Protiviti Sees $3.50 Return on AI Dollar as Top Adopters Beat Targets

Bridging the AI Expectation Gap With Protiviti, Inc — Ai4

Ai4

Enterprises generate predictable returns from artificial intelligence only after shifting from open-ended experimentation to rapid two-week prototypes and centralized governance.

AI-assisted summary of Bridging the AI Expectation Gap With Protiviti, Inc — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published by Ai4 on YouTube on August 24, 2025, this presentation features Christine Livingston, Managing Director and Global AI Leader at Protiviti, Inc. According to the video description, the session outlines how organizations can maximize their AI investments through a value-driven approach, select high-impact use cases, and resolve common expectation and ROI measurement gaps.

## TL;DR Christine Livingston addresses the pervasive enterprise anxiety of being 'behind' in AI adoption, presenting benchmark data that shows the vast majority of organizations are still in early exploratory stages. She explains that bridging the AI expectation gap requires shifting away from viewing AI as magic and instead adopting a structured, value-driven framework categorized into individual productivity, embedded software, and purposeful integrated workflows. To drive sustainable returns, enterprises must narrow their focus to three to five high-value use cases, prioritize two-week rapid prototyping over months of analysis, and establish centralized governance bodies to manage use-case intake and legal risk.

## Atmosphere & style The presentation carries a pragmatic, reassuring, and seasoned tone that swiftly dismantles corporate FOMO with grounded research and executive realism. Livingston's delivery blends professional authority with candid, lighthearted humor regarding corporate jargon, steering committees, and legal bottlenecks, drawing on twelve years of hands-on deployment history. Watching the talk feels like an actionable strategy briefing that clears away generative AI hype to offer clear, structured operational clarity for enterprise leaders.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) Enterprise leaders broadly suffer from the false anxiety that they are behind, yet benchmarks show only 1% to 8% of organizations have reached mature or transformative AI stages. - [01:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=78s) Perceived ROI is tightly linked to organizational maturity: 85% of surveyed organizations overall report AI meets or exceeds expectations, jumping to 96% among mature enterprises. - [01:45](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=105s) Implementation bottlenecks evolve as maturity grows, transitioning from early use-case definition to legacy technical integration and complex data governance requirements. - [02:55](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=175s) Enterprise GenAI adoption divides into three distinct archetypes: individual productivity tools, embedded SaaS features, and purposeful integrated solutions that weave into proprietary workflows for maximum return. - [04:16](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=256s) Organizations learn far more in two weeks of rapid prototyping against real data than in two months of theoretical analysis and planning. - [04:45](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=285s) Applying a four-stage framework—Envision, Prioritize, Align Technology, and Execute—ensures AI initiatives are grounded in measurable business value rather than hype. - [07:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=424s) Enterprises should narrow wide ideation funnels down to a focused shortlist of 3 to 5 high-value, low-to-medium complexity use cases with predefined KPIs. - [08:51](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=531s) Scaling AI beyond ad-hoc pilots requires a formal intake engine and centralized governing body (CoE or steering committee) to manage backlogs that can exceed 1,000 use cases. - [11:02](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=662s) Modern generative AI yields an average observed return of $3.50 for every dollar invested alongside roughly a 40% performance gain for proficient workers. - [11:35](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=695s) Deployment cycles have accelerated dramatically, reducing prototyping from months to 3–4 days and compressing enterprise production rollouts to under 6 to 12 months. - [12:29](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=749s) As AI architectures shift toward agentic workflows, organizations should mitigate security and legal risks by embedding controls into existing data governance pipelines and vetted vendor ecosystems.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) Dispelling the "We Are Behind" Myth in Enterprise AI

Opening with a quick show of hands on who has heard endless buzz about artificial intelligence over the past two years, Christine Livingston shares that she has worked in the AI ecosystem for approximately twelve years, beginning with the early IBM Watson platform. Having guided deployments across that span—some successful, others decidedly not—she addresses the single most pervasive anxiety facing enterprise leaders today: the nagging feeling that "we think we're behind; we're not as far along as everybody else."

Livingston emphasizes that industry benchmarks show enterprise adoption is still at the ground floor of the generative AI curve. A McKinsey study conducted roughly a year prior found that only 1% of organizations considered themselves mature in the generative AI space, even as the overwhelming majority planned aggressive investments. A survey released by Protiviti a month prior reinforces this reality: only 8% of enterprises classify themselves in the transformative stage, while the vast majority remain stuck in initial discovery and early experimentation.

- Christine Livingston’s background: 12 years in enterprise AI, originating with the IBM Watson platform. - McKinsey benchmark: Only 1% of organizations categorized themselves as mature in GenAI adoption. - Protiviti study benchmark: Only 8% of organizations have reached a truly transformative stage of AI implementation.

