Many Swiss companies still select their future ERP through catalogues containing hundreds of detailed requirements. Today, language models create and answer such lists within minutes, and the established cloud systems cover the requested standard functions to a largely comparable degree. The comparison therefore mainly produces paper. Research also shows that projects rarely fail because of the software itself, but because organisations are unwilling to change. In our view, other criteria belong at the centre of an evaluation: the quality and industry experience of the implementation team, the willingness to work within the standard, and the vendor's investment and AI roadmap, including its openness to integrate different language models. Evaluating this way compares what actually determines the value of a system over the years.