Contributors: Ruvisha Pillay, Bill Tonetti
Pivot Forecasting is a three-step process. Our article “The Process, an Overview” introduces the basic Pivot Forecasting workflow. If you haven’t had a chance to check it out, it might be useful to give it a read before you continue with this article.
The three-step process of Pivot Forecasting:
The system generates automatic Calculated Forecasts;
Users review and adjust forecasts, which modifies the Synchronized Forecast;
Users finalize the forecast, which copies it into the Final Forecast measure.
In this article, we aim to dig more deeply into Step 2. In particular, we will discuss how to find the things that cause inaccurate forecasts. This is very important, as better forecasts enable improvements in customer service, inventory/asset turnover, and many other important business performance indicators.
To start, let’s look at some of the most common causes for inaccurate forecasts.
Items without enough historical data are more difficult to forecast simply because there isn’t enough data.
Unusual demand such as promotions, non-seasonal events, outliers can cause inaccurate forecasts. Examples include one-time buys, supply chain disruptions, pipeline fills, etc. Large outliers, whether high or low, will cause significant changes in the forecast if they aren’t corrected for. For example, a large one-time order will cause increased forecasts even though it is not likely to reoccur.
Human error, such as optimistic, pessimistic, or old and re-used adjustments. Some companies only produce forecasts a few times each year, which means they will still be working with forecasts that may have been produced several months ago, even though recent demand could have increased or decreased since the forecast was first produced.
The forecast method could be incorrect, for example, the Expert method would continue to forecast for a product for a few months or weeks after it’s discontinued.
Demand may be extremely low, sporadic or random. This might happen when forecasting less popular items, or forecasting smaller customers independently instead of grouping them.
Now let’s explore how to find these situations in Pivot Forecasting.
There are four primary ways to find and prioritize things in Pivot Forecasting, namely:
Attributes
Filters
“Top N” Panels
Pivot Maps
Attributes
There are attributes that can help you to discover and prioritize your work. For example, all Pivot Forecasting models have “Forecastability” and “Forecast Results” attributes which identify potentially challenging items or forecasts. The models also include the ABC and HML categories that come from your inventory planning app.
The Forecastability groupings are determined by comparing the series variability to the average. When the variability is higher, items are classified as having low forecastability.
Forecast Results are the forecasting methods that are in use. If an item is using a method that’s inappropriate for the series, this attribute can help you find it. For example, a highly forecastable item shouldn’t use a simple method like a moving average.
ABC is useful for prioritizing your work. A items are more important since they account for a larger portion of demand. HML is similar to this. It’s a measure of the consistency of demand. Low means it doesn’t sell as regularly as other products.
Filters
Filters are accessed just above the attribute picker on the left navigation side of the Pivot Forecasting user interface. FIlters limit the items that are loaded into the Tree. All Pivot Forecasting models have standard filters for identifying items with outliers, recent promotions, new items, and several others as shown in the image below. These are extremely valuable because they identify situations that can cause significant forecast inaccuracy.
The number of historical periods that are “Recent” is a model option that can be configured by an administrator. The default value is 1 period, but an administrator can change this. The image below shows a user has opted to change it to 2 to allow more time for review and correction.
You can also create your own filters using the “Custom Filter” option. Any user can make custom filters, but administrators can create public filters and share them with other users. Custom filters can use any attributes or measures in a model, as well as many of the most important statistics and forecast settings. The filter in the image below shows items with future forecasts but no history in the last few months.
Top N Panels
The Forecast Performance dashboard in the image below includes an example of the Top-N and Exception panels. The Demand - Exception Sheet Panel can display items that meet a number of previously configured exception criteria, including many that are also available as standard filters on the navigation Tree.
Pivot Maps®
In the “Forecast Performance” standard dashboard, you’ll also find a very useful Pivot Map. In this Pivot Map, the size of the boxes represent sales volume and the colors indicate the magnitude of the forecast deviations for the previous month.
You can also use Pivot Maps to compare Current Sales and Final Forecast for the current month. Current Sales are the sum of month-to-date (or week-to-date) sales, plus open orders. When we compare it to the current month’s Final Forecast, we can see where we have over- or under-sold.
Now that you know how to find the problems that lead to inaccurate forecasts, head over to our article on “Filters and Advanced Workflows: Correcting Problems” to learn how to correct the things that cause inaccurate forecasts.
Have you read the related articles? Check out our Pivot Forecasting collection!