To enable the AI to detect and evaluate business-relevant anomalies, the machine-learning algorithms are trained on the basis of process data and then concentrated to determine which anomalies have an impact on the economic efficiency of the plant. The plant operator themself then defines the further focus of the AI using the app dashboard, where anomalies can be selected, evaluated and commented. This evaluation phase is accompanied by several feedback loops, so that the plant operator ends up with well-trained, focused AI that is able to evaluate anomalies, based on the process data, for their business relevance. The AI Anomaly Assistant app is installed either as a cloud application or within the user's own infrastructure, for example on a Simatic Box PC or a virtual machine. The cloud-based solution is particularly advantageous during the training and evaluation phase, since it supports efficient collaboration between data analysts and plant operators. In addition, it also allows the results of anomaly detection to be combined with other services, such as predictive asset management, as part of the Asset Performance Suite (APS).