If you assume the key to successfully deploying enterprise AI is model capability, you're thinking about it the wrong way. Most models are capable enough.
The problem is the context that's missing - the exceptions nobody documents and the judgment calls that live in people's heads. Basically, the kind of knowledge that isn't in any dataset.
The only way to capture it is through execution. The systems that do the work are the systems that learn the business.