The continuing shift to demand-driven or "pull" supply chains has intensified business's focus on demand management-the science and art of understanding, predicting and even shaping customer demand. The reason is clear: As companies restructure operations to make their supply chains more customer-responsive, their reliance on demand signals grows.

A good reading of the tea leaves translates into synchronized supply chains with little unsold inventory and more complete, error-free orders, benefiting the bottom line. AMR Research, Boston, has quantified this effect, estimating that profits go up 1 percent for every 3 percent improvement in perfect order fulfillment, while a 10 percent improvement plumps earnings by 50 cents per share.

This potential gain is not easily realized, however. An explosion in the number of new product introductions, shorter product life cycles and more special-events selling make today's customers notoriously hard to predict, adding complexity and risk to the demand management challenge.

"The dilemma for most companies is that increasingly fewer supply chains have normal demand distributions," says Lora Cecere, research director at AMR. As a result, she says, the value of traditional statistical forecast modeling "gets increasingly limited each year."

Moreover, the combination of lean inventory management and global sourcing of parts makes companies less able to recover from planning mistakes. "In the past, if you missed badly on the forecast, you could run to the plant and ask them to slot something in," says Ann Grackin, CEO of ChainLink Research, Cambridge, Mass. "Those days are gone."

Indeed, constrained supply and capacity is a second key factor driving companies to improve demand management capabilities, says Ian Watson-Jones, supply chain planning consultant at IBM Global Business Services, Armonk, N.Y. "We are seeing a lot of supply constraints in different raw materials, components and sub-assemblies, depending on the industry," he says. "We believe that's one reason our work in this area has increased noticeably over the past six months, where before it had not been a focus of investment."

This is a driver for IBM internally as well. "Our current focus is more around coming up with what is really a constrained demand plan that maximizes profitability for the corporation," says Joe DiPrima, manager of supply chain planning and optimization for the IBM Integrated Supply Chain. "We want to understand where the supply constraints are and make sure that we apply those supply constraints to get the right product mix. That's different from what did in the past, which was using all sorts of different modeling techniques to figure out the future unconstrained demand plan."

The next advanced step is to then begin shaping demand, DiPrima says, "seeing what levers we can pull to make sure that the plan that we have can support the customer demand that is out there. There are things we can do from a pricing and promotions standpoint to try to generate demand around the supply that we have. That's the focus that we are on now and, to my mind, this really is what defines the difference between demand planning and demand management."

IBM has long used a solution from i2 Technologies, Dallas, and recently upgraded its global operations from i2's Demand Planner to its latest, more advanced Demand Manager solution.

	"The dilemma for most companies is that increasingly fewer supply chains have normal demand distributions." - Lora Cecere of AMR Research

Paula Natoli, director of solutions management, at JDA Software, Scottsdale, Ariz., agrees that business is at the beginning of an evolution toward true demand management, characterized by intelligent demand shaping. In retail, particularly, "the promotion piece plays a heavy, heavy role in that," she says. "How can I help determine what my customers are going to buy in order to hit my planned revenue and margin targets?" JDA has a number of solutions in this area, including pricing and revenue optimization acquired with its recent purchase of Manugistics.

SmartForecasts, from Smart Software, Belmont, Mass., also has a special-event modeling tool, says President Charles Smart. "We actually have had this capability for quite a while and it has worked very well for customers," he says. By looking at demand history of products, down to the SKU, along with the dates of past promotions, "we can measure the effects statistically and automatically," says Smart. "If the same trade promotion is expected in the future, we can incorporate a promotional factor into the demand forecast and predict how much that trade promotion will increase demand, either in absolute numbers or as a percentage. This is extremely helpful to companies, especially in CPG and retail, where a lot of demand is driven by promotions. Instead of scratching their heads and crossing their fingers, companies can have a pretty good idea in advance of a promotion as to what demand will be, apart from other trends like seasonality that may be operating simultaneously."

The evolution to demand management also involves more market analysis, often using outside data sources, says Grackin. A great example of this is CVS pharmacies, which regularly receive data from a healthcare industry service showing emergency room visits by region, she says. "If an emergency room in Miami is getting an influx of people who are sick from mosquito bites, area stores will know to stock more bug repellent."

Short Cycle S&OP
Consensus decisions about how to meet and shape demand increasingly come out of a corporate sales and operations planning (S&OP) process. "S&OP is about understanding projected demand along with financial implications for the company and the impact on inventory and manufacturing, and to compare that against any constraints at your suppliers, so you can agree on how to move forward to achieve corporate goals," says Cyrus Hadavi, CEO of Adexa, Los Angeles. "The demand plan is the first step in that process."

"The move toward S&OP is a recognition that forecasting shouldn't happen in isolation," adds Natoli. "We see companies taking a strong initiative to collaborate internally on a consensus forecast, whether that is through an S&OP process or in some less formal way."

