In a previous blog (Should you be Demanding IBP?), I highlighted some of the high level differences between SAP APO Demand Planning and SAP Integrated Business Planning (IBP) for demand.
With the impending release of IBP v6.3 due later this year, I intend to go under the bonnet of the IBP for demand statistical forecasting engine to see if all that glisters is really gold.
Originally released as part of IBP v5.0, the IBP for demand module is SAP’s next generation forecasting application for supply chain. Since SAP Integrated Business Planning was developed natively on the SAP HANA® platform, its performance outpaces that of systems that have merely been modified to use in-memory technology.
Fig 1: Applications of SAP IBP
Note that the focus of this blog will be on the “traditional” statistical forecasting algorithms as opposed to detailing Demand Sensing functionality which will be covered in a later blog.
The underlying process for statistical forecasting between APO DP and IBP for demand has in essence not changed. Putting the input data and output forecast to one side, as per the Figure 2 below, the subtle difference is that SAP have split out the process in to three key elements on the system – Preprocess, Forecast, Postprocess.
Fig 2: IBP for demand – “traditional” statistical forecasting steps
A lot of the features in the three steps will already be familiar to you from APO DP, but let’s take a look in detail at what is now included within IBP for demand.
Pre-Processing Steps
With the help of algorithms built into IBP, the pre-processing step is in essence preparing the historical data for forecasting. Simple! In addition to outlier correction that was part of APO, a nice addition that comes with IBP for demand is an additional step to correct any missing values that you may have from the historical data called ‘Substitute Missing Values’. The user can select in the model whether the missing value is substituted with either the mean or median of the existing values, and the result can be stored against a separate key figure as standard.
Fig 3: Substitute Missing Values
There is now a third option within Outlier Correction to correct with Tolerance as opposed to Mean or Median, and the detection method is either a Variance (Standard Deviation) or Interquartile Range test (difference between the third quartile and the first quartile of the data).
Note that there is also ‘Promotion Sales Lift Elimination’ available in relation to Demand Sensing (requires a daily time bucket).
Forecasting Steps
For the Demand Planning community, it is fair to say to that there have been no major developments in SAP APO since SCM 5.0 was released. The statistical algorithms and functionality have remained largely the same, yet, Demand Planners are expected to forecast more Product/Location combinations with normally only a token understanding of what statistics is all about. So with the reality of Cloud based forecasting available today with IBP for demand, what are SAP providing over and above what is already in APO DP?
Putting Demand Sensing to one side, the current reality is that IBP (up to v6.2) has not delivered the revolution many were hoping for compared to its APO forefather from a statistical forecasting algorithm perspective. As an example, one of the more frequently used models that I have seen over the years, Seasonal Linear Regression, is not yet available with IBP for demand, although I suspect this omission will be rectified in the near future. The table below shows a comparison between APO DP and IBP for demand;
Fig 4: Comparing APO DP with IBP for demand algorithms
It is good to see a composite forecasting type approach available as standard within IBP for demand. Similar to the existing functionality within APO DP, the system can choose either the best forecast according to the chosen forecast error, or calculate a weighted average forecast where the user can apportion weighting factors to the various models.
Fig 5: Composite Forecasting
It is also worth stressing here the key differences between IBP for sales and operations and IBP for demand as that is an often asked question from our clients. Demand Sensing is the obvious call out that is not part of IBP for sales and operations, but there are also no pre-processing methods available (see below), and only RMSE is available as a post processing method. In terms of the statistical algorithms available, there are only the following four methods available;
Simple Moving average
Single Exponential smoothing
Double Exponential smoothing
Triple Exponential Smoothing
The disclaimer here is that although I cannot see a way how SAP can technically restrict IBP for sales and operations customers from using the full range of models available, it is the licensing implication that allows the users of IBP for demand the full statistical forecasting suite.
Post-Processing Steps
These are the steps to be performed after forecasting, such as the calculation of error measures that reflect the level of forecast accuracy or model fit. The following error methods are offered within IBP for demand, and can all be stored against key figures as standard;
MPE | Mean Percentage Error | |
MAPE | Mean Absolute Percentage Error | |
MSE | Mean Square Error | |
RMSE | Root of the Mean Square Error | |
MAD | Mean Absolute Deviation | |
ET | Error Total | |
MASE | Mean Absolute Scaled Error | |
WMAPE | Weighted Mean Absolute Percentage Error |
Note the following points;
- Two new additional measures of MASE and WMAPE compared to APO
- The absence of any user defined measure. This is currently not possible within IBP for demand, which is purely down to IBP being cloud based, with no option for customers to develop their own user exits or BAdIs.
Fiori User Interface
It is worth calling out the fundamental shift in the user interface that comes with recent versions of IBP (from a setup perspective as opposed to viewing planning results in the Excel Add-In). Compared to the traditional user menu, I do like the shift to a more intuitive and tablet friendly view that comes with the Fiori tile based UI.
Fig 6: Fiori User Interface
Note the tiles for Forecast Model Assignment & Forecast Model Maintenance in the figure above to take the user (assuming relevant authorisations) to the maintenance/setup transaction.
The Future?
From various pieces of SAP literature and webinar sessions, it is clear that the future is centred on the SAP HANA Predictive Analysis Library (PAL) - refer to this PAL Help URL for further detail. All of the algorithms within IBP for demand all come from the PAL. What is not known (by me at least!) but hoped for by a lot of the SAP forecasting community is that methods such as Arima, Seasonal Linear Regression and Unified Demand Forecast (UDF), along with tests such as Chi-Squared are all incorporated in to IBP for demand in the near future. Several of these methods are already present in PAL after all.
One object I have briefly seen before during an SAP session is a KPI Performance tile bringing pre-defined forecast accuracy metrics and reporting under one roof. I had hoped to see this delivered as part of release 6.2, but can only wait until later releases.
Conclusion
In the short term, all that glisters is not gold compared to APO DP. Why? The main reason would not necessarily be from a purely statistical algorithm angle, but I would have to call out the current lack of supporting lifecycle modelling and realignment functionality on offer. Granted both of these areas are not without their limitations within APO DP, but at least the presence of user defined ABAP and/or macro functionality allows us to at least enhance SAP standard functionality.
But! If I remember where APO DP was two years after release, then SAP is definitely heading in the right direction with IBP for demand and PAL - tipped to become the best of breed repository.
The future is bright. The future is IBP for demand.
I hope this has been a useful read, and please keep an eye out for further blogs on IBP for demand!
Ian Brister
Demand Practice Lead - Olivehorse Consulting