tuned in on it.

Q: Is this what is called supply-chain network optimization?

Shapiro: Yes and "network" is the key word. You have a geographical dispersion of facilities and you have dispersion of activities across this network and so you need to holistically optimize across the network. As I said at the start, globalization is carrying that network idea another step. Often, when companies expand globally, they do so with arm's length arrangements. Retailing or apparel companies don't have subsidiaries in China or India or France. They may contract for manufacturing in those countries, but it is at arm's length. On the surface, that makes planning easier because you have a well-defined contract and you know what to expect. But you also have much less flexibility.

Q: In general, are companies getting better at synchronizing supply and demand?

Shapiro: Consciousness is expanding about coordinating demand management with supply management, but this depends heavily on the type of industry. Coordinating supply and demand management in a commodity industry, which is pretty much price driven, is challenging. But it is easier than in consumer products, where demand management has to do not just with prices, but with advertising, promotion, sales efforts and so on.

One of my prejudices is that it is much easier to measure supply-chain quantities than it is to measure peoples' behavior. It is not that easy to understand what all your physical costs are, but at least that is a fairly objective question. But in a CPG company, which is very marketing driven, what happens is the marketing people decide what they think they are going to sell. They then throw that over the transom and say to the supply-chain people, "here is what we want you to do."

But I am optimistic. I think the whole spirit of more and more managers wanting to do fact-based decision-making will penetrate marketing and sales efforts. And there are analytic tools in marketing science to help them understand, if they spend $20m on a television campaign, what they are getting for it. Then you can ask the supply-chain question: What is it going to cost to meet the extra demand? At some point, you get to diminishing returns because supply-chain costs incrementally increase and the marketing effects incrementally decrease. So there is some qualitative desire to find the point at which it is not worth it anymore. To a certain extent that type of analysis is still ahead of us. Even a major company with very good branding and very good marketing and also very good supply-chain management is unlikely to have these capabilities seriously integrated yet, though I think we will see a lot more of that over the next five or 10 years.

I have the definite impression, though, that even very forward-looking companies are not devoting enough resources to actually managing and planning supply-chain integration. I have been really surprised that at some very large, leading-edge companies, the people responsible for having a global view of outsourcing and final product assembly seem not to have sufficient time or resources to really get on top of this. In fact, that is actually a fairly general phenomenon. They just don't yet appreciate the payback from supply-chain network optimization.

Q: What is the typical payback?

Shapiro: Let me give you a quick example. I worked with an industrial chemical company in the middle '90s and we did a pilot study to see the impact of making some changes in global sourcing. This company had about 20 plants around the world. By doing a supply-chain network optimization over all 20 plants and markets for a major product line, we came up with a scheme to reduce costs by about 9 percent on $100m. The solution was based on the fact that some plants were more efficient at various processes needed to make the intermediate products, partly because volumes were high and economies of scale were there. If you have a nice, long run making products or intermediate products you save a lot of money. So the strategy to save $9m had these plants working more efficiently and shipping intermediate products to places nearer the intended market for final processing. That certainly proved the point.

I have to say, though, that there were downstream political issues, in the sense that negotiations had to take place with the plant managers about what they were going to make or not make. This illustrates a hidden issue-the need to set up managerial incentives that reflect results you want from the entire supply chain, rather than doing things myopically.

When doing network optimization, companies often come up with plans that are politically infeasible, even though economically they make a lot of sense. So if you find 15 percent savings, sometimes you have to ask yourself whether the middle managers running the company will go along with the changes. That is where you need senior management to step in and say, "you guys are not a bunch of barons running your own plants and your own fiefdoms. We have to pull together on this."

Q: If the results are so dramatic, why is there not more recognition of this approach?

Shapiro: Because it is still new. In the '90s we had enterprise resource planning come in, which is getting key data centered at one place in the company, communicating it well, making it accessible to people across the company. That is the foundation for doing analytical planning. Then the bubble burst and not too much happened for two or three years. I think now companies are more aware that they need to go beyond the transactional data and develop these planning tools.

There are two types of modeling using ERP databases. The first is descriptive modeling, which lets you understand what the costs in your manufacturing plants are and lets you develop accurate forecasts of what you are liable to sell next year. A wide range of descriptive activities start out by telling you where you are now, but mainly you need to project that into the future.

The descriptive modeling then provides inputs to prescriptive or optimization modeling. Now that you understand your costs, your capacities, your bills of material and what the forecasts look like, then you want to explore how to run the company next year. Maybe you need more outsourcing, maybe you need to close down a plant. Maybe you need to drop a product line that is not making any money. So that is where prescriptive or optimization modeling comes in.

