Artificial Intelligence (AI) offers promising capabilities for many industrial applications and the use of AI has gained traction within manufacturing processes. However, working procedures lack criteria to decide about the benefit of AI systems for particular use cases and do not describe any uncertainty analysis in the development stage, since they assume readily available significant data. The objective of the present study is to introduce a process model, which includes decision criteria to determine the necessity of AI and evaluate the uncertainty of the system, consisting of suitable sensors and algorithms, in early development phases. The procedure is verified by analyzing a battery production line and identifying the prediction of success for MIG-welding as a suitable point of improvement by an AI system. A qualitative extraction of domain knowledge and data source selection leads to a fuzzy model and a clear distinction between the epistemic and aleatoric uncertainty of the system to estimate the prediction capability. The presented method allows the distinction between such uncertainties in a systematic development process and thus allows for a targeted optimization of AI systems. Extending the introduced systematic to other industrial fields can help to increase the implementation rates of AI in manufacturing.