Issues I could imagine range from obtaining/labeling training data, communicating with peers about datasets, testing model quality, and updating models in production. I would love to hear your specific thoughts and experiences.
A lack of understanding of what building machine learning in production entails. How to manage a machine learning project. How to understand and interpret data. How it's not magic. How it depends on the organization's own management and curation of data quality. And even simple things like how to run a SQL query and analyze the data yourself, etc etc.
It reminds me of 10-20 years ago when a major problem at technical companies was lack of software literacy. Now that hump has been overcome by an increasing proportion of the economy (nowhere near perfect, but it's better than it was!).
Trying to convince people 20 years ago that software had a different development lifecycle than hardware was an uphill battle because leadership trusted what worked for them in the past. Now we have that problem with software aware leaders that don't want to change how they work to learn how to productionize machine learning.
What is it about the nature of data that makes this different (and harder to overcome) from a typical software problem for businesses?