The transactional databases on which you do these two types of analysis have really only shaped up in the last 10 years. Today we have much more powerful data acquisition and management tools, and more flexible ones, so the groundwork is laid for this descriptive and prescriptive modeling.

Beyond that is stochastic analysis. The way planning is done for next year using the supply-chain network optimization model is that you run scenarios. You might run 50 or 100 scenarios with different data in them about what product demand is going to be, what diesel fuel will cost, what transportation rates will be-a whole host of variables. Then you look at the results from the 100 or 50 scenarios and you summarize it. Then management has to make a decision about what to do. With the stochastic programming approach, instead of running 100 separate scenarios, each one saying this is what it will look like next year under these circumstances-you do them all together and you assign probabilities that each scenario will occur. Then you optimize over all the scenarios together.

You also can take the approach, when planning for the future, of using multi-period models. This means you are not simply taking a snapshot of next year or the next five years, but you look out the next three years, by quarters, so you see how the business evolves and how your decisions should evolve over time. You probably only implement the plans for next quarter. Then three months from now you update all the data and do it again, but you already have established some continuity. So that gives you a chance, without doing the stochastic programming, to adapt to changing conditions. It doesn't do it as rigorously or as powerfully as stochastic modeling, but that is really more what the state of the art is now.

Q: If you run network optimization every month, how much of what is in place can you really change every month?

Shapiro: That gets to the difference between strategic planning and tactical planning. At the strategic planning level, much more is up for grabs than at the tactical planning level. However, the network optimization models, from a technical point of view, are the same. You are fixing the structure and existence of facilities from a strategic point of view. For shorter-term planning, you still have a lot of leeway-for example, how many shifts a week are you going to run, where are you going to deploy inventories. There are questions how you build up inventories for peak seasons and how you use and locate peak storage facilities. So on a monthly basis you are adjusting your plans continuously and you are refining your resources and your strategies to deal with changes that you hadn't expected.

Q: Are there other developments in supply-chain management that you are excited about?

Shapiro: One situation that I ran into recently that has some of the same spirit is to use modeling tools to do data mining on retail point-of-sale (POS) data. Some of my colleagues at MIT have been doing research on an approach where you simultaneously cluster, say, 50,000 retailing products. If you take information from 500 stores, that is a huge amount of data. But maybe you can cluster it so that, from a statistical point of view, you look at only 100 or 500 distinct families of products. All the items in a family would behave the same way in terms of how people buy the products. For example, you might cluster products that are seasonal or that have no season. So you go from this fantastic complexity of 50,000 SKUs in 500 stores to something that is easier to understand. The human mind can grasp it. This approach also is very useful for forecasting because, if you better understand how all the items in a particular family behave in terms of demand, then you can develop better forecasts. And inventory management policy also is very much a function of what those demand patterns look like-you have slow-moving items and fast-moving items. You then can base decisions on when to order, how much to order, on very good analytics. The innovative thing here is that the data mining is using point-of-sales information. A lot of retailing companies have POS data but it is not clear what they are doing with it. That was part of the "looks like a good idea so lets buy it" trend of the '90s. Some of it actually was a good idea; it just needed some more work.

Personally, I recently have been drawn to looking at how we can relate ideas of organizational behavior to rational tools that would help us run companies better. There is a large quantity of literature on this with some really interesting ideas that, to a certain extent, pre-dated ERP systems and supply-chain network optimization. What these ideas suggest is that companies need to do business process expansion to exploit IT and planning tools. We see some primitive versions of this in some companies. Actually, the oil companies in their own way have been doing this for a long time but most industries have not gotten to it yet. So, what I mean by this, is that what companies need to do to better manage their supply chains is to create what you might call a supply-chain team. This would include a supply-chain manager who is responsible for holistic analysis of the supply chain. And this type of analysis needs to be done on a repetitive basis. Senior management has to insist that managers who are executing the supply chain participate appropriately in these supply-chain reviews. For example, for tactical planning you might do it once a month and the supply-chain team managers would have to devote maybe four to 16 hours a month to understand how what they are doing fits in with what everyone else is doing. They would discuss problems that any member of the team sees coming up and discuss potential solutions-rational solutions, not "I get my way because I can scream louder than you." So as problems come up and you have to make concrete decisions you are going to make much better decisions than if just suddenly had to react.

