Franklin Templeton AI Chief Sees Multi-Agent Systems Driving Growth

Scaling Enterprise AI in Asset Management With Franklin Templeton — Ai4

Ai4

Enterprise AI value requires redesigning end-to-end workflows and operating models rather than deploying isolated automation tools.

AI-assisted summary of Scaling Enterprise AI in Asset Management With Franklin Templeton — original source. Check the original for full context.

Source published: 2026-08-19

Detailed notes

## Background This talk, titled "Scaling Enterprise AI in Asset Management With Franklin Templeton," was presented by Lily Li (Head of AI Adoption and Solutions at Franklin Templeton) and published by the Ai4 channel on YouTube on August 19, 2026. The session highlights how a large asset management firm moves past isolated AI pilots to drive end-to-end organizational transformation, integrating agentic AI, multi-agent platforms, and trust-based adoption across research, operations, portfolio management, and client engagement.

## TL;DR Lily Li outlines Franklin Templeton's strategic roadmap to transition from simple AI experimentation to enterprise-wide scaling. She presents live production cases across back-office operations and front-office sales and research, where AI workflows deliver verified high accuracy and substantial labor reduction. The presentation details a multi-year timeline leading up to Decision AI and multi-agent systems, emphasizing that true competitive moats lie in redesigned operating models, institutional memory, and compounding workflow value rather than underlying AI models. Finally, she advises organizations to embed governance into platform architectures, focus on cost-per-outcome metrics, and continuously upskill talent through internal academies.

## Atmosphere & style The presentation carries an authoritative, pragmatic, and highly structured tone suited for an executive audience. Beginning with an interactive room survey that highlights the stark drop-off between AI pilots and production deployment, the narrative quickly shifts into an energetic, forward-looking roadmap. Li's delivery feels experienced and grounded, blending high-level strategic frameworks with concrete operational data to convey both urgency and disciplined execution.

## Key takeaways - [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) The primary challenge in enterprise AI is bridging the massive drop-off between running initial pilots and achieving production-grade scaling that delivers measurable business value. - [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) Frontier models are universally accessible table stakes, so competitive moats no longer come from raw models but from embedded controls, agentic usability, and total workflow value. - [02:41](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=161s) Enterprise AI scaling requires a three-part mandate focused on compounding long-term value, end-to-end workflow transformation, and workforce capability evolution. - [05:12](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=312s) Franklin Templeton follows a multi-year AI roadmap progressing from foundational strategic bets (pre-2024–2025) to platform scaling (2026), multi-agent value (2027), and real-time 'Decision AI' (2028+). - [08:41](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=521s) Production AI systems in middle and back-office operations have achieved a 95% verified accuracy rate and an 80% to 90% reduction in manual labor capacity. - [08:41](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=521s) A central Research Intelligence Platform and Common Agent Store allow cross-asset insights to be reused across the organization instead of building isolated point solutions. - [11:21](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=681s) In distribution, the Sales Intelligence Hub consolidates 15 disparate systems into a single interface, yielding expanded client coverage and higher value-add engagements. - [11:21](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=681s) Automated research agents deliver comprehensive earnings summaries directly to portfolio managers within one minute of an earnings call concluding. - [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) The firm executes a 'build and partner' technology strategy, co-developing with hyperscalers while strictly safeguarding internal domain knowledge, security, and agent orchestration frameworks. - [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) To successfully transform, leaders must focus on shifting entire operating models, tracking 'cost per outcome' versus token spend, and actively guiding employees into AI-native roles. - [18:59](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1139s) Organizations must transition from chasing quick-fix pilots with linear ROI to establishing durable foundations that fundamentally redesign how work operates.

## Detailed walkthrough ### [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) The Gap Between AI Pilots and Enterprise Value

Opening the session by engaging the audience, Lily Li asks attendees to raise their hands if their organization shipped AI pilots last year—prompting a large room response. However, when she asks them to keep their hands up if those pilots are currently in production and actively delivering measurable business value, almost every hand drops. "Very, very minimum," she notes, pointing out that this dramatic drop-off illustrates the exact problem her talk addresses: bridging the massive gap between initial experimentation and production-grade enterprise scaling.

As Head of AI Adoption and Solutions at Franklin Templeton—an asset management firm handling approximately $1.8 trillion in assets under management (AUM)—Li focuses on embedding artificial intelligence directly into the daily operating core of where employees work. Presenting industry data, she explains that while most organizations have moved beyond the "start small" or piloting stage, scaling across the enterprise remains the central challenge. The core obstacle isn't simply making AI work technically, but capturing tangible real-world business value.

Behind the scenes, four key shifts are quietly redefining the enterprise AI landscape: - **Multi-reasoning as a table stake:** Frontier foundational models are widely accessible to everyone, meaning the raw underlying model no longer provides a competitive edge. - **Agentic AI usability:** Organizations are heavily experimenting with autonomous AI agents, but most are still struggling to establish the usability and governance required to scale agents beyond isolated functions. - **Embedded control frameworks:** Risk and governance frameworks are shifting from late-stage "gatekeepers" or final approvers into fundamental design parameters that govern product workflows and system economics from day one. - **Shift from linear ROI to workflow value:** With industry events saturated by discussions around return on investment (ROI), leading firms are moving beyond simple linear ROI to measure total organizational throughput and systemic workflow value. Achieving this requires completely redesigning enterprise value chains.

---

[02:41](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=161s) Strategic Framework for Enterprise Scaling

To guide leaders looking to scale AI enterprise-wide, Li lays out a three-part mandate centered on value, end-to-end transformation, and talent development:

- **Value front and center:** Organizations must look past quick wins or single use-case ROI and invest in initiatives that compound value over time. This requires long-term commitment and shared accountability with domain and functional leaders. - **End-to-end transformation:** AI should serve as a catalyst to enable and accelerate complex processes that traditionally could not be automated. Crucially, this exercise surfaces the exact decision points and accountabilities where human expertise must shine, allowing machine automation to handle the rest. - **Talent and capability evolution:** Identifying the optimal split between human and machine work defines the next generation of required organizational skills, establishing a roadmap for hiring, upskilling, and transitioning the workforce into an AI-native enterprise.

Systemic execution requires robust organizational foundations. Organizations must harness enterprise knowledge in structured memory spaces supported by a standardized platform that provides data integration, cross-firm orchestration, and embedded governance. Operating this framework creates a continuous feedback loop: strategic vision sets direction, direction shapes execution, empowered talent takes accountability, standardized platforms enable delivery, and embedded governance ensures responsible, measurable outcomes.

---

[05:12](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=312s) Franklin Templeton's Multi-Year AI Transformation Roadmap