---

[01:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=78s) Maturity, Expectation Gaps, and Shifting Implementation Bottlenecks

A widespread expectation gap persists because business leaders often view AI as "magic"—assuming it will deliver immediate wonders and outsized financial returns with minimal friction. However, Protiviti's research indicates that perceived return on investment is directly linked to organizational maturity. Overall, around 85% of surveyed organizations reported that AI met or exceeded their expectations. That figure climbs to an impressive 96% among enterprises in the advanced "transformative" stage. A marked leap occurs when moving from ad-hoc experimentation into formal definition, where organizations establish structured operating models and begin realizing predictable business returns.

As maturity deepens, the operational roadblocks shift noticeably. During initial stages, data infrastructure is rarely the primary barrier; but as models become more advanced and "data hungry," stage-one through stage-five organizations encounter increasingly complex data availability and governance hurdles. Meanwhile, technical integration remains a persistent obstacle across the lifecycle, hitting its peak during the definition and optimization phases as engineering teams attempt to integrate AI solutions into legacy tech stacks at scale.

```mermaid flowchart TD accTitle: AI Maturity vs Shifting Bottlenecks accDescr: Progression from experimentation to transformation showing changes in ROI realization and technical roadblocks.

subgraph Maturity ["Maturity Progression"] A["Stage 1-2: Experimentation<br/>(Ad-hoc testing, high curiosity)"] B["Stage 3-4: Definition & Optimization<br/>(Structured use cases & systems integration)"] C["Stage 5: Transformation<br/>(Enterprise-wide scaling, 96% meet/exceed ROI)"] A --> B B --> C end

subgraph Challenges ["Evolving Bottlenecks"] D["Initial Hurdles: Use-case clarity & basic buy-in"] E["Integration Hurdles: Core systems & API middleware peak"] F["Data Hurdles: Deep data hunger & governance requirements"] A -.-> D B -.-> E C -.-> F end ```

---

[02:55](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=175s) Three Archetypes of Enterprise GenAI Adoption

Enterprise generative AI initiatives typically fall into three distinct operational categories, each serving a different purpose and carrying a different return profile:

1. **Individual Productivity Tools:** Designed as personal digital assistants to help individual knowledge workers execute tasks faster and more accurately. Examples include Microsoft Copilot and Google Gemini. These tools enhance day-to-day workflow efficiency rather than overhauling enterprise operations. 2. **Embedded Enterprise Software AI:** Core SaaS platforms—including ServiceNow, Jira, Salesforce, and SAP—have embedded generative AI capabilities directly into their interfaces. Enterprise leaders must evaluate which of these out-of-the-box vendor features actually provide value, what it practically means to turn them on, and whether they justify associated licensing tiers. 3. **Purposeful Integrated Solutions:** Rather than custom-building models from scratch, this category focuses on deeply weaving AI capabilities into proprietary workflows and internal data assets. This category consistently produces the highest measurable financial and operational returns.

---

[04:16](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=256s) Aligning AI Initiatives to Business Value and Rapid Prototyping

Too many companies chase AI simply for the excitement of using cutting-edge technology. Livingston advocates for rigorous, upfront value definition: identifying clear targets, deciding how those returns will be measured, and evaluating technical dependencies before committing capital. Teams must determine where model data originates, which core systems need to integrate, whether outputs require transparency and explainability, and how change management will be handled across the workforce.

To move past analysis paralysis, Protiviti applies a structured four-stage framework: Envision and identify, Prioritize based on value versus complexity, Align technology (choosing whether to build, buy, or borrow), and Execute. Prototyping should begin almost immediately. As Livingston highlights, "You will learn more in two weeks of prototyping than you will in two months of analysis." Testing early concepts against real enterprise data inside the target environment reveals true technical viability far faster than theoretical planning.

```mermaid flowchart LR accTitle: Four-Stage Value-Driven AI Framework accDescr: Flowchart outlining the process from opportunity identification to rapid prototyping.

subgraph Stage1 ["1. Envision & Identify"] I1["Analyze industry use cases"] --> I2["Collect cross-functional ideas"] end

subgraph Stage2 ["2. Prioritize"] P1["Score Value vs Complexity"] --> P2["Select top 3-5 opportunities"] end

subgraph Stage3 ["3. Align Tech"] T1["Decide: Build, Buy, or Borrow"] --> T2["Set model & data governance"] end

subgraph Stage4 ["4. Action"] A1["Launch 2-week prototypes"] --> A2["Measure impact & iterate"] end

Stage1 --> Stage2 Stage2 --> Stage3 Stage3 --> Stage4 ```

---

[07:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=424s) Ideation, KPI Definition, and Prioritizing the Top 3 to 5 Use Cases

Successful organizations bring stakeholders together—either in person or through digital workshops—to surface process pain points and brainstorm creative AI applications. From these sessions, broader themes emerge. Rather than leaving concepts as abstract ideas on a sticky note, teams should quickly build conceptual sketches illustrating user interfaces, target personas, and user interaction loops.

