Automation of decisions - IT must think interdisciplinary

UCompanies waste the important resource of time of their highly specialized professionals for decisions that are made again and again according to the same pattern. From another perspective, one could also say that there is a great deal of potential for productivity increases hidden here.

Thanks to new methods of data management and the possibilities of efficient data access, as well as the current developments in the field of machine learning and artificial intelligence, it is obvious to automate such routines.

In addition, business processes with routine decisions, which are already mapped digitally with the help of a workflow system, provide an important basis for the automation of decisions.

For example, decision-relevant process data can be collected by executing a process workflow and stored temporarily via interface on a Hana system for model training.

Data from real human decisions is then available via this workflow. Based on this data collection and special machine learning algorithms from the Hana Predictive Analysis Library or, alternatively, the well-known cloud offerings on the market, the machine learning models are trained for automated decision making and these models are also stored on Hana.

As a result, the process workflow was adapted to allow automated decision making by the system when running these models and to evolutionarily improve itself through automated retraining.

Challenging for this decision automation is on the one hand the analysis, selection as well as parameterization of the right machine learning algorithm and on the other hand the selection as well as preparation of the relevant data and the attributes which are to be used for a decision.

This is because these form the foundation of the decisions and lead to incorrect decisions in the event of incorrect choices, data gaps or incorrect data. Therefore, it is essential to deal with the anatomy of the individual decisions in detail.

The amount of data is also of great importance. Without an appropriately large amount of data, it may not be possible to learn an exact pattern from the machine learning model, which in turn negatively influences the decision quality.

Those responsible for implementation must control and support the training of the machine learning model. The necessary know-how must be built up beforehand.

In addition to the technical possibilities, however, it is also important to understand the anatomy of human decisions in order to be able to relate them to the technical knowledge.

Here, knowledge about the different decision theories and the behavioral economics of humans in decision-making plays an important role. In relation to this, knowledge about the prescriptive decision theory, which deals with making the rationally correct decision, and the descriptive decision theory, which deals with the typical mistakes that humans make as decision-makers, is necessary.

Especially knowledge of so-called cognitive biases, which reflect unconscious tendencies of humans in remembering and thinking, is important in this context and must be considered in the technical implementation.

These are completely new competencies that otherwise rather technically versed teams have to deal with in the course of the new developments. IT departments must work in an interdisciplinary manner here and, in the course of increasing digitization, deal with people as subjects within business processes.

The automatic control of decisions in business processes by means of algorithms and data is not just some specter of the future. It is worth breaking up the old structures and making IT departments more interdisciplinary. Only in this way can digitization and the associated increases in productivity succeed.