Full Answer

Exact figures depend on what you count as failure, so treat any single percentage with care. The most defensible, primary-source number is Gartner's: it predicts that through 2026, organisations will abandon 60 percent of AI projects that lack AI-ready data. Separately, Gartner reported that only about 12 percent of organisations have data of sufficient quality to support AI, which tells you how common the underlying problem is.

Broader industry estimates run higher. A frequently repeated figure, attributed to Gartner and echoed by HBR-style analyses, is that up to 80 to 85 percent of AI projects fail before or after deployment, roughly double the failure rate of conventional software. Analysts at Deloitte and McKinsey have likewise linked the majority of failures to data problems rather than algorithmic shortcomings. The headline number shifts, but the diagnosis doesn't.

For a store owner the takeaway is practical: the risk in an AI initiative sits in the data layer, not the model. Incomplete event capture, inconsistent schemas, duplicate conversions, and missing history all show up later as predictions that quietly miss. That's why the useful work happens before any model is trained, in auditing whether the data can be trusted at all. Getting the data right isn't a prerequisite to the real work; for AI, it is the real work.