These concepts are then evaluated across two primary axes: business value and technical complexity. Instead of attempting dozens of simultaneous rollouts, leadership should narrow the funnel down to a shortlist of three to five high-value, low-to-medium complexity initiatives. Crucially, each selected project must have defined, measurable KPIs established from the outset to evaluate effectiveness and justify downstream infrastructure investments.

- High-yield ideation brings cross-functional business and technical teams together to spot recurring workflow patterns. - Conceptual mockups clarify who interacts with the AI output, how they consume it, and where friction might occur. - Initiatives are filtered to a concise roadmap of three to five low-to-medium complexity, high-value projects with defined KPIs.

---

[08:51](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=531s) Scaling Beyond Pilots: Intake Backlogs and AI Centers of Excellence

While deploying one or two isolated pilots is manageable, delivering AI value consistently and repeatably at scale requires formal operational infrastructure. Acknowledging that business leaders often dread corporate jargon—joking that teams can call it a steering committee, working group, or Center of Excellence (CoE) depending on their tolerance—Livingston notes that a centralized governing body is essential for managing intake.

Some enterprise AI steering committees face backlogs of over 1,000 use cases awaiting evaluation and prioritization. Without a structured intake and triage engine, organizations become overwhelmed. A functioning governing body coordinates across legal, privacy, and compliance departments, decides between build, buy, or borrow strategies, recruits necessary technical talent, provides developer tooling, and tracks value realization across business units.

- Backlog pressure: Enterprise backlogs can exceed 1,000 AI use cases, requiring systematic evaluation frameworks. - Governance structure: Steering committees or CoEs unite cross-functional leaders to manage compliance, privacy, and legal risks. - Operating posture: The governing body coordinates talent strategy, enterprise tooling, and ongoing KPI tracking.

---

[11:02](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=662s) The Speed of Modern AI ROI: From Ideation to Impact

Modern generative AI delivers rapid time-to-value compared to historical software cycles. Individual productivity implementations reliably yield around a 40% performance boost for proficient employees. Macro-level ROI figures show organizations generating an average of $3.50 in return for every dollar invested.

Deployment timelines have also compressed significantly. While building an enterprise machine learning prototype on early platforms like IBM Watson once took three to four months, modern generative solutions can often be prototyped in three to four days. Most production deployments now wrap up in under twelve months, with many finishing in less than six months. Livingston concludes by quoting the distinction between creativity and innovation: creativity is merely thinking about new things, whereas innovation requires actually executing them.

- Productivity boost: Around 40% performance improvement observed among employees using generative AI tools effectively. - Return on Investment: Average observed ROI of $3.50 for every $1.00 spent. - Time-to-prototype: Down from 3–4 months in early Watson deployments to 3–4 days with modern generative tools. - Production cycles: Full enterprise deployments typically take under 12 months, with many deploying in under 6 months.

---

[12:29](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=749s) Q&A: Navigating Legal Hurdles, CISOs, and Agentic Security

During the audience Q&A, an attendee asks how organizations should handle mounting legal and security concerns when deploying AI. After joking that she has "no advice," Livingston unpacks the current corporate dynamic, explaining that CIOs, CTOs, and CISOs increasingly route complex questions directly to General Counsel and legal teams for final risk determinations.

On the security side, risk must be assessed platform by platform. The landscape becomes significantly more complex as architectures advance into agentic systems and Model Context Protocol (MCP) servers, which require heightened vigilance around permissions and system-level execution. Livingston advises enterprises to prioritize vendors whose baseline security postures have already been vetted and approved internally. Furthermore, security controls should integrate directly into existing data governance platforms that monitor enterprise data pipelines, while legal and compliance strategies remain an evolving discipline requiring close partnership with executive legal teams.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) we think we're behind; we're not as far along as everybody else. — Christine Livingston

> [01:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=78s) magic — Christine Livingston

> [01:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=78s) transformative — Christine Livingston

> [01:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=78s) data hungry — Christine Livingston

> [04:16](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=256s) You will learn more in two weeks of prototyping than you will in two months of analysis. — Christine Livingston

> [04:16](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=256s) build, buy, or borrow — Christine Livingston

> [11:02](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=662s) creativity is merely thinking about new things, whereas innovation requires actually executing them — Christine Livingston

> [12:29](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=749s) no advice — Christine Livingston

## Mentioned Christine Livingston, Protiviti, Inc, Ai4, IBM Watson, McKinsey, Microsoft Copilot, Google Gemini, ServiceNow, Jira, Salesforce, SAP, Model Context Protocol (MCP)

## Video description Christine Livingston, Managing Director - Global AI Leader, Protiviti, Inc

Abstracts

Have you ever wondered why your AI investments aren't delivering the expected results? This session will explore how to maximize AI investments by taking a value-driven approach. Many companies have implemented AI solutions without fully understanding the expected value or how to measure it, leading to mismatched expectations and results. This talk will cover how to select the right use cases, align them with business goals, and address common challenges like ROI qualification before experimentation. You'll learn how to develop AI strategies to ensure your AI initiatives deliver real business impact.

