Prudential Cuts AI Risk Reviews to One Day Using ModelOp
Ai4
Prudential Financial accelerated its AI risk-assessment process from two weeks to less than one day by codifying quantitative risk scoring into ModelOp's automated governance platform.
## Background
This content was published on YouTube by Ai4 on August 27, 2025. It features a fireside chat between Omodunni Ayodele, Vice President and Product Owner of Prudential Financial's Global Responsible AI Program, and Jennifer Ward, VP of Customer Success at ModelOp. The video details how Prudential built its Responsible AI framework on ModelOp's platform to scale innovation, drastically reduce risk-rating times, and streamline governance in a regulated enterprise.
## TL;DR
In this discussion, Prudential Financial demonstrates how operationalizing Responsible AI (RAI) transforms governance from a friction point into a competitive advantage. By establishing a multi-disciplinary coalition and utilizing ModelOp's platform, Prudential automated its risk assessment process, cutting evaluation cycles from two weeks down to less than one day. The enterprise manages an inventory of over 200 active models using Gartner's framework to categorize ROI across internal productivity, business integration, and future innovation. Additionally, the conversation addresses managing third-party vendor AI risks and traces the evolution of technology risk management into dedicated AI governance roles.
## Atmosphere & style
The conversation carries a highly professional, collaborative, and pragmatic tone, reflecting a mature corporate environment navigating complex regulatory landscapes. Jennifer Ward and Omodunni Ayodele share an engaging, low-ego rapport that makes intricate governance topics feel accessible and strategic rather than overly bureaucratic. Listeners experience an insightful narrative progression, moving from high-level principle definitions to practical operational victories and forward-looking discussions on vendor risk and career evolution.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) Responsible AI (RAI) in a regulated enterprise means enforcing a risk-aware framework across the full lifecycle—from design and procurement to deployment and monitoring.
- [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) High-level RAI principles like transparency, security, and privacy only become actionable when translated into everyday operational workflows.
- [01:12](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=72s) Operationalizing RAI requires a multi-disciplinary coalition across legal, compliance, risk, technology, and business units to embed governance from the ground up.
- [01:12](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=72s) For trust-based financial institutions, safety and data privacy must be built in 'by design' to serve as a competitive asset rather than a point of friction.
- [03:10](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=190s) Prudential intentional chose the term 'Responsible AI' over 'AI Governance' to move away from negative perceptions of restrictive authority toward shared enterprise commitment.
- [03:10](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=190s) Prudential uses Gartner's ROI framework to categorize AI use cases into Defend (employee efficiency), Extend (workflow integration), and Upend (new revenue models).
- [06:38](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=398s) Prudential manages a centralized AI 'assembly line' inventory of over 200 active use cases spanning in-house, vendor, classical AI, and Generative AI systems.
- [06:38](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=398s) Codifying a deterministic, quantitative risk scoring methodology into ModelOp's platform slashed Prudential's inherent risk rating cycle from two weeks to under one day.
- [10:47](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=647s) Prudential anchored its Responsible AI Program in Global Technology and Operations (GTO) because core AI risks originate within the underlying technology stack.
- [10:47](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=647s) The boom in Generative AI catalyzed executive leadership to fund centralized governance infrastructure, enabling control partners to approve high-value use cases faster.
- [13:54](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=834s) Managing third-party vendor AI risk requires rigorous evaluation of vendor data usage rights, inferencing practices, and preventing enterprise data leakage.
- [16:02](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=962s) The rapid advancement of AI technology is actively creating new strategic governance roles and career pathways even as it transforms traditional work.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) Defining Operationalized Responsible AI
The fireside chat opens with Jennifer Ward welcoming Omodunni "Omi" Ayodele, Vice President and Product Owner of the Global Responsible AI Program at Prudential Financial. Ayodele sets the stage by clarifying what Responsible AI (RAI) actually means in practice. In a regulated enterprise, RAI means ensuring that during the entire lifecycle—from design, development, and procurement to deployment and ongoing usage—AI operates within a risk-aware framework. This framework is anchored in key principles such as transparency, explainability, security, privacy, and trust.
However, Ayodele emphasizes that high-level principles can easily sound like vague "motherhood and apple pie" concepts to business leaders. Principles only become real and actionable when they are translated into everyday operational processes that align with the goals of the business.
- **Core Principles:** Transparency, explainability, security, privacy, and organizational trust.
- **Key Takeaway:** Governance must be embedded into real operational workflows to deliver business value rather than remaining theoretical.
---
[01:12](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=72s) Balancing Risk and Innovation Across a Multi-Disciplinary Team
Ayodele explains that operationalizing RAI requires building a broad, multi-disciplinary coalition because "it takes a village." Governance cannot succeed if it is isolated inside Governance, Risk, and Compliance (GRC), legal, compliance, or technology teams acting in a vacuum, nor can it succeed if the business side does not understand how to integrate use cases into their value chain. Every stakeholder must have a seat at the table to contribute their specialized domain expertise from the ground up.
```
+-----------------------------------------------+
| Responsible AI (RAI) Coalition |
+-----------------------------------------------+
| | | |
v v v v
+------------+ +------------+ +------------+ +------------+
| Law | | Enterprise | | Global Tech| | Business |
| & Compliance| | Risk | | & Ops | | Units |
+------------+ +------------+ +------------+ +------------+
```
Because Prudential operates in the financial services sector—a fundamental trust-based industry—the program must balance risk management with rapid innovation. Protecting customer and institutional client data cannot feel like a late-stage "bolt-on" or a temporary band-aid; it must be an integrated, "by design" expectation. When designed properly, this trust-based risk framework becomes a true competitive asset and advantage.
- **Cross-Functional Partners:** Legal, compliance, risk, technology/architecture, data science, and business units.
- **Design Philosophy:** Safety and privacy must be built in "by design," positioning responsible governance as a competitive advantage rather than a friction point.
---
[03:10](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=190s) Categorizing ROI with the Gartner Framework
Addressing the challenge of measuring Return on Investment (ROI), Ayodele points out that fast-moving AI cannot be governed with slow, manual methods. Prudential intentionally chose the term "Responsible AI" over "AI Governance" because "governance" often carried a negative connotation of an authority imposing restrictions on the business. RAI signals a shared, enterprise-wide commitment spanning corporate functions and line-of-business value chains.
To evaluate ROI across different types of initiatives, Prudential adopted Gartner's framework to categorize use cases into three tiers:
* **Defend ("Return on Employee"):** Focused on internal workforce efficiency and productivity. While hard to measure strictly in financial terms, these low-risk, entry-level use cases allow teams to quickly test, learn, and build organizational knowledge.
* **Extend:** Direct integrations into core business value chains that streamline customer-facing or operational workflows. These offer clear efficiency metrics by measurably reducing process execution times.
* **Upend ("Return on the Future"):** Cutting-edge initiatives that create entirely new product streams or revenue models by extracting novel insights from massive datasets (e.g., leveraging insurance and claims data).
---
[06:38](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=398s) Building the AI Assembly Line and Automating Risk Rating
To execute governance at scale, Prudential built a centralized AI "assembly line" to inventory and process every use case. The enterprise currently oversees an inventory of **200+ active use cases and models**, encompassing both in-house developments and vendor solutions, as well as both traditional/classical AI and Generative AI. Ayodele stresses that traditional AI models continue to deliver immense business value and that traditional and generative models frequently work in tandem to improve product roadmaps.
