Terra Technology: Souping Up the Forecast

With real-time analysis to complement its long-range system, Campbell stirs up a recipe for demand-planning success.

After selling condensed soup for more than one century, the Campbell Soup Company knows pretty well how to forecast demand for classic flavors such as tomato and chicken noodle.

But even for these culinary icons, predicting demand down to the item and location level can be a challenge, admits Mike Mastroianni, vice president, North American planning and operations at Campbell, Camden, N.J.

Imagine what it takes to foretell the market’s hunger for individual Campbell products today. Several years ago, the company launched a major transformation initiative; part of that plan is introducing a cornucopia of new choices to the market.

As microwavable soups, green and purple Goldfish snacks, pop-top lids, and other innovations started hitting store shelves, Campbell found it had multiplied the variables it needed to consider in its forecasts. Would new products cannibalize older ones? Would competitors respond with novel offerings of their own? How do you predict future demand for an item that barely has a past?

“We recognized that in a much more volatile environment, with our current capabilities and tools, predicting future demand was going to be very problematic,” Mastroianni says. Along with historical information, Campbell needed to consider new factors, such as point-of-sale data, in its calculations.

Campbell used Manugistics’ Demand Management solution to create its forecasts, and that system still forms the solid basis of the company’s demand planning strategy. This spring, though, Campbell started supplementing Manugistics’ capabilities with a second tool—Real-Time Forecasting from Terra Technology, Orange, Conn.—specifically designed to increase forecasting accuracy in the short term.

Complementary System

Real-Time Forecasting complements existing demand planning systems by considering data they don’t, and using it to paint a more accurate picture of demand over the next four to six weeks.

Traditional planning systems look at factors such as seasonal demand patterns, rises and falls in average sales, and the expected effects of scheduled promotions, says Robert Byrne, president, Terra Technologies. But when unexpected activity occurs, those systems don’t adjust their forecasts for the near term.

“For example, if Campbell normally sells 2,000 cases a week, but just sold 20,000 at Shop Rite because it got the end aisle, that would kill Shop Rite demand for the next four weeks. That’s what our planning system is doing,” he says.

Using traditional demand forecast tools, the average rate of forecast error, even for a long-established product, may be 40 percent. “That error rate wreaks havoc with manufacturing schedules,” Byrne says. “It wreaks havoc with transportation planning, warehouse requirements, and inventory levels. Companies have to carry a lot of inventory because there’s so much uncertainty.”

Terra’s system uses data that a company already has at its disposal, pulling it together from several sources to gain a finer-grained picture of fluctuating demand.

“We look at forecasts and we look at recent orders,” Byrne says. “We ask, ‘Are orders coming in in a way that makes us believe the forecasts are right?’ If my weekly forecast is 5,000, and I only have 500 orders, that forecast of 5,000 is wrong.”

The software also considers how recent sales might affect near-term demand. “To most systems, each week is independent of all other weeks. But in consumer packaged goods, particularly, that’s not realistic,” Byrne says.

In the consumer market, an unexpected surge in demand this week often heralds a corresponding drop in the near future. “You eat only so much soup or use so much toilet paper,” he notes.

Mandatory Test Run

Officials at Campbell decided to implement Real-Time Forecasting after conducting a pilot. This test run is mandatory for every prospective customer, Byrne says.

“A company gives us three years of history, and we actually simulate the last year. We say, ‘If you’d been running our system last year for all your locations, here’s what you would have seen, compared to what you actually did.’ You can’t buy the software without being convinced that it works for your data.”

Campbell found the results of its pilot “compelling,” Mastroianni says. “We saw a dramatic improvement—about a 50-percent reduction—in item/location weekly forecast error.”

Terra has seen similar results in pilots with other customers, such as Georgia Pacific and Ventura Foods. “We cut error in half,” Byrne says.

Mastroianni says he was skeptical of the system at first. While Real-Time Forecasting might do a good job with mature products, he thought, it couldn’t make accurate predictions for items that don’t have a long track record.

To his surprise, “we saw substantial improvements across all the segments of our SKUs within the product life cycle: SKUs that were on the back end of the life cycle, new products that were being launched, as well as the icons in the mix. We saw dramatic improvements on all fronts.”

Feed the Plan

Running under the Microsoft Windows or Unix operating system, Real-Time Forecasting uses links to the company’s enterprise resource planning (ERP), forecasting and order management systems to obtain the data it needs each night, then feeds the results of its analysis into the planning process.

“From a demand planning standpoint, there’s no change to the company’s process,” Byrne says. “From a supply planning standpoint, they just have more accurate forecasts.”

At Campbell, after Real-Time Forecasting runs an analysis, it sends the results to the Manugistics planning system. “We strengthened the logic in Manugistics around dealing with close-in information and point-of-sale information,” Mastroianni says.

Running this kind of analysis, based on the latest information, is better than “executing against a forecast that was generated four or six weeks ago, based on the best information you had at that point,” he says.

It is not Terra’s goal to replace traditional demand planning systems, Byrne says. “We’re like a post-processing step to the forecast. Because our system only looks out six weeks, a company needs a forecasting system if it wants to plan further ahead. We don’t look at factors such as seasonality. We expect to deduce that from the forecasting system.”

With more accurate near-term forecasts, Campbell can better decide when to ramp up production as demand increases. It can determine, for example, whether a rise in orders this week simply means the need to increase production now, or the need to put out extra product next week as well.

“With the precision we’ve seen in this pilot, to the extent that we’re 80- percent accurate in that item/week, we can make that decision once and make it right,” Mastroianni says. When demand falls, the company can cut back production appropriately, with less risk of falling short on inventory.

Better Use of Carriers

Mastroianni also expects Real-Time Forecasting to help Campbell make more efficient use of its carriers. “If we have 80-percent precision in terms of what our day is going to look like for the next four weeks—every day, at every plant—we can estimate the number of trucks we will need. This allows us to crew our warehouse more effectively.”

More accurate forecasts should also help Campbell make better use of drivers’ time under the federal government’s new Hours of Service rules, Mastroianni says. “If we can adjust crewing, docks, and carriers more proactively, then we can expect improvements on turnaround time.

“The ancillary benefit is, at the end of the day, we will have less firefighting,” Mastroianni continues. “We will have more efficiency on the shop floor as a result of fewer fires, and we’ll certainly see improved service levels.”

Better inventory management is “a given” he says. “We’ll have reduced safety stock levels, and forecast accuracy is going to get the right product to the right place at the right time.”

Along with the operational improvements Campbell gains from Real-Time Forecasting, Mastroianni says he is pleased with the software’s ability to adapt to his special needs. “We have not come across an issue or made a custom request that Terra Technology wasn’t able to solve fast. They’ve demonstrated breadth and depth in this space, as well as flexibility.”

Leave a Reply

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