The Scale of the Problem
Gartner, MIT, RAND, BCG and McKinsey each published AI project failure research in 2025-2026 and converged on the same range: 70-85% of enterprise AI initiatives fail to deliver expected business value — roughly double the failure rate of ordinary IT projects. RAND's analysis of over 2,400 enterprise AI initiatives found the rate essentially unchanged over three years. Separately, MIT found 95% of AI pilots show zero measurable P&L impact, and 61% of projects were approved on a projected ROI that was never actually measured after launch.
The Number That Actually Explains It
73% of failed AI projects had no agreed definition of success before the project started. Projects that did define quantified success metrics upfront hit a 54% success rate versus 12% for those that didn't — a bigger gap than any model or vendor choice produced in these studies.
What This Means in Practice
The failure mode usually isn't the AI underperforming a reasonable bar — it's that no reasonable bar was set. An automation project scoped as 'add AI to this process' without a specific, measured outcome (hours saved, error rate reduced, cost per transaction) is most of the way to becoming one of the 70-85%, independent of which model or vendor sits behind it.
The Actual Takeaway for Scoping a Project
This is why an AI Automation engagement should start by naming the specific metric it needs to move before any tooling decision gets made — the data says that single step correlates more with success than anything that comes after it.