How SAP S/4HANA Machine Learning foreshadows the Intelligent Enterprise

Successive versions of SAP S/4HANA increasingly integrate artificial intelligence. There are numerous out-of-the-box scenarios embedded in the application that are gradually introducing AI and Machine Learning. This step-by-step introduction prepares users for an in-depth transformation of their daily job and the rise of intelligent enterprise. 

S/4HANA Machine Learning

Machine Learning is based on analysing large volumes of historical data to make predictions, with the probability of these events happening — a bit like weather forecasts. In the field of ERP (Enterprise Resources Planning) software, this technique has multiple applications. It can facilitate data input, detect anomalies, and help with forecasting.  
Over the years, SAP has embedded more features based on Machine Learning into successive versions of S/4HANA for this functionality. 
For example, a new function related to cash management helps reconcile actual payments with expected payments on a probabilistic basis. Previously, these reconciliations were made using deterministic rules. For the processes to work, each payment had to match one of these predefined rules. Using Machine Learning, the reconciliation can be proposed by the software, even without a 100% match.  
In logistics, Machine Learning enables the software to forecast purchased goods delivery delays. And, more importantly, they give users the ability to assess the impact of these delays. This forewarning means the user is better able to focus on the most critical aspects of their work. 

Standard scenarios implemented for a minimal investment  

Machine Learning in S/4HANA is primarily aimed at generating productivity gains by allowing users to save time on routine tasks. The detection of data entry errors would normally be classed as generating quality gains. But it’s worth noting that these gains in quality also result in increased productivity. You eliminate the time that would be spent managing the impact of data errors. 
Despite impacting both quality and productivity, accurately assessing the direct gains of Machine Learning remains difficult. This makes its value and ROI (return on investment) hard to gauge. 
The first reason is because it is not widely deployed. The results are not universal. The gains are also subtle and intertwined with other gains generated by the new S/4HANA User Interface and the integration of analytics at the heart of transactions.  

However, one of the benefits of deploying the standard Machine Learning scenarios offered by SAP is that it only requires minimal investment. SAP provides the algorithmic models required, as well as integration with target data, and a Fiori application which allows the models to be trained on historical company data.  
A few clicks and a couple of hours of work are enough to “train” the algorithms — provided you work with relatively clean data and reasonable volumes (for example 50,000 orders) without either developers or Data Scientists being needed. 

A “super algorithm” chooses the best algorithm for each application  

SAP has come a long way since S/4HANA’s integration of its first Machine Learning library (Predictive Analytics Library). PAL was complex, only usable by a Data Scientist. Following the acquisition of Kxen, a predictive analytics specialist, SAP was able to integrate a second library, APL (Automated Predictive Library) which improved the usability of Machine Learning in day-to-day applications.  
APL uses a ‘super-algorithm’ capable of selecting and training the most suited Machine Learning model for a type of data and expected result. Fundamentally, this is the integration that puts Machine Learning functions directly into the hands of S/4HANA key users.  

This is also the mechanism SAP uses to increase the number of out-of-the-box use cases with each new S/4HANA release. We are now at the point where the number of use cases for Machine Learning doubles with each new release.  
Furthermore, the ‘super algorithm’ makes the process more intuitive. All these uses of Machine Learning are completely embedded in the existing applications. The user does not need to switch to a different system or transaction. Nor do they need to know the algorithms behind the forecasts.  
Users just need to be aware that these are statistical calculations that should not be accepted blindly. They need to be followed with critical thinking — like the instructions from your sat-nav, or car traffic forecasts in Google Maps. 

Implementing your own S/4HANA Machine Learning models  

In addition to the preconfigured applications of Machine Learning, SAP also provides a toolbox that enables customers to go further. Using the APL library, it is now possible to develop new scenarios similar to the preconfigured ones, all within S/4HANA.  

There are also options for more complex applications. If more advanced algorithms (such as neural networks) are needed, or the user needs access to data outside of S/4HANA, or large volumes of information (for example from data-lakes), SAP recommends using a “side-by-side” scenario.  
In this approach, the analysis engine is built using a tool called SAP Data Intelligence, on SAP Cloud Platform (SCP). However, the results can still be exploited from within S/4HANA applications.  

These developments represent a gradual but profound change in SAP software, a change that will accelerate in the coming decade.  

Whereas the previous generation of SAP’s ERP software, ECC, was — above all — a data capture system. S/4HANA aims to be a source of information.  
That’s why SAP has endeavored to maximise data entry automation, thanks to anomaly detection, semi-automatic data entry and, increasingly, full automation via intelligent software robots. These are all applications for which Machine Learning is proving to be a highly effective tool. 

Moving towards automated decision making

SAP also aims to leverage artificial intelligence in the information retrieval part of the software, by offering forms of automated decision-making. For example, when matching invoices with payments received, the user can decide to completely trust the algorithm for this operation.  

However, the transition to fully automated decisions poses a double challenge. 
Firstly, there is a substantial technological challenge to make the predictions made by the algorithms sufficiently reliable. But also, there is a cultural challenge. Companies and users must gradually get used to these tools to fully exploit them and to derive the expected benefits, both in terms of productivity and quality. This requires a cautious and patient approach.  
An effective way to get user buy-in and achieve a successful transition is to start by deploying machine learning in S/4HANA. 
To talk to our team about S/4HANA machine learning, or starting your conversion, click here and arrange a call about your goals.