A major operational milestone was revamping their inherent risk rating process. Originally, risk assessment was a qualitative and manual process that utilized Microsoft Forms and complex macro spreadsheets. Evaluating a use case took roughly **two weeks**.
By establishing a quantitative, deterministic scoring methodology based on variables like data sensitivity, privacy, security, and algorithmic complexity, Prudential codified and deployed the logic directly into ModelOp's governance platform. This automation dramatically slashed the risk assessment cycle time from **two weeks down to less than one day**.
- **Model Inventory:** 200+ models across vendor, in-house, traditional AI, and Generative AI systems.
- **Previous Bottleneck:** 2-week manual risk rating process managed via MS Forms and macro spreadsheets.
- **Automated Benchmark:** Reduced evaluation time to less than 1 day using codified, quantitative scoring in ModelOp.
---
[10:47](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=647s) Cultural Drivers and Grounding Governance in Technology
When asked about key success factors and organizational challenges, Ayodele highlights Prudential's low-ego, highly collaborative culture. Rather than engaging in "hot potato" debates over which department should own the program, stakeholders agreed to anchor the Responsible AI Program within the Global Technology and Operations (GTO) organization.
This structural decision reflects the reality that AI risks—such as operational risk, information security, data privacy, and data management—are inherently generated within the technology stack. At the same time, the program maintains deep, formal partnerships with Enterprise Risk (to ensure executive-level visibility), Legal, and Compliance.
Rather than slowing teams down, the program was catalyzed by the rise of ChatGPT, which prompted top executive leadership to invest heavily in a centralized governance infrastructure. The goal is to provide a clear, supportive framework so control partners don't have to say "no" when high-value use cases are ready for production.
> "AI is a technology... The implications of AI is operational risk. That's fraught in technology: information security, data privacy, data management... It sits in our Global Technology and Operations organization."
---
[13:54](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=834s) Navigating Third-Party AI Risks and Vendor Data Rights
Looking toward future challenges on the horizon, Ayodele highlights the rapid expansion of third-party vendor AI integrations as a critical focus area. As third-party software vendors incorporate generative and automated features into their tools, enterprise architecture and governance teams must carefully evaluate the underlying consumption patterns and data flows.
Because model inferencing typically takes place at the source—often hosted on major hyperscalers or foundation model providers—Prudential proactively questions vendor data practices before onboarding new solutions:
- **Data Rights & Usage:** Are third-party vendors or hyperscalers processing enterprise data at the source, and do they retain any rights to re-use or train on that data?
- **Data Protection:** How can institutional and individual customer data be prevented from leaking across multiple external vendor environments into shared cloud models?
- **Architectural Control:** Ensuring that the organization's core "gold fold" data assets remain fully controlled and protected under strict corporate governance standards.
[16:02](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=962s) The Evolution of Risk Management: From Infrastructure to Responsible AI
Omodunni Ayodele reflects on the professional journey that led to her current leadership role in AI governance, tracing how risk management in large organizations has evolved over time. She describes her early work handling fundamental information security and infrastructure risks—managing systems like Unix, Winintel, distributed setups, and networking infrastructure. Over time, her scope shifted higher up the technology stack into business application risk management. A major turning point occurred when Prudential decided to establish a dedicated data office to accelerate analytics and data management. At that moment, it became clear that traditional technology risk frameworks were ill-equipped to address emerging, novel data risks or the business risk of failing to leverage data effectively to deliver value.
As classical and traditional machine learning technologies began to emerge nearly a decade ago—well before the rise of generative AI—her team started actively engaging with these technologies to manage their unique operational risks. Because her career had always centered on risk and technology governance, taking ownership of the Responsible AI ecosystem was a natural progression to support Prudential's central AI initiatives. Jennifer Ward wraps up the discussion by observing how fascinating this evolution is, pointing out that while many worry about AI displacing jobs, it is simultaneously creating entirely new domains of specialization and career paths.
- Ayodele's career progression moved from foundational infrastructure (Unix, Winintel, networks) up to application risk and enterprise data governance.
- The establishment of Prudential's data office highlighted a critical gap in traditional risk frameworks regarding emerging data management and AI/ML risks.
- Prudential began tackling classical AI and machine learning risk management nearly a decade ago, long before generative AI gained mainstream prominence.
- Ward highlights that despite widespread anxiety about AI eliminating roles, it is actively creating new strategic governance jobs and opportunities across industries.
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) motherhood and apple pie — Omodunni Ayodele
> [01:12](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=72s) it takes a village — Omodunni Ayodele
> [01:12](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=72s) by design — Omodunni Ayodele
> [03:10](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=190s) Return on Employee — Omodunni Ayodele
> [03:10](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=190s) Return on the Future — Omodunni Ayodele
> [06:38](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=398s) two weeks down to less than one day — Omodunni Ayodele
> [10:47](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=647s) AI is a technology... The implications of AI is operational risk. That's fraught in technology: information security, data privacy, data management... It sits in our Global Technology and Operations organization. — Omodunni Ayodele
## Mentioned
Omodunni Ayodele, Jennifer Ward, Prudential Financial, ModelOp, Gartner, Microsoft Forms, ChatGPT, Global Technology and Operations, Enterprise Risk, Legal and Compliance, Unix, Winintel
## Video description
Omodunni Ayodele, Vice President and Product Owner of the Global Responsible AI Program, Prudential Financial
Jennifer Ward, VP of Customer Success, ModelOp
Abstracts
What does AI ROI look like in a global, regulated enterprise like Prudential Financial? In this fireside chat, Omodunni Ayodele, VP and Product Owner of Prudential’s Global Responsible AI Program, joins ModelOp's VP of Customer Success, Jen Ward, to unpack how they built a Responsible AI program on top of ModelOp's AI lifecycle automation and governance platform to drive AI innovation and better business outcomes. Learn how Prudential applies its Responsible AI framework to scale innovation, cut a risk-rating process from two weeks to one day, and streamlined governance processes to keep pace with the rapid evolution of AI. This session offers a real-world look at how Responsible AI, when operationalized through ModelOp’s AI control tower, can deliver both speed and scale without sacrificing trust. If you’re innovating with AI and trying to measure ROI , this conversation is for you.
保德信利用ModelOp将AI风险审查缩短至一天
保德信金融（Prudential Financial）通过将量化风险评分规则化并纳入ModelOp自动化治理平台，将AI风险评估流程从两周缩短至不到一天。
在本次讨论中，保德信金融（Prudential Financial）展示了如何通过推进负责任AI（RAI）的落地运营，将治理从阻力点转变为竞争优势。通过建立跨领域联盟并运用ModelOp平台，保德信实现了风险评估流程的自动化，将评估周期从两周压缩至不到一天。该企业管理着包含200多个活跃模型的资产库，并采用Gartner框架从内部生产力、业务整合和未来创新三个维度划分ROI。此外，对话还探讨了第三方供应商AI风险的管控，并梳理了技术风险管理向专职AI治理角色的演变历程。
保德信金融（Prudential Financial）建立跨法务、合规、风险及技术的跨职能协作机制，将责任 AI（Responsible AI）原则内嵌于全生命周期运营流程中，把安全与数据隐私作为核心竞争优势。 (1:12)
公司采用 Gartner 的 ROI 评估框架，将 AI 用例划分为防御（Defend，员工效率）、扩展（Extend，流程整合）与颠覆（Upend，新收入模型）三层，并以“责任 AI”替代具有限制色彩的“AI 治理”表述。 (3:10)
保德信集中管理超过 200 个涵盖自研、外部供应商、传统及生成式 AI 的模型用例，借助 ModelOp 平台将固有风险量化评估周期从原本的两周大幅压缩至不足一天。 (6:38)
针对加速集成的第三方 AI 软件，团队重点审查外部模型与超大规模云服务商的数据使用权及推理路径，严格防范企业核心数据泄露或被用于未经授权的二次训练。 (13:54)
Reading summary
Prudential Financial operationalizes Responsible AI by integrating automated risk evaluations into its technology infrastructure and aligning use cases with a tiered business-value framework, according to Global Responsible AI Program head Omodunni Ayodele.