[Source](https://www.youtube.com/watch?v=KJEkRxRjEo4)

中文简报

Protiviti：领跑企业超额达标，AI每投入1美元获3.50美元回报 企业只有在从开放式试验转向两周快速原型设计与集中化治理后，才能从人工智能中获得可预测的回报。 Christine Livingston针对企业普遍存在的AI落地“掉队”焦虑指出，基准数据显示绝大多数组织仍处于早期探索阶段。她解释称，弥合AI预期差距需要摒弃将AI视为魔法的看法，转而采用结构化、以价值为导向的框架，将其分为个人生产力、嵌入式软件以及针对特定目标的集成工作流三类。为获得可持续的回报，企业必须将重心收窄至三到五个高价值用例，优先开展为期两周的快速原型构建而非耗时数月的理论分析，并建立集中式治理机构以统筹用例准入与法律风险。 多数企业领导者普遍担忧自身在 AI 上落后，但 McKinsey 与 Protiviti 的基准调查显示，仅 1% 至 8% 的机构达到成熟或变革阶段，绝大多数仍处于早期探索期。 (0:04) AI 的实际 ROI 感受与组织成熟度高度相关：85% 的受访企业表示 AI 达到或超出预期，成熟企业该比例达 96%，其实施瓶颈也会逐步从用例定义转向遗留技术栈集成与数据治理。 (1:18) 企业生成式 AI 落地主要分为个人生产力工具、嵌入式 SaaS 功能以及与专有工作流深度绑定的定制集成方案，其中与自身数据资产深度融合的方案能带来最高的商业回报。 (2:55) 针对真实业务数据进行两周的快速原型开发，比耗费两个月的理论规划更有成效；企业应将海量创意聚焦至 3 到 5 个具备明确 KPI 的高价值、中低复杂度用例。 (4:16) 面对可能超过 1,000 个用例的待办积压，企业需依靠 CoE 等集中治理机制统筹推进；如今生成式 AI 原型开发仅需 3 至 4 天，平均每投入 1 美元可产生约 3.50 美元的回报。 (11:02)

Reading summary

Protiviti's Christine Livingston argues that enterprise anxiety over lagging behind in artificial intelligence is largely unfounded, as true organizational maturity remains rare and measurable returns depend on moving from ad-hoc experimentation to disciplined execution. Dispelling the Adoption Myth and Shifting Bottlenecks (0:04) Despite intense public hype, benchmark data reveals that enterprise adoption of generative AI remains in its earliest stages. A McKinsey survey cited by Livingston found that only 1% of organizations considered themselves mature in the generative AI space, while a Protiviti study reported that only 8% have reached an advanced, transformative stage. An expectation gap often arises when corporate leaders treat AI as an effortless solution; however, satisfaction correlates closely with maturity. While 85% of surveyed organizations reported that AI met or exceeded expectations, that figure rose to 96% among enterprises at the transformative stage. As organizations progress from ad-hoc experimentation to formal definition and enterprise scaling, implementation roadblocks shift. Initial hurdles center on identifying viable use cases and securing internal buy-in. As initiatives mature, technical integration into legacy enterprise stacks peaks during the definition and optimization phases. Finally, advanced stages encounter steep data governance, availability, and infrastructure hurdles as models grow increasingly data-hungry. Three Archetypes of Enterprise GenAI Adoption (2:55) Enterprise generative AI applications generally fall into three operational categories with distinct risk and reward profiles. The first category comprises individual productivity tools, such as Microsoft Copilot and Google Gemini, which act as personal assistants to boost personal workflow speed without altering core business operations. Livingston notes these tools reliably yield around a 40% performance gain for proficient workers. The second category involves embedded AI within standard enterprise software suites, including Salesforce, ServiceNow, Jira, and SAP. Organizations must evaluate whether these native vendor capabilities justify their associated licensing costs and operational adjustments. The third category, purposeful integrated solutions, weaves AI capabilities directly into proprietary workflows and internal enterprise data assets. Livingston emphasizes that this integrated tier consistently delivers the highest measurable operational and financial returns. Prioritizing Value and Compressing Prototyping Timelines (4:16) To prevent analysis paralysis and unfocused spending, Protiviti applies a four-stage delivery model: envisioning opportunities, prioritizing based on value versus complexity, aligning technology choices (build, buy, or borrow), and executing. Livingston advises against deploying dozens of concurrent projects. Instead, cross-functional teams should ideate, map user interaction workflows, and select a focused pipeline of three to five high-value, low-to-medium complexity initiatives with clearly defined KPIs from the outset. Livingston advocates moving rapidly into real-world testing, stating that organizations learn more from two weeks of prototyping with enterprise data than from two months of theoretical planning. Modern generative AI has radically accelerated delivery cycles: functional prototypes that once took three to four months on early platforms like IBM Watson can now be built in three to four days. Most modern production rollouts conclude within twelve months, with many finishing in under six months and generating an average observed return of $3.50 per dollar invested. Centralized Governance and Emerging Security Challenges (8:51) Delivering AI value across an enterprise requires formal governance infrastructure, such as a steering committee or Center of Excellence (CoE). Centralized oversight is essential for managing intake demand, as some enterprise backlogs exceed 1,000 proposed AI use cases. A dedicated governing body standardizes evaluation criteria, coordinates talent and tooling, monitors ongoing KPI realization, and bridges technical teams with legal, privacy, and compliance departments. During implementation, technical and security leaders increasingly partner with General Counsel to navigate evolving compliance and risk questions. Livingston notes that security risk must be evaluated platform by platform. This dynamic grows increasingly intricate with the emergence of agentic systems and Model Context Protocol (MCP) servers, which demand strict oversight over permissions and automated execution. To mitigate these exposures, enterprises are advised to favor pre-vetted vendors and integrate AI security controls directly into existing enterprise data governance pipelines. Livingston concludes that meaningful AI progress requires shifting from creative ideation to execution, while acknowledging that managing compliance and platform security—especially across emerging agentic architectures—remains an evolving challenge requiring tight alignment between technical leadership and legal counsel.

