There was a thread yesterday about how data science is done and my answer was that it's just too broad of a question to fit in a single comment. Same here - there is a lot of variability in how a data science / machine learning / AI project is defined in different companies.

Some teams do research - it could be research in machine learning, or it could be research in using ML in some other problem domain. Depending on how that team is set up "production" might not even be a thing for them.

Some teams build software with ML features. They may do some in-house research or reuse existing models/architectures. The ML feature can be a critical part of the product, or it can be a nice-to-have.

Some teams neither do research, nor build software - they run their analysis and models interactively so to speak and the results get saved to a database/file and/or are used to create some report or inform some decision. This process can be internal to the operations of the company, applied to some other domain or it could be a piece of consulting.

I'm sure people can come up with more ways to do this type of work. My point is that in each of those cases success is defined differently. So when some online article starts throwing numbers around, I'm very skeptical about how deep they actually looked into these parameters.