As professor emeritus, Shapiro continues to teach courses at MIT in supply-chain management and related topics. For nine years, he served as co-director of MIT's Operations Research Center and has lectured and consulted extensively in Europe, Asia, Australia and South America. Prior to his appointment at MIT, Shapiro was employed by Procter & Gamble, Hughes Aircraft Co., and the Port of New York Authority. Shapiro received his BME and MIE degrees from Cornell and a Ph.D. in operations research from Stanford. He has published more than 50 papers and is the author of two books: Mathematical Programming: Structures and Algorithms, published by John Wiley and Sons in 1979; and, Modeling the Supply Chain, published by Duxbury Press in 2001.

Q: What are some of the key supply-chain issues you see companies grappling with today?

Shapiro: One issue that I am particularly sensitive to is this: IT developments have enabled companies to collect, transmit and communicate transactional data about what is going on extremely well, but there is more to running a supply chain than making sure the orders get shipped out today. Companies need to be planning ahead for next week, next month and next year. I think many are beginning to realize this and are looking for the right tools and processes. This becomes more difficult, by the way, if a company relies heavily on outsourced manufacturing in places like the Far East, because they don't have as much control over contracted operations. As companies increasingly perceive the need to do a better job of planning ahead, they are running into issues of understanding what kind of capacity they have at contract manufacturers, what their future costs will be and perhaps how they might renegotiate contracts. Without this information, they have a lot of exposure.
And that touches on some of the ongoing and classic issues of risk management that have to do with longer-term planning - things like currency exchange rates. Exchange rates could now be very favorable to a company manufacturing in a Far East plant, but if the exchange rates vary significantly next year, that arrangement might not turn out to be such a great deal. So it comes down to what your contract looks like and how you plan around that.

There also is the issue of cycle time when goods are coming from so far away. A company might be able to procure a product really cheaply in the Far East, but the lead time to get it to distribution centers and stores in North America might be very long and less predictable. So companies have to ask what the tradeoff is of cost vs. lead time. I know in the retail business, they are starting to ask those questions.

Q: What kind of answers are they coming up with?

Shapiro: Well, for example, I worked with one large Canadian retailer that is considering leasing or building one or more DCs in China in order to do inbound consolidation. That would give them better control over how long it would take goods to get to ports of entry and then on to North American DCs and on to stores, so this retailer is facing up to it. These are complicated issues because some of the classical analytic approaches for looking at cycle time and lead time vs. inventory really look at items one at a time and don't consider the complication that items travel together. If you do inbound consolidation you can control that. Of course, this also depends on the quantities you buy. If you are buying container loads or truckloads it's less of an issue, but smaller quantities may be kept waiting for up to 20 days, until there are enough orders for a full container.

Another interesting example is Obermeyer, the ski apparel manufacturer. Obermeyer makes a lot of products in the Far East-this was written up in the Harvard Business Review a while back. They focus on getting early forecasts of which products are going to be really popular. Instead of just using a no-information forecast, they get some information early in the season and use that to fine-tune their orders. Of course, they had to structure their contracts with clothing manufacturers to allow them to do this. It turned out to be very successful for them.

So going back to my original theme, you want to manage the execution of the business well, but you also need to develop longer-term plans because, while things may be going great guns right now, they can easily get off track.

Q: Is visibility a big part of that?

Shapiro: Visibility helps a lot, but I would say that more important are modeling tools that allow companies to run scenarios of the future. This lets them see how sensitive and cost effective their plans are in different circumstances. These tools allow you to change what the future looks like and compare different ways of responding. For example, you can evaluate issues such as having redundancy in product or component production, so that if you have a surge in demand or if you have a plant outage for a week, you can shift production from one place to another and not lose a lot of business. A modeling tool also can help you understand what kind of inventories you should carry and where you should carry them. Or, if you want to look at using airfreight as a backup, you can easily see the tradeoff of occasional airfreight vs. other contingency plans that might seem to be less expensive but really aren't. So you can build these holistic models that look at sourcing, manufacturing and distribution together. Then, as a result of (hopefully) having accurate and comprehensive cost and capacity information about various parts of the supply chain, you optimize for next year's plans, or maybe optimize to 10 versions of next year's plan. Then you use management judgment to say how you shall actually run the supply-chain next year. So in that sense, you are exploring the options for running the company next year as opposed to just empirically extrapolating numbers from this year to next year and waiting to see how the metrics go. But it takes more work to develop these exploratory models. Given the payoffs, it is well worth it, but I would say that that type of modeling is new and companies are just getting tuned in on it.

