In many organisations, it is very common to use the statistical forecast as a baseline forecast. Understanding and analysing of historical data are very important to generate a reliable statistical forecast. In most of the cases, demand planners rely on their team’s experience and conduct visual check (graphs) to understand historical sales pattern such as seasonality, trend, etc. SAP adds another feature in its SAP IBP software under Forecast Automation application, and through this application, demand planners can analyse historical demand data statistically without being too mathematical. In this blog, I will walk you through how it works and discuss some of its advantages and drawbacks of it.
What is Time Series Data?
It is a sequence of numerical data points taken at a successive equally spaced point of time. In demand planning world, it is the sales history. Over the period, time-series data will develop pattern and characteristics such as seasonality, trend, etc., and through time series analysis, those patterns could be understood to develop a robust statistical forecast.
SAP IBP Time Series Analysis
SAP released time series analysis as part of IBP 1811 release to help analyse the data pattern at various levels of planning hierarchies. These hierarchies can be a combination of any attributes from Product, Customer and Location dimensions. A data pattern may change according to the combination of attributes looked at (such as product and customer, or product and customer region, or product family and customer group, etc). For example, a product might be a trendsetter in one market, but not in the other. The aim of the time series analysis is to run through the data streams at various aggregated and disaggregated levels and perform some key check to help planner understand the pattern of the time series data.
For example, sales pattern at a product family level may show an upward trend, but, the same data at a customer level may exhibit a downward trend for some customers for the same product family.
With SAP IBP Time series analysis, a user can check for the below data patterns:
- Intermittency & Volatility
How does it work?
In the below example, I have set up the SAP IBP system to analyse the pattern of a historical data at 6 different levels (or attribute combinations) of Product, Customer and Location dimension.
The conditions must be set to test each of the data pattern and how the system should analyse data to provide the results. It is flexible, so planners can experiment by choosing different thresholds for each section. Some real good features here are to get results with the direction of a trend in the data (upward / downward), also to let the system analyse the length of the seasonality rather than assuming the products to have a 12-month seasonal cycle. A drawback would be, the system cannot be forced to use the seasonality length as 12 months and detect if the data pattern is seasonal.
For this example, I have used 1.33 for Average demand interval to check whether the data is continuous or intermittent/sporadic. If the average gap between data points in the sales history is more than 1.33, then it will be considered as intermittent or sporadic data.
Result of analysis
Result of the analysis will be the pattern of data as a pre-defined text and the average demand interval and seasonal indices at each level. In this example, I can see the results at each of the 6 levels.
Display Results in Excel
Analysis result at a Customer and Product combination level (at a very basic level).
As you can see below, sales history pattern of product AP_F002 at Customer AP1001 is continuous and additive seasonality and (12) indicates the length of seasonality. Likewise, product AP_F003 has both seasonality and trend. However, if you notice product AP_F006, it has lumpy sale history and is not continuous as the average gap between two data point in its sales history is 1.83, which is more than the threshold setting 1.33.
If the same data is analysed from Customer group and Product Group level, Strategic customers for AP-DIATEX products (aggregated level) do not exhibit intermittent or sporadic or trend or lumpy pattern, but Continuous data with additive seasonality.
Display Results in Browser
Same results can be seen via web UI.
The results of time series analysis can be used in
Selecting the right forecast model:
For example, a simple weighted moving average forecast model or single exponential smoothing forecast model would be good for continuous data pattern, which does not have any seasonality and trend. Similarly, Triple Exponential Smoothing model would be good for a data that exhibits both seasonality and trend pattern.
It is no secret that this is where the demand planner will struggle if they lack the statistical forecasting knowledge. Not to worry! SAP brings a new feature in the upcoming IBP1908 release (SAP IBP 2019 quarterly update in August), where the results of time series analysis can be used with a best-fit forecast model profile and let the system to choose the appropriate best fit model. A planner can then use the filters in SAP IBP Assign Forecast Models application to see results of Time Series Analysis and to assign the right forecast model.
Segmentation of products
In SAP IBP system, the XYZ segmentation profiles can be defined to use the results of time series analysis to classify the assortments (products or product customers, etc) based on the volatility of the data. Then, the system will ignore the seasonality and trend in the data pattern of sales history before generating the XYZ segmentation.
Through time series analysis, it is much simpler to classify the assortments by the data pattern from multiple dimensions without spending a huge amount of time on gathering and analysing data. This will help the demand planner to spend more time on finding the right forecast model and understanding the root cause of changes in the sales pattern. However, there are a couple of points which the planners need to consider while using this application.
- First, if sales history of the product (NPDs for example) is less than the number of historical periods specified in the forecast automation profile (36 months for example), then the results of this application may be incorrect. So, there is no option to set the system to analyse the data pattern only from the first period where sales occurred. There are other options to achieve this, but they are not simple or straight forward!
- Second, this application still does not help a planner in finding the right aggregation level to run the statistical forecast. Of course, this issue can be solved by finding the forecast accuracy at each aggregation level and assigning a forecast model to find the optimal level. But patience and time is the key!
Should you be interested in seeing how this or any other features can help your organisation, please contact us to enquire about our FREE IBP taster session.