从试点到生产：如何大规模操作人工智能（英）

An Executive Summary · Coastal AI Operations Report 2026 A B O U T Oxford Economics Oxford Economics specializes in evidence-based thought leadership, forecasting, and economic impactanalysis. Our economists use sophisticated analytical models and have access to a rich database of figures,forecasts, and analysis on 200 countries, 100 industrial sectors, and 8,000 cities and regions. Headquarteredin Oxford, with offices around the world, we employ more than 600 people, including over 400 economists,industry experts, and business editors. The rigor of our analysis, caliber of staff, and best of-class globaleconomic models and analytical tools make us a trusted resource for over 2,500 corporations, financialinstitutions, government organizations, professional service firms, and universities. For more information,visitwww.oxfordeconomics.com. Contents 01The Expectation/Performance Gap04 02AI Misconceptions05 Three Markers That Separate Leaders from the Field06 04The Value Leaders Profile09 05Partnerships and Closing the Gap12 The Readiness Gap Is Measurable and Bridgeable13 08Continue Reading14 The Expectation/Performance Gap01 Common misconceptions about AI can lead decision-makers to choose unproductive solutions when betteroptionsare within reach.Research conducted byOxford Economics and Coastal in the spring of 2026looks at how 800 US-based executives from diverse in-dustries are approaching AI implementation, navigatingpilotdeployments,and integrating tools into work-flows. We found that AI value is often limited by data-access barriers and poor change management, and lessso by funding or technology shortfalls — and that theorganizations with stronger performance take a specificset of steps to keep initiatives on track. While some or-ganizations and industries start with particular advan-tages, businesses of all types can follow a clear path toAI success. As faster, smarter, and more integrated AI ca-pabilities power an expanding suite of applica-tions, business leaders are encountering a fa-miliar problem: Friction. Executives struggle torealize expected ROI, managers run into IT-im-posed blockades, and employees still preferfamiliar ways of working over vetting AI out-puts of variable quality. The technology is sup-posed to make things easier — so why doeseverything still feel so complicated? Maturity Snapshot: Where Enterprises Stand AI integration across U.S. enterprises remains relativelyshallow for most organizations. Most use AI for dataanalysis and reporting (54% mostly or completelyintegrated into daily workflows), but integration dropssharply for use cases that involve more AI autonomy.Conversational interfaces like chatbots and copilots(39%) and AI embedded into existing systems to shapeuser experiences (30%) see occasional use, while onlyjust over one-quarter augment human decisions withrecommendations or AI insights (27%). Most build AI capabilities using a hybrid methodology(42%), but some make use of standalone tools thatwork alongside existing systems (31%). Only 8% buildon a custom infrastructure — an indicator of thelargest, most mature organizations in the sample. AI has a more meaningful impact within certainfunctions. Customer service and support (54%), IT andinternal tooling (49%), and sales and revenueoperations (44%) all have well-established use cases,consistent with where organizations have been pilotingthe longest. Only 11% of respondents have fully autonomous AI usecases integrated into workflows today, while most arein less mature stages. Pilot volume tells a similar story:on average, organizations have 3.28 AI pilots in playtoday, and about a fifth have just one (22%). AI Misconceptions02 The commitment to AI spending and innova-tion is nearly universal across our survey sam-ple, with over half of respondents saying lim-ited budgets from leadership are essentially anon-issue(56%),and stakeholder buy-inranked as the lowest-cited constraint on suc-cessful AI pilots at 23%. In fact, investment ap-petite is accelerating: 74% plan to increase AIspending over the next 12 months and essen-tially no one plans to cut back. Executivesare approving budgets because of theplethora of practical business applications for AI tech-nologies. Over half of respondents (54%) say their AIoutputs fit real-world workflows — and just 27% cite alack of clear use cases as a primary constraint whengetting AI pilots off the ground. Yet there remains a persistent gap between expectedvalue and realized performance. While 84% of surveyrespondents say their AI investments have improvedtheir competitive positioning, only 54% say those sameinitiatives have actually met or exceeded the expecta-tions set going in. Business leaders are still seeking outthe root cause of this shortfall. The gap between the share of executives saying AI improvedtheir competitive positioning (84%) and the share sayinginitiatives met or exceeded expectations (54%). 30pts Three Markers That SeparateLeaders fro