According to our experts, however, the traditional monthly schedule for S&OP meetings is not sufficient to manage demand in today's dynamic environments. "A month is too slow," says Ajay Chidrawar, director of solutions marketing at i2. "You have to be able to reassess frequently and quickly employ demand shaping levers." i2 endorses a "plan, do, check, act" approach, he says. "For example, say six weeks into a season, you realize something is not selling. Our tool allows companies to optimize their response based on such factors as whether they want to run this inventory to the end of the season or its price elasticity. "They may drop the price 10 percent for the next two weeks, check the response, then drop the price again," Chidrawar says. On the other hand, he says, if a company is running short of a particular option or model, the application lets them see the impact of other levers, such as running a promotion on alternate items.

IBM used to review a lot of this data monthly, says DiPrima, but "depending on what hardware brands are in play, we have gone to weekly or bi-weekly sales and operations planning. We also run daily cycles inside of that, so really we are refreshing demand information daily," he says.

The ability to constantly readjust demand plans also plays into flexible manufacturing strategies many companies are trying to adopt, says Robert Byrne, CEO of Terra Technology, Norwalk, Conn. "Companies in the consumer packaged goods area are trying to shrink their manufacturing cycles and lead times by scheduling only a day or two in advance," Byrne says. "To do that you need a very good short-term demand forecast that reflects all of your current information. We have customers that are publishing a new forecast every day, really reflecting all the information from yesterday, including orders that came in and shipments that went out." Forecasts with a horizon of a few days naturally have fewer errors than ones looking out three weeks or more, he says.

Ironically, however, this trend intensifies the need to have accurate long-term forecasts for raw materials supply. "To be able to schedule production of whatever you want tomorrow or the day after, you have to have the materials on hand," says Byrne. This goes for packaging materials as well, which often have fairly long lead times, he says.

To achieve both these goals, different forecasting methodologies are required-a challenge that permeates demand planning. Whether companies are dealing with long-term vs. short-term horizons, slow-moving vs. fast-moving products, or core stock vs. promotional items, the flexibility to employ different approaches is essential.

"Retailers like Wal-Mart or Target will forecast demand for everyday household items like paper towels very differently from how they forecast apparel," says David Johnson, vice president of replenishment and allocation at JDA Software. JDA has a toolbox of different statistical methods and systems that can discover and pick the right method to apply to the specific demand patterns of an item, he says, "whether it is a slow mover or a fast mover, a seasonal product or a lumpy product. But demand also is driven by what replenishment strategies and deployment strategies are in place," he says, "so you have to be able to consider replenishment strategies and shipping patterns as well."

Unpredictable Demand
Adexa offers what it calls a "Best Bet" tool that automatically tries many different forecasting methods, says Hadavi. "Based on the objectives that the user states, it comes back and gives you the method that works best," he says. "This enables companies to deal with very unpredictable environments as far as demand is concerned," he says. "You can experiment with all sorts of different factors."

Increasingly, companies want solution providers to help them tailor a mix of different approaches to specifically match their business, says Kai Trepte, vice president of sales and service at John Galt Solutions, Chicago. For one of its customers, Galt created a methodology to assess a product's "forecastability," or how easy it is to accurately forecast demand. This assessment can then be used to set business rules, Trepte explains. "If you have a very valuable product that is not forecastable, you might require a collaborative step to get a better forecast; if it is not forecastable but of low value, you might decide to produce a certain quantity to hold," he says. "Using these kind of prescriptions that are very specific to a company's operations is extremely useful."

For some customers i2 provides tailored forecast optimization as a consulting-type service. "Clients don't have to be using i2 technology," says Chidrawar. "We go in and analyze their forecasting processes and use our domain expertise to apply more sophisticated techniques. It's really about understanding all the factors particular to that business that are driving demand and how to tweak the model to get a more accurate picture. "

Very often customers would like to do demand planning at the SKU level, rather than at the product and family levels, says Hadavi. "It may not be too hard to predict how many sweaters you will sell, but when you get to questions of what color and style of sweater will sell in what store or region, it becomes much more challenging. Unless you can address these issues, there isn't much gain from the demand planning process."

This can be particularly important when tracking promotions, says Natoli. "At a granular level, you may find that one distribution center or sub-region is overperforming where another is underperforming. At an aggregate level, you can think you are doing great, but by the time you get to the end of the promotion, you will have an imbalance of stock and lost sales."

A continuing challenge is how to manage the huge amount of data that results from this desire for a more granular forecast, coupled with the vast and growing number of SKUs. "It's a fairly recent development that we have had enough memory and hard-drive space and processing speed to throw at these problems to make it theoretically feasible to do SKU forecasting," says IBM's Watson-Jones. But being able to do something doesn't necessarily mean that you should. "There is a real risk of getting bogged down in the details and not achieving the accuracy you were hoping for," he says. "You think better accuracy will automatically come from having that level of detail, but that's not a reliable assumption."

Outside of retail, some companies are approaching this problem through attribute-based forecasting. Lenovo, which bought IBM's personal computer business and which is still being supported by IBM's Integrated Supply Chain, provides an example. "Instead of planning the actual part number that customers order, we plan how many 40-gig or 60-gig or 80-gig hard drives we expect to sell, how many different processors and so on," says DiPrima. "Before, we were looking at each combination of those technologies and there were just too many items to accurately or effectively plan." The result of this change has been far fewer items to manage and far greater forecast accuracy, he says.

