2026年AI治理基准报告：AI项目组合的爆发式增长

When Activity Createsthe Illusion of AI Value Contents Click below to navigate C H A P T E RT H R E EIndustrializingAI delivery 2Key Findings C O N C L U S I O NScaling AI and realizingvalue means operating AIas a managed portfolio I N T R O D U C T I O NA new phase forenterprise AI D A T AI N S I G H T SWhat’s Blocking AIat Scale — And WhyAgentic AI ChangesEverything23 C H A P T E RO N EThe AI portfolioexplosion5 C H A P T E RT W OWhy is there anillusion of AI value?11 Contributors Jeannie FurlanVice President,Financial Data andAnalytics Strategy,Mutual of Omaha Farid SheikhiSenior Manager,Analytics Innovationand Data Enablement,Royal Bank of Canada Ramila PeirisGlobal Head Data,ML and AI Platform,MSAT, Sanofi Barbara WidholmVice President,Emerging Technologies,State Street Key Findings 1AI portfolios are exploding The most common enterprise AI portfolio nowcontains 101–250 proposed use cases—upsharply year over year. 2Success at scale remains rare Despite hundreds of proposed use cases,most enterprises operate fewer than 25 AIsystems at production scale. 3Time-to-market is accelerating forGenAI use cases that reach production The majority of GenAI and agentic use cases canreach production in under 6 months—much fasterthan at the beginning of 2025 when more thanhalf took 6 to 18 months to reach production. ROI tracking is still manual More than two-thirds of organizations rely on manual orprojected ROI metrics, even for production AI systems. Agentic AI multiplies third-party risk Most enterprises connect agentic systems to 6 to 20external tools or services, expanding dependencyand cost exposure. Lifecycle management adoption has surged Use of commercial AI lifecycle management andgovernance platforms jumped from 14% to nearly50% year over year. I N T R O D U C T I O N A new phase for enterprise AI After heavy investment, hundreds of AI use cases are proposed,developed, and piloted each year in large enterprises.But progressing from AI pilots to production remains ahuge challenge, and as a result, leadership attention isshifting to a consequential question: what measurablevalue is AI delivering once it is live, at production scale? cross industries, executives face growingpressure to explain AI outcomes in financial andoperational terms. Boards want transparency.Regulators expect accountability. AI leaders, CIOs, andCFOs demand clarity on costs, risk, and returns. In thisenvironment, activity alone no longer signals progress.Leaders must be able to deliver AI use cases through arepeatable, governable, and economically accountableoperation across the enterprise portfolio.A The problem is not a lack of ambitionor innovation. Enterprises are notfailing at building AI. They are failingat industrializing AI delivery andrunning AI as a managed portfolio.Leading organizations are respondingby rethinking AI delivery itself—treating it as an operating discipline,not a deployment milestone. Byembedding governance, monitoring,and accountability directly intodelivery workflows, they movebeyond the illusion of value andtoward industrialized AI portfoliomanagement. This tension is intensified by therise of generative and agenticsystems. A single use case cannow reach production in months,often accelerated by third-partyplatforms and prebuilt capabilities.While this speed lowers barriersto innovation, it also expandsdependency, cost exposure,and third-party risk. Systems thatoperate continuously, interact withmultiple tools, and evolve overtime cannot be understood—orcontrolled—through periodicreviews or manual reporting. past the pilot stage, and confidencecollapses without operationalintelligence. Visibility fades as costsaccumulate without clear attributionand ownership fragments acrossteams and vendors. What initiallyappears successful during a pilotoften becomes difficult to explain,measure, or defend months later. Based on a global survey of 100senior AI, data, and technologyleaders, this report examines whyAI value fails to scale even as usecases proliferate and deploymentaccelerates. It explores wherefriction can limit scale, why ROIremains difficult to track, and howenterprises become overwhelmedby expanding portfolios rather thanempowered by them. The result is a growing value illusion.Activity creates confidence: moremodels deployed, more agentsconnected, more integrations added.But many use cases still fail to move This report explores that transition—and why those that fail to make it willremain bottlenecked by explodingportfolios, trapped in the illusion ofvalue they believe AI is creating. C H A P T E RO N E The AI portfolio explosion than 100 proposed AI and machinelearning use cases. In 2026, portfoliosizes have shifted dramatically: 67% oforganizations report between 101 and250 proposed AI and ML use cases,while a growing share report more than250. Only 21% now report fewer than100 proposed use cases. Over the past fewyears, enterprises havelaunched a historic wa