Operationalizing Cross-Functional Governance by Design (0:04)
Ayodele notes that high-level ethical principles such as transparency, privacy, and explainability risk becoming abstract unless embedded into daily business workflows. Operating in a regulated, trust-based financial industry, Prudential aims to build safety and privacy into systems by design rather than as an afterthought. Ayodele argues that when risk controls are properly integrated, they function as a competitive asset rather than an operational bottleneck.
To prevent governance from operating in an organizational vacuum, Prudential established a cross-disciplinary coalition encompassing legal, compliance, enterprise risk, technology, data science, and business units. Rather than using the term 'AI Governance'—which Ayodele states carries negative, restrictive connotations—the company adopted 'Responsible AI' to reflect shared enterprise ownership across business value chains.
Categorizing AI Return on Investment (3:10)
Prudential adopted Gartner’s three-tiered framework to evaluate return on investment across different AI initiatives. 'Defend' initiatives focus on workforce productivity; although Ayodele acknowledges these are harder to measure purely in financial terms, they serve as low-risk entry points to build institutional knowledge. 'Extend' initiatives integrate into core operational and customer workflows, generating measurable ROI by decreasing process runtimes. 'Upend' initiatives target transformative models and new revenue streams by extracting insights from extensive datasets, such as claims and insurance records.
Automating Risk Ratings Across Model Inventories (6:38)
To manage its footprint of more than 200 active models—comprising both classical machine learning and generative AI developed internally or procured from vendors—Prudential developed a centralized governance intake process. Ayodele notes that traditional AI models continue to provide substantial business value alongside newer generative systems.
Prudential replaced a manual, qualitative risk rating workflow based on Microsoft Forms and macro spreadsheets, which previously took approximately two weeks to review. By codifying quantitative scoring criteria—including data sensitivity, privacy, security, and algorithmic complexity—into the ModelOp governance platform, the enterprise reduced risk assessment cycle times to less than one day.
Structural Placement and Third-Party Risk Oversight (10:47)
Prudential anchored its Responsible AI Program within the Global Technology and Operations organization while maintaining formal oversight partnerships with Enterprise Risk, Legal, and Compliance. Ayodele explains that this structure reflects the origin of AI operational risks, such as information security and data management, inside the technology stack itself.
Looking ahead, Ayodele identifies third-party software integrations and external vendor models as primary operational challenges. Governance teams must scrutinize vendor architectures and cloud hyperscalers to verify whether enterprise data is processed at the source, prevent proprietary or customer data leakage into shared models, and ensure external partners retain no training rights over corporate assets.
Reflecting on the evolution of technology risk from infrastructure to data management, Ayodele and host Jennifer Ward emphasize that managing machine learning risks began years before generative AI gained prominence. While automation raises concerns about workplace displacement, Ward notes that institutionalizing AI risk management simultaneously creates new specialized governance roles across regulated industries.
## Background
本内容由 Ai4 于 2025 年 8 月 27 日在 YouTube 平台发布。视频记录了保德信金融集团（Prudential Financial）全球负责任 AI 项目副总裁兼产品负责人 Omodunni Ayodele 与 ModelOp 客户成功副总裁 Jennifer Ward 的一场炉边对谈，重点探讨在强监管金融企业中如何借助 ModelOp 自动化平台构建负责任 AI 治理框架，以实现风险控制与商业 ROI 的双赢。
## TL;DR
保德信金融集团将“负责任 AI”确立为兼顾风控与创新的核心竞争优势，摒弃孤立被动的合规管控，推动跨职能协同共建。企业引入 Gartner 的三层价值框架（防御型、拓展型、颠覆型）来多维度量化评估 AI 的投资回报率。通过在 ModelOp 平台上实现治理流水线代码化与自动化，保德信成功将 200 多个模型的固有风险评级周期从两周大幅压缩至一天以内。未来，保德信将治理边界进一步延伸至第三方供应商 AI，从源头牢牢守护数据主权与合规底线。
## Atmosphere & style
整场对话洋溢着专业、务实且极富前瞻性的行业洞察氛围。两位讲者配合默契，对话节奏流畅明快，褪去了枯燥理论的生硬感，洋溢着基于实战经验的松弛与自信。听众能清晰感受到大型传统金融机构在拥抱前沿科技时那种既严谨审慎又雷厉风行的战略定力，整个交流既有底层架构演进的纵深感，又充满了对新型职业机遇的热忱与启发。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) 负责任 AI（RAI）的核心是在全生命周期中置于风险意识治理框架下，且必须将透明度、可解释性与隐私等原则转化为可落地的具体业务流程。
- [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) AI 治理绝非单一团队的职责，必须由法律、合规、风控、技术与前线业务部门从一开始就同桌共商，实现多元协同治理。
- [01:58](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=118s) 风险与创新是同一枚硬币的正反面，在金融机构中，内生设计的负责任 AI 治理是维系客户信任的核心竞争优势，而非事后补丁。
