Companies’ AI Push Is Stalling Because Their Data Is Broken

Overview

• Reports published Monday showed that only a small share of firms judge their data fit for AI and many generative-AI projects are dropped after proofs of concept because inputs are incomplete, duplicated or untraceable.
• Surveys and studies cited across the coverage found concrete numbers: one research partnership put data-ready companies at about 12 percent and others reported that roughly half of generative-AI projects fail to move past proof-of-concept or produce measurable financial returns.
• Experts advise starting with the problem and success metrics by defining clear objectives and KPIs before choosing tools so pilots test value rather than technology.
• Practical fixes include document management, process redesign, automation and AI-assisted data cleaning plus governance measures that assign owners, log provenance and monitor models to keep systems reliable in operation.
• Leaders must also address human limits by training teams, reducing dependence on single experts and guarding against cognitive biases that can make organizations use AI to reinforce old, flawed decisions rather than to improve them.