Machine learning is a branch of artificial intelligence that enables IT systems to use algorithms to recognize patterns and regularities in data sets and develop solutions independently. For this to work, the systems must first be supplied with the relevant data and algorithms. Depending on the application, the time horizon of the data should be one year or more in order to obtain reliable statements.
Rules must also be defined for analyzing the data and recognizing patterns. Once this is done, the systems can find, extract and summarize relevant data, make predictions based on the analyzed data, calculate probabilities for certain events, adapt independently to developments and optimize processes based on recognized patterns.
Five types of algorithms
Algorithms are at the heart of machine learning, as they are responsible for recognizing patterns and generating solutions. Basically, they can be divided into five areas:
- Supervised learning: Here, the algorithm is trained on a target. This requires a data analyst and a user to evaluate the data as correct in advance. This creates a so-called ground truth, on the basis of which the algorithm will automate approval workflows in future.
- Unsupervised learning: In this technique, the algorithm divides the data into clusters or categories based on patterns, but without knowing which categories are involved. This method is often used to classify customers.
- Semi-supervised learning: This is a mix of supervised and unsupervised learning.
- Reinforcement learning: Reinforcement learning does not require any prior data material, but generates solutions and strategies on the basis of rewards received in a trial-and-error process.
- Active learning: In this variant, the algorithm is given the opportunity to ask for the desired results for certain input data. In order to minimize the number of questions, the algorithm itself selects relevant questions with high relevance to the results.
SAP portfolio can be expanded
SAP is also increasingly relying on AI and machine learning in its solutions. This is particularly evident in the fact that each new version of S/4HANA either introduces new machine learning functions or improves existing ones. What many users do not know, for example, is that PP/DS and IBP already use machine learning techniques in the background to optimize planning and capacity requirements in manufacturing and production respectively.
In S/4HANA in particular, there are various machine learning approaches that are already linked to different applications and are also explicitly advertised. These include, for example, predicting the delivery date for order items in purchasing, predicting cash discount losses for invoices with a payment block or predicting the fulfillment rate of purchasing contracts. Examples in the supply chain area include the prediction of delivery delays for plant relocations, the early detection of out-of-stock materials and stockists and the prediction of the lead time for stock transfer products. But sales can also benefit. For example, SAP has integrated the prediction of delivery delays for outgoing deliveries, the prediction of the quotation conversion rate and the prediction of sales results.
All of these scenarios are fully integrated into the processes and generate forecasts based on historical process data. The forecast results are displayed in the operational and analytical FIORI apps and enable employees to react to specific situations in a timely manner.
For example, the SAP FIORI app “Predicted delivery delay” shows the delay in days for delivery creation and processing as well as the predicted overall status of the delivery item for each sales order item in addition to the planned delivery date. It also graphically displays statistics on all order items. This enables sales staff to identify delays in the supply chain at an early stage so that they can take action on critical items in good time and thus increase customer satisfaction.
CONSILIO has already gained experience with SAP ML standard solutions in various projects and therefore knows where the stumbling blocks lie during implementation and which solution is the best economic fit for the company.
Partners are essential
Although the number of solutions that SAP integrates into its systems grows with every new release, not all applications benefit from the new technologies to the same extent. Pioneers who are already toying with the idea of using machine learning to optimize their business processes on a large scale will still be disappointed. This is because they have to invest in development work themselves. However, the interfaces they need for this are already integrated in S/4HANA with the Predictive Analysis Library (PAL), the Automated Predictive Library (APL) and the external Machine Learning Library (EML). In order to use these programming libraries for the development of ML models and thus automate existing processes or even develop new business models, they require expert knowledge that often exceeds the traditional know-how of the company's internal IT department. SAP customers must therefore either rely on employees who are trained in data science or on the support of partners with the relevant expertise. SAP partners such as CONSILIO maintain close links with universities and therefore have the expertise to optimally combine state-of-the-art science with application know-how. As a result, they are able to develop the right questions regarding data analysis with the user, then collect the necessary data and prepare it accordingly. Because only when the triad of question, data and data quality is right will the analysis work.
In order to use the programming libraries for the development of ML models and thus automate existing processes or even develop new business models, users require expert knowledge that often exceeds the traditional know-how of in-house IT.