精炼阅读摘要

Protiviti董事总经理Christine Livingston指出，企业界普遍存在的“在AI领域已经落后”的焦虑与实际现状脱节，目前绝大多数组织仍处于早期实验阶段；企业亟须破除将技术视作“魔法”的过高预期，通过聚焦高回报的深度集成方案、建立敏捷原型验证机制与集中化治理中枢，将AI投入转化为可衡量的业务价值。 破除落后迷思与按成熟度弥合投资预期 (0:04) 基于其在企业级AI领域深耕十二年的实战经验，Livingston指出企业管理者普遍深陷“自身进度远落后于同行”的焦虑，但多项权威基准调研揭示企业对生成式AI的应用仍处在初始阶段。麦肯锡此前一年的研究表明，仅有1%的受访企业自认在生成式AI领域达到了成熟阶段；Protiviti最新公布的调研数据也显示，仅8%的企业跨入了具有颠覆性意义的变革期，绝大多数企业依然在初期探索与临时性测试中徘徊。 企业内部预期间隙的产生，主要源于决策层容易将AI视作无需磨合即可迅速产生超额回报的“魔法”。Protiviti的研究表明，企业感知到的投资回报与组织的数字化成熟度紧密挂钩：全样本中约85%的企业认为AI达到或超过了预期，而在最前沿的变革阶段企业中，这一比例高达96%。随着企业从偶发性尝试步入具备正规运营模型的定义阶段，实施瓶颈也呈现动态演进——初期主要受限于场景选择与内部认同，进入定义和优化阶段后核心系统的技术集成成为最大挑战，而到高成熟度阶段，模型对丰富数据的极度渴望则会引发更为严峻的数据治理与供给壁垒。 生成式AI的三大形态与业务价值锚定 (2:55) 企业推进生成式AI主要呈现三种运作形态，各自具备差异化的定位与回报表现。第一类是面向个人的生产力辅助工具，例如Microsoft Copilot与Google Gemini，这类助手能将熟练员工的工作效率提升约40%，但重点在于个体工作流提速而非整体业务重塑；第二类是核心SaaS软件的内置AI功能，如ServiceNow、Jira、Salesforce和SAP直接整合的模型能力，企业需要理性评估各厂商开箱即用功能的实际价值及相应的软件授权许可层级；第三类则是针对特定业务场景深度集成的专属解决方案，此类方案并不追求从头构建基础大模型，而是将AI能力深度织入企业专有工作流与私有数据资产中，实践表明该类别能够持续带来最高的财务与运营回报。 为了避免因盲目追逐技术概念而导致资本浪费，Livingston主张在立项之初就进行清晰严苛的价值界定。企业团队必须预先明确业务回报目标与量化度量指标，并在资金投入前全面评估技术依赖，这包括排查模型数据的输入来源、需对接的核心遗留系统接口、技术输出的透明度与可解释性要求，以及贯穿全员的组织变革管理配套方案。 四阶段框架与以两周敏捷原型打破空想 (4:16) 针对企业常见的“过度分析导致决策瘫痪”问题，Protiviti推行了一套结构化的四阶段实施框架：愿景识别、基于价值与复杂度的优先级排序、技术路线对齐（自主构建、外部采购或技术借用）以及敏捷执行。在构想阶段，业务与IT团队共同研讨痛点并形成概念草图，具象化交互界面与使用者闭环；管理层随后需根据业务价值与技术复杂度进行漏斗筛选，将精力和资源严格聚焦在三至五个具备中低复杂度与高商业价值的项目上，并从第一天起为每个项目确立可量化的核心KPI。 在行动阶段，快速启动原型验证是检验可行性的核心。Livingston强调，“两周的原型开发往往比两个月的案头分析能带来更多认知”；在真实生产环境下接入企业私有数据进行小步快跑，能够以最快速度暴露潜在的技术阻碍。得益于生成式AI开发链条的成熟，现代企业项目的交付周期已显著压缩：相比早年在IBM Watson平台上开发机器学习原型需耗时三至四个月，如今构建概念验证仅需三至四天，多数企业级系统在六至十二个月内便能全面投入生产运行。 治理枢纽运营与应对智能体安全风险 (8:51) 零散推进一至两个试点项目尚可应付，但若要在整个企业实现持续、可复现的价值产出，就必须建立正规化的治理机制。不论被称为指导委员会还是卓越中心（CoE），集中化的协调中枢对于应对海量需求积压至关重要——部分大型企业等待评估的AI候选用例库甚至超过了1000个。该机构负责跨法务、隐私与合规部门进行综合分流把关，权衡构建或采购方案，调配技术人才与研发工具，并持续追踪各业务条线的真实成效；整体宏观数据显示，企业在AI上的平均回报率可达到每投入1.00美元创造3.50美元回报。 面对快速上升的法务合规与网络安全挑战，企业管理层在实践中正将复杂的风险边界裁决越来越多地委托给总法律顾问（General Counsel）及合规部门。在系统安全防护方面，随着架构向自主智能体（Agentic systems）和模型上下文协议（MCP）服务器迈进，系统级执行权限与数据调用控制面临着更复杂的暴露面。Livingston建议企业优先采用已通过内部安全基线准入审查的成熟供应商方案，并将防护控制直接嵌入现有的企业数据治理管线，以此在守住合规底线的同时稳步推动规模化创新。 Livingston引用名言指出，创意的本质是思考新事物，而创新则必须将其切实付诸执行。尽管生成式AI展现出了数倍于传统软件时代的原型构建效率与平均3.5倍的投资回报，但随着自主智能体等高权限架构的引入，企业如何在敏捷落地与严苛的安全法务治理之间建立动态平衡，仍是一个需与法务高管紧密协作且处于演进之中的核心议题。

