面向客户至上组织的代理流程自动化

Disclaimer This point of view reflects our current understanding and practical experience. It is provided forinformational purposes and does not claim to be error-free. To the extent permitted by law, liabilityand recourse claims are excluded. Where charts or tables do not cite external sources, the data isderived from Capgemini surveys or internal analysis. Executive summary03 The evolution of agentic processautomation04 Our vision for customer first in theagentic era06 Agentic process automation forcustomer first-organizations.08 Capgemini value proposition for agenticprocess automation12 Industry-specific challenges and agenticsolutions18 Proven success in agentic customeroperations22 Key takeaways for your agentic journey26 Call to action for organizations28 Appendix A: Delivery framework foragentic automation (Phases 0–8)30content Appendix B – Sector-specific challengesand agentic solutions (examples)35 Sources38 About Capgemini38 The evolutionof agentic processautomation frameworks for orchestrationand interoperability dramaticallyexpanding their capabilities notonly to answer questions, but toplan, collaborate, reason, executetasks, and improve continuously.AI had already surpassed humanperformance across many tasks,including several previouslychallenging benchmarks, withonly a few exceptions. Even inthose areas, the performancegap between AI and humans isshrinking rapidly. Technology providers areproductizing agent platforms,orchestration protocols, andagent builders that integratewith enterprise applicationssuch as customer relationshipmanagement (CRM), enterpriseresource planning (ERP), andinformation technology servicemanagement (ITSM). At thesame time, cost dynamics andmodel availability are improving,making experimentation easierto start – but not easier toscale responsibly. Automationhas evolvedfrom workflowstandardizationand roboticprocessautomation (RPA) into intelligentautomation augmented bygenerative AI and agenticsystems. The defining changeis not related to technologyas such; it is the ability to plan,coordinate, and act acrosssystems under new possibilitiesand explicit constraints. At the same time, thedeployment of agents requiresgovernance, guardrails, andthe ability to facilitate human-in-the-loop for agent andhuman collaboration. The coreleadership question is thereforenot “Can we build an agent?”but “Can we delegate decisionssafely at scale?” Organizationsthat succeed treat agenticautomation as an operating-model change – designed aroundorchestration, governance, andcontinuous optimization – ratherthan a one-time tooling upgrade. Agents were at the center ofthe global technology narrativein 2025. The past year hasmarked a turning point inautomated software agents,with breakthroughs in largelanguage models (LLM) and Our vision forcustomer first in theagentic era Customer-firstorganizationscontinuouslythrive to improvevalue propositionsand improvecustomer journeys. Thedifferentiator in the agentic erahas become the ability to sense,decide, and act quickly acrosschannels and functions – withoutsacrificing trust. to document as-is processesand incrementally design to-beflows remain useful but are nolonger sufficient on their own.New AI-powered tools helpincrease efficiency and changehow enterprises build futureblueprints to run their business.With any tools and collaborationformats, most importantly,leaders must reimagine customeroutcomes end-to-end andthen decide where autonomyis appropriate, where human judgment must remain, andhow the organization willgovern both. After early experimentation,clients are increasingly focusedon agentic “control planes”:governance structures andtechnical layers that overseecustomer flows across front,middle, and back office, ensuringauditability, recoverability andaccountable decision-making. Traditional transformationapproaches with workshops Figure 1Capgemini Agentic Enterprise Stack Our vision of the newenterprise architecture The glue breaking organizational and IT silos, with semantic layer, agent fabric andagent control plane as vital organs. Interactions Agentcontrol plane Agent fabric The combination oftools and platformsfor building,evaluating anddeploying agents andtheir integration withenterprise resources •Identity& TrustGovernance •Action& PolicyGovernance •Agents Gallery &Market Places •Agentic workflows,contexts &orchestrations –low/pro code tools •Audit,Compliance &Evidence •Agentic integrationwith enterpriseresources - AImodels, data& knowledgeproducts, businessoperations systems Systems of record Composable Core Enterprise &operations systems (incl. transactiontruth and compliance) Agentic processautomationfor customer-firstorganizations workflow automation, rulesengines, and RPA. Agents caninterpret intent, assemblecontext, recommend options,and execute actions wherepermitted – while humanspecialists retain authorityfor high-risk decisionsand exceptions. Leading practices for agenticcustom