At one company where I worked, over the course of four years, the Materials team was able to improve inventory turns by 300% using statistical forecasting and other inventory management techniques. During that same interval, revenue increased by more than 50% without adding a single square foot for material storage. In addition, the value of excess and obsolete inventory (“rottage”) decreased from over 8% of inventory value to less than 0.5% because Sales and Operations coordinated very closely to avoid overdriving customer demand.
There are software vendors that sell extremely expensive demand forecasting software that will integrate with an ERP system. Initial costs to implement these systems include the purchase price of the software, training users and managers, setup and debugging, possible additional software license fees, hiring staff to maintain the software, and hiring staff to use the software. Thereafter, there is the expense of support contracts, renewing licenses, and buying and training for new releases. These expense elements can very quickly total into the hundreds of thousands or even millions of dollars. A small- or medium-sized company simply cannot justify such a large expense.
Economical, PC-based software – and more recently Cloud-based software as a service (SaaS) – with good user interfaces are readily available. These affordable systems greatly reduce one significant barrier to implementing forecasting tools.
I very strongly advocate that companies implement a Sales and Operating Procedure (S&OP) to manage both sales demand planning and order fulfillment performance at the executive level. Simplifying this complex topic, S&OP means that the CEO, COO, or site General Manager schedules the Sales, Marketing, and Operations heads to meet at least monthly to discuss their forecasts and commitments, and to track performance. Implementing S&OP is almost always a major cultural change for a company. Only the top executives are able to drive the deep cultural changes necessary, so their full commitment and active leadership is absolutely essential for successfully implementing S&OP.
During S&OP meetings, the VP of Sales should make commitments as to what the sales numbers will be in the coming months, with granularity by product line. He or she will also compare last month’s sales to the previous forecast. The rest of the executive staff will ask questions about major differences between the previous forecast and the actual performance and the reasons for deviations. In this way, there is closed-loop feedback and accountability for the customer forecast. Over time, surprises like missed sales or unplanned urgent sales orders diminish.
Before the monthly S&OP meeting, the VP of Operations should use statistical tools to independently forecast the customer demand based on actual past history. Common statistical forecasting software will automatically try 20 or 30 or even more different forecasting algorithms to arrive at the best forecast.
As an example, during the S&OP meeting, the Sales VP might forecast a near-term demand of 230 widgets. The Operations Director might say that the statistical forecast predicts a near term customer demand of 212 widgets. Then the CEO can lead a discussion of why there seems to be an overdrive forecast from the Sales VP that could result in excess inventory. Maybe there are very good reasons for this increase; on the other hand, maybe the customer demand forecast needs to decrease to some other figure. Over time, open discussion like this one among the Sales VP, Operations VP, and CEO will result in better economic decisions for the company.
Adopting S&OP supported by statistical demand forecasting often yields results within months. It is not unusual for the customer demand forecast to improve from +/- 50% variances each month to consistently within +/- 10%, or even better. As a point of reference, it might take an analyst half of a day to prepare the statistical demand forecast for 15 to 20 key items (or categories of items) for the coming month. As a practical consideration, the participants in the S&OP meeting are only able to strategically and operationally focus on a dozen or so major customer demand items or categories anyway. There is no need for an analyst to forecast hundreds of demand items to prepare for the S&OP meeting.
Improvements from implementing S&OP, including statistical customer demand forecasting, go a very long way toward assuring that Operations is only buying the inventory and paying for resources that the business truly needs to satisfy its customers. Air freight, expediting, and overtime expenses often drop sharply as a result, releasing more cash for the company to pursue other value added ventures.
Once the key decision makers in the S&OP meeting determine and publish the demand forecast, then the Operations VP should use that information to statistically calculate the demand for material. In practical terms, it only makes sense to perform this step for the most expensive inventory items and those with the longest lead times – the “Class A” items – plus possible a few other “Class B” (less critical) items. Statistical demand forecasting software could also easily forecast Class C and Class D items (readily available commodities), but what benefit is there to saving cash on material having a small financial impact to the company?
Depending on the nature of the products a company sells to customers, there may well be important options and variations that appear in the actual sales orders for the items for which the company prepares customer demand forecasts. It is certainly possible to use statistical material demand forecasting to prepare for those eventualities as well.
Good ERP systems have the ability to load planning Bills of Material and to use these BOMs to purchase raw material. For example, if the company forecasts that customers will buy 230 Widgets in the coming month, but is unsure which exact configurations customers will buy, it is possible to use a planning BOM in which the proportions of options are previously defined based on order history. By statistically forecasting the demand for Widgets and using the percentages in the planning BOM, the ERP system can automatically generate purchase orders for the estimated material requirements to satisfy the demand for the probable mix of options.
Since Class A inventory items require very tight purchasing control, the Operations VP might require that the planning manager adjust the actual purchase quantities individually and independent of the ERP system. Statistical forecasting software can help with this exercise by using the demand history for each of these Class A items to forecast the probable demand for the near term. In addition, if any other purchased items are physically large and consume a great deal of storage space, the Operations VP might also want to use statistical forecasting to manage these purchases and inventory to prevent consuming too much space to store them.
Once a company is accustomed to using statistical demand forecasting for individual purchased parts, it might take an analyst two working days to convert the customer demand forecast into actual purchasing quantities for the buyers to act upon. Using spreadsheet macros or ERP scripts, completing this task in half a day is certainly possible.
Like any other tool, tossing statistical forecasting software into untrained hands and into an unprepared organization is a recipe for disaster. At a minimum, an organization will need two people in Operations, a manager and an analyst, who both have a good practical understanding of basic applied statistics. The manager can know less about statistics and be less capable than the analyst, but does need to be able to distinguish nonsense from what is realistic. The company needs to guard against creating an esoteric priesthood of statisticians who can buffalo others into making wrong decisions and make themselves seemingly indispensable. (“Well, you had better give me a big raise or I’ll resign and your ability to control inventory will evaporate!”)
For a fast start, some companies can find people with the applied statistics skills they need within the QA department, particularly among those who have passed the Certified Quality Engineer or Certified Reliability Engineer exams, are 6-sigma Black Belts, have learned Design of Experiments, or already have strong analytical skills and are motivated to learn more skills with additional guidance or training. If the company will hire new staff, then young professionals with a math degree, or a major or minor in statistics can be good candidates. Since a small- to medium-sized company does not need full-time statisticians, hires such as these can make excellent material planners who can develop into effective master schedulers and beyond.
Starting from scratch, it is certainly possible to assign and train staff who are delivering some effective results into the S&OP meeting within a few months. The breadth and depth of their effectiveness, as measured by inventory reduction and operational expense control, will rapidly increase within the first year.
How to select software? There are a number of good software packages available at a cost of under $2000 per license. These reside on a laptop or desktop PC. Other systems hosted in the Cloud generally cost in the range of $20 to $40 per month per user, and you will probably need just two or three seats. Preferred software will be able to automatically download demand data from mainstream ERP systems and import data from spreadsheets. The software will have imbedded within it at least 20 to 40 different forecasting algorithms, and be able to assess and present a forecast that minimizes unavoidable statistical variances based on the data available. The software will also be able to display, for those who both need and understand them, relevant statistics to enable deeper analysis if that is ever required.A good S&OP process creates a strong platform for companies to improve their business results by using statistical forecasting. Inventory values and operating expenses will decrease. Executives will experience greater control over their resources, and will quickly learn why and how to exercise that control. The implementation costs and time to achieve effectiveness for statistical forecasting have decreased to levels that small- and-medium sized companies can pay back their investment in well under a year.