Over the past six decades, many companies have discovered the potential of computer-controlled systems in the manufacturing industry. Overall, digitization can be identified as one of the main drivers of cost reduction in the manufacturing industry. However, recent advances in Artificial Intelligence indicate that there is still untapped potential in the use and analysis of data in industry. Many reports and surveys indicate that machine learning solutions are slowly adapted and that the process of implementation is decelerated by inefficiencies. The goal of this paper is the systematic analysis of successfully implemented machine learning solutions in manufacturing as well as the derivation of a more efficient implementation approach. For this, three use cases have been identified for in-depth analysis and a framework for systematic comparisons between differently implemented solutions is developed. In all three use cases it is possible to derive implementation patterns as well as to identify key variables which determine the success of implementation. The identified patterns show that similar machine learning problems within the same use case can be solved with similar solutions. The results provide a heuristic for future implementation attempts tackling problems of similar nature.