Why AI pilots fail - and how manufacturers can break the cycle (www.techradar.com)

Manufacturers are watching nearly 9-in-10 AI pilots stall before they scale because projects are too often “technology-first” rather than outcome-driven. The article argues that most failures aren’t algorithmic but structural: fragmented, low‑quality and siloed data, plus separated IT and OT teams, prevent lab successes from becoming production wins. Without clear business cases, measurable ROI (throughput, yield, energy, downtime) and executive buy‑in, AI risks remaining a proof‑of‑concept with little impact on factory floors. The remedy is a shift to trusted, scalable data foundations and cross‑functional ownership. Practically this means investing in unified data infrastructures that ingest and integrate data across the value chain (including digitized unstructured sources), robust IT systems able to manage industrial-scale data, and workflows that let generative and analytic models produce actionable insights for predictive maintenance, supply‑chain optimization and real‑time production control. Treat AI like a capital project: define KPIs tied to operational goals, bridge IT‑OT divides, and prioritize use cases with clear ROI to turn pilots into sustainable, scalable improvements.