On its first day of operations in 1914, Philadelphia-based Tasty Baking Co. produced 100 cakes. Now it bakes nearly five million cakes, donuts, cookies, and pies every day and generates sales of $270m annually, making it the nation's fourth-largest bakery. The company is best known for its Tastykake brand and for the individually packaged, lunchbox-sized treats that it introduced to the market in the 1930s.

Scheduling production of Tasty Baking's 176 different products is a challenge in complexity. "We can't produce every product every day because of constraints related to allergens, colorants and so on," says Jeff Marthins, demand manager. "Some items we will bake only every other day or only once a week." Since Tasty Baking products have a shelf life of only eight to 21 days, making the right product in the right amount is crucial to product freshness. "Any excess inventory means problems with aging products and we obviously don't want that," Marthins says.

Also because of the short shelf life, "hours matter" when it comes to distribution, he says. Tasty Baking distributes finished goods from its two plants in Pennsylvania to 40 sales centers around the country via its private fleet of trucks. From these hubs, owner-operator route salesmen deliver products to convenience and grocery stores. Each sales center receives from one to three truckloads daily. Big-box retailers typically are served direct, says Marthins.

To ensure that it produces and delivers the right products, Tasty Baking needs reliable forecasts. Orders are shipped within two days of receipt, but production scheduling is frozen two weeks out, so manufacturing is largely based on projected numbers. Before 2004, these numbers were pulled together without a formal forecasting process using Excel spreadsheets. "This resulted in a lot of scrambling around if the order did not come in as expected," Marthins says.

In 2004, Tasty Baking implemented SAP as its backbone enterprise system. This major project prompted the company to look more closely at a number of issues. "It was pretty obvious as we prepared to go into SAP that we needed a demand forecasting process," Marthins says. After going live with SAP, the company attempted to implement a forecasting tool built into the R3 model, "but that just didn't seem to fit our needs," he says. One reason was that it required users to "understand a lot of the statistical formulas, which would have required a good deal of training," he says. "Our regional managers can understand Excel and read spreadsheets, but getting them to work with statistical formulas would have been quite a hurdle."

At that point the company began looking for a forecasting software package. Tasty Baking had several criteria that a solution needed to meet. "First, it had to easily integrate with SAP," Marthins says. "Also we needed a solution that was Excel-based since most of our guys already knew how to work with that."

Space constraints were an issue, so the company did not want a web-based solution that would require another server. Additionally, it was looking for a solution that was geared for a company its size and one that had a fast response rate and was easy to use. "Obviously, after coming off a major implementation, we wanted something that was economical, but that also would grow with us and offer regular updates," Marthins says. We didn't want something that we would install and that would be the end of it."

After looking at several vendors, Tasty Baking decided on the ForecastX Wizard from John Galt Solutions, Chicago. "We thought John Galt was the best fit for us for a several reasons," Marthins says. "It is Excel-based, so very little training was required for our guys. It has a quick response time so we are able to create a forecast much faster-we now can generate our basic statistics in less than a minute-and it was economical and just right for our size."

Marthins liked John Galt's "walk, drive, fly" approach that enables companies to start where they are. "In our case, I think there should have been an additional step labeled 'crawl,' he says. " But we are in the walking stage-maybe close to driving."

Tasty Baking now creates a new, 12-week forecast every month. Marthins says it is not a true 16-week rolling forecast "because we only do 12 weeks at a time, but we always keep the current month in hand, so when we first load it we have 16 weeks, then at the end of the month we are down to 12," he explains.

The statistical forecast created by ForecastX Wizard is easily translated into an Excel format and distributed to Tasty Baking's regional mangers, Marthins says. The managers review the projections and recommend changes based on their market knowledge, such as a planned promotion by a retailer that may not have been in the data available to the software. "We always have something on promotion and that complicates the forecasting task in a big way, which is why it is important for the regional managers to be able to tweak the statistical forecast," Marthins says. Having the statistical forecast as a base, however, is a big improvement in the process. "As recently as six months ago, the regional managers would just get a blank spreadsheet that they would have to fill it in with all the numbers," he says. "With John Galt, we give them a base forecast and they just have to analyze it, rather than populate the data. That alone has cut the time they spend on this task by 75 percent."

Another area where the John Galt system helps is with new product introductions. Tasty Baking introduces new products four times a year, with three to five new products each time. "Whenever you introduce a new product, you have to consider the impact of cannibalization," Marthins says. "For example, we have a core item that is a chocolate cupcake with chocolate icing. Then we introduced a 'chocolate lovers' cupcake that also had chocolate filling and a dark chocolate stripe on top. It is unlikely a consumer would buy both, so we have to try and figure out how many will switch from the core item to the new one." Before, managers had to make this calculation on their own, but now the John Galt solution forecasts new products based on historical launches of similar products. It also calculates a percentage decrease for the product that is expected to experience some cannibalization. "We can now generate a forecast for a new product in less than five minutes," Marthins says.

Tasty Baking supports its forecasting process with a weekly consensus meeting. The company does not have a formal sales and operations planning process, but this gathering is similar in purpose. "This is an upper level meeting, mostly vice presidents and directors from across the business," explains Marthins. "Everyone is able to review the forecast and have input on any constraints they see that might impact production schedules and demand for the upcoming 12 to 16 weeks. Decisions then can be made on any action that may be required. When we leave that meeting everybody has agreed to one set of numbers and that is what gets loaded into SAP," he says.

The John Galt software also is often used during the consensus meeting to do "what-ifs," Marthins says. "If an executive asks during the meeting what the impact would be of taking a 2 percent increase on Category B products, for example, we can enter that in Wizard right there and have an answer in less than a minute. It gives a graphical view of what the impact of that change would be on our business."

In the first year after going live with the John Galt ForecastX Wizard in 2005, Tasty Baking reduced its finished goods inventory by 16 percent, says Marthins. Part of the inventory savings was due to better management of safety stock. The company adopted a dynamic approach to calculating safety stock, changing the level weekly based on the forecast. "If the forecast goes up, so does our safety stock requirement and vice versa," says Marthins. To make this calculation, the company feeds data from John Galt into SAP, which calculates required safety stock for each product. "I think it is pretty unusual to have a flexible safety stock number that changes every week," says Marthins. "This works for us because it helps us control product aging. When things slow down, we want our safety stock to decrease. This really lets us flex our supply chain in line with consumer patterns."

In the second year of using John Galt, the company held inventory steady while improving its order fulfillment from 97.7 percent to 98.4 percent. "Another quantifiable result in 2006 was a reduction in our MAPE [mean absolute percentage error]," he adds. "Our forecast error percentage dropped from 25.6 to 21.4, so our forecast accuracy definitely improved."

Recently, Tasty Baking upgraded to the Premium version of the software, which will allow it to begin forecasting at the SKU level. "The Premium product will allow us to do a bottoms-up forecast instead of a top-down forecast," Marthins says. "If a region sells a lot of cream-filled products, the manager will be able to drill down into individual SKUs within that category and make adjustments, which will all roll up to the final number." The company hopes to realize savings on packaging as a result of this change since packaging is SKU-specific.

Other projects for 2008 are to more fully share forecasts with vendors and to develop vendor managed inventory programs with larger suppliers.