从探索到实践：面向金融服务领导者的AI新前沿

Fromdiscoveryto doing Contents Executivesummary Banks and fintechs have a long history ofexperimenting with and deploying artificialintelligence (AI). Now the industry is entering anew wave of adoption. After initially leveraginggenerative AI tools offering prompt‑basedassistance, the sector is embracingagentic The rise of agentic AI is poised to fundamentallyreshape the future of financial services,requiring a clear strategic vision combinedwithdisciplined change management.Successfully navigating this transition will opena new frontier, where AI’s full capabilities unlock The emergence of threenew agentic economies From helping to doingThe two waves of AI Generative AI for helping Three transformative customer paradigmshifts driven by agentic AI will define the yearsto come, reshaping customer experiences, Banks and fintechs are investing in generative AIto improve operational efficiency and innovatecustomer interactions. Common applications TheAssistance Economywill seeagents delivering entire customerexperiences — across a range of engagement Adaptive Customer Experienceswill enablebanks and fintechs to deliver interfaces thatadjust in real‑time to preferences and context, Agentic AI for doing Agentic AI builds on generative capabilitiesby autonomously completing multi‑stepworkflows such as performing deep research, Agentic Twinswill represent their owners withagency and trust, becoming a key part of the Banks and fintechs are adopting AI alongdifferent trajectories but both face a criticalchoice: either incrementally extend existingpropositions, or fundamentally rebuild processesand offerings. Both approaches have potential,but success will depend on decisive leadership This report serves as a guide through thetransformative journey from initial generativeAI adoption to the age of agentic autonomy.It highlights the critical enablers required to section oneInflection From helping to doing Current tools are the tip of the AI iceberg. The financialservices industry will see fundamental reimaginingof service and execution as today’s AI Assistants are The journey so farFrom Wave 1 to Wave 2 Generative AI for helping based on the customer’s financial profile, andcoordinate with third parties on the client’sbehalf. It would then submit applications andpresent validated loan options for final user The first wave of AI adoption has seenexcitement and the rapid rollout of AI ChatAssistants, such as Gemini, Copilot, andChatGPT. They offer detailed answers to user The agentic era will drive a profoundtransformation not only in customer experiencebut also in the fundamentals of how financialinstitutions operate. By leveraging AI agents,banks and fintechs will unlock underserved For example, a customer looking for a mortgagemight ask an AI Assistant to lay out thenecessary steps and then present and compareoptions. The tool will summarise advice and Agentic AI for doing Much like how the gig economy emergedfrom the convergence of mobile technologyand GPS, new economic models will arisearound agentic AI. Success will not come frommerely layering these capabilities onto legacy In contrast to the current wave’s focus onisolated point solutions, the next frontier willbe defined by autonomous action rather thanadvisory support. AI Assistants will be capable Why is the transition happening now? The ongoing transition from point solutionsto agentic AI is being driven by continuoustechnological advancements, and business’ ofrespondents working in financial servicesexpress willingness to work with agents as a “co‑worker.”2The transition will be gradual,but disruptive. Consider the emergence ofmobile banking, which shifted the retail banking Note 2: “Finance” score reflects the average score across three finance occupations evaluated — “Financial Managers”,“Finance and Investment Analysts”, “Personal Financial Advisors;” ‘Government’ score reflects the average score acrossfive government‑related occupations evaluated — “Administrative Services Managers,” “Child, Family, and School Social The four tech advances makingagentic AI possible Reasoning The performance of LLM models has improvedsignificantly, unlocking new use cases andagentic applications. In financial services,however, keeping a human in the loop will “The interaction we aretrying to create here is AIworking for you — moving Memory Both short‑and long‑term memory is nowavailable to retain user information (e.g., /Melissa Ann-ChanChief Marketing Officer, ArtaFinance Data Data has become more valuable as largelanguage models (LLMs) have expanded contextwindows — with the largest growing around Action Secure connections to other services, systemsand models are now possible, so agents canaction tasks by calling on external tools such sectiontwoDual pace of adoption Dual AI strategies are emerging — incumbent banks havea lot to gain from back‑office efficiencies, while fintechs are Incumbent banks KnowYour Customer (KYC)