- [03:10](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=190s) 保德信借助 Gartner 框架将 AI 用例划分为防御型（Defend）、拓展型（Extend）与颠覆型（Upend），以差异化标准衡量各自的投资回报率（ROI）。
- [06:38](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=398s) 保德信全面纳管了 200 多个模型与用例，并主张将自研与外部供应商模型、生成式 AI 与传统经典 AI 有机结合以释放最大价值。
- [06:38](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=398s) 借助 ModelOp 自动化平台，保德信将固有的 AI 风险评级周期从过去耗时两周的人工宏表流程大幅缩短至 1 天以内，实现高低风险分流审查。
- [10:47](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=647s) 治理体系设立在全球技术与运营组织之下，依托“低内耗（low ego）”的企业文化以及高管层的支持，转型成为赋能业务加速上线的引擎。
- [13:54](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=834s) 第三方供应商 AI 的纳管成为当前治理重点，企业需严格审查模型推理源头的数据处理与权限，牢牢守住核心数据资产与数据主权。
- [16:02](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=962s) 风险管理重心经历了从底层基础设施、业务应用到数据管理与 AI 生态的演进，体现了科技风控与 AI 业务深化的必然趋势。
- [17:17](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=1037s) AI 的发展在带来挑战的同时也在重塑职业发展路径，催生出诸如负责任 AI 产品负责人等新兴的专业治理岗位。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) 负责任 AI 的定义与业务落地
对话伊始，ModelOp 客户成功副总裁 Jennifer Ward 与保德信金融集团（Prudential Financial）全球负责任 AI 项目副总裁兼产品负责人 Omodunni Ayodele（昵称 Omi）展开交流。在被问及保德信的“负责任 AI”（Responsible AI，简称 RAI）项目究竟意味着什么时，Omodunni 指出，其核心定义是在设计、开发、采购以及部署使用 AI 的全生命周期中，始终置于一个具备风险意识的治理框架之下。该框架以透明度（transparency）、可解释性（explainability）、安全性（security）、隐私保护（privacy）以及信任（trust）为基石。
Omodunni 坦言，如果这些原则仅仅停留在宏大叙事或“无可挑剔的空泛道理”（motherhood and apple pie）层面，对业务部门而言便毫无实际意义。只有当这些高层原则真正转化为具体、可落地的业务运营流程时，治理才会发挥真实的价值。在保德信，构建这套体系的关键认知在于“众人拾柴火焰高”（it takes a village）：AI 治理绝不是某一个孤立团队的工作，不能单纯交给风险合规（GRC）、法律、技术部门，也不能由不清楚如何将用例融入价值链的业务团队单打独斗，而是必须让所有相关利益方从一开始就同桌共商，将多元专业能力自底向上地融入整个机制中。
- **核心原则**：透明度、可解释性、安全性、数据隐私与建立信任。
- **协同治理主体**：法律、合规、企业风险管理、技术架构、数据科学以及前线业务团队。
---
### [01:58](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=118s) 风险与创新的平衡：将信任打造成核心竞争优势
讨论转向风险控制与业务创新之间的博弈。Omodunni 强调，在保德信看来，风险与创新是“同一枚硬币的正反面”，而非互斥的对立面。保德信作为一家老牌金融服务机构，其立足根本正是基于信任的行业属性。无论是使用机构客户还是个人客户的数据来提供金融服务，企业都必须在持续、高效交付的同时牢固维护这份信任。
因此，负责任 AI 框架的职责不是给创新设卡，而是提供一套标准机制以支持大规模快速扩张。当客户询问“你们如何保护我的数据”时，数据保护绝不能是一个事后打补丁的附加项（bolt-on），而必须是一开始就贯彻始终的“内生设计”（by design）。Jennifer 对此深表认同，并指出保德信团队从未把合规当成创可贴式的补救手段，而是切实将其视为企业在市场中的差异化竞争优势与核心资产。
---
### [03:10](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=190s) 衡量 AI ROI：摒弃老旧治理与 Gartner 分类模型
在探讨 AI 的投资回报率（ROI）与成功衡量标准时，Omodunni 指出一个根本矛盾：“你无法用陈旧、手动的治理手段去监管一项飞速发展的技术。”保德信在推行项目时特意选用“负责任 AI”而非“AI 治理”（AI Governance）这一术语，正是为了避免给内部留下“某一个职能部门在单向约束业务”的被动刻板印象，进而强化“全员共建、贯穿全价值链”的协同理念。
由于业务单元（创造价值）与支持性职能部门（赋能业务）对价值的诉求截然不同，保德信借助了 Gartner 提出的三层用例价值框架来定义并衡量不同场景下的 ROI：
```
┌──────────────────────────────────────────────────────────┐
│ 保德信 AI 用例价值分类框架 (Gartner) │
└──────────────────────────────────────────────────────────┘
│
┌─────────────────────────┼─────────────────────────┐
▼ ▼ ▼
【Defend 防御型】 【Extend 拓展型】 【Upend 颠覆型】
· 员工效率回报 (ROE) · 深入业务价值链 · 面向未来的回报
· 知识库检索、日常提效 · 缩短流程周期、客户触达 · 挖掘海量理赔/保险数据
· ROI 较难直接量化 · 从立项起跟踪效率指标 · 孵化全新产品与业务线
```
- **Defend（防御型用例）**：关注员工效率回报（Return on Employee），例如内部知识库查询和低门槛的探索性场景。此类用例风险较低、利于快速试错，但直接量化经济 ROI 的难度较高。
- **Extend（拓展型用例）**：直接切入业务价值链并可能触及终端客户，核心价值在于端到端流程的大幅缩短与效率跃升。此类项目必须从立项阶段就明确衡量指标与持续的投入产出对比。
- **Upend（颠覆型用例）**：代表“面向未来的回报”（Return on the Future）。保德信利用自身积累的海量保险与理赔数据资产，通过前沿 AI 技术提炼深度洞察，进而开辟出全新的数字化产品线与业务增长点。
---
### [06:38](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=398s) 200+ 模型盘点与自动化风控流水线
要实现精准治理与价值分类，首要基础是全面掌握企业内部的资产底账。保德信已经完成了涵盖 200 多个模型及用例的全局盘点，且在纳管范围上兼收并蓄：既包含自研模型，也涵盖外部供应商模型；既覆盖生成式 AI（Generative AI），也高度重视传统经典 AI（Classical AI）。Omodunni 强调，企业绝不是“喜新厌旧”，因为传统 AI 模型至今仍在业务中释放巨大价值，而传统 AI 与生成式 AI 的有机结合更能持续推动现有产品路线图的演进。
在流程提效方面，保德信借助 ModelOp 平台实现了治理流水线的彻底变革。此前，企业内部的固有风险评级（Inherent Risk Rating）采用人工主观评估与宏表（Macro Spreadsheet）流转，从 Microsoft Forms 登记、人工打分到沟通缓解措施，整套周期长达两周，效率低下且极易出错。
```
【传统人工流程】(耗时约 2 周)
MS Forms 提交 ──> 宏表格手动计算 ──> 人工风险定级 ──> 沟通缓解措施
【ModelOp 自动化流水线】(耗时缩短至 < 1 天)
自动接入注册 ──> 多维量化算法测算 ──> 自动化定级分流 ──> 高风险"深度拥抱" / 低风险极速放行
(隐私/安全/算法复杂度)
```
通过将数据敏感度、网络安全、底层数据质量、算法复杂度等参数提炼为确定性的量化算法，并将其代码化部署在 ModelOp 的 AI 治理流水线中，保德信将原本长达两周的风险评级周期直接压缩至 1 天以内。这不仅实现了“低风险用例极速放行、中高风险用例深度审查（bear hug）”，更在组织内部建立了高度透明、环环相扣的信任链条。
---
### [10:47](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=647s) 组织架构与低 Ego 的跨职能协同文化
谈及项目落地过程中的阻力与关键成功要素，Omodunni 认为保德信“高度协作、低内耗（low ego）”的企业文化起到了决定性作用。面对全新且复杂的 AI 治理挑战，团队成员普遍秉持“如果我不懂，就向懂的同事请教，共同攻克难题”的务实态度，有效避免了部门之间争夺主导权或推诿责任的“烫手山芋”现象。
在治理职能的最终归属上，保德信将其设在**全球技术与运营组织（Global Technology and Operations）**之下。其内在逻辑在于：
1. **风险源头在技术端**：AI 本质是技术创新，伴随的操作风险、信息安全、数据隐私与数据管理均扎根于技术底层；
2. **价值实现在业务端**：最终的业务用例由业务部门落地；
3. **顶层连接在风控与合规端**：必须与企业风险管理（Enterprise Risk）深度联动，向高管层直观呈现全局风险态势，同时法律与合规伙伴紧密驻场以满足严格的金融监管要求。
此外，高管层的强力支持也是关键催化剂。ChatGPT 的爆发让业务高管切身体会到了前沿 AI 的颠覆性力量，纷纷要求加速技术引入。管理层的真金白银投入也向控制与技术团队提出了要求：治理体系必须证明自身价值，不能一味对业务说“不”，而是要成为既能保驾护航又能加速上线的赋能引擎。
---
### [13:54](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=834s) 拓展边界：第三方供应商 AI 治理与数据主权控制
在谈到负责任 AI 的下一步演进方向时，Omodunni 明确指出，**第三方供应商 AI（Third-Party Vendor AI）的纳管**是当下的重中之重。随着市面上几乎所有软件供应商都在产品中嵌入 AI 能力，企业面临着全新的架构与数据安全挑战：