Detailed notes — zh

## Background 本视频由 Ai4 于 2025 年 8 月 24 日发布在 YouTube 平台。内容为 Protiviti 全球 AI 业务负责人 Christine Livingston 的主题演讲，主要探讨企业如何通过以价值为导向的方法弥合“AI 期望落差”，有效评估投资回报率、筛选高价值用例并制定确保业务实效的 AI 战略。 ## TL;DR 演讲指出目前绝大多数企业仍处于生成式 AI 的初期探索阶段，常因不切实际的幻想而产生期望落差，但成熟度越高的企业价值兑现度越显著。Christine 将 GenAI 场景归纳为个人生产力、平台内置与深度集成解决方案三大类，强调应聚焦能产生最高可衡量价值的集成方案。她提出了一套涵盖机会识别、价值复杂度矩阵筛选、概念设计与敏捷原型验证的四步落地框架，主张通过两周敏捷冲刺快速验证实效。最后，她建议企业建立卓越中心（COE）以应对海量用例池、协同法务与安全合规治理，实现从单点实验到规模化业务价值的跨越。 ## Atmosphere & style 整场演讲呈现出专业、沉稳且极具实战导向的咨询顾问风格。Christine 凭借超过 12 年的 AI 落地实操底蕴，以共情式的现场互动开场，有效化解了企业领导者普遍存在的“落后焦虑”。随后的分享条理清晰、数据详实，没有空洞的行业概念炒作，充满去伪存真的实干精神，给身处转型迷茫期的听众带来了清晰可行的战略指引与行动信心。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) 尽管企业普遍存在‘落后于他人’的 AI 焦虑，但整个市场仍处于生成式 AI 采纳周期的极其早期阶段，仅有约 8% 的企业进入全面变革期。 - [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) 企业 AI 成熟度与投资回报高度正相关：在进入全面变革阶段的企业中，高达 96% 的组织确认 AI 达成或超越了预期目标。 - [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) 随着 AI 方案演进，核心痛点会动态转移：系统集成在‘定义与优化阶段’成为最大瓶颈，而对高质量数据的渴求度则随成熟度提升持续走高。 - [02:55](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=175s) 企业 GenAI 应用主要分为个人生产力工具、平台内置 AI 以及目标明确的集成化解决方案三大类，其中与业务深度集成的方案能产生最高且最可衡量的商业价值。 - [02:55](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=175s) 企业切忌盲目追逐时髦的技术，必须从‘预期的业务结果’与‘战略使命对齐’出发，建立明确的价值定义与价值衡量指标。 - [05:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=318s) 倡导快速从构想走向原型验证：在两周的原型敏捷开发中所沉淀的认知与经验，远多于两个月的纸上分析。 - [05:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=318s) 建议企业通过‘价值 vs 复杂度’矩阵筛选出 3 到 5 个高价值、中低复杂度的用例作为首选切入点，并在‘自研、外购、微调’间权衡技术路径。 - [09:23](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=563s) 规模化落地的关键在于建立卓越中心（COE）或专项治理机制，以应对企业需求池中数以百计乃至上千个 AI 诉求的吞吐与优先级排序。 - [09:23](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=563s) 生成式 AI 的投资回报与交付速度表现显著：平均每投入 1 美元可带来约 3.50 美元回报，利用基础模型仅需 3 至 4 天即可快速构建原型。 - [12:29](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=749s) 安全与法务合规成为 AI 决策落地的核心关卡，随着智能体架构（Agentic AI）演进，企业需将法务及安全团队前置纳入流程，并依托平台化工具实现数据管道的全程治理与审计。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) 生成式 AI 的采纳现状与“期望落差” 演讲者 Christine Livingston（Protiviti 全球 AI 业务负责人）以现场互动开场。当她询问全场过去几年有多少人频繁听到关于 AI 的讨论时，几乎全场起立或举手。Christine 拥有超过 12 年的 AI 领域经验，早在第一代 IBM Watson 平台推出时便已涉足该领域，亲历了无数 AI 项目的落地与成败。她指出，当前企业领导者最普遍的焦虑就是“感觉我们落后了，没有跟上别人的步伐”。然而事实是，整个市场仍处于生成式 AI（GenAI）采纳周期的极其早期阶段。