Li outlines Franklin Templeton’s multi-year journey to transform its global organization of roughly 10,000 employees across diverse asset classes and investment strategies:

``` +-----------------------------------------------------------------------------------+ | FRANKLIN TEMPLETON AI ROADMAP | +-------------------+--------------------+--------------------+---------------------+ | Pre-2024–2025 | 2026 | 2027 | 2028+ | | Foundational Runway| Scaling Year | Tangible Value & | Decision AI Era | | & Strategic Bets | Platform Expansion | Multi-Agent Systems| Real-Time Moat | +-------------------+--------------------+--------------------+---------------------+ ```

- **Pre-2024–2025 (Foundational Runway):** Franklin Templeton established its dedicated AI organization early, prior to the launch of ChatGPT in late 2022. Over the following two years, the firm built a proven operating model, testing strategic bets that demonstrated clear production value across core business functions. - **2026 (The Scaling Year):** The firm focuses on replicating its proven operating model across flagship platforms in distribution, investments, operations, and enterprise governance to scale impact by "10 times and 100 times." This effort is backed by unified public and proprietary data architectures, clear governance policies, top-down steering from executive committees, and bottom-up employee upskilling through an internal AI Academy. - **2027 (Measurable Bottom-Line Impact & Multi-Agent Systems):** As multi-agent orchestration platforms become the standard operational norm, harnessed institutional knowledge will surface as top-of-the-house intelligence. Expected results include measurable bottom-line growth, expanded market share driven by AI-empowered frontline sales teams, and disruptive applications like autonomous "agentic investment teams" operating within controlled sandbox environments. - **2028+ (The Era of Decision AI):** Li envisions a transition into "Decision AI," where early-moving organizations realize durable competitive moats through real-time autonomous actions, dynamic intelligence, and entirely new AI-driven product streams.

---

[08:41](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=521s) Production AI Systems in Middle/Back Office and Investment Core

Moving from strategy to concrete implementation, Li highlights several modular AI platforms that have been live in production across Franklin Templeton for 9 to 10 months, delivering verified user adoption and measurable business outcomes:

#### Operations & Corporate Functions - **Reconciliation & Trade Operations:** Custom modular platforms handle tedious, manual reconciliation and data matching end-to-end. Reusable components are extracted and retrained across different operational models, including trade operations. - **AI Communication & Triage Modules:** Intelligent communication engines automatically triage incoming messages across back-office operations and corporate functions. - **Automated Review Engines:** Systems automatically evaluate complex financial statements, contracts, or unstructured data inputs against regulatory and institutional rulebooks, flagging anomalies down to specific footnotes.

#### Investment Core & Research Intelligence - **Research Intelligence Platform:** A central platform ingests public filings, internal proprietary research, and meeting transcripts across investment communities to create a unified institutional knowledge base. - **PM Copilot & Multi-Agent Orchestration:** A multi-agent system acts as an extension of the portfolio management team. Specialized agents serve as virtual investment analysts or portfolio coaches, digesting the same information streams as human team members to highlight fresh analytical perspectives and augment investment decisions. - **Common Agent Store:** Distills cross-asset insights across equities, fixed income, alternatives, and digital assets into reusable, centralized agents (covering macroeconomics, sensitivity analysis, news catalysts, and market events). This prevents disparate teams from redundantly building isolated point solutions.

- Middle and back-office AI operational workflows operate with human-in-the-loop oversight, where humans focus on exception review and final approvals. - Production operational AI systems have maintained a verified 95% accuracy rate over 10+ months of live deployment. - Automated operational workflows deliver an 80% to 90% reduction in manual labor capacity requirements.

---

[11:21](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=681s) Frontline Sales, Marketing, and Enterprise Enablement

Li details how AI transformation extends directly into client-facing distribution and corporate marketing functions:

- **Sales Intelligence Hub:** Historically, sales representatives had to navigate up to 15 disparate software systems daily to manage client relationships. The unified Intelligence Hub consolidates territory management, Salesforce advisory engagements, targeted client focus lists, pre/post-meeting summaries, and actionable analytics into a single AI-driven workflow interface. - **RFP & DDQ Platform:** An intelligent Request for Proposal (RFP) and Due Diligence Questionnaire (DDQ) platform automates complex subject matter expert engagement, eliminating back-and-forth email streams. The system projects accurate draft responses based on historical repository records. - **Rep Response:** Fast-tracks localized investment viewpoints and market insights directly into the hands of frontline sales reps and clients. - **Componentized Marketing Content:** Moving beyond basic translation, proofreading, tone dialing, and compliance checks, marketing teams leverage AI to transition toward componentized, dynamic content management systems for large-scale campaigns.

- The Sales Intelligence Hub has generated net-measurable gains compared to historical benchmarks, including expanded account coverage, measurable sales lift, and higher value-add client engagements. - Research agents automatically generate and deliver comprehensive earnings summaries and actionable insights directly to portfolio managers within one minute of an earnings call concluding.

---

[15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) Execution Strategy, Enterprise Stack, and Core Takeaways

Explaining how Franklin Templeton executes its technology strategy, Li emphasizes a "build and partner" philosophy. The firm forms strategic co-development partnerships with major cloud hyperscalers and frontier AI startups while strictly keeping internal memory, proprietary domain knowledge, specialized workflows, and agent orchestration frameworks within its own security fence. This hybrid approach preserves competitive differentiation while providing the flexibility to seamlessly plug and play evolving third-party models.

Li warns against building isolated point solutions, running perpetual pilots, or generating flashy demos for innovation's sake. Every initiative must be co-created directly with domain experts to ensure fit-for-purpose delivery. Furthermore, enterprise control and auditability must be embedded directly into every layer of the enterprise AI architecture, ensuring smooth alignment with regulatory authorities.

