When an AI project stalls, the usual suspects are the model, the vendor or the use case. More often the problem is lower down: the AI cannot reach the data it needs, because the data is locked inside the application…
Your systems record what happened: the invoice was issued, the price was changed, the order was approved. They almost never record why.
Companies worry about being locked in to an AI provider through its model or its interface. Those can be changed: models are swapped, interfaces are wrapped.
When a company connects an AI agent to a process, the easiest path is to give it broad access and narrow it later. Later rarely comes.
The first objection an IT manager raises against an AI pilot is fair: will it copy our data somewhere we don't control? The answer should be no. A well-run pilot does not get a copy of the company's data.
When an AI provider says it "uses your data", it can mean two very different things. Using your data to answer a question, then forgetting it, is access.