- **推理源头的数据暴露风险**：由于大模型推理通常发生在源头端（如各类 Hyperscalers 云巨头或 OpenAI），保德信在引入新厂商方案时，架构与安全团队会严格审查数据在源头端的处理机制与使用权限。
- **捍卫核心数据资产**：对于金融企业而言，数据就是立足的核心底牌。如果缺乏控制，机构与个人客户的数据就会通过五花八门的第三方工具外流至公共云端。保德信正在将这套成熟的内研模型治理框架延伸至全球生态及供应商采购链条中，确保在源头牢牢锁住数据主权。
---
### [15:20](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=920s) 职业背景：从德勤咨询到科技风险治理
在本部分结尾，Jennifer 邀请 Omodunni 分享个人职业发展路径。Omodunni 回顾称，自己的职业生涯起步于德勤（Deloitte）的管理咨询业务，长期深耕网络安全、数据隐私和数据治理领域，为众多大型金融机构搭建风险管理框架。加入保德信后，她始终聚焦于技术风险领域，并逐步从传统的科技风控演进为主导保德信全球负责任 AI 项目的产品负责人，将深厚的风控积淀与前沿 AI 运营深度融合。
### [16:02](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=962s) 从底层基础设施到负责任 AI 的风险管理演进
Omodunni Ayodele 回顾了自己在风险管理领域的职业发展轨迹与思维演进过程。他最初专注于信息安全与基础设施风险框架，主要涵盖 Unix、Wintel、分布式系统风险管理以及网络风险控制等底层 IT 架构；随后，他的重心沿着技术栈向上移动，转向了业务应用风险管理（Business Application Risk Management）。当保德信金融（Prudential Financial）决定成立数据办公室（Data Office）以大力推动分析技术与数据管理的应用时，团队意识到现有的传统风控框架无法覆盖这些新兴领域——面对层出不穷的新型风险，必须建立一套专属的管理框架。
这促使他们深入思考如何定义和应对真正的数据风险与数据管理风险，包括因无法有效利用数据而错失业务价值的潜在风险，以及如何应对人工智能与机器学习（AI/ML）带来的挑战。大约在十年前，团队就开始涉足传统经典 AI 领域（远在生成式 AI 爆发之前）。由于整个职业生涯始终深耕于风险管理领域，因此当企业构建“负责任的 AI（Responsible AI）”生态体系时，由他来支持并对接 AI 业务组织可以说是顺理成章的自然演进。
```
技术栈与风险管理重心的演进路径：
┌─────────────────────────────────────────────────────────┐
│ 负责任 AI 治理 (Responsible AI & ML Ecosystem) │
├─────────────────────────────────────────────────────────┤
│ 数据办公室与数据管理风险 (Data Office & Emerging Risks) │
├─────────────────────────────────────────────────────────┤
│ 业务应用风险管理 (Business Application Risk Management) │
├─────────────────────────────────────────────────────────┤
│ 基础架构与信息安全 (Unix, Wintel, Distributed, Network) │
└─────────────────────────────────────────────────────────┘
```
### [17:17](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=1037s) AI 带来的职业重塑与对谈总结
Jennifer Ward 对这段经验分享表示赞叹，并指出大众普遍担心“AI 会取代工作岗位”，但事实上它同时也在催生全新的职业方向与专业角色，负责任 AI 的治理体系就是极佳的例证。最后，Jennifer 对过去一年中与 Omodunni Ayodele 的紧密合作以及他在本次炉边谈话中倾囊相授的宝贵见解表达了由衷的感谢，并热情邀请现场有疑问的参会者在当天下午或次日前往展区（Expo Area）继续进行深入交流与探讨。
- AI 治理的发展不仅是防范风险，更是在重塑组织架构并创造全新的专业岗位。
- 两位讲者在大会结束后将驻守展区，为参会者提供关于企业级负责任 AI 落地与 ModelOp 平台的进一步咨询。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) 如果这些原则仅仅停留在宏大叙事或“无可挑剔的空泛道理”（motherhood and apple pie）层面，对业务部门而言便毫无实际意义。只有当这些高层原则真正转化为具体、可落地的业务运营流程时，治理才会发挥真实的价值。 — Omodunni Ayodele
> [00:04](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=4s) 众人拾柴火焰高（it takes a village）：AI 治理绝不是某一个孤立团队的工作，不能单纯交给风险合规（GRC）、法律、技术部门，也不能由不清楚如何将用例融入价值链的业务团队单打独斗，而是必须让所有相关利益方从一开始就同桌共商 — Omodunni Ayodele
> [01:58](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=118s) 风险与创新是“同一枚硬币的正反面”，而非互斥的对立面。保德信作为一家老牌金融服务机构，其立足根本正是基于信任的行业属性。 — Omodunni Ayodele
> [01:58](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=118s) 当客户询问“你们如何保护我的数据”时，数据保护绝不能是一个事后打补丁的附加项（bolt-on），而必须是一开始就贯彻始终的“内生设计”（by design）。 — Omodunni Ayodele
> [03:10](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=190s) 你无法用陈旧、手动的治理手段去监管一项飞速发展的技术。 — Omodunni Ayodele
> [06:38](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=398s) 低风险用例极速放行、中高风险用例深度审查（bear hug） — Omodunni Ayodele
> [10:47](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=647s) 如果我不懂，就向懂的同事请教，共同攻克难题 — Omodunni Ayodele
> [10:47](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=647s) 治理体系必须证明自身价值，不能一味对业务说“不”，而是要成为既能保驾护航又能加速上线的赋能引擎。 — Omodunni Ayodele
> [17:17](https://www.youtube.com/watch?v=g-Te5nS4-6I&t=1037s) 大众普遍担心“AI 会取代工作岗位”，但事实上它同时也在催生全新的职业方向与专业角色，负责任 AI 的治理体系就是极佳的例证。 — Jennifer Ward
## Mentioned
Omodunni Ayodele, Jennifer Ward, Prudential Financial (保德信金融集团), ModelOp, Gartner, Deloitte (德勤), OpenAI, ChatGPT, Microsoft Forms, Generative AI (生成式 AI), Classical AI (传统 AI)
## Video description
Omodunni Ayodele, Vice President and Product Owner of the Global Responsible AI Program, Prudential Financial
Jennifer Ward, VP of Customer Success, ModelOp
Abstracts
What does AI ROI look like in a global, regulated enterprise like Prudential Financial? In this fireside chat, Omodunni Ayodele, VP and Product Owner of Prudential’s Global Responsible AI Program, joins ModelOp's VP of Customer Success, Jen Ward, to unpack how they built a Responsible AI program on top of ModelOp's AI lifecycle automation and governance platform to drive AI innovation and better business outcomes. Learn how Prudential applies its Responsible AI framework to scale innovation, cut a risk-rating process from two weeks to one day, and streamlined governance processes to keep pace with the rapid evolution of AI. This session offers a real-world look at how Responsible AI, when operationalized through ModelOp’s AI control tower, can deliver both speed and scale without sacrificing trust. If you’re innovating with AI and trying to measure ROI , this conversation is for you.
