供应链人工智能准备报告：为什么运营纪律决定代理人工智能的成功

Why Operational DisciplineDetermines Agentic AI Success By Michael DuValland Dr. Timothy M. Laseter ABOUT THE AUTHORS Michael DuVall:Mike is the Global Head ofStrategy for GEP, a leader in the advancementof AI in the supply chain. With over 20 years instrategy consulting, he specializes in redefiningoperational frameworks through market-backapproaches across industries. His mission is tomove organizations beyond fragmented pilotstoward agile, intelligent supply chains that leverageAI to unlock sustainable competitive advantageand new levels of value creation. Dr. Timothy M. Laseter:Tim is a Professor ofPractice at the University of Virginia’s DardenBusiness School and the LaCross AI Institute.A former partner at Booz Allen & Hamilton andManaging Director at PwC-Strategy&, he hasauthored four books and over three dozenpractitioner articles on business model innovation.His work bridges academic research with decadesof global consulting experience, helping Fortune50 corporations and startups alike navigate theevolution of supply chain operations. INTRODUCTION More than half of supply chain professionals areprobably using generative AI. But fewer than 1 in10 of supply chain AI pilots have identified what ittakes to translate that enthusiasm into industrial-scale results. Majority of the market, however, remains in pilotpurgatory. AI tools proliferate, but enterprise capabilities donot. Individual productivity improves while end toend organizational performance barely moves. This gap between individual adoption andenterprise-wide performance defines the centralchallenge facing supply chain leaders today. The missing ingredient is not model accuracyor computing power. It’s operational disciplineand Lean thinking. The results achieved by thePerformance Elite require more than a betteralgorithm. They demand what we call “IntelligentValue Streams.” Through our research and interviews with seniorsupply chain executives, we have identified agroup we call the “Performance Elite.” These leaders have successfully moved beyondAI pilots to disciplined operational automationand achieved remarkable results. They aren’tjust seeing incremental gains. They are doublingproductivity, reducing error rates, and compressingresponse time. These leaders recognized a truth that remainshidden from most: to scale agentic AI, you mustfirst redesign the work itself. They chose to slowdown to speed up — clarifying strategic intentand Lean logic before delegating decisions tomachines. Across the board, these elites attribute theirsuccess to a fundamental redesign of the workitself. While others wait for AI to solve their problems,the elite have built a new operational architecturedesigned for an autonomous future. THE SCALING GAP Drawing on decades of combined experience inoperations strategy and ongoing conversationswith supply chain executives, we developeda framework to assess agentic AI deploymentmaturity. TheGEP Agentic Scaling Framework, aproprietary maturity matrix, evaluates 10 criticalvectors across three developmental horizons:identify and plan, pilot and refine and, ultimately,operationalize and scale. with published research from Anthropic1showingthat generative AI adoption is highest in codingand other high-skill tasks, with greater tractionin individual workflows than in deploymentsaccessed through corporate APIs. To test the framework, we conducted a broadsurvey of practitioners and further interviewsfocused on supply chain operations. (See theAbout the Survey on page 18) We first fielded the survey at GEP INNOVATE 2025in November at New Orleans, drawing respondentsfrom companies spanning retail, automotive, CPG,and pharmaceuticals. For most organizations, operational scaling ofagentic AI has failed to launch. Procurement and sourcing show the widest gapbetween ambition and execution: only 11%of respondents have no plans, yet just 4% areoperating at scale. The following month, we refined the surveythrough conversations at the University of Virginia’sEthical AI in Business conference sponsoredby the LaCross Institute, where Peter McCrory,Anthropic’s head of economics, presentedresearch confirming the pattern: individualadoption runs ahead of enterprise deployment. Demand planning/forecasting has the highest levelof scaling, though still a mere 10% of respondentshave progressed to that stage; nearly three timesas many remain in the pilot stage. In total, 180 executives participated in the study,and among them only a dozen reported thatthey had launched and broadly scaled AI pilotsin supply chain operations. Our findings align Exhibit 2 summarizes how respondents describeadoption across a range of supply chain functions. EXHIBIT 2: Status by Function Separately, we asked which function had yieldedthe most successful AI use cases and which hadproved the most challenging. high—consistent with the broader EconomistImpact survey, commissioned by GEP in 2024.2 But the same functions that some respondents