SAP SCM 5.0 provides functionality to improve forecast accuracy by using a combination of increased forecast model availability, safety stock computation, and system performance. Because of the sheer magnitude of planning run exceptions that result from poor forecast accuracy, I’ll provide several best practices and tips to help you achieve accuracy within acceptable parameters.

A planner experiences significant consequences when a bad forecast leads to inadequate replenishments. This scenario is compounded by a multitude of exceptions after running SAP materials resource planning (MRP) and distribution requirements planning (DRP). For example, a bad forecast triggers many backorders and some of them are cancelled, resulting in lost sales. Backorders are exceptions that force a planner to reevaluate the planning run and safety stock levels to meet unplanned demand.

Managing these exceptions is labor-intensive and typically requires resequencing or expediting replenishment objects (such as purchase orders, production orders, and transfers) to partially offset poor delivery compliance.

SCM 5.0 addresses common forecasting issues and creates more realistic demand plans in five steps (Figure 1). The five steps of the forecasting process are:

Step 1. Upload historical data to SCM-BW/BI InfoCubes

Step 2. Prepare and analyze the data

Step 3. Generate the base forecast

Step 4. Fine tune the base forecast

Step 5. Generate custom reports

Figure 1

Overview of the five steps to create a demand plan in SCM 5.0

The Five Steps

Step 1. Upload historical data to the SCM-BW/BI InfoCubes. Perform this step to identify the necessary attributes to load the data. This historical data can come from sources such as an external SAP BW 3.x/SAP NetWeaver BI system, a flat file, an R/3, or a legacy system. For example, specific details of the attributes consist of the product, location (plant/warehouse), region, and sales organization for which consumption data needs to be loaded. This is a typical design decision identified early in the project.

The SAP BW/SAP NetWeaver BI component, standard functionality in SCM 5.0, provides multidimensional data storage. Multidimensional data storage allows you to view the data at different levels and correct any discrepancies. This functionality identifies missing data or a potential outlier to improve data quality. SCM Advanced Planning and Optimization (APO) Demand Planning (DP) supports automatic outlier control so you can easily correct the historical data, as described in step 2.

Algorithms, standard in SCM 5.0, detect anomalies that would otherwise go unnoticed and fix them in a timely manner. You typically load historical data into SCM-BW/SAP NetWeaver BI through standard transactions, such as /SAP-APO/RSA1.

Step 2. Prepare and analyze the data. In this step, you correct any faulty historical data, outliers, or data distortion due to sales events. For example, one of the settings for an APO forecast model is the automatic outlier correction functionality. This functionality takes into consideration the average workday and then corrects the historical data if there is an inconsistency in the number of workdays in a month. The logic behind correcting historical data is based on the daily sales rate.

For example, because a one-time sales event does not have a pattern, it can distort the result of the statistical forecast. On the other hand, if your business has a predictable summer promotion, your SAP system detects and adjusts the forecast to reflect a seasonal pattern. You define the outlier parameters and maintain the forecast profile using transaction /SAPAPO/MC96B for like-modeling. Like-modeling is the approach used to create new data based on SAP profiles. Use the SAP profiles to accelerate and simplify new data creation and maintenance. You should perform thorough analysis to identify which existing product’s historical demand data you can use for new products.

Next, execute standard APO transaction RSATTR (attribute/hierarchy realignment run) to access the data realignment tool. This tool automatically corrects data altered by master data changes. You can set it up to run daily, for example, in the transaction RSATTR screen. Master data edits in the form of attribute/hierarchy changes lead to inconsistent forecast results. Analyze the data by slicing and dicing it at different levels. SCM’s Forecasting and Replenishment module helps to identify the null values, possibly due to stockouts, and to replace them with meaningful mean values.

When the outlier correction is switched on, users can see their corrections in red font, as circled in Figure 2. You activate the outlier correction by selecting the on check box in the forecast profile. This correction enables the system to run forecast calculation again with the amended historic data. See the sidebar, “Best Practices to Address Common Challenges of Forecast Accuracy,” below for data preparation best practices.

Figure 2

Outlier corrections are displayed in red font in the system

Step 3. Generate the base forecast. In this step, you identify the correct forecast model (statistical or causal forecast model) and then generate the forecast.

To identify the correct forecast model and generate consistent results, follow these four steps:

1. Identify and segment your products into the proper demand categories. For example, you can divide them into categories such as sunrise (products that are new to the business and are ramping up in volume), sunset (products with a ramp-down in volume), slow movers (products with sporadic demand patterns), and seasonal (products with specific demand patterns during certain times of the year).

2. Define business criteria associated with these demand categories. For example, you may establish a policy whereby all new product introductions are classified as sunrise products and a ramp-up rate is applied based on the expected growth rate within the first few months of the products’ life.

