人工智能如何塑造资产管理

Contents Executive summary3Survey overview4AI integration in the investment process5 Nature of AI capability7Resourcing dedicated to AI development9Benefits observed from AI integration so far10Main barriers to further AI adoption11 Conclusion13 Executivesummary Key findings: 55% report AI is integrated in atleast one of their strategy’sinvestment processes 18% have not yet integrated AI into anypart of their investment process This white paper summarizes the results of aMercer survey of 131 asset managers conductedin February 2026 on how artificial intelligence (AI)is being integrated into the investment process.According to the survey findings, the industryis in an active early-adoption phase: more thanhalf of firms report live AI integration in at leastone strategy, and a substantial cohort reportspilot or proof-of-concept activity. Firms reportthe strongest measurable benefits in operationalefficiency and in the speed/quality of insights,rather than in improved returns orriskreduction.1 91% plan to increase their useof AI in the next 12 months 69% report data constraints as asignificant barrier that preventsfurther AI adoption in theirinvestment process AI adoption today is concentrated in ideageneration, processing unstructured data,and signal generation. For trade executionand portfolio construction, AI utilizationremains uncommon. The primary barriers to AIadoption are data quality/access, regulatory andcompliance concerns, and systems integration.Respondents identify data governance andsystem-level risks (including herding) as themost significant regulatoryblind spots. 73% are using AI for operationalefficiency in their existingteams (for example,automating routine tasks) 28% have internally developed AI models Survey overview The Mercer 2026 AI in Asset Management Surveywas conducted via an online survey in February 2026.Responses were provided by 131 asset managersfrom acrossthe world. 1. Firm-size distribution by assets undermanagement USD$, as at December31, 2025 Respondents were asked about the AI integration stage,use cases across the investment lifecycle, the nature of AIcapabilities, resourcing, data inputs, measurable benefits,staff education, barriers, and regulatoryblind spots. AI integrationin the investmentprocess Artificial intelligence has arguablybecome an essential componentof our daily lives, and assetmanagement firms are no exceptionto this trend. The adoption of AI isno longer merely a trend that firmsfollow to maintain market presence.Rather, we believe it has becomea decisive factor for operationalefficiency and faster decision-making. The primary objective ofthis survey was to gain insight intohow asset management firms arepresently incorporating AI into theirinvestment processes. The asset management industryis in an active early-adoptionand scale-up phase: 55% of firmshave integrated AI in at least onestrategy’s investment process, and27% of firms have integrated it asa pilot or proof of concept. Thebalance of 18% of firms have notyet integrated AI. We asked all firmswhether they plan to increase AI usein the investment process, and 91%said they do, so we would expect the18% figure to decrease significantlyover the next 12 months. In terms of the areas of theinvestment process that currentlyuse AI, the chart below showsthat current deployments areconcentrated on upstream,research-oriented workflows and onunlocking value from data. Amongthe areas that remain nascent inAI adoption are trade executionand portfolio construction. Nature of AI capability AI is being used predominantly to seekimprovements in productivity and augmenthuman decision-making (co-pilot). 73% of firms are using AI to improve operationalefficiency within their existing teams (for example,automating routine tasks), while 68% of firmsgo beyond that, using AI as a partner in theinvestment process (for instance, providinginsights and analysis). Few firms (only 5%) entrustAI with autonomous or semi-autonomousdecision authority in investment decisions suchas trade orinvestment recommendations. complexity, and speed-to-market considerations.Only 9% of firms are currently operating a fullyproprietary AI platform. While understandablefrom a resourcing and cost perspective, this doespotentially raise the issue of vendor (and perhapseven model) concentration risk (identified as aconcern by 9% of respondents —seepage 12) 4. Nature of AI capability (multipleanswers permitted; percentages basedon the total number ofasset managers) 73% of firms areusing AI to improveoperational efficiencywithin their existingteams We use vendor/off-the-shelf AI tools We use vendor tools with proprietary tuningor custom model configurations We have internally developed AI models The market currently relies heavily on vendorsolutions, with 63% of firms using off-the-shelfAI tools and 51% of firms applying proprietarytuning to vendor models. Fully in-house platformbuilds are less common (28%), reflecting