The Core Strategy
Enterprises are pouring real budget into AI pilots right now — and by most independent research, the large majority of that spend produces nothing measurable. A widely cited MIT study found that the vast majority of generative AI pilots deliver zero measurable return on the bottom line. RAND has put failure rates for AI projects broadly at more than double that of conventional IT projects. Different researchers, different methodologies, same conclusion: most AI initiatives never make it past the pilot stage, and even fewer show a dollar of proven return.
This isn't a story about AI not working. Task-level, the technology performs. The failure is organizational, and it clusters around a consistent set of causes:
No metric defined up front. Success gets described after the project launches, not before — which means there's nothing concrete to measure against, and nothing concrete to defend when budget season arrives.
Data infrastructure treated as an afterthought. Pilots get built on top of data that was never organized to support them, so the model performs well in a demo and falls apart the moment it meets real, messy operations.
Adoption assumed, not designed. Leaders assume people will simply start using the new tool once it's deployed. They don't, unless the workflow was built with the people doing the work — not around them.
The exciting use case beats the useful one. Companies pilot the flashiest possible application instead of the boring, well-defined one with a clear before-and-after number.
The financial impact compounds. A shelved pilot isn't just a sunk cost — it's the team-hours redirected to build it, and the credibility tax on every AI proposal that comes after. Once leadership has watched one initiative fail to show results, the next one faces a much harder budget conversation, which slows the whole organization's ability to adopt the use cases that would actually work.
The businesses landing in the minority that succeed tend to sequence things differently: they define the financial metric before choosing a use case, fix the underlying data before piloting rather than after, and treat the rollout as a change to how people work — not a tool dropped on top of it.
Executive Takeaway
Before approving another AI pilot, demand one number it has to move, and how it will be measured. "Let's explore AI" is not a success metric.
Treat data readiness as the real gate, not the model. If the underlying data is disorganized or inaccessible, no amount of model quality will fix that downstream.
Sequence beats ambition. The organizations seeing real returns fixed their foundations first and picked the unglamorous, well-scoped use case — not the most impressive one.
Inside Xylora
This is exactly why we push clients to define the ROI number before we build anything. A pilot without a measurable target is just an expensive demo. If you've got an AI initiative sitting in limbo — or you're about to approve one without a clear before-and-after metric — reply and we'll help you pressure-test it before it becomes a write-off.
The Tuesday Briefing is published weekly by The Xylora Digest.