3. Make use of like-modeling product functionality. For example, the historical data of a product with similar sales patterns is used to create a like-modeling forecast for a new product. To facilitate this analysis, use the business criteria defined in the second step above.

4. Learn about the cause-and-effect relationship between different configuration settings. During the blueprint phase, perform unit testing and become comfortable with the cause-and-effect relationship between the different configuration settings available for automatic model selection.

The creation of the base forecast is fairly straightforward for most products. SAP SCM 5.0 offers automatic forecast model assignment functionality. However, depending on your product mix and the product life cycle attributes, identifying the correct forecast model can be challenging. In some cases, the applicability of automatic forecast assignment is overrated; the automatic forecast model may not yield the desired result. This is particularly true in the case of some scenarios, such as sunrise, sunset, and slow movers.

These scenarios require additional analysis and effort — such as identifying the correct like- modeling to use or performing several simulation runs to determine the correct forecast model — to ensure reliable results. As a rule, determining the appropriate forecast model for new product introductions poses significant challenges and you must take several factors into consideration.

If you rely purely on guesswork for selecting the best forecast model, high forecast variability usually occurs. In a worst-case scenario, this produces very low customer service levels for new products and obsolete stock for phased-out products. The net result on the inventory management side is that poor forecast accuracy leads to higher inventory levels to compensate for variability. Furthermore, the higher stock levels may have the wrong mix. The potential for additional losses increases due to obsolescence write-offs. To get the most out of this step, understand your data and the various limitations of automatic forecast assignment.

Step 4. Fine tune the base forecast. In this step, you adjust the base forecast according to new market information. For example, your competitor planned a sales promotion that has not been officially announced, but your district sales manager knows about it. Step 4 encourages collaboration with all relevant parties to ensure the accuracy of your base forecast. Collaborating with internal and external planning groups to fine tune the forecast can be very tricky.

For example, the sales and marketing team may have conflicting objectives with the manufacturing and operations group. Based on our experience, we’ve encountered several situations in which the sales team tried to increase the forecast to meet sales targets while the operations team tried to reduce it to minimize changeovers and match available capacity. This is an example in which a sound approach to change management and an organizational reward system structure is critical to success. To alleviate some of these issues, follow these guidelines:

1. Ensure interdepartmental communication by scheduling periodic sales and operations meetings in which the proposed sales promotions and production line scheduling plans are shared and discussed

2. Align departments to focus on collective key performance indicators (KPIs), such as first pass order fill, by realigning the reward system

3. Use automatic error correction features to enhance the base forecast. SCM 5.0 supports the automatic correction of errors in the forecast within the replenishment lead time. Various forecast error parameters, such as mean absolute deviation (MAD), mean average percentage error (MAPE), and root of mean squared errors (RMSE), assist in determining errors and misalignment. These parameters provide the necessary information to increase accuracy (Figure 3). In the example shown in Figure 3, the user is notified because the MAPE exceeded the preconfigured 40% threshold. This error prompts the user to investigate why the forecast accuracy is poor. The troubleshooting tips outlined in step 3 (Generate the base forecast) are applicable in this scenario.

Figure 3

Forecast error display

Step 5. Generate custom reports. As the planner’s maturity and experience increase, the use of custom BI reports further ensures consistent, higher levels of forecast accuracy.

These custom reports extend beyond the standard functionality outlined in step 4. Now, I’ll describe two custom reports, forecast accuracy and demand plan variance, in detail.

Forecast accuracy custom report. A BI custom forecast accuracy report compares the actual sales data with the forecast data. For example, you can compare the actual sales quantity for Jan08 with the forecasted quantity for Jan08 that was generated during the forecasting run of Aug07. The analysis in Figure 4 illustrates that the forecast accuracy for material BLADES is a low 51.2%. This indicates that the forecast issued in Aug07 for the month of Jan08 of 5,000,000 units is only 51.2% accurate, compared to the actual Jan08 sales of 3,360,000 units.

Figure 4

Forecast accuracy custom report example

Significant differences prompt you to analyze the situation to assess faulty data, incorrect use of forecast models, unforeseen promotions, and failed promotions, for example.

Demand plan variance custom report. This BI custom report tracks the month-by-month changes to the forecast (Figure 5). For example, the forecasted quantity for Jan08 generated in the forecasting run in July07 is compared with the forecasted quantity for Jan08 generated during the forecasting run in Aug07.

Figure 5

Demand plan variance custom report example

For day-to-day analysis, APO DP provides forecast accuracy reporting capabilities by using the forecasting errors transaction screen in Figure 3. You can configure alerts to capture exceptions during the forecasting run. Planners access the alerts through the alert monitoring screen.

After you complete these five steps, your system becomes fully capable to deliver higher forecast accuracy, improved inventory levels, higher customer service levels, and smoother exception management.