# Pilot-to-Production Failure Rates ## Most AI pilots do not reach production Public studies give a wide range. Some studies say 80 percent or more of AI projects fail to deliver the planned value. Other studies say 88 to 95 percent of pilots do not reach production. A third group says 20 to 30 percent reach production at scale (secondary summaries, mid-2026). > [!warning] The numbers disagree > Each study defines "pilot", "production" and "success" in a different way. Do not average the numbers. Use them only for direction. The studies agree on the causes. Data quality, integration work, change management and unclear ownership cause most failures. Model capability does not. ## Why it matters The gap is the market. A buyer who has a stalled pilot has a budget and a pain. The risk is that a high failure rate also proves that buyers do not value the fix. Test with buyer behaviour, not with market statistics. > [!info] Verification status > All figures here come from secondary web summaries. Check the primary reports before you quote any number. ## Related - [[The Deployment Gap]] - [[Evidence Quality and Vendor Claims]] - [[AI Agent Vertical SaaS DD MOC]]