"This is another area where the capability to do this type of forecasting just wasn't available in software tools a few years ago," says Watson-Jones. "And it's not just IBM that is using it. We are seeing a lot of interest among our consulting clients."

Three Short Lessons In Demand Planning

Company: Hercules Bulldog, Clearwater, Fla. Business: Global distributor of hydraulic seals and parts Goal: Improved forecasting of individual parts Solution Provider: IFS Hercules Bulldog needed to improve forecasting for the many different parts that it carries, says Chris Breierly, purchasing manager. "Our prior system could give us a minimum stocking point or an average monthly demand, but we still had to do a lot of calculations almost by hand," he says. "We struggled with it and the results just weren't as good as we wanted them to be." Hercules already was using an MRP system from IFS and decided to implement IFS Demand Planner. "The IFS solution allows us to forecast each part individually and we can use several different forecasting models, which is a nice feature," Breierly says. "We have categorized products so we can forecast some categories with one method and then use a different method for other categories." Different forecasting methods also can be used for different sites. "For example, our satellite facilities are much smaller and have more intermittent demand than our main facility in Florida, and I am able to separate those facilities and use a forecasting method that is better suited to the wide variability in their sales patterns," he says. Also, the IFS solution is able to filter out spikes in sales that do not represent typical demand, Breierly says. "We might sell 50 pieces a month of something for five years and then, out of the blue, someone shows up and buys all 800 that we have. The IFS solution allows us to filter out those kind of exceptions. This is a nice feature and certainly something that wasn't in our prior software." Overall, Hercules' inventory management has significantly improved. "When we started with IFS, we typically would be out of stock on around 2,200 of the 26,000 items we carry. Now, at any given time, we are out of about 450 or 500 items," he says. Company: Vicor, Andover, Mass. Business: Manufactures custom power conversion components Goal: Reduce raw materials inventory and improve delivery performance through better demand forecasting Solutions Provider: SmartForecasts Vicor makes more than 8,000 highly configured products involving such options as input and output voltages and power levels. The company sells to about 10,000 customers worldwide. "The first thing to understand is that we are a make-to-order business," says Doug Richardson, vice president of information systems. "We don't carry finished-goods inventory. Second, the components that we use to make our products have longer lead times than our quoted lead times to our customers. That creates a situation where we have to rely on our forecasts for decisions about purchasing components." In the past Vicor has dealt with this by buying lots of raw materials, Richardson says. "What we wanted to do was minimize raw materials inventory, while ensuring that we have the right components on hand to make whatever we need to make." Vicor selected SmartForecasts in 2001 to work with a PeopleSoft ERP system that also was being implemented. "We looked for a moderately priced package that could handle our base-line business, which is more statistically forecastable, as well as the variable input we get from sales. Before, we were doing this pretty much manually and not very well," he says. SmartForecasts enabled Vicor to integrate its statistical and sales forecasts and it now is able to adjust this figure as real demand becomes known. "For example, Richardson says, "if the sales folks forecast that we will sell 1,000 units this month, when we get 1,000 units booked as orders we adjust the forecast going forward." Richardson says this approach, along with the PeopleSoft implementation, has helped Vicor improve on-time delivery, "which was a big issue with our customers," while significantly reducing raw materials inventory. Company: Poclain Hydraulics, Paris, France Business: Makes hydraulic motors Goal: Improved production planning and sourcing through a centralized picture of demand Solutions Provider: John Galt Solutions Poclain has manufacturing facilities in the U.S., the Czech Republic and at its headquarters in France, which support 13 sales subsidiaries around the world. "A great deal of coordination is required to properly plan which plant will manufacture which product for a given customer," says Bob O'Neill, North American planning manager. "In addition, we source raw materials on a global basis and have to designate which of the three factories will receive which materials, so a lot of coordination is required there as well." Factory allocations are based on a number of factors, including a product's time sensitivity and cost competitiveness. "We used to manually plan products into factories and raw materials purchasing, both based on sales predictions," O'Neill says, "It was a slow and painful process. And, since salesmen are eternally optimistic, very often our numbers were missing by large sums, both high and low. We really needed to find a way to statistically forecast what our existing client base would be doing in the next 18 to 36 months and to calculate probable new business." Poclain implemented The Atlas Planning Suite demand management solution from John Galt a little more than a year ago. Since then, forecast accuracy has improved 14 percent, says O'Neill. Over the same period, inventory has gone down significantly and the company has improved on-time performance. "For the past year, our on-time delivery performance has been in the high 90s, which is just phenomenal given the engineering and assembly that goes into our product," O'Neill says. "At the same time, we have reduced inventory to the lowest it has been in three years, so right now our planning is more precise than has ever been." These improvements have improved cash flow, "allowing us to go out and find new customers and invest in new designs," he says.