And from experience, they are the ones that are less plug and play. Lots of business alignment, data archeology and scientiffic know how + personalization. So difficult scenario for pure products as well. Maybe I'm thinking wishfully since it is our job, but I do believe this ;)

ML 'traditional' methods are still king in tabular tasks. And those are where internal company data is more valuable and ROI is higher (time series, fidelization, pricing...) and where explainability/control is a must. So I agree but there is space for many other options as well. Isn't it?