Why 95% of AI Projects Are Failing to Generate Returns (medium.com)

A member-only piece argues that up to 95% of enterprise AI projects are failing to generate returns, illustrating the point with everyday examples: an “AI-powered” scheduler that mangled calendars and chatbots that replaced customer-service staff only to drive satisfaction down and force rehires. The story paints a picture of an AI bubble where tools sold as efficiency multipliers instead create new, unexpected work—technical debt, manual fixes, and thorny edge cases that erase promised savings and cost companies millions. For practitioners and leaders this matters because it exposes predictable failure modes: poor training data and domain mismatch, brittle models that hallucinate or mishandle edge cases, lack of integration with existing workflows, and absent monitoring or metrics to measure true ROI. The technical implications are clear—deploying models without human-in-the-loop safeguards, continuous monitoring for data drift, robust evaluation against real operational metrics, and clear change management almost guarantees value erosion. The antidote is less hype and more engineering: instrumented deployments, governance, iterative evaluation, and pragmatic automation that augments rather than replaces human expertise.