To conclude, Li leaves enterprise leaders with three fundamental action items: 1. **Transform operating models, not just tools:** Focus on updating organizational structures, end-to-end workflows, and talent development strategies rather than deploying isolated software tools or chasing quick wins. 2. **Focus relentlessly on compounding value:** Align leadership around value creation, measuring ROI strictly by tracking "cost per outcome" and evaluating output metrics against total token and compute spend. 3. **Bring your people along:** Address enterprise friction, fear, and FOMO by actively guiding employees into newly defined roles, providing clear career paths and continuous learning through internal AI academies.

[18:59](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1139s) Frameworks for Enterprise Scaling and Strategic Focus

Lily Li emphasizes that scaling AI effectively is currently where leading enterprises stand at this stage of adoption. To successfully achieve scale, organizations must incorporate key structural components into their framework and strategic planning. It is vital not to overlook crucial foundational elements, such as business values, the overall operating model, and the people executing the work, rather than getting caught up in isolated point solutions.

Moving beyond pilots requires a fundamental shift in how organizations approach AI initiatives. Organizations should actively "start operating and stop piloting". Instead of chasing quick, point-solution wins that deliver merely linear ROI—which will likely be discarded or obsolete by the following year—businesses need to build durable, scalable foundations that transform how the organization functions

## Notable quotes > [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) Very, very minimum — Lily Li

> [05:12](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=312s) 10 times and 100 times — Lily Li

> [05:12](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=312s) Decision AI — Lily Li

> [05:12](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=312s) agentic investment teams — Lily Li

> [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) build and partner — Lily Li

> [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) cost per outcome — Lily Li

> [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) Transform operating models, not just tools — Lily Li

> [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) Focus relentlessly on compounding value — Lily Li

> [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) Bring your people along — Lily Li

> [18:59](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1139s) start operating and stop piloting — Lily Li

## Mentioned Lily Li, Franklin Templeton, Ai4, YouTube, ChatGPT, Salesforce, Sales Intelligence Hub, AI Academy, PM Copilot

## Video description Lily Li - Head of AI Adoption and Solutions, Franklin Templeton

AI is delivering on its promise, and leading enterprises are already on the journey to scale its impact. As companies move beyond siloed pilots, the focus shifts to integrating AI into how the business actually runs. This session explores how a large asset manager positions AI as a driver of end-to-end transformation, catalyzing new ways of working that deliver real business value across research, portfolio management, distribution, operations, and client engagement. It highlights how agentic AI, multi-agent platforms, business co-creation, and trust-based adoption are embedded from day one, enabling the scaling of common, reusable AI capabilities across the organization.

What You'll Take Away: Beyond the pilot — why the winners redesign how work happens, not just automate tasks. Reusable by design — the platform approach that compounds value instead of multiplying copilots. The real moat — why it isn't the model. Trust from day one — scaling AI where the stakes, and the regulators, are real. What's next — the shift to AI-native operating models, and what it means for your business.

Who Should Attend: Investment, distribution, technology and operations leaders ready to move past experimentation and drive enterprise-wide AI transformation.

[Source](https://www.youtube.com/watch?v=JO2WtYrsfbM)

中文简报

Franklin Templeton AI主管：多智能体系统将驱动增长实现企业级AI价值需要重构端到端工作流与运营模式，而非仅仅部署孤立的自动化工具。 Lily Li概述了Franklin Templeton从简单的AI实验迈向企业级规模化落地的战略路线图。她展示了涵盖后台运营以及前台销售与研究的实际投产案例，证明AI工作流能够带来经验证的高准确率并大幅减少人工投入。演讲详细介绍了通向决策AI（Decision AI）与多智能体系统的多年期时间表，并强调真正的竞争护城河在于重构的运营模式、机构知识沉淀以及复合工作流价值，而非底层AI模型本身。最后，她建议各企业将治理机制内嵌于平台架构中，重点关注单位成果成本指标，并通过内部学院持续提升员工技能。企业AI面临的首要挑战，在于弥合初期试点项目与交付可衡量业务价值的生产级规模化落地之间的巨大断层。 (0:08) 前沿模型已成为普遍可及的基础门槛，竞争护城河不再来自底层模型，而是来自内嵌控制机制、智能体可用性以及工作流的整体价值。 (0:08) 企业级AI规模化落地需要聚焦于三项核心任务：实现长远复利价值、推进端到端工作流变革，以及推动员工能力的演进。 (2:41) Franklin Templeton遵循多年期AI路线图，依次推进基础战略投入（2024–2025年前）、平台规模化（2026年）、多智能体价值转化（2027年）以及实时“决策AI”（2028年及以后）。 (5:12) 中后台运营中投产的AI系统已实现95%的验证准确率，并减少了80%至90%的人力负荷。 (8:41)

Reading summary

Lily Li, Head of AI Adoption and Solutions at Franklin Templeton, argues that capturing enduring value from enterprise artificial intelligence requires moving past isolated pilots to transform end-to-end operating workflows, anchor institutional memory, and embed governance from day one. Bridging the Pilot-to-Production Value Gap (0:08) While most financial enterprises have deployed AI pilots, Li observes that very few successfully transition these experiments into production systems that generate measurable business value. She identifies four structural shifts redefining the enterprise landscape: frontier multi-reasoning models have become table stakes rather than proprietary differentiators; autonomous agents require mature usability frameworks to scale; governance must shift from late-stage review into upfront system design; and evaluation metrics must transition from simple linear ROI to broader systemic workflow throughput. Multi-Year Roadmap from Scaling to Decision AI (5:12) To modernize Franklin Templeton's 10,000-employee global footprint and roughly $1.8 trillion in assets under management, the firm designed a phased multi-year transition. Following foundational capabilities and functional tests established before 2025, the firm designated 2026 as its year of platform expansion, aiming to scale operating models tenfold to a hundredfold across distribution, investments, and operations. Looking further ahead, Li forecasts that 2027 will establish multi-agent systems and top-of-house intelligence as the baseline operational norm, predicting measurable bottom-line growth and autonomous agent investment teams in controlled sandboxes. By 2028 and beyond, Li anticipates the emergence of 'Decision AI,' envisioning real-time autonomous execution and dynamic intelligence as durable competitive moats for early adopters. Production Deployment in Operations and Investment Research (8:41) Franklin Templeton has moved several modular AI systems into production, maintaining them for 9 to 10 months. In middle- and back-office operations, the firm deployed automated modules for reconciliation, message triage, and document review against rulebooks. Operated with human-in-the-loop exception handling, these operational workflows have demonstrated a verified 95% accuracy rate over 10-plus months while reducing manual labor capacity requirements by 80% to 90%. In the investment core, the firm built a centralized Research Intelligence Platform to ingest public filings, meeting transcripts, and proprietary internal research. Alongside this, a multi-agent PM Copilot system acts as a virtual investment analyst for portfolio managers—delivering actionable earnings call summaries within one minute of an event's conclusion—while a Common Agent Store standardizes macroeconomic and catalyst analyses across equities, fixed income, and alternatives. Unified Sales Intelligence and Frontline Distribution (11:21) In client-facing operations, Franklin Templeton targeted fragmentation among sales representatives who previously navigated up to 15 disconnected software systems daily. The firm's unified Sales Intelligence Hub consolidated territory management, client focus lists, and meeting analytics into a single interface, yielding measurable gains in account coverage and sales lift compared to historical benchmarks. Distribution enablement also includes automated RFP and DDQ platforms that generate initial drafts from historical data repositories, systems that accelerate localized investment viewpoints to sales teams, and componentized content infrastructure that supports dynamic, compliance-checked marketing campaigns. Hybrid Architecture and Compounding Economic Value (15:19) Franklin Templeton executes its technology strategy via a 'build and partner' model, co-developing with cloud hyperscalers and frontier AI startups while strictly retaining internal institutional memory, proprietary data, and orchestration logic within its internal security boundary. Li emphasizes that every project must be co-created with domain experts and audited across all architectural layers. To ensure sustained financial return, Li advises measuring AI success through cost per outcome and token compute spend rather than pursuing one-off, linear wins. Furthermore, she stresses the necessity of mitigating workforce anxiety by providing clear career pathways and continuous education through initiatives like an internal AI Academy. Li concludes that organizations must actively 'start operating and stop piloting' to avoid building short-lived point solutions that quickly become obsolete. However, realizing the firm's projected future milestones—particularly autonomous multi-agent investment teams and Decision AI by 2028—remains contingent on navigating evolving regulatory standards and maintaining technical control as agentic systems expand.

精炼阅读摘要

富兰克林邓普顿（Franklin Templeton）AI采用与解决方案负责人Lily Li指出，资产管理行业虽然普遍启动了AI试验，但在将试点推向生产环境并创造可衡量商业价值上面临严重鸿沟。该机构通过将AI嵌入日常运营核心、重构端到端业务流程，并践行“自研与外部合作并重”的系统架构，致力于实现从单点线性ROI向复合工作流价值的全面跨越。突破试点瓶颈与企业AI范式转移 (0:08) Lily Li结合现场调研与行业数据指出，多数企业虽已跨过概念验证阶段，但极少数试点能真正进入生产环境并带来可衡量的业务成效；核心阻碍并非技术可行性，而是能否在现实中捕获业务价值。她总结了正在重构企业AI格局的四大转变：首先，前沿大模型已成为普及的基础设施，底层基础模型本身不再构成竞争护城河；其次，智能体（Agentic AI）虽受热议，但多数机构在跨职能扩展时仍受制于可用性与治理框架；第三，风控与合规正在从后置的事后审批，转变为在产品工作流设计之初就内嵌的系统设计参数；第四，考核标准正从孤立的单点线性投资回报率（ROI），转向衡量组织整体吞吐量与工作流重塑价值。为推动企业级规模化，Li提出三大战略原则：必须将创造长期复利效应的业务价值置于核心，而非追逐短期速赢；必须推动端到端业务转型，明确识别人类专业智慧必须主导的决策节点，将复杂繁琐的其余流程交由机器自动化；必须持续升级组织人才结构，通过明晰人机协作分工，为员工提供AI原生企业的技能进阶与转型路径。多年转型路线：从底层跑道迈向决策AI (5:12) 针对管理约1.8万亿美元资产、拥有约10,000名全球员工的庞大体系，富兰克林邓普顿制定了多阶段演进路线图。Li介绍，该机构早在ChatGPT于2022年底推出前便成立了专职AI团队，并在2024至2025年的“基础跑道”阶段建立了经检验的运营模型，在核心业务职能中验证了生产级价值。按规划，2026年被定为“规模化扩张之年”，重点在于将成熟运营模式复制至分销、投资、运营及治理等旗舰平台，目标是将影响力提升“10倍至100倍”，并依托高管委员会的自上而下统筹与内部AI学院的员工赋能提供支撑。按照其阶段预测，2027年多智能体协同编排将成为标准运营范式，沉淀的机构知识将转化为顶层智能，从而带动底层净利润增长，甚至在受控沙盒中探索自主运行的“智能体投资团队”；到2028年及之后，机构预计进入“决策AI（Decision AI）”时代，先行企业将通过实时自主行动与动态智能构建起持久的竞争壁垒并开拓全新的AI驱动产品线。中后台与投资核心的生产级落地成效 (8:41) 在落地层面，富兰克林邓普顿已在中后台与投资核心部门部署了多套运行9至10个月的模块化生产系统。在中后台运营方面，定制平台实现了对账与交易操作的端到端自动化匹配，智能审查引擎可根据监管与内部规则自动比对复杂财务报表和合同，精准标注细至脚注的异常。在中后台系统中，人类员工保留在闭环中负责异常审核与终审，运行数据显示，这些系统在过去10个月以上的上线期内保持了经核实的95%准确率，使相关手工人力负荷缩减了80%至90%。在投资核心与投研领域，机构搭建了汇聚公开文件、专有研究与会议纪要的统一研究智能平台；同时部署多智能体“投资组合经理副驾驶（PM Copilot）”系统，让专业智能体担任虚拟分析师与辅助教练，为人类经理提供独立分析视角；此外还设立了跨资产类别的“公共智能体商店（Common Agent Store）”，避免各团队重复开发分散工具。目前，投研智能体已能在财报电话会议结束后1分钟内自动生成关键摘要并直接推送至投资经理。前端业务赋能与“构建与合作”架构原则 (15:19) 在面向客户的前端分销与营销职能中，转型同样产生了直接业务效益。销售智能枢纽（Sales Intelligence Hub）将以往销售代表每天需切换的多达15个独立软件系统统一整合至单一AI工作界面，带来了客户覆盖面扩大和可量化的销售提升；智能RFP与DDQ平台则基于历史存储库自动拟定问卷初稿，消除了跨部门往复邮件沟通；营销团队亦从基础校对翻译迈向动态组件化内容生成。Li强调，这些工具必须与一线业务专家协同设计，以确保契合实际应用场景。在技术架构执行策略上，富兰克林邓普顿坚持“构建与合作（Build and Partner）”哲学，即积极与主流云服务巨头及前沿AI初创公司开展联合研发，但严格将机构内部记忆、专有领域知识、特色工作流及智能体编排控制在自身安全边界之内。Li告诫企业切勿停留在孤立工具或炫目演示上，必须把可审计的合规控制内嵌至技术栈每一层，并严格依据“单项产出成本”以及对照Token与算力消耗来审视业务回报。 Lily Li总结指出，企业必须从无休止的概念验证转向常态化实际运营（start operating and stop piloting），把重心放在改造运营模式与人才培养上，而非追逐短期容易过时的单点工具。虽然富兰克林邓普顿展示了运营效率的大幅提升与前台协同成果，但在高度严苛的资产管理行业中，未来如何将沙盒中的自主智能体投资决策推广至生产核心，仍有赖于全行业在多智能体治理标准与责任分界上的进一步验证。

Detailed notes — zh

## Background 该内容由行业峰会频道 Ai4 于 2026 年 8 月 19 日发布在 YouTube 平台。视频记录了富兰克林邓普顿（Franklin Templeton）AI 落地与解决方案负责人 Lily Li 的主题演讲，深入探讨大型资产管理机构如何跨越孤立试点，将 Agentic AI、多智能体平台与端到端业务协同深度融入投研、分销与运营等核心场景，实现企业级 AI 的规模化价值沉淀。 ## TL;DR 演讲直击企业 AI 从实验走向规模化生产落地的断层，指出通用前沿模型已成行业标配，企业真正的护城河在于专有知识编排与端到端流程重构。Lily Li 详细拆解了富兰克林邓普顿在生产环境中已跑通的标杆案例，包括释放 80%~90% 负荷的自动化对账审查平台、聚合 15 个系统的分销端 Intelligence Hub，以及模拟专业投研团队的多 Agent PM Co-Pilot。她系统阐述了“自建与合作并重”的架构布局与 2026 至 2028 年演进蓝图，倡导企业从第一天起内嵌治理风控，坚决告别低效单点试点，全面转向具有复合收益的 AI 原生运营模式。 ## Atmosphere & style 演讲以直击痛点的现场举手互动开局，犀利揭示了行业内“试点遍地开花但实际投产寥寥”的严峻现实，迅速奠定了务实而富有共鸣的专业基调。Lily Li 展现出干练沉稳、行云流水的高管台风，全篇摒弃花哨的 Demo 噱头与空洞概念，完全以生产级数据指标和深度架构洞见支撑论点。整场演讲逻辑紧凑、节奏利落，在冷静剖析组织恐惧与变革阻力的同时，传达出极具穿透力的战略紧迫感与行动号召力。 ## Key takeaways - [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) 企业正面临从单点试点向生产级规模化扩展的跨越，随着顶尖模型的多步推理能力成为行业标配，模型本身已不再构成企业竞争壁垒。 - [02:50](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=170s) 企业规模化 AI 转型需要三大支柱支撑：聚焦长期复合业务价值、推动端到端业务重塑以明确人机职责，以及沉淀标准化平台底座与内嵌治理机制。 - [05:12](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=312s) 富兰克林邓普顿规划了 2026-2028 年 AI 演进路径：从 2026 年核心业务规模化扩展，迈向 2027 年多智能体协同，最终在 2028 年形成决策型 AI 护城河。 - [08:41](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=521s) 搭建跨资产类别的通用 Agent 共享中心（Common Agent Store），将宏观分析与催化剂监控等通用能力模块化封装，避免各业务部门重复造轮子。 - [11:16](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=676s) 分销端推出 Intelligence Hub 平台，将销售人员原本需穿梭的约 15 个分散系统全面整合，并利用 AI 自动化重塑 RFP/DDQ 协同与营销合规流程。 - [13:09](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=789s) 中后台运营实现约 95% 准确率的全流程自主对账与异常处理，释放了 80% 至 90% 的手动负荷；投研端实现财报电话会结束后 1 分钟交付深度摘要。 - [13:09](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=789s) 投资端推出 PM Co-Pilot 多智能体协同架构，通过集成研究、分析、交易与风控 Agent 模拟完整团队，让人类基金经理专注于高阶决策。 - [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) 践行“自建与合作并重（Build and Partner）”战略：依托外部云厂商与前沿模型，但将专有知识资产、业务记忆与核心编排层严格保留在企业边界内以打造真正护城河。 - [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) 坚决摒弃为了展示而做的孤立 Demo 和用例陷阱，坚持与业务专家共同创造（Co-create），并将合规审计与风控从第一天起前置内嵌于架构中。 - [17:03](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1023s) AI 规模化落地需追求平台能力的复合增长效应，建立基于“每个业务产出的单位推理成本（Cost per outcome）”等严肃量化度量体系。 - [17:03](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1023s) 消除员工不安全感与错失恐惧，管理层应主动重构岗位职责画像并设计人机协同职业路径，全面推动组织从“无休止试点”转向“实质化运营”。 ## Detailed walkthrough ### [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) 现场互动与企业 AI 演进拐点：从试点走向规模化扩展 Lily Li 开场向在场的行业领袖致谢，并邀请观众进行现场举手互动："去年你们机构上线过试点项目的请举手。" 此时台下许多人举手；紧接着她追问："如果你们的试点项目已经在生产环境中运行并实际创造了价值，请继续保持举手。" 结果举手的人寥寥无几。Lily 直言，从第一轮举手到第二轮举手之间的巨大鸿沟，正是本场演讲的核心议题。作为富兰克林邓普顿（Franklin Templeton，管理着约 1.8 万亿美元 AUM 的全球资产管理公司）的 AI 落地与解决方案负责人（Head of AI Adoption and Solutions），她的使命是让 AI 真正融入员工日常工作场景，并植入企业的运营核心。行业报告显示，大部分企业已经跨越了小步试水与试点探索阶段，当前面临的核心命题是如何在整个企业层面实现规模化扩展（Scaling）。AI 的演进背后有四个关键趋势正在悄然发生： ```mermaid flowchart TD accTitle: AI 企业级落地的四大核心支柱 subgraph Foundation ["底层技术与形态"] A["多模态/多步推理 (通用前沿模型成标配)"] B["智能体 AI (探索跨职能可用性)"] end subgraph GovernanceAndValue ["运营与商业机制"] C["内嵌式管控体系 (贯穿设计全流程而非事后卡点)"] D["工作流价值度量 (超越单一 ROI，重塑价值链)"] end A --> C B --> D ``` - **多步推理成为行业标配（Table Stakes）**：各大顶尖前沿模型人人可用，仅凭模型本身已无法构建竞争壁垒。 - **智能体 AI（Agentic AI）拓展**：企业普遍在探索 Agent，但关键在于摸索如何让 Agent 超越单一职能并在规模化场景中发挥实用性。 - **管控框架前置嵌入**：风控与合规不再是流程末端的审批把关者，而是直接嵌入到产品设计与业务流程中。 - **超越线性 ROI 的经济学考量**：不再局限于单一用例的快速回报，而是关注如何在端到端工作流中创造可量化的业务价值，重构企业的价值链与组织吞吐量（Organizational Throughput）。 --- ### [02:50](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=170s) 企业规模化转型的落地战略：价值、端到端变革与组织赋能面对如何规模化落地的挑战，Lily Li 提出了指引转型方向的三大核心组件： 1. **将价值置于核心（Value Front and Center）**：关注长期可复合增长的影响力，而非单个用例的短期收益。