[Source](https://www.youtube.com/watch?v=g-Te5nS4-6I)
Source quotes
[0:04] Omodunni Ayodele: 如果这些原则仅仅停留在宏大叙事或“无可挑剔的空泛道理”（motherhood and apple pie）层面，对业务部门而言便毫无实际意义。只有当这些高层原则真正转化为具体、可落地的业务运营流程时，治理才会发挥真实的价值。
[0:04] Omodunni Ayodele: 众人拾柴火焰高（it takes a village）：AI 治理绝不是某一个孤立团队的工作，不能单纯交给风险合规（GRC）、法律、技术部门，也不能由不清楚如何将用例融入价值链的业务团队单打独斗，而是必须让所有相关利益方从一开始就同桌共商
[1:58] Omodunni Ayodele: 风险与创新是“同一枚硬币的正反面”，而非互斥的对立面。保德信作为一家老牌金融服务机构，其立足根本正是基于信任的行业属性。
[1:58] Omodunni Ayodele: 当客户询问“你们如何保护我的数据”时，数据保护绝不能是一个事后打补丁的附加项（bolt-on），而必须是一开始就贯彻始终的“内生设计”（by design）。
[3:10] Omodunni Ayodele: 你无法用陈旧、手动的治理手段去监管一项飞速发展的技术。
[6:38] Omodunni Ayodele: 低风险用例极速放行、中高风险用例深度审查（bear hug）
[10:47] Omodunni Ayodele: 如果我不懂，就向懂的同事请教，共同攻克难题
[10:47] Omodunni Ayodele: 治理体系必须证明自身价值，不能一味对业务说“不”，而是要成为既能保驾护航又能加速上线的赋能引擎。
[17:17] Jennifer Ward: 大众普遍担心“AI 会取代工作岗位”，但事实上它同时也在催生全新的职业方向与专业角色，负责任 AI 的治理体系就是极佳的例证。
Source quotes
[0:04] Omodunni Ayodele: motherhood and apple pie
[1:12] Omodunni Ayodele: it takes a village
[1:12] Omodunni Ayodele: by design
[3:10] Omodunni Ayodele: Return on Employee
[3:10] Omodunni Ayodele: Return on the Future
[6:38] Omodunni Ayodele: two weeks down to less than one day
[10:47] Omodunni Ayodele: AI is a technology... The implications of AI is operational risk. That's fraught in technology: information security, data privacy, data management... It sits in our Global Technology and Operations organization.
[13:54] Omodunni Ayodele: gold fold
Original transcript
[0:04] Thank you. It's great to be here. Um,
[0:06] Omi, it has been such a pleasure working
[0:08] with you. Thank you for being here. I'm
[0:10] excited to have this conversation.
[0:12] Ditto. Just dig in. All right. You're
[0:14] the product owner for Credential's
[0:16] responsible AI program. Can you tell us
[0:18] a bit more about what that program is
[0:20] and what it means to credential?
[0:22] >> Certainly. Um, I think we'll start with
[0:24] maybe defining what responsible AI is.
[0:26] Uh it means that as we are designing um
[0:30] developing procuring using before using
[0:33] deploying using AI that we're doing so
[0:35] within a a riskaware framework um that
[0:39] underpins uh principles such as
[0:41] transparency, explanability, security
[0:43] and privacy um trust um uh uh and and
[0:47] and really bringing that into a way that
[0:49] can be operationalized. Right? While
[0:52] those principles are very high level and
[0:54] motherhood and apple pie, it doesn't
[0:56] make sense to business, right? When you
[0:58] really try to land it eventually into
[1:00] operational processes in a way that
[1:02] makes sense, that's when it becomes
[1:03] real. That's right. And this is like how
[1:06] have you operationalized this for
[1:08] business value because you're doing this
[1:10] with an eye towards the business.
[1:12] >> Yes, absolutely. Um I think one of the
[1:15] first way we started was understanding
[1:17] that it takes a village. Uh I think
[1:19] going into it and realizing that it's
[1:22] not one team. It's not a a GRC risk or
[1:26] law or compliance or technology alone or
[1:30] business not actually understanding how
[1:32] to bring the use case to become alive in
[1:34] their value chain. Um but rather to say
[1:37] you need all of those folks to be at the
[1:39] seat at the table. Um because they bring
[1:42] different types of expertise. And so the
[1:44] RAI program realized we needed to have
[1:46] this from the ground up and that's um
[1:48] that's how we began.
[1:49] >> Yeah. And I know you know when working
[1:51] with the team, one of the things that
[1:52] really stood out to me was that
[1:54] balancing act that you've that you've
[1:56] been doing. You want to talk a little
[1:57] bit about that?
[1:58] >> Yes. Um uh it's what Jen is referring to
[2:01] is uh the balancing risk and innovation,
[2:04] right? Um they're they're two sides of
[2:06] the same coin in my opinion, right? you
[2:08] want use cases um that continue to uh
[2:12] reinforce the trust industry that we are
[2:15] in. So credential is a financial
[2:16] services uh organization uh we are
[2:19] trustbased industry and so it means that
[2:21] as we think about using our customer
[2:23] institutional clients and individual
[2:25] client data um in the in pro in
[2:28] providing them the services that we
[2:29] offer to them. How do we ensure that
[2:31] we're able to do that continuously,
[2:33] efficiently, uh, but still maintaining
[2:36] that trust? And so that balance and act
[2:38] is all about how do you establish a
[2:40] framework to enable scale really quickly
[2:42] so that when we get the question, how
[2:44] are you protecting my data? It's not a a
[2:47] bolt-on, right? It's a it's a by design
[2:50] uh expectation.
[2:51] >> I love that like building it. I I
[2:53] remember the team talking about it not
[2:55] wanting to be an afterthought, not
[2:57] wanting it to be bandated on afterwards
[2:59] and that you really credential really
[3:01] looks at it as I think the phrase was
[3:03] just as a competitive competitive
[3:05] advantage, a competitive asset.
[3:07] >> Absolutely. I love that.
[3:09] >> Um
[3:10] what how how are you thinking about
[3:12] since we're on the ROI track, how is
[3:14] credential and the and the responsible
[3:15] AI program? How are you thinking about
[3:17] ROI? How do you measure success?
[3:19] >> Certainly. Uh the first thing that we
[3:21] thought about was you can't govern a
[3:24] technology that's moving so fast with
[3:26] old governance. What does that mean?