Q: Is this what is called supply-chain network optimization?

Shapiro: Yes and "network" is the key word. You have a geographical dispersion of facilities and you have dispersion of activities across this network and so you need to holistically optimize across the network. As I said at the start, globalization is carrying that network idea another step. Often, when companies expand globally, they do so with arm's length arrangements. Retailing or apparel companies don't have subsidiaries in China or India or France. They may contract for manufacturing in those countries, but it is at arm's length. On the surface, that makes planning easier because you have a well-defined contract and you know what to expect. But you also have much less flexibility.

Q: In general, are companies getting better at synchronizing supply and demand?

Shapiro: Consciousness is expanding about coordinating demand management with supply management, but this depends heavily on the type of industry. Coordinating supply and demand management in a commodity industry, which is pretty much price driven, is challenging. But it is easier than in consumer products, where demand management has to do not just with prices, but with advertising, promotion, sales efforts and so on.

One of my prejudices is that it is much easier to measure supply-chain quantities than it is to measure peoples' behavior. It is not that easy to understand what all your physical costs are, but at least that is a fairly objective question. But in a CPG company, which is very marketing driven, what happens is the marketing people decide what they think they are going to sell. They then throw that over the transom and say to the supply-chain people, "here is what we want you to do."

But I am optimistic. I think the whole spirit of more and more managers wanting to do fact-based decision-making will penetrate marketing and sales efforts. And there are analytic tools in marketing science to help them understand, if they spend $20m on a television campaign, what they are getting for it. Then you can ask the supply-chain question: What is it going to cost to meet the extra demand? At some point, you get to diminishing returns because supply-chain costs incrementally increase and the marketing effects incrementally decrease. So there is some qualitative desire to find the point at which it is not worth it anymore. To a certain extent that type of analysis is still ahead of us. Even a major company with very good branding and very good marketing and also very good supply-chain management is unlikely to have these capabilities seriously integrated yet, though I think we will see a lot more of that over the next five or 10 years.

I have the definite impression, though, that even very forward-looking companies are not devoting enough resources to actually managing and planning supply-chain integration. I have been really surprised that at some very large, leading-edge companies, the people responsible for having a global view of outsourcing and final product assembly seem not to have sufficient time or resources to really get on top of this. In fact, that is actually a fairly general phenomenon. They just don't yet appreciate the payback from supply-chain network optimization.

Q: What is the typical payback?

Shapiro: Let me give you a quick example. I worked with an industrial chemical company in the middle '90s and we did a pilot study to see the impact of making some changes in global sourcing. This company had about 20 plants around the world. By doing a supply-chain network optimization over all 20 plants and markets for a major product line, we came up with a scheme to reduce costs by about 9 percent on $100m. The solution was based on the fact that some plants were more efficient at various processes needed to make the intermediate products, partly because volumes were high and economies of scale were there. If you have a nice, long run making products or intermediate products you save a lot of money. So the strategy to save $9m had these plants working more efficiently and shipping intermediate products to places nearer the intended market for final processing. That certainly proved the point.

I have to say, though, that there were downstream political issues, in the sense that negotiations had to take place with the plant managers about what they were going to make or not make. This illustrates a hidden issue-the need to set up managerial incentives that reflect results you want from the entire supply chain, rather than doing things myopically.

When doing network optimization, companies often come up with plans that are politically infeasible, even though economically they make a lot of sense. So if you find 15 percent savings, sometimes you have to ask yourself whether the middle managers running the company will go along with the changes. That is where you need senior management to step in and say, "you guys are not a bunch of barons running your own plants and your own fiefdoms. We have to pull together on this."

Q: If the results are so dramatic, why is there not more recognition of this approach?

Shapiro: Because it is still new. In the '90s we had enterprise resource planning come in, which is getting key data centered at one place in the company, communicating it well, making it accessible to people across the company. That is the foundation for doing analytical planning. Then the bubble burst and not too much happened for two or three years. I think now companies are more aware that they need to go beyond the transactional data and develop these planning tools.

There are two types of modeling using ERP databases. The first is descriptive modeling, which lets you understand what the costs in your manufacturing plants are and lets you develop accurate forecasts of what you are liable to sell next year. A wide range of descriptive activities start out by telling you where you are now, but mainly you need to project that into the future.