这需要与业务及领域主管达成共识并共同承诺。 2. **端到端业务转型（End-to-End Transformation）**：将 AI 作为催化剂，加速并赋能传统手段难以自动化的复杂流程。在此过程中，明确定义人类员工的核心价值、关键决策点与责任归属，让机器接管其余繁重工作。 3. **沉淀组织能力与标准化底座**：明确未来所需的人才技能画像并进行针对性培养与招募；同时搭建标准化的知识平台、数据接入与编排能力，将责任制与治理机制深度内嵌。 ```mermaid flowchart LR accTitle: 愿景驱动的持续扩展循环 V["战略愿景 (明确方向)"] --> D["业务流程重塑 (机器与人协作)"] D --> P["人才与责任 (人类决策与掌控)"] P --> T["数据与统一平台 (知识沉淀与编排)"] T --> R["可量化成果与合规 (复合价值交付)"] R --> V ``` --- ### [05:12](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=312s) 富兰克林邓普顿的演进路线图：2026 至 2028 年演进规划作为一家拥有约 10,000 名员工、覆盖多元资产类别的全球资产管理巨头，富兰克林邓普顿在 2022 年底 ChatGPT 问世前就已启动了 AI 组织建设。过去两年间，团队在若干战略级方向上跑通了运营模型，并交付了切实的生产成果，为未来的跨越式发展奠定了坚实跑道。 | 年份 | 阶段定位 | 核心举措与预期成果 | |:--- |:--- |:--- | | **2026 年** | **规模化扩展元年 (Scaling Year)** | 将成熟的运营模式复制到分销（Distribution）、投资（Investments）、运营（Operations）及核心治理等旗舰平台，实现 10 倍至 100 倍的业务影响力；统一公私数据资产，配合 AI Academy 培训、应用赋能体系以及高管/运营委员会组成的自上而下顾问委员会（Advisory Board）。 | | **2027 年** | **智能体协同与业务突破 (Multi-Agent Era)** | 多智能体系统与 Agent 编排成为日常规范；内部知识沉淀转化为顶层企业级智能（Top of the House Intelligence）；赋能一线销售拓展市场份额；沙盒中的 Agent 投资团队（Agentic Investment Teams）逐步落地，直接在财报底线（Bottom Line）体现可量化收益。 | | **2028 年** | **决策型 AI 新时代 (Decision AI Era)** | 领先企业将依托深厚智能沉淀、实时行动能力构建出真正的竞争护城河，形成持久的业务成果，开拓全新的产品形态与收入增长曲线。 | --- ### [08:41](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=521s) 生产环境真实落地案例：中后台运营与核心投资平台 Lily Li 强调，展示的并非概念验证，而是过去 9 到 10 个月内已经在生产环境中稳定运行、服务真实用户并带来可量化价值的实际模块： ```mermaid flowchart TD accTitle: 生产级 AI 模块应用架构 subgraph Operations ["中后台运营赋能"] M1["对账与数据匹配平台 (模块化设计，复用于交易运营)"] M2["智能通信分流模块 (覆盖中后台并扩展至公司级职能)"] M3["自动化审查引擎 (财报、合同与数据审查，精确识别附注异常)"] end subgraph InvestmentCore ["核心投资能力赋能"] I1["投研情报底座 (整合公开数据、专有研究与会议纪要)"] I2["PM Copilot 多智能体系统 (作为投资分析师/投资教练辅助决策)"] I3["通用 Agent 共享中心 (跨股票、固收、另类资产，提供宏观分析与催化剂监控)"] end Operations -.-> InvestmentCore ``` - **中后台运营与对账（Reconciliation & Operations）**：针对极其繁琐的手工对账流程构建了端到端平台。其高度模块化的设计支持更换不同模型进行微调，并成功迁移复用至交易运营（Trade Operations）中。 - **智能沟通分流与智能审查（Communication & Review）**：AI 通信模块能够自动执行跨部门信息分流；自动化审查引擎则结合业务规则与指标，全面审查财报、合同与各类数据集，能敏锐捕捉小到财务报表附注（Footnotes）级别的细微异常。 - **核心投研智库（Research Intelligence）**：搭建了集中式投研知识底座，深度捕获公开市场资讯、内部专有投研成果以及投资团队内部会议内容。 - **投资经理副驾驶与智能体团队（PM Copilot & Agentic Investment Teams）**：针对特定投资团队落地了多智能体协同的 PM Copilot，让 Agent 以"投资分析师"或"投资教练"的角色融入投研工作流，消化相同的信息流并提供全新视角。 - **跨资产类别的通用 Agent 商店（Common Agent Store）**：将宏观经济分析（Macroeconomics）、敏感性分析（Sensitivity Analysis）、突发新闻与催化剂（Catalysts）追踪等通用能力封装为开箱即用的 Agent，供股票、固定收益、另类投资及数字资产团队直接调用，避免各业务部门重复造轮子。 ### [11:16](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=676s) 分销与销售赋能：Intelligence Hub 与自动化响应平台在资产管理公司的前端销售与分销团队中，销售人员过去往往将大量宝贵时间耗费在繁琐的非客户沟通事务上。为了解决这一痛点，富兰克林邓普顿（Franklin Templeton）打造了统一的智能化 AI 工作流平台——**Intelligence Hub**。过去，销售代表在日常工作中需要反复穿梭并操作大约 15 个不同的业务系统；如今，他们只需登录这一个平台，就能搞定辖区管理、Salesforce 顾问互动记录、重点与非重点客户名单追踪、前置准备与行动建议报告，以及各类分析师互动的前置和后续跟进工作。与此同时，团队还构建了高度智能化的 RFP（征求建议书）与 DDQ（尽职调查问卷）协同平台。该平台不仅终结了传统冗长、碎片化的往返邮件链，将各领域业务专家（SME）的协同完全整合在一个界面内，还能基于公司历史记录智能推导并生成 AI 建议回复。此外，通过被称为“Rep Response”的工具，投资观点能够以极快的速度直接触达销售代表与客户；而在市场营销板块，AI 已经深度融入翻译、校对、语调调整（tone of voice）以及合规审查等全流程，当前团队正进一步探索组件化内容管理（componentized content management）的未来形态。 - **多系统整合**：将销售代表原本每日需切换的约 15 个孤立系统合并为一个统一的 Intelligence Hub。 - **全流程覆盖**：整合事前准备、事后分析、顾问互动管理与行动报告，大幅降低上下文切换成本。 - **SME 协同与响应**：RFP/DDQ 平台消除邮件拉扯，基于历史资产数据自动生成专业应答。 - **营销与合规沉淀**：内容生成、多语言翻译、语气把控与合规审查全面实现 AI 赋能。 --- ### [13:09](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=789s) 业务场景量化收益：智能运营与多 Agent 投资组合协同从各个业务条线的实际落地成效来看，AI 已经展现出极具说服力的净增量价值。在投研端，AI Agent 能够在财报电话会议（Earnings Call）结束仅仅一分钟后，就将提炼出的核心财报摘要与深度洞察直接交付到基金经理（PM）和研究员手中。在分销端，Intelligence Hub 自上线以来，已经在客户覆盖率（account coverage）、销售额提升（sales lift）以及高价值互动频次上带来了显著且可衡量的业务增长，目前正从美国率先落地的团队向全公司其他业务板块迅速推广。 ```mermaid flowchart TD accTitle: PM Co-Pilot 多 Agent 协同架构 subgraph InputSources["多源数据输入"] MKT["市场行情数据"] SELL["卖方研报"] INT["内部投研资产与观点"] end subgraph AgentOrchestration["PM Co-Pilot 专业 Agent 编排"] RS["投研 Agent (Research)"] AN["分析 Agent (Analysis)"] TR["交易 Agent (Trading)"] RK["风控 Agent (Risk)"] end subgraph HumanDecision["人机协同决策"] PM["基金经理 (PM) 把控投资逻辑 / 动态调仓"] end InputSources --> AgentOrchestration RS & AN & TR & RK --> PM ``` 在运营（Operations）领域，AI 实现了从交易对账匹配、异常个案处理、归因推理、跨部门与外部沟通，到最终系统调整的全流程自主处理，整个链条几乎无需人工干预，人类仅需保持“在环”（Human-in-the-loop）负责最终审批和偶发的异常复核。该系统在过去 10 至 11 个月的持续监测中，准确率高达 95% 左右，直接释放了 80% 至 90% 的人工工作负荷。而在投资管理端，**PM Co-Pilot** 绝非单一的聊天机器人（Bot），而是一个模拟完整投资组合管理团队的多 Agent 编排系统——集成了研究 Agent、分析 Agent、交易 Agent 和风险管理 Agent，它们深入理解基金经理的投资逻辑，让人类 PM 可以将全部精力集中在更高维度的策略调整与资产配置决策上。 - **投研响应速度**：财报电话会议结束后 1 分钟内生成专业提炼与洞察。 - **运营自动化指标**：实现约 95% 的端到端处理准确率，节约 80%~90% 的手动运营人力负荷。 - **销售可度量增长**：Intelligence Hub 显著拉动了客户覆盖面与实际销售转化。 - **多 Agent 拟人化架构**：PM Co-Pilot 聚合研究、分析、交易、风控及市场数据 Agent，形成端到端投资编排体系。 --- ### [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) 落地战略与护城河：自研与合作并进的企业级 AI 架构在落地路径上，富兰克林邓普顿坚持“自建与合作并重（Build and Partner）”的战略。团队一方面与超大规模云厂商（Hyperscalers）和前沿 AI 初创企业建立深度战略合作伙伴关系，开展联合研发；另一方面，坚决将核心业务记忆、专有知识库、独门工作流逻辑以及最关键的系统编排层保留在公司安全边界（Fence）之内。这种架构确保了底层模型和第三方供应商的即插即用与灵活切换，使企业既能时刻紧跟最前沿的技术迭代，又能牢牢锁死自身的核心竞争壁垒。 Lily Li 强调，企业必须告别为了创新而创新的心态，“千万不要为了做演示而搞花哨的 Demo（Don't innovate for the sake of delivering cool fancy demos）”，也不要陷入沉迷于成百上千个孤立用例的“用例陷阱”。AI 必须以端到端体验为核心，与垂直领域的业务专家共同创造（Co-create），直接对最终业务指标和底层商业价值负责。与此同时，完善的治理架构（Governance）、审计追踪机制和响应能力必须从第一天起就深度嵌入架构的每一层，唯有将合规与内控部门全程带入变革旅程，企业才能从容应对日益严格的金融监管要求。 --- ### [17:03](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1023s) 复合效应与三大行动指引：迈向 AI 原生运营模式在总结演讲时，Lily Li 指出，企业规模化应用 AI 的终极思维在于构建具有“复合效应（Compound）”的系统能力——这才是生成式 AI 与前沿智能带来的真正价值所在。零散的单点工具只会带来一次性的线性回报，而平台化的能力沉淀则会随时间呈指数级释放红利。为了帮助企业真正跨越从实验到规模化落地的鸿沟，她向在场的技术与业务领袖提出了三项关键行动倡议： 1. **重构端到端流程与运营模式**：颠覆原有的组织架构、人才策略与业务流程，坚决放弃修修补补的工具级微调与单纯追求短期单点成效的 quick wins。 2. **将业务价值置于一切的核心**：促使所有高管团队坐到一起，展开围绕真正商业价值的严肃对话，并建立明确的量化考核机制——精准计算“每一美元投入所产生的价值”以及“每个业务产出的单位推理/Token 成本（Cost per outcome）”。 3. **消除恐惧，带领员工共同进化**：企业级 AI 面临的最大阻力往往来自员工内心的不安全感与错失恐惧（FOMO）。管理层不仅要提供前沿工具和技能培训，更要主动为员工重塑未来的岗位职责说明（Job Description），为他们设计从当前岗位平滑过渡到未来人机协同新角色的职业成长路径。最后，Lily Li 呼吁所有与会者：“停止无休止的试点，全面启动实质化运营（Start operating and stop piloting）。”资产管理及各行各业的领跑者不应再将精力耗费在那些明年就可能被淘汰、仅能带来微弱线性 ROI 的临时性方案上，唯有秉持规模化与复合增长思维，才能在这场由 AI 驱动的运营模式重塑中赢得长久的主动权。 ## Notable quotes > [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) 去年你们机构上线过试点项目的请举手。 — Lily Li > [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) 如果你们的试点项目已经在生产环境中运行并实际创造了价值，请继续保持举手。 — Lily Li > [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) 从第一轮举手到第二轮举手之间的巨大鸿沟，正是本场演讲的核心议题。 — Lily Li > [00:08](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=8s) 多步推理成为行业标配（Table Stakes）：各大顶尖前沿模型人人可用，仅凭模型本身已无法构建竞争壁垒。 — Lily Li > [02:50](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=170s) 将价值置于核心（Value Front and Center）：关注长期可复合增长的影响力，而非单个用例的短期收益。 — Lily Li > [15:19](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=919s) 千万不要为了做演示而搞花哨的 Demo（Don't innovate for the sake of delivering cool fancy demos） — Lily Li > [17:03](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1023s) 企业规模化应用 AI 的终极思维在于构建具有“复合效应（Compound）”的系统能力——这才是生成式 AI 与前沿智能带来的真正价值所在。 — Lily Li > [17:03](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1023s) 精准计算“每一美元投入所产生的价值”以及“每个业务产出的单位推理/Token 成本（Cost per outcome）”。 — Lily Li > [17:03](https://www.youtube.com/watch?v=JO2WtYrsfbM&t=1023s) 停止无休止的试点，全面启动实质化运营（Start operating and stop piloting）。 — Lily Li ## Mentioned Lily Li, 富兰克林邓普顿 (Franklin Templeton), ChatGPT, AI Academy, Intelligence Hub, Salesforce, PM Copilot, Rep Response ## Video description Lily Li - Head of AI Adoption and Solutions, Franklin Templeton AI is delivering on its promise, and leading enterprises are already on the journey to scale its impact. As companies move beyond siloed pilots, the focus shifts to integrating AI into how the business actually runs. This session explores how a large asset manager positions AI as a driver of end-to-end transformation, catalyzing new ways of working that deliver real business value across research, portfolio management, distribution, operations, and client engagement. It highlights how agentic AI, multi-agent platforms, business co-creation, and trust-based adoption are embedded from day one, enabling the scaling of common, reusable AI capabilities across the organization. What You'll Take Away: Beyond the pilot — why the winners redesign how work happens, not just automate tasks. Reusable by design — the platform approach that compounds value instead of multiplying copilots. The real moat — why it isn't the model. Trust from day one — scaling AI where the stakes, and the regulators, are real. What's next — the shift to AI-native operating models, and what it means for your business. Who Should Attend: Investment, distribution, technology and operations leaders ready to move past experimentation and drive enterprise-wide AI transformation. [Source](https://www.youtube.com/watch?v=JO2WtYrsfbM)

Source quotes

[0:08] Lily Li: 去年你们机构上线过试点项目的请举手。 [0:08] Lily Li: 如果你们的试点项目已经在生产环境中运行并实际创造了价值，请继续保持举手。 [0:08] Lily Li: 从第一轮举手到第二轮举手之间的巨大鸿沟，正是本场演讲的核心议题。 [0:08] Lily Li: 多步推理成为行业标配（Table Stakes）：各大顶尖前沿模型人人可用，仅凭模型本身已无法构建竞争壁垒。 [2:50] Lily Li: 将价值置于核心（Value Front and Center）：关注长期可复合增长的影响力，而非单个用例的短期收益。 [15:19] Lily Li: 千万不要为了做演示而搞花哨的 Demo（Don't innovate for the sake of delivering cool fancy demos） [17:03] Lily Li: 企业规模化应用 AI 的终极思维在于构建具有“复合效应（Compound）”的系统能力——这才是生成式 AI 与前沿智能带来的真正价值所在。 [17:03] Lily Li: 精准计算“每一美元投入所产生的价值”以及“每个业务产出的单位推理/Token 成本（Cost per outcome）”。 [17:03] Lily Li: 停止无休止的试点，全面启动实质化运营（Start operating and stop piloting）。

Source quotes

[0:08] Lily Li: Very, very minimum [5:12] Lily Li: 10 times and 100 times [5:12] Lily Li: Decision AI [5:12] Lily Li: agentic investment teams [15:19] Lily Li: build and partner [15:19] Lily Li: cost per outcome [15:19] Lily Li: Transform operating models, not just tools [15:19] Lily Li: Focus relentlessly on compounding value [15:19] Lily Li: Bring your people along [18:59] Lily Li: start operating and stop piloting

Original transcript

[0:08] Good morning, everyone, and thank you for joining the room. I'm very [0:11] honored to spend the time with you guys for all the industry leaders. [0:15] Now, engage with me. Raise your hands if your organization have shipped at pilots last year. [0:23] Okay, and keep your hand up if you already have your pilots in production and delivering value. [0:31] Very, very minimum. The gap between the first and the second hand is what this talk is about. [0:37] My name's Lilly Li. I'm head of AI adoption and solutions for Franklin Templeton, [0:41] where we manage about 1.8 trillion AUM. My focus is actually trying to bring [0:47] AI to meet where people work and also make it into operating core. But before I get into the [0:52] details, let's look into some of the industry reports of where we stand in the AI journey. [0:59] As you can see, um I think some of the rooms already amplified that as well is [1:04] the industry is already projecting that most of the organization has passed the [1:08] stage of start small or piloting space, and actually what comes next is scaling. [1:15] How do you scale across enterprise? Now, the other stats shows you the challenge of the scaling is [1:22] where you see the value that actually AI can bring to real world instead of just getting AI to work. [1:28] What secretly happened behind the scenes is the four little things that show on the screen. [1:34] Number one, the multi-reasoning is actually becoming a table stakes. So, everybody have [1:40] access to the big models, the frontiers. So, it's not going to be your edge going forward. [1:45] Everyone is actually experimenting into agenda AI. However, everyone is still also figuring out what [1:51] is the usability of those agents to scale and go beyond just one functions. The control framework [1:58] is something that instead of uh gatekeeper or approver at the end, it becoming the design of [2:04] our product processes and more importantly economics. How many of you been invited to [2:10] networking event and conference this year with a label saying ROI? And here we're talking about [2:16] beyond just the linear ROI concept but more of a what do you deliver a measurable workflow value? [2:23] For leading companies at this stage when you see the inflection of the industry core is you [2:29] want to be in a scaling stage, you want to think about organizational throughput and for enterprise [2:34] AI except for just consumer AI that means you have to consider redesigning your value chains. [2:41] So how? As a leader in the room you're thinking, great, we want to get scaling. [2:46] How do we do that? What is the mandate behind it? How do you framework it? [2:50] My personal advice is the three key components for you to think about what set the direction for you. [2:56] First, it's value. Front and center. Here what I meant by value again is beyond a use [3:02] case ROI. It's beyond quick wins. It's actually talking about things can compound over time. [3:08] You have to think about all the long term impact that might have some initial investment [3:13] and jointly committed and agree with any functional or domain leaders that you work with. [3:19] With that in mind, naturally the prioritization will lead to the initiatives that needs to [3:24] come to end-to-end transformation. And the transformation I meant not surrounding around AI [3:29] but using AI as a catalyst or the enablement for the things that you traditionally cannot automate [3:35] nowadays we can actually speed up that process for you. And that's also in the process for you while [3:41] you're doing those end-to-end transformation, you will surface up what is actually key important [3:46] for your human workers. What is the decision points? What is the accountability that your [3:50] human team needs to really be a style at while let the machine take the rest of the work. [3:57] And that frames the next round of capabilities of skill set of the organization talents you [4:03] want either hire or develop. And then that sets the stage for you to get to the next future of [4:09] genetic enterprise. Doing that multiple times it would give you the foundation of production [4:15] at scale not only considering the technology but product also adoption, quality and also economics. [4:22] To make it happen obviously you have to set the foundation and I think the previous call also [4:26] talk about the foundations from Siemens. And needless to say you need to really [4:31] create your knowledge and harness in memory space supported by a standardized platform to give you [4:37] integrations and access to data and orchestration across the firm as well as having your governance [4:43] always embedded in the process instead of being an overarching or at the end of process. [4:50] So as a leader think about that. Now you want to scale. Now you have a framework how to guide your [4:54] scaling and you have to do it in the ongoing loops where your vision sets your direction, [5:00] your direction shapes your work and your people take accountability. Data platform enable that. [5:05] Eventually you will have the measurable outcome with a responsible governance. [5:12] Now how do we do it? How does it as a manager or Franklin Templeton asset global firm