Global Lighthouse Network 2026

Highlights from the 2025 cohort FIGURE 5 ML-power ed pr ecision setup for legacy machines Manual setup of parameters on 60-year -old machines for 5k+ SKUs with tight tolerances was significantly impacting yield, lead time and quality. The solution uses ensemble ML models trained on 2+ years of golden batch data to pr escribe optimal machine settings. Operators transfer these first-time-right settings directly to the machine pr ogrammable logic contr oller (PLC) system and a continuous ML pipeline r efines models based on desired outcomes and operator feedback. Cost leadership and agility in a competitive packaging marketAAward categories Site and location sesac esu tuodnat S Productivity ytilauq & tso CShirwal, Indi a In-house development %73 + tsri F pass yield%34 - ytilau Q setup time%31 - noitcudor P lead time AI agents for war ehouse performance management Wher e fragmented, high-volume data previously delayed action, an AI agent now str eamlines performance management by automating issue detection, r oot cause analysis and r esolution workflows. Leveraging LLMs for self-lear ning and knowledge standar dization, it integrates data fr om 12+ systems to trigger r eal-time alerts, generate r eports and gover n the PDCA cycle. A GenAI copilot enhances decision-making with advanced r easoning. Factory-to-customer agile model for e-commer ce gr owthB niahc ylppuS resilience ytiliga & ecivreS Hefei, ChinaJoint-development %08 - eussI fixing time+40% tuphguorhT (parcels/hour)+6% erots enilnO satisfaction rating Simulation and genetic algorithm (GA)-based workstation r econfiguration for ETO Floor scales have millions of configurable variants and engineer ed-to-or der (ETO) solutions, with 66% one- piece or ders pr ocessed daily . Through multi-system integration, discr ete event simulation (DES) and GA, modular cluster workstations ar e dynamically r econfigur ed via r eal-time scheduling, supporting complex line- balancing and scalability constraints and impr oving r eliability .Agile design & fulfil- ment of complex ETO customer r equir ementsC Customer centricity tekram ot deepS & customizationChangzhou, China Joint-development %83 + ytivitcudorP (UPPH)+13% gnirutcafunaM on-time delivery-54% emit elcyC AI-enhanced R&D testing optimization for energy efficiency A neural network automates optimal plans for pr oduct testing to addr ess energy inefficiencies, while a deep lear ning model r efines plans using historical data. Multi-constraint optimization determines ideal test sequence and r einfor cement lear ning allocates lab capacity efficiently . A load calculation model and a PLC system maintain pr ecise contr ol, including fine tuning for temperatur e accuracy .Emissions and resour ce efficiency across pr oduct lifecycleD ytilibaniatsuS & ytiralucriC decarbonizationQingdao, China Joint-development -20% gnitseT cycle time%23 - Energy consumption in new product testing %23 - Scope 1 & 2 emissions in new product testing Intelligent workfor ce or chestration for demand volatility To addr ess fluctuating demand (up to 200%) and limitations of experience-based planning, a deep lear ning- power ed system enables skill-based matching and dynamic workload balancing via r eal-time dashboar ds. The system integrates upskilling into task allocation, aligning with training pr ogrammes to pr ovide exposur e to new skills while leveraging existing expertise. An AI assistant delivers r ecommendations and generates 12- month training and hiring plans to pr omote workfor ce readiness.Digital appr enticeships and integrated, AI- enabled upskillingE Talent Workforce empowerment & stability Wuhan, China In-house development .p.p4.7- seicnacaV -51% rep emitrevO employee per week.p.p8.6+ emit nO delivery Notes: LLM = large language model, ML = machine learning, PDCA = plan-do-check-act, SKU = stock keeping unit. Source: Global Lighthouse Network.