Introduction to the Use case: Forecasting of Infrastructure expenses
As a controller within the CoE central forecast (part of Corporate Controlling) at SAP, I’m a curious person who likes to try new things to solve problems. In this blog I’d like to share one of those innovations I’ve done recently. Let me start with a short introduction of its purpose:
The CoE central forecast team has one major goal: To accurately forecast the SAP Group result without any bias and being data driven. This forecast is used to steer the company over the year and to enable the management to make early decisions.
Data-based forecasting cannot go without Predictive Analytics. Predictive Analytics really helps to create non-influenced forecasts, and this can be done in an automated way, much faster than the manual process.
One problem I often see is that not everyone has the time to dig into Predictive Analytics, to understand the whole mathematical background of it and to test each algorithm and its application on a particular problem. Everyone who was involved in such a process knows that it can take a lot of time until you find a reasonable fit for your problem and then roll it out productively.
SAP Global Controlling is currently running an initiative within the Global Finance transformation to further standardize and automate processes. One ambition is to provide automated forecasting tools that can be used by every Controller to support the forecast process. Those tools should help to reduce the manual workload (freeing time for other activities) and increase the quality of our forecasts.
SAP Analytics Cloud (SAC) can help with that. With this blog I want to illustrate how I leveraged SAC Predictive Planning for automation in the forecasting process, starting with no experience with SAC beforehand.
SAC Predictive Planning helps to eliminate a lot of repetitive steps in the process. I tried to push that to the limit. With just one click I can generate multiple forecasts. After that I just need to select a scenario that I am most comfortable with, and the forecast is done – six parallel scenarios in under five minutes! Sometimes you are not even able to grab a coffee in the meantime. Let’s get started with the different elements that are needed for running SAC Predictive Planning so you can achieve this yourself as well.
SAC Planning Model
I have an SAC model where all data is stored, and that model is automatically updated every time there are new Actuals available. It is important that those models are as lean as possible and only contain data that you are going to use. This is particularly true when it comes to the granularity levels of dimensions.
Figure 1: The model
As soon as there is new data it is important to investigate. We built an easy review Page in an SAC story for that. With just a few clicks it is possible to scan every level that is stored within the hierarchies.
Raw numbers are great, but for most people it is easier to see a graph to get the feeling of development and direction. You can easily spot outliers and follow them up with your colleagues and so much more. You can also compare your old forecast against the new data to get a taste of how good your forecast really was and improve accuracy along the way.
Figure 2: Review page
SAP Analytics Cloud Predictive Planning
As soon as the general understanding of the data is done, you press the button to kick off the process. End of blog -
Just kidding. Predictive Scenarios need to be built next. This is something that everyone must set up at least once. Because in the end every problem is different and might change over time.
For my case, I created six different Predictive Planning Scenarios with differing historical time inputs and different levels of granularity for the other dimensions.
You can also insert different Influencers (drivers) that Predictive Planning should test. There are a lot of possibilities with Predictive Planning. Please have a look at the blogs mentioned below to gain more knowledge.
As mentioned, I strive for automation. Therefore, I created a Multi Action which is an action framework that triggers: Data loads, data manipulations, allocations and running predictive scenarios. Ultimately, this is the “magic” button I talked about. Learn all about multi actions:
After you grabbed your coffee and got back to your workplace the entire process is done. I created a second page for reviewing purposes. Here one can reflect on all the different outcomes that all have their merit in their own way. However, in our process we only need one final forecast to commit to. Therefore, I will compare them against each other and select the most fitting one.
As soon as that is done, I load the scenario of choice in the definitive public version to store it for later usage and review. Naturally I store the other versions as well as I want to retrospectively analyze performance of all scenarios to actuals as they come in to improve accuracy over time.
Figure 3: Reporting page
What I learned along the way:
Before concluding this blog, I want to point out what I recognized as most important after creating this project.
Do it yourself. Every tool has its niches and that is also true for SAC and Predictive Planning. There is nothing better than finally figuring out that one thing that was bugging you. And if you struggle with something, get an expert to help you. I had Nick Verhoeven at my side, a great Solution Manager, who supported me with every tedious question and laughed together with me when I made the same mistake a few times.
It also helped that I have a general learning appetite. This helps especially when it comes to Data Actions and the like. But other than that, there are great resources to get started and with the right mindset everyone can get into SAP Analytics Cloud. From my experience I can only encourage everyone to try it out and just do it!
With all of that I was able to cut down our workload a lot (I must generate a forecast seven times a year) and get a complete data driven Forecast presented that is directly reportable out of the system. Beyond that, I plan to inject business conditions (Influences) into the model and resulting out of that, test different scenarios for the future.
But even more important: This is just a framework. We can take my results and use this framework for other problems as well (with small adjustments). The future user of this framework will not have to set up all the different Data and Multi Actions by himself and does not have to think the entire process through.
With a bit of extra work, you can create your own predictive forecasts independently and in a self-sufficient way, just like myself now. Would you like to get the perspective from Nick that supported me from solution management, then I’d motivate you to read his blog here.