根据麦肯锡近一年前的调研，仅有 1% 的企业自认为在该领域达到了成熟阶段；而 Protiviti 最新发布的调查报告也显示，仅有 8% 的企业进入了“变革期”（Transformative stage），绝大多数组织仍停留在初始探索与实验阶段。 很多企业对 AI 抱有不切实际的幻想，误以为“AI 是万能的魔法”，能够瞬间创造巨大价值，从而产生了期望与现实之间的落差。Protiviti 的研究揭示了企业 AI 成熟度与投资回报之间的强关联性： ```mermaid flowchart TD accTitle: AI 成熟度与价值交付关系 subgraph S1["实验与探索阶段"] A["初始实验 / 零散用例"] --> B["数据治理处于基础期"] end subgraph S2["定义与优化阶段"] B --> C["系统集成与流程融合"] C --> D["集成难度达到峰值"] end subgraph S3["全面变革阶段"] D --> E["96% 达到或超越预期价值"] E --> F["对高质量数据极度饥渴"] end ``` - **整体回报率**：约 85% 的受访企业表示 AI 带来的成效达到或超出了预期。 - **成熟度跃升**：在进入全面变革阶段的企业中，高达 96% 的组织确认 AI 达成或超越了预期目标；从“实验阶段”跨越到“定义阶段”时，价值实现度出现了最显著的跃升。 - **演进中的挑战**：随着企业成熟度提升，核心痛点会发生转移。AI 对数据具有极强的“饥饿感”（Data hungry），随着模型向纵深演进，数据层面的挑战反而呈现上升曲线；而系统集成（Integration）则在“定义与优化阶段”成为最大瓶颈。 --- ### [02:55](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=175s) 企业 GenAI 采纳的三大类别与价值定位 在思考生成式 AI 的落地价值时，企业应将应用场景划分为三个层次，避免盲目跟风。Christine 强调，当前企业最忌讳为了追求时髦而盲目追逐 AI，必须回归到明确的“价值定义与价值衡量”上。 | 应用类别 | 典型形态与代表产品 | 核心价值与特点 | |:--- |:--- |:--- | | **个人生产力工具**<br>*(Individual Productivity)* | Microsoft 365 Copilot、Google Gemini 等内置个人助手 | 充当员工的个人副驾驶，提升个人工作速度与效率，通常可带来约 40% 的个人绩效提升。 | | **核心业务平台内置 AI**<br>*(Embedded Platform AI)* | ServiceNow、Jira、Salesforce、SAP 等原生 AI 功能 | 几乎所有主流企业级 SaaS 均已嵌入 GenAI 功能；企业需要评估何时开启这些功能以及其具体业务价值。 | | **目标明确的集成化解决方案**<br>*(Purposeful Integrated Solutions)* | 深度结合企业内部业务流程与数据的定制化集成系统 | **产生最高且最可衡量业务价值的领域**；不意味着必须从头自研底层模型，而是将 AI 有意义地融入业务运作。 | 在确立 AI 项目时，业务团队必须先回答一系列核心战略问题：我们究竟想通过 AI 达成什么目标？预期的业务结果是什么？它如何与组织的战略使命对齐？随之而来的才是技术与执行层面的落地考量：数据从何而来、需要打通哪些旧有系统、模型是否需要可解释性与透明度、依靠哪些技术与技能落地，以及如何推进组织内部的变革管理。 --- ### [05:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=318s) 敏捷落地的 AI 规划与优先级排序框架 为了帮助企业高效迈出第一步，Protiviti 总结了一套从构想到行动的标准化落地框架。该流程主张快速从理论走向原型，正如 Christine 指出的：“在两周的原型开发中你所学到的东西，远比进行两个月的纸上分析要多得多。”企业应尽快在自身数据与真实环境中运行原型，验证其是否能真正推动业务指标。 ```mermaid flowchart LR accTitle: AI 用例评估与落地推进框架 subgraph P1["机会识别"] A["行业内外用例扫描"] --> B["合规与负责任 AI 审查"] end subgraph P2["优先级评估"] B --> C["价值 vs 复杂度矩阵"] C --> D["筛选 3-5 个高价值/中低复杂度用例"] end subgraph P3["方案选型与行动"] D --> E["技术选型: 自研 / 外购 / 借用微调"] E --> F["敏捷原型验证 (2周冲刺)"] end ``` - **第一步：识别机会（Identify Opportunities）**：汇聚行业内外的创新实践，在工作坊中集中头脑风暴，挖掘能够推动组织价值的可行想法，并提前结合行业合规要求与“负责任的 AI”准则进行筛选。 - **第二步：优先级评估（Prioritize）**：通过“价值 vs 复杂度”矩阵对想法进行解构。