In ten years of building AI systems, I can count on one hand the projects that failed purely because the technology didn't work.
Almost every stalled project I can think of stalled for the same handful of reasons. It may come as a surprise to you that not a single one of those reasons is technical.
1. Ownership
Someone on the client’s side champions the project, gets it approved, and gets it moving. But if that person changes roles or leaves, there’s either no one to pick it up, or the person that picks it up is not as passionate about it as their predecessor because it’s not their baby. These projects don’t always fail loudly, they just stop getting attention until they fall away.
2. Access
I've seen clients wait years for internal approval to open up the API access we need to integrate. That's not a criticism of any one organisation, it's usually politics, security policy, or a process that was never built with this kind of project in mind. But from the outside, it looks identical to a stalled deployment, even though the AI itself was never the bottleneck.
3. Governance
The third is a quieter one: excitement that doesn't survive contact with governance. A business champion sees the value, gets it signed off, and assembles a project team. Then their colleagues get pulled in, but those people weren’t part of the original decision. The first question is always some version of "why wasn't I consulted?" The excitement that got the project approved in the first place evaporates the moment it has to go through proper checks and balances. That's not necessarily dysfunction, but it does tend to happen when you implement something inside an organisation with its own rules.
