95% of AI pilot projects never reach production, according to an MIT study quoted on stage. Why — and above all, how do you join the 5%? In this session recorded at the Tech for Industry Show, Karl Fontenau (Nine Altitude, an ERP and PLM integrator for industry, Microsoft and PTC partner) and Jean-Jacques Galliano (PTC) deliver a pragmatic, product-data-centric method.
The talk covers: the four reasons AI pilots fail (generic AI fed on web data, standard workflows ignorant of internal processes, poor or erroneous data, no industrialization); structuring data in four steps — raw data, semantics (status, version, history), ontology (linking CAD, requirements, quality sheets and tests through knowledge graphs) and governance (who accesses what); the "digital red thread" across ALM, CAD, PLM, ERP, MES, CPQ, CRM and SLM; and concrete AI agents: optimizing millions of automotive requirements (redundancies, conflicts, test-plan gaps), generative design for sustainability, part deduplication (up to €6,000 per year per managed part number), a dormant-stock agent, and a change-management demo from Teams through PTC's PLM to Microsoft's ERP — with human validation at every step. Plus audience questions: MCP, data location, encryption keys, cutting web access, and choosing one governed AI against shadow AI.
A concrete roadmap for manufacturing, engineering and IT leaders.