优先锁定 3 到 5 个具有“高商业价值、低至中等复杂度”的切入点，并同步制定衡量成效的关键绩效指标（KPIs）。 - **第三步：概念原型设计（Conceptualization）**：为核心想法绘制概念图或交互线框图，明确使用者身份、交互方式与交付价值，避免停留在抽象讨论。 - **第四步：技术对齐与路线图（Align Technology & Roadmap）**：在“自研（Build）、外购（Buy）、借用/微调（Borrow/Fine-tune）”之间做出权衡，明确技术前置依赖条件，迅速进入实验与敏捷构建阶段。 --- ### [09:23](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=563s) 规模化推广与卓越中心（COE）建设 当企业成功交付一两个用例后，真正的核心挑战便浮出水面：如何将零散的成功转化为持续、可重复且规模化的生产力？许多大型企业的 AI 待办需求池（Backlog）甚至堆积了上千个潜在用例，单靠临时项目组根本无法应对大规模的评估与运营需求。 无论企业将其称为“卓越中心（Center of Excellence, COE）”、“指导委员会（Steering Committee）”还是“专项工作组（Working Group）”，建立一个集中化的赋能与治理机制都势在必行： - **需求吞吐与准入机制**：建立标准化通道，支持对全公司数以百计乃至上千个 AI 诉求进行敏捷评估、ROI 测算与排期。 - **跨部门治理协同**：将法务（Legal）、隐私（Privacy）、合规（Compliance）与信息安全团队纳入决策流，提前防范合规风险。 - **人才与赋能体系**：根据“自研、外购或微调”的技术战略，精准配置工程、算法与产品变革人才。 - **价值衡量闭环**：在立项初期即设定明确的成效验证方式，持续向利益相关者证明 AI 投资的实际回报。 在投资回报表现上，生成式 AI 正在展现惊人的速度与效益。数据显示，在善用 AI 的员工群体中，个人生产力提升幅度可达 40%；在 ROI 方面，企业在 AI 上每投入 1 美元，平均能获得 3.50 美元的回报。更关键的是，如今的项目交付周期大幅压缩：过去从零构建原型需要 3 到 4 个月，而现在利用现成基础模型仅需 3 到 4 天即可完成原型；多数落地项目在不到 6 个月内便能投入生产。Christine 引用名言总结道：“创意是去思考新事物，而创新则是真正去践行新事物（Creativity is thinking about new things. Innovation comes when we actually do new things）。” --- ### [12:29](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=749s) 现场问答：应对 AI 的安全与法务合规挑战 在演讲后的问答环节中，现场观众针对“如何应对 AI 落地过程中的安全（Security）与法务（Legal）顾虑”进行了提问。 Christine 分享了当前的行业观察与应对策略： - **决策终局在法务**：如今首席信息官（CIO）、首席技术官（CTO）以及首席信息安全官（CISO）在推进项目时，大量决策最终都会流转至法务总监与总法律顾问（General Counsel）处进行最终权衡。法务环境仍在快速演变且极为复杂，高度依赖总法律顾问团队的深度介入与政策定调。 - **安全架构复杂化**：随着技术向智能体（Agentic AI）解决方案与模型上下文协议服务器（MCP servers）演进，系统的安全架构正在变得异常复杂。 - **平台化与数据治理协同**：企业在应对安全问题时，主要策略包括优先采购已有安全背书的合规平台，或者部署专门的 AI 安全防护层；这些工具通常与企业现有的数据管道治理监控平台紧密打通，以实现数据流向的全程受控与审计。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) 感觉我们落后了，没有跟上别人的步伐 — Christine Livingston > [05:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=318s) 在两周的原型开发中你所学到的东西，远比进行两个月的纸上分析要多得多 — Christine Livingston > [09:23](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=563s) 创意是去思考新事物，而创新则是真正去践行新事物（Creativity is thinking about new things. Innovation comes when we actually do new things） — Christine Livingston > [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) AI 是万能的魔法 — Christine Livingston > [00:04](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=4s) Data hungry — