AI has some very common roadblocks:
1. Most data scientists are expected to be their own product owners. Meaning, data scientists - who are programmers and mathematicians by training - are expected to become students of macroeconomics, supply and demand, marketing, customer qualification, pain points and value propositions, market definition, and the many other nuances of product strategy. This usually happens because most companies don't have a discipline of placing a product strategist/owner/manager as the head of the AI efforts. Product management has very well-defined frameworks for building web-based/mobile apps (SaaS apps). But very little has been done to articulate how to design a good algorithm, how to define metrics and dimensions and ML objectives so that a data scientist can hit the ground running, armed with clarity. Hence, most AI initiatives in non-AI companies fall flat on their faces.
2. AI doesn't make intuitive sense to statisticians, or people with a basic understanding of math, so there is a big resistance to some of its messaging, which can come across as oversimplification. For instance, whereas in traditional business-applied stats you can't just add more data in (it has to be cleaned and preprocessed), machine learning allows you to infuse messy, half-complete data and still keep improving the algorithms. I have seen many projects get halted by those in power - who have a vested interest in maintaining an old-school approach to regression modeling and predictions that is vastly outpaced by today's ML/AI capabilities.
I'm sure there are more examples, but hopefully this helps.