with [5:18] about 10,000 employees various of asset classes investment strategy deliver We always approach [5:24] it with a long-term vision. And as a firm we actually started AI organization pretty early [5:29] before introduction of ChatGPT in 2022. More importantly the last two years we have built [5:35] a very solid runway where we've proven the case of operating model that we run on some strategic [5:41] bets that are delivering production valuable incomes for everyone to see. Now it sets off [5:48] the foundation for us to kick off in 2026 being our scaling year. How we replicate those operating [5:54] model across the flagship platform we already built in distribution, investments, operations, [6:00] and all the core governance across the enterprise and making sure that we can actually 10 times and [6:06] 100 times the impact. And having the foundation of those meanings you have a prepared to data, public [6:12] data, and all the data harnessed in the same way that can be accessible by various different use [6:17] cases, as well as having the governance framework with policies, guidelines, do's and don'ts, [6:23] and the education of the people leveling up all the employees at Academy, adoption, enablements, [6:29] and also advisory board from top down advising from our executive committee, operating committee, [6:35] pretty much bringing along the entire enterprise. And that is something you have to do if not [6:39] already. Bottom up, top down. In 2027, when everything is really scaling the model that [6:46] we have proven case this year, we're going to see the tangible number. Hopefully, we can come back [6:52] and report to the audience with actual delivery numbers in the bottom line of the companies. [6:57] At that point, multi-agent or orchestration agent system will become a norm. [7:03] And with all the knowledge that we've harnessed and all the internal insights that we've built [7:07] up will become surface at the top of the house intelligence that we can leverage and [7:12] unlock additional opportunities that was never explored before. We also have seen [7:17] our footprint of the sales front line can actually get to the next level with AI-like platforms and [7:23] create more additional market shares and growth that is leading to us into the competitions. [7:29] We have also a very exciting disruptive um engagement currently going on in the sandbox [7:34] on agent investment teams and like Derek colleague in the audience um at Goldman's, tomorrow is going [7:40] to dive further into those details around what we are delivering. However, that kind of shapes the [7:46] future of 2027 to you. We're going to have bottom line impact, which can be reportable. We're going [7:51] to have top of the house intelligence. We're going to have disruptions in the industry to lead house. [7:56] Obviously, all the front line investment core is only part of it. For you to servicing your [8:02] clients during institutions, we need to have a consistency, accuracy, all the servicing behind [8:06] the scenes. That's what will make sure that the end-to-end operations behind the scenes [8:10] will be working well as well. Going into 2028, personally, I believe that's when we [8:16] going to a new era of decision AI. And having the foundation that we built in the past few years, [8:22] we will set to the stage where winning or leading organizations will be able to actually realize [8:28] the mode that we built from intelligence, from real-time actions, and actually coming up with [8:33] durable kind of outcome and new products and revenue stream hopefully can set us apart. [8:41] What does it mean by real things? Real examples in production. Everything that I show here [8:46] are just a portion of things that we do, but those are already live in production with [8:51] user and measurable outcome for the past 9 or 10 months. Um ranging from the [8:58] middle and back office, you see operations. We have built platforms that can handle very tedious [9:03] manual work end-to-end for reconciliation. In other industry, that means probably data matching, [9:08] etc. And some of the components is built really modulars on the platform so that we can take them [9:14] that out and actually reuse them and train with different set of model and proven the case that [9:19] can be actually utilized in trade operations. And the AI communication module that we built is so [9:25] powerful that can actually triage beyond just middle and back office and getting into other [9:30] corporate functions. Same thing with the review. And here review can be a financial report review, [9:35] it can be any type of contract review or data review, but the guideline behind it [9:40] or the engine that we build is you have a set of data, a set of guidelines rules, [9:45] and they can automatically review them and point out the anomaly as detailed as in the footnotes. [9:52] Coming into investment core as the asset management, this is really our edge. [9:56] We have built a very robust platform that can capture all the public, internal, preparatory [10:02] research and meetings from our investment communities into a mode of research intelligence. [10:09] And we also build on a proven case for certain investment teams utilizing a something called [10:14] PM Copilot, which is a multi-agent orchestration proven case. We also can introduce agents that [10:21] as a member of the investment teams, and that means it could be an investment analyst or or [10:25] could be an investment coach. The goal behind it is really digest the same set of information that [10:31] your human team can have and come up with more new points between augment the investment process [10:36] that you can actually deliver in the future. We also kind of distill all the information across [10:43] asset classes, meaning equities, fixed incomes, alternatives, digital assets, [10:48] and understanding what matters to them across the board and put them into a common agent store where [10:54] everybody can access. Give you example such as macroeconomics. Everybody needs to understand [10:58] that. Um sensitivity analysis, news, catalysts. And those agents are available for people to [11:05] go towards it without building their own individual productivity tools. Again, further [11:11] into the agentic investment teams, we're very exciting to see how that progress we can make. [11:16] However, if that's something we probably won't distill to you for another session or future. [11:21] Coming into our sales front, our distribution team and sales team, they normally spend a lot of their [11:27] times doing something that's not focusing on their client engagements. So we built Intelligence Hub [11:33] as their to-go intelligence AI workflow platform, where they in the past have to navigate about [11:39] 15 systems in a day, now they just go to one to manage their territory, to manage their salesforce [11:45] advisory engagements, focus non-focus list, actionable actionable reports, and also their [11:52] pre-reports. You name it, like pre-post engagements with any analyzers, all in one place. [11:59] And we also built an RFP and DDQ platform that is smart enough that can do your subject matter [12:05] expert engagement all in one platform without back and forth email chain, and also smart enough [12:09] to project AI responses based on historical records with the same company or without. [12:16] We also built something that will lead the investment viewpoint into the hands of our clients [12:21] or sales people's hands in a very fast manner, so that's what we call Rep Response. Our marketing [12:26] funds also have been using AI for the past few years going through things like translations, [12:33] proof readings, tone of voice, dialing, compliance check, everything that you name off in content and [12:38] campaign areas has been utilizing AI for that. Now it's lifting up them to think about the future, [12:45] how to do componentized content management in the future. And last but not least, [12:50] to level up the entire enterprise, we provide the resources and we bring frontier tools in-house and [12:57] let them to be more AI-enabled workforce in the future. Obviously, everything has to build on a [13:02] shared foundation and build on all the controls and leadership that we all share the same vision. [13:09] I'm going to skim through some of the slides or examples that I already talked about. For [13:13] research, besides people can interact with it, we also have our agents able to deliver earning [13:19] reports and insights into the hands of PM or analysts in their hands a minute after [13:24] earning call happen. Intelligence Hub, as I not only it's a workflow's transformation, [13:31] as well as AI workflow AI insights driven intelligence for them to sell further, [13:36] but also it already shows since launch the increase contributed because of intelligence [13:42] hub of account coverage, sales lift, and well adding engagements. And those are very tangible [13:49] and net measurable against the benchmark of the things before they had the intelligence hub. We're [13:55] looking for massively expanding that beyond the already adopted team in US, well into all other [13:59] segments, and deepening the engagements within the platform. Operations AI, as I mentioned already, [14:06] it can actually deal with imagine from the matching to exception casing to reasoning [14:13] to actual internal external communicating to finally adjusting to the system, all without human [14:18] intervention. Human just on the on the loop trying to approve and make some exceptions um reviews. [14:25] And that already delivered about 95% accuracy that would be monitoring the past [14:30] 10 months or 11 months. Also, that will represent about 80 or 90% of manual work capacity savings. [14:39] PM Kokali, as I mentioned, behind the scene is not a single agent, it's not a bot. It has [14:45] a orchestrating of all the functional team that would compose a portfolio management team look [14:50] like, your analysis agent, research agent, your trading agent, your risk management agent, and [14:56] the agent who understand all the market data and different internal sources and sell-side research [15:00] to compose. And this agent also fully understand the investment thesis this portfolio manager [15:05] deliver. So, the human PM can only focusing on how to augment their existing adjustment [15:11] or allocations based on recommendations. How do we do that? How do we make everything happen? [15:19] The executable strategy that we go about are the key things that we call build and partner. And [15:26] utilizing the strategic partnership in hyperscaler and also frontier AI startups to make sure we [15:32] can have a co-development process, but while keeping the memory, the knowledge, the you know, [15:38] proprietary workflows and all the orchestrations within our fence so that we keep our edge, [15:44] but mean meanwhile keeping very flexible. Being flexible we can plug and play different models [15:49] and different vendors in a very stay on the top manner and also making sure that we remain our [15:55] edge in our competition. The other thing that I already mentioned earlier in the talk is [16:00] end-to-end experience, always front and center, always think about what is mattering for the [16:05] end business to deliver the bottom line measurable values. So, not point solutions, no pilots, [16:12] not hundreds of use cases or thousands pages. Fit for purpose delivery means you always co-create [16:19] with your domain experts. Don't build for the sake of building, don't innovate for the sake [16:23] of delivering cool fancy demos. Obviously having governance always in the picture, [16:29] so not only you have effective usage, you have transformation, but you also have [16:34] responsiveness in the picture so that you can deal with regulations, you can deal with audit, [16:38] you can deal with all the control functions that be them bring them along the journey with you. [16:45] I will flash this something for people are more early [16:48] early adopter of AI. This is something very standard enterprise AI stack that [16:52] for you to consider. I'm not going to spend much time on it, but key is [16:56] every layer needs to be considered with the right commissioning, right audibility and governance. [17:03] So, hearing up to now that you already realize scaling is some key mindset, [17:10] having scaling framework in place that you are already transforming your operating model. And [17:15] what you'll be used to for the past many years is about to change. And there are all key components [17:20] that you need to consider behind the scenes. But the really key message I want to deliver is [17:26] build something can compound. And that is the power of generative AI, generative AI, whatever [17:32] AI you've been hearing nowadays. The compounding effect is where the true value will come. [17:39] And to wrap up this talk, I really like you guys to walk away with three key actions. One, [17:46] transform your processes. Transform your operating model, your organization, your talent strategy, [17:52] not the tools, not single use cases, not quick wins. Think about value. Put value for for funds [18:00] and center of everything. And invite conversations for all your leaders just coming into the value [18:05] conversations. Get them commit along with you. And measure the value per dollar spent. [18:10] Measure your cost per outcome. So that kind of resonating with a lot of conversations. [18:15] I'm pretty sure you guys getting to about the recent months token management, etc. [18:19] All of those are all comprehensively working together. And last but not least, enterprise AI, [18:25] one of the key components from consumer AI, is some of the frictions, is some of the fear, [18:31] is some of the things that people are FOMO. Bring your people along with you. And that means [18:36] beyond just resources and education, but also try to design the future job description for [18:41] them. Try to design the future path for them to lead from their current role to the future. And [18:46] that will give you a guarantee of a new operating model and also a success of realizing the value. [18:55] At the end of the day, I would say, to recap the entire talk, [18:59] the scaling is where we are as a leading company at the stage. Having scaling in your visions [19:05] means that you have to consider key components of framework to do your scaling planning. [19:10] And do not forget about anything that you need to worry about the values [19:15] and the operating model, your people, and do not try to focus on any point solutions. [19:20] Start operating and stop piloting. Stop chasing anything that is giving you some sort of ROI, [19:27] which is linear, and you'll probably throw them away next year. Thank you very much and [19:32] I'm Lily Li. I'm driving scaling enterprise AI at Franklin Templeton. Hope that's going to be [19:38] inspiring for you for not only asset management but also industries. Thank you very much.