[3:28] Governance that's manual, right? And so
[3:30] the the the the governance life cycle
[3:32] and responsible AI is fancy for AI
[3:34] governance, but we were intentional
[3:36] intentional about not saying AI
[3:38] governance because AI governance comes
[3:40] with a at least for us at credential
[3:42] came with a connotation of it's a it's
[3:44] an organization that's doing something
[3:46] to the business but rather responsible
[3:48] says we're all in this together because
[3:50] it's going to cut across all our value
[3:51] chain corporate function to the business
[3:53] units. Uh and so how do we then
[3:56] understand ROI relative to understanding
[3:59] what everybody's needs are? Because ROI
[4:01] is different, right? You're measuring
[4:03] ROI from a business perspective
[4:05] that has a value chain is quite
[4:07] different from a corporate function that
[4:08] enables a business, right? And then
[4:10] being able to show the value of the the
[4:12] use cases that they've deployed. And so
[4:14] how we did that was saying that the RAI
[4:16] program needs to be automated. uh we
[4:18] need to understand really quickly each
[4:19] of the use cases that are coming to us
[4:21] and what risk it presents to the
[4:23] organization. Is it a low risk? And if
[4:25] it's a low risk, can we allow it to sail
[4:27] and fly? And if it's a higher risk and a
[4:29] medium risk, that's where we need to
[4:30] give it a a bear hug, right, of of
[4:33] governance. And so, how do you enable
[4:34] innovation and and efficiency for use
[4:37] cases that just the prior panel talked
[4:40] about the low entry use cases that
[4:42] allows you to have a quick win? we're
[4:45] able to test and learn, right, with a
[4:47] with a knowledgeb knowledgebased uh
[4:49] ecosystem and it's low risk versus a use
[4:53] case that you can then start to measure
[4:55] um and and I'll talk about the Gartner
[4:57] framework that helped us here uh where
[5:00] they had the defend append extend uh
[5:03] which started to become something that
[5:04] we liked. Gartner did a research and
[5:06] it's it's uh publicly available. Uh but
[5:09] really they talked about defend use
[5:10] cases are use cases that are return on
[5:13] employee right there are use cases where
[5:15] you're getting efficiency um employee
[5:18] time it's hard to manage ROI uh hard to
[5:21] measure ROI and those are the use cases
[5:23] that we see a lot of people learning
[5:24] around and then you've got the the
[5:26] extend use cases uh which are use cases
[5:29] that are more connected to the value
[5:30] chain right that allows you to create
[5:33] efficiencies that may touch customers
[5:35] and so you can truly measure the
[5:36] efficiency of cutting a process from x
[5:39] number of time to something shorter and
[5:42] those are extend use cases and you have
[5:43] to measure it from the beginning as
[5:45] you're going into those use cases you
[5:46] need to understand what's the measure of
[5:48] success right not only the money that's
[5:50] being committed to it but like how do
[5:52] you measure success from time to time to
[5:55] time and then you've got the upend use
[5:57] cases and those are use cases that say
[6:01] return on the future right um cut an
[6:04] edge we have masses amount of data like
[6:07] uh the the keynote that talked about
[6:09] from MongoDB and says we're getting new
[6:12] insights because we're a financial
[6:14] services company and we're generating
[6:17] insurance data with whatever claims data
[6:20] we have and we can actually create new
[6:21] product streams from this because we are
[6:24] able to tap into our data and that was
[6:27] something that the RAI program began to
[6:29] see right and wasn't it to to even be
[6:32] able to get to that place of being able
[6:34] to categorize it step one was even just
[6:35] getting a view of the inventory Yes.
[6:38] >> Well, it sounds so simple. That was a
[6:39] big That's a big part of your
[6:40] foundation. Absolutely. You've got over
[6:43] >> 200
[6:43] >> 200 models.
[6:44] >> 200 plus use cases. Yeah.
[6:46] >> Um you want to talk a bit more about
[6:48] what it even took to get to that place
[6:50] and the the assembly line. I know I know
[6:52] a big part of of your story is taking
[6:55] that process and reducing even the cycle
[6:57] time around that process.
[6:58] >> Absolutely. Um the the assembly line is
[7:01] the it takes a village line which is um
[7:06] democratize
[7:07] um as as chat GPT came in and
[7:09] democratized access to AI for everybody.
[7:12] Then REI comes in and says we are the
[7:14] orchestra. We're not going to own um
[7:17] everything but rather we need to bring
[7:19] in the law partners, the compliance
[7:21] partners, the data scientists, um the
[7:24] technologies, the the architecture uh uh
[7:28] experts to the table. Um so as a use
[7:31] case comes in to be registered, how do
[7:33] we automat automate um the risk rating
[7:36] process so that that's for inherent risk
[7:38] rating. So we did that. We established a
[7:41] quantitative methodology and then really
[7:43] quickly automated and deployed in model
[7:46] op. Um and then uh
[7:48] >> and you got over 200 models both vendor
[7:52] and in-house. I bring that up because I
[7:54] know there's been a lot of questions
[7:55] from the group about
[7:56] >> Yes.
[7:57] >> looking at both.
[7:58] >> So the the the vendor model journey is
[8:00] still ongoing. uh but we started with
[8:02] our in-house use cases and we didn't
[8:04] discriminate against generative or or
[8:05] classical traditional AI because we
[8:07] actually still find that our classical
[8:09] traditional AI models were still very
[8:11] much generating value for the
[8:12] organization. So it's not in with the
[8:15] new out with the old definitely not I
[8:17] think it's both and both both present
[8:18] risk you actually find that classical
[8:20] traditional AI models and um enabled
[8:23] with generative AI use cases can
[8:25] actually continue to evolve that
[8:27] existing roadmap uh of of a product. So
[8:30] we we look at both um and so yeah
[8:34] >> this used to take a piece of that used
[8:36] to take two weeks.
[8:38] >> Yes. Um the the inherent risk rating
[8:41] aspect was about two weeks manual. So
[8:44] you also have to think about automation
[8:45] as a key part of AI adoption right um
[8:49] and and an automation a manual process
[8:52] is very painful and froth with risk. And
[8:55] so an AI governance process that's
[8:56] manual cannot govern something that's
[8:58] going so fast as AI. And so um one of
[9:01] the key part of the process automation
[9:03] was our inherent risk rating methodology
[9:05] which was manual. It was first quantit
[9:07] qualitative and then it became
[9:09] quantitative and then that quantitative
[9:11] measure was in a macro spreadsheet that
[9:14] took about two weeks for us to process
[9:16] from intake registration from MS forms
[9:20] um and and actually generate a risk
[9:22] rating and then communicate it to folks
[9:23] and then come back and say hey let's
[9:24] talk about what you need to put in place
[9:26] in order to mitigate this risk. Then we
[9:28] took that and said like now we have a
[9:30] quantitative deterministic thing that
[9:32] can calculate inherent risk based on
[9:34] different parameters like I talked about
[9:36] before privacy security right the data
[9:39] the underlying data the underlying
[9:40] complexity of the algorithm that's being
[9:42] used and then we said like let's let's
[9:45] codify that and then codified it and
[9:48] deployed it into model and that cut down
[9:50] a two weeks process to about one less
[9:52] than one day in a good in a good day.
[9:54] >> It's amazing. It's it's interesting the
[9:56] foundation that you've laid and the ROI
[9:58] that you've already achieved just
[10:00] getting everybody on the same page and
[10:02] coming together and and creating even
[10:04] trust within the organization of I've
[10:06] got a step in this process and I know
[10:08] that someone after me is going to be
[10:09] looking at this piece and somebody after
[10:10] me is going to be looking at that piece
[10:12] for the benefit of risk management but
[10:14] also speed of innovation.
[10:16] >> Absolutely.
[10:16] >> So you've you've gone through you set a
[10:18] solid foundation. You've got an
[10:19] inventory of over 200 models. you're
[10:22] also able to step back and analyze those
[10:24] and say, you know, what's happening with
[10:26] those? Um, everybody kind of is singing
[10:28] from the same sheet of music. Sounds so
[10:31] great. What have been the challenges for
[10:34] anybody in the room who's either
[10:36] embarking on a responsible AI program or
[10:38] governance program or is maybe already
[10:40] underway? You know, what what were the
[10:42] challenges you you ran into and or the
[10:44] the success key success factors that you
[10:47] would recommend?
[10:47] >> Yeah, absolutely. The challenge is I
[10:50] think there's one thing to note about
[10:51] credential that's like we we have a very
[10:53] collaborative culture. Um and I'm sure
[10:56] different organizations have their their
[10:57] own in innate um character that makes
[11:00] them who they are. For Peru, I think
[11:03] it's a highly collaborative ecosystem
[11:05] that says, "Hey, if I don't know it, I'm
[11:08] going to tap the next person who does."
