Demanding a Better Forecast

A demand forecasting methodology that blends input from multiple sources is the right formula for one chemical manufacturer.

When software crunches historical data then serves up a useful demand forecast, that’s good. When you supplement that forecast with the insight of people who understand your market, that’s even better.

Hexion Chemicals, a Columbus, Ohio-based manufacturer of chemicals used for binding, bonding, and coating, can attest to that.

In the four years since Hexion began using the Zemeter demand forecasting suite from Supply Chain Consultants (SCC), Wilmington, Del., its forecasts have grown more accurate.


“Forecasting has improved by about 25 percent from the baseline,” says Paul Johnson, Hexion’s director of supply chain planning, North America.

But a big question still hangs in the air for companies such as Hexion that use demand forecasting solutions: how much credence to give to each source of knowledge that contributes to the forecast?

The answer: it depends. One month, a demand planner might have the best sense of upcoming customer demand; another month, it could be a sales manager who has the inside track; and the next month, a statistical analysis might offer the best view of the future.

Companies using Zemeter can take advantage of collaborative forecasting, a process that bases forecasts on input from multiple sources. Until recently, however, that process did not include a way to make sure the most trustworthy source had the biggest impact on the forecast.

“A hierarchy existed within these collaborative forecasting levels,” says Sujit Singh, vice president of supply chain solutions at SCC.

A company might, for example, produce a statistical forecast and give the resulting figures to sales representatives to tweak. The reps might pass their results to the sales manager, who would make further changes and forward those results to the demand planner, who would modify the figures once again.

“The business process that companies used to determine the final forecast was simply to select it from the last level in the forecasting hierarchy,” Singh says.

While Zemeter used input from all sources, and most companies operate with similar hierarchies, “the last level always won,” he notes.

This method worked well for most SCC customers, especially when compared with the results they got without using forecasting software.

Such was the case for Hexion, whose forecasting hierarchy varies from one line of business to the next. Its demand forecasting process often runs from statistical analysis to demand planner to sales rep to sales manager. And like many other companies, Hexion’s final forecast was usually the one from whomever put data in last.

But Hexion wondered if this was the best way to reach the final figure. “Maybe the sales rep had the most accurate forecast all along and we’re ignoring that data in the hierarchical approach,” Johnson says.

Look at Accuracy

“Our clients started asking us for a way to look at the accuracy of these individual inputs—the sales rep’s input versus the demand planner’s input, for example,” Singh says.

Clients also requested a way to combine the inputs once the system calculated those measures, giving each input more or less influence on the forecast based on its accuracy score.

In response to these demands, SCC produced Adaptive Collaboration, a feature available in Zemeter 4.0.

The new module generates forecasts from multiple inputs, which it weights for accuracy. Periodically, it compares each source’s recent forecasts with actual performance for that period, calculates its accuracy, gives each source a weight, and applies those weights the next time it runs a forecast. Weights go up or down each time the engine runs, based on each source’s latest performance.

By combining different inputs this way, SCC hopes users will gain greater forecast accuracy. Companies that have used Zemeter for a few years won’t see the same dramatic improvements as companies implementing forecasting software for the first time.

But Singh expects even existing customers to improve their forecast accuracy by 3 to 5 percent.

That’s a significant boost. “Small improvements in forecast accuracy can lead to lower inventories, which frees up working capital,” he explains.

Hexion has not yet implemented Adaptive Collaboration, but Johnson is considering it.

“In looking out one or two months, the sales team’s input should be more accurate than statistics, especially considering situations such as hurricanes Katrina and Rita,” which affected the chemical market in ways statistics couldn’t possibly predict, he says.

But it’s hard to tell if the sales team will do as well predicting demand three or four months out. If Adaptive Collaboration works when Hexion tests it against the old method, “it should show the best of all worlds,” he says.

Who’s to Blame?

But Johnson does express some concerns about the new forecasting technology. With the current method, if Hexion makes a decision based on projected demand, “we can always go back to the person accountable for that forecast,” he says.

But if the new system blends inputs to produce a composite forecast, it’s harder to know where to lay blame if the prediction turns out badly, or give credit if it turns out well.

“If no one is held responsible for the forecast, people might not put as much effort into it, which could be dangerous,” says Johnson.

Hexion and SCC have not yet discussed this question. SCC’s new module, however, does help companies determine which forecast was the most accurate for different time frames.

“That is one key issue we always wonder about,” Johnson says. Next year, he plans to investigate the new module and determine if it’s right for Hexion.

An early warning system is another Zemeter feature Johnson is exploring for the near future.

“When a metric meets a certain criterion, the system automatically sends an e-mail to the person responsible for the relevant process,” he explains.

If inventory of a certain material falls to a pre-determined level, for example, the system alerts the appropriate scheduler or buyer before a shortage can interrupt the manufacturing process.

In addition to its forecasting capabilities, Zemeter offers numerous other supply chain management modules, including a business analyst for analyzing historical data; inventory planning; supply planning; finite scheduling, which manages day-to-day manufacturing operations; a capable-to-promise engine; a knowledge repository, which preserves information about why the company took certain actions; and a supply chain historian, which helps identify trends hidden in historical data.

Along with Zemeter 4.0 and its new Adaptive Collaboration feature, SCC recently introduced a version of the supply chain suite designed for small- and mid-sized companies—those in the $100-million to $500-million range. Called Zemeter S&OP (sales and operations planning), it is an out-of-the-box solution that users can implement in five steps, with a fixed price for each.

Though companies often implement supply chain management software using the “big bang” approach—spend one year on implementation then introduce new processes all at once—this might not be the best approach for demand forecasting technology.

“In our experience, that doesn’t work well. What does work is implementing in small steps,” Singh says.

Five Steps

Implementing Zemeter S&OP involves five steps. “Every step has deliverables and benefits. And payments are tied to completing each step,” Singh explains.

Initially, SCC creates links to a company’s ERP system, pulls in data, and provides ways to analyze the data so companies can better understand demand variability.

“This helps companies clean up their data and get it in the right format. This step also allows companies to look at problems—a customer that always makes late shipments, for example—and begin to control them,” Singh explains.

Before moving on to step two, companies use the system to get familiar with its components. “Even at this early stage, companies can derive monetary benefits by reducing late shipments or doing manual forecasting, for example,” Singh notes.

Because each step provides a return on investment, smaller companies can reap benefits without committing a large sum up front.

“After you pay for step one,” Singh says, “every subsequent step can be funded from the success of the previous one.”

Leave a Reply

Your email address will not be published. Required fields are marked *