The descriptive modeling then provides inputs to prescriptive or optimization modeling. Now that you understand your costs, your capacities, your bills of material and what the forecasts look like, then you want to explore how to run the company next year. Maybe you need more outsourcing, maybe you need to close down a plant. Maybe you need to drop a product line that is not making any money. So that is where prescriptive or optimization modeling comes in.

The transactional databases on which you do these two types of analysis have really only shaped up in the last 10 years. Today we have much more powerful data acquisition and management tools, and more flexible ones, so the groundwork is laid for this descriptive and prescriptive modeling.

Beyond that is stochastic analysis. The way planning is done for next year using the supply-chain network optimization model is that you run scenarios. You might run 50 or 100 scenarios with different data in them about what product demand is going to be, what diesel fuel will cost, what transportation rates will be-a whole host of variables. Then you look at the results from the 100 or 50 scenarios and you summarize it. Then management has to make a decision about what to do. With the stochastic programming approach, instead of running 100 separate scenarios, each one saying this is what it will look like next year under these circumstances-you do them all together and you assign probabilities that each scenario will occur. Then you optimize over all the scenarios together.

You also can take the approach, when planning for the future, of using multi-period models. This means you are not simply taking a snapshot of next year or the next five years, but you look out the next three years, by quarters, so you see how the business evolves and how your decisions should evolve over time. You probably only implement the plans for next quarter. Then three months from now you update all the data and do it again, but you already have established some continuity. So that gives you a chance, without doing the stochastic programming, to adapt to changing conditions. It doesn't do it as rigorously or as powerfully as stochastic modeling, but that is really more what the state of the art is now.

Q: If you run network optimization every month, how much of what is in place can you really change every month?

Shapiro: That gets to the difference between strategic planning and tactical planning. At the strategic planning level, much more is up for grabs than at the tactical planning level. However, the network optimization models, from a technical point of view, are the same. You are fixing the structure and existence of facilities from a strategic point of view. For shorter-term planning, you still have a lot of leeway-for example, how many shifts a week are you going to run, where are you going to deploy inventories. There are questions how you build up inventories for peak seasons and how you use and locate peak storage facilities. So on a monthly basis you are adjusting your plans continuously and you are refining your resources and your strategies to deal with changes that you hadn't expected.

Q: Are there other developments in supply-chain management that you are excited about?

Shapiro: One situation that I ran into recently that has some of the same spirit is to use modeling tools to do data mining on retail point-of-sale (POS) data. Some of my colleagues at MIT have been doing research on an approach where you simultaneously cluster, say, 50,000 retailing products. If you take information from 500 stores, that is a huge amount of data. But maybe you can cluster it so that, from a statistical point of view, you look at only 100 or 500 distinct families of products. All the items in a family would behave the same way in terms of how people buy the products. For example, you might cluster products that are seasonal or that have no season. So you go from this fantastic complexity of 50,000 SKUs in 500 stores to something that is easier to understand. The human mind can grasp it. This approach also is very useful for forecasting because, if you better understand how all the items in a particular family behave in terms of demand, then you can develop better forecasts. And inventory management policy also is very much a function of what those demand patterns look like-you have slow-moving items and fast-moving items. You then can base decisions on when to order, how much to order, on very good analytics. The innovative thing here is that the data mining is using point-of-sales information. A lot of retailing companies have POS data but it is not clear what they are doing with it. That was part of the "looks like a good idea so lets buy it" trend of the '90s. Some of it actually was a good idea; it just needed some more work.

Personally, I recently have been drawn to looking at how we can relate ideas of organizational behavior to rational tools that would help us run companies better. There is a large quantity of literature on this with some really interesting ideas that, to a certain extent, pre-dated ERP systems and supply-chain network optimization. What these ideas suggest is that companies need to do business process expansion to exploit IT and planning tools. We see some primitive versions of this in some companies. Actually, the oil companies in their own way have been doing this for a long time but most industries have not gotten to it yet. So, what I mean by this, is that what companies need to do to better manage their supply chains is to create what you might call a supply-chain team. This would include a supply-chain manager who is responsible for holistic analysis of the supply chain. And this type of analysis needs to be done on a repetitive basis. Senior management has to insist that managers who are executing the supply chain participate appropriately in these supply-chain reviews. For example, for tactical planning you might do it once a month and the supply-chain team managers would have to devote maybe four to 16 hours a month to understand how what they are doing fits in with what everyone else is doing. They would discuss problems that any member of the team sees coming up and discuss potential solutions-rational solutions, not "I get my way because I can scream louder than you." So as problems come up and you have to make concrete decisions you are going to make much better decisions than if just suddenly had to react.