[11:10] Uh and that's like a low ego ecosystem.
[11:12] and that low ego eco ecosystem says,
[11:15] "I'm not going to own it, but I'm going
[11:17] to work with you for us to to to move
[11:19] forward together." You're going to
[11:21] probably need that because like this is
[11:23] a let's think because we've not solved
[11:25] this before. And in order for our
[11:27] organization to actually take a use case
[11:29] from ideation to production, like we're
[11:31] going to actually have to help them. And
[11:33] so everybody needs to come into the
[11:35] table and just like reduce the ego. And
[11:37] so like the hot potato discussion of
[11:39] does it sit in risk, does it sit in law,
[11:41] does it sit in technology like it ended
[11:45] up being in technology because that's
[11:47] where the risk is being created. AI is a
[11:49] technology, right? The implications of
[11:52] AI is operational risk. That's froth in
[11:55] technology. Information security, data
[11:58] privacy, data management, right? And
[12:00] then the usage of it is business for
[12:02] sure. And so we we realized that really
[12:04] quickly and so it sits in our global
[12:06] technology and operations organization.
[12:09] And then we know that we need deep
[12:11] partnership with enterprise risk in
[12:13] order for us to convey the implications
[12:15] to senior leaders and make that visible.
[12:18] We need to connect to our risk
[12:20] ecosystem, right? And then like law
[12:23] needs to be at the table, compliance
[12:25] needs to be at the table because we're
[12:26] also a regulated industry. And so so
[12:29] yeah, it it it's an orchestra
[12:32] for sure. Uh but we understood this
[12:35] because our culture also requested that
[12:37] we we be aware of that.
[12:38] >> Yeah. It isn't something you just built
[12:40] separately. Yet another kind of
[12:42] governance process, something that you
[12:44] you built in and kind of bolted on to
[12:46] things that were already in place.
[12:48] >> No, it's it's not a bolt-on. It's a by
[12:51] design. it's ground up and it's saying
[12:53] like how do I if I I had a blanket slate
[12:56] u that said here which was also a
[12:59] benefit like top leadership saw the
[13:02] value of having a program now the
[13:04] program needed to show value really
[13:06] quickly and when they saw the the need
[13:08] the need for the program was when chat
[13:09] GPT came alive right it made it real
[13:13] right for a business an executive to say
[13:15] like how do I get this in-house
[13:18] like yesterday
[13:20] >> make it
[13:21] and and that that really was the
[13:23] catalyst of like top leadership putting
[13:26] their mouth their money where their
[13:27] mouth is and then saying okay well now
[13:30] control partner ecosystem technology
[13:33] prove your worth and show me what you
[13:34] can do with this governance program so
[13:36] when I come to you with a use case you
[13:37] don't tell me no you can't put that in
[13:38] production because it's going to violate
[13:39] X12 356 right and so that was great
[13:44] >> yeah you want this to be something folks
[13:45] are attracted to or find value from
[13:48] don't perceive this to be slowing them
[13:50] down.
[13:51] >> Correct. Balancing innovation and risk.
[13:53] >> Absolutely. What do you think is next?
[13:55] What's on the horizon?
[13:57] >> Third party.
[13:59] Uh it's it's been on our list um as well
[14:02] as kind of the global uh roll out of the
[14:04] program. But the third party piece is
[14:06] not to be u ignored uh because of course
[14:10] every third party vendor wants to embed
[14:12] AI in their product. And so what does
[14:14] that mean for us from an architecture
[14:16] consumption pattern, right? Why? Because
[14:18] inferencing happens at the source. If
[14:21] the source is let's say the the the
[14:23] hyperscalers or the open AIs of the
[14:25] world then the question says like are we
[14:27] comfortable with our data being
[14:29] processed at that source and do they
[14:31] have the right to do that and so with
[14:33] that question at the head of you know
[14:35] the team and architecture as new vendor
[14:38] solutions are coming into our ecosystem
[14:40] we're already asking certain questions
[14:42] that traditionally the technology
[14:43] companies are not used to us asking that
[14:45] but the AI world has changed so much
[14:47] that you have to ask those questions
[14:49] because my goal fold is my data and if I
[14:52] don't have control over my data which is
[14:55] mine and my clients institutional
[14:57] clients are expecting us to protect that
[14:59] well then it's walking out the door
[15:01] across multiple different vendors not
[15:03] one but like multiple to the same
[15:06] probably hyperscaler that we're all on
[15:09] right so we have to think about how to
[15:12] control like what the third party
[15:13] ecosystem looks like it's an amazing
[15:16] foundation that you've laid to like
[15:18] create a jumping off point for this No,
[15:20] we just have a couple minutes left. I'm
[15:21] going to kind of add hawk if you'll
[15:22] indulge me. Ow me. Totally. Um, a lot of
[15:24] folks have talked, you know, have come
[15:26] up talking about careers. You mind just
[15:28] giving a background? You've had an
[15:29] interesting just how you've navigated to
[15:30] this place. Do you mind just covering
[15:32] real quick your background and how you
[15:33] you found yourself in this role?
[15:35] >> Totally. I I started with uh management
[15:37] consultant with Deote. Um, and and that
[15:40] was really focused on cyber security,
[15:42] privacy and and and data. And over time
[15:45] I think when I came into credential it
[15:47] was always a risk role uh but really
[15:49] technology risk. I gravitated from that
[15:51] side of the house of like serving
[15:53] multiple financial services and
[15:54] establishing kind of their risk programs
[15:56] and frameworks to then credential
[15:58] adopting um uh an expanded operational
[16:02] risk uh framework for information
[16:04] security infrastructure risks. So
[16:06] thinking about um um uh your Unix,
[16:10] Winintel, your distributed uh system
[16:12] risk management, network risk,
[16:14] networking risk management and helping
[16:16] in the infrastructure side shifted to
[16:20] business application risk management. So
[16:21] moving up the the tech stack to then
[16:24] when we decided we wanted a data office
[16:26] right to increase the use of uh
[16:29] analytics data management realized that
[16:32] this framework really didn't have a
[16:35] riskmanagement framework the risks are
[16:37] are new they're emerging risks and we
[16:39] needed to have a framework to manage it
[16:41] and so that's where kind the the cogs
[16:44] start to spin of like how do you
[16:45] understand kind of real data risks data
[16:47] management risk the lack of being able
[16:49] to use your data to actually deliver
[16:51] business value. Um, how do you think
[16:53] about like AI and machine learning
[16:54] risks? Because then at the time about 10
[16:56] years ago like the the the the
[16:59] technology started and we started to
[17:00] dabble into it in classical traditional
[17:02] AI before generative AI even came to
[17:05] bear and so that's it's always been in
[17:07] risk. So when the the the responsible AI
[17:09] ecosystem came about was you know it was
[17:12] rightfully so that I was supporting the
[17:13] organization that owned the AI
[17:15] organization. um and to
[17:17] >> I think it's fascinating. It's been
[17:18] quite a like we everybody's worried
[17:20] about you know AI removing jobs but it's
[17:22] also creating new ones. So it it's been
[17:25] great to partner with you um over this
[17:27] past year as well as at this event.
[17:29] Thank you so much for your time and for
[17:31] your for all of the the great knowledge
[17:33] that you shared with us about this. You
[17:34] have questions. We'll um we'll be out in
[17:36] the expo area this afternoon as well or
[17:38] tomorrow.
