Though its business sprang from an accident, nowadays Kellogg Co. leaves little to chance. One hundred and fourteen years after W.K. Kellogg's botched granola-baking session resulted in wheat flakes -- and soon after, the flagship Corn Flakes, the company has turned heavily to IT for planning, production and distribution. This month it is turning part of its forecast over to demand sensing software to better deliver the right products...
Based in Battle Creek, Mich., the maker of other iconic cereals such as Rice Krispies -- and of snack and frozen food lines that include Cheez-Its and Eggo waffles -- pushes a huge amount of inventory through its global supply chain. With $13 billion in sales last year, it operates 104 plants and warehouses on six continents. "Demand planning and forecast accuracy are always key topics of discussion," said Richard Dregne, the company's vice president of sales and operations planning (S&OP).
Kellogg went live this month with Demand Sensing software from Norwalk, Conn.-based Terra Technology LLC. Demand Sensing is essentially a pattern-recognition system that identifies which demand signals, such as inventory and sales data, are best at predicting demand. It then creates a forecast that replaces those prepared by demand planners with their usual toolkit of Microsoft Excel spreadsheets and ERP-based demand-planning software, SAP Demand Planning (DP).
Kellogg hopes Demand Sensing will give it a better handle on "where in our warehouse network product needs to be to best meet customer needs," Dregne said.
S&OP tools bolstered by demand sensing software
In the fall, Kellogg entered the early phases of a parallel run of Demand Sensing, which Dregne said provided approximately a 30% improvement in location-level inventory accuracy compared to the company's existing forecasts. The effects of improved short-term forecasts ultimately will cascade back through Kellogg's production planning for further savings and inventory optimization.
"We saw the same improvement with the pilot," Dregne said. "The degree of improvement is directly proportional to the forecast lead time. Terra's Demand Sensing is focused primarily on the next six weeks with the biggest improvements in the next one to three weeks. This allows for a significant change in how we deploy to support demand at the location level and the corresponding inventory we would carry in those locations."
More on demand sensing software
Understand the hardware side of demand sensing
Read about trends in SCM software
Learn about other types of manufacturing business intelligence
The company uses SAP Advanced Planning & Optimization (APO) and SAP DP software that feeds into a Manugistics production planning system, though it expects to move the latter to SAP in the future. SmartOps Corp. inventory optimization software helps to predict safety stock requirements by location and "window of time." Unlike demand sensing tools that work with point-of-sale data, Kellogg's pulls in open orders along with the company's forecast. The company is considering piloting Terra's Multi-Enterprise Demand Sensing (MDS), which "adds more indicators of future demand," Dregne said.
"The forecast is pretty much coming out of the software provided by Terra and directly integrated to Manugistics," said Efrain Altamirano, Kellogg's manager of IT projects. "The information is then brought into SAP Demand Planning and the SAP Business Warehouse analytics tool for further analysis."
The Terra deployment will be widespread. "We have a large number of locations that are going to be operational across our direct store delivery [DSD] system and our warehouse system," said Mike Wojcik, senior manager of demand planning processes at Kellogg. "Our demand planners will have this coming in as a report for their forecast" alongside their current six-week forecasts, he said. "The demand planners will see the forecast so they can compare the Terra forecast to what they had predicted."
Focus on data cleansing and change management
Much of the deployment work involved data cleansing, according to Dregne. "Clean data is important to create accurate forecasts. In the early stages of implementation, you learn very, very quickly how clean your data is. It could be duplication. It could be items for a certain customer that we didn't have visibility into." Kellogg provided a representative data sample for Demand Sensing to use for its calculations, a process Terra helped with.
Kellogg also worked closely with Terra to explain its distribution scenarios so its networks could be accurately represented in the model. "We have some unique distribution characteristics," Dregne said. "We operate a connected warehouse network, a frozen assets network, and also a DSD network."
Terra CEO Rob Byrne elaborated on the modeling process. "The issue here was that they did not have historical order-creation dates stored for DSD orders, so we had to estimate when the orders were placed, as order-creation date is a required field," Byrne said. "Going forward, those dates are available."
As with any system change, Dregne anticipates some training needs. "The Terra solution replaces the Demand Planning forecast in SAP. To get the maximum benefit, we need to drive the acceptance of using the new forecast rather than simply overriding it. There will be a lot of training activity with the supply planners to ensure they understand the types of changes they will be seeing and are comfortable that the data coming out of the Terra tool is sound."
With the parallel run consistently demonstrating a 30% improvement in forecast accuracy, Dregne is confident about going live: "We're very optimistic about where we are at."