Every executive survey says AI budgets are going up. A quieter set of numbers says most of that money is not paying off. Only about a quarter of enterprise AI initiatives deliver the ROI they promised. That gap is the most interesting, and least discussed, story in corporate AI, because the reason for it is almost never the thing everyone blames.
It is not the model
When an AI project underdelivers, the post-mortem usually blames the technology: the model was not smart enough, the tool was not mature, the vendor overpromised. Sometimes that is true. Usually it is a comfortable lie. The models are genuinely capable now. When a deployment fails to produce value, the failure is almost always upstream of the model, in the process it was bolted onto.
What actually goes wrong
- No clear problem. "We need an AI strategy" is not a problem. It is a press release. AI applied to a fuzzy goal produces fuzzy results.
- No baseline. If you never measured how long the task took or how often it went wrong before, you cannot prove the AI improved it. So you cannot show ROI even when it exists.
- Integration reality. The demo runs on clean data in a vacuum. The enterprise runs on messy data across ten systems that do not talk to each other. The model is fine. The plumbing is the project.
AI does not fix a broken process. It photographs it in high resolution and charges you for the print.
The pattern in the winners
The companies getting real returns are boring about it. They pick a specific, measurable, high-volume task. They know the baseline cold. They deploy narrowly, measure honestly, and expand only what proves out. They treat AI as a tool applied to a well-understood job, not a magic strategy sprinkled over a vague ambition. It is unglamorous, and it works.
The takeaway I would put on a slide
If your AI initiative is not delivering, resist the urge to buy a smarter model. Go look at the process first. Nine times out of ten, the ROI was never blocked by capability. It was blocked by the fact that nobody could say precisely what success looked like. Fix that, and the technology, which really is good now, does the rest. Refuse to, and you will fund a very expensive lesson in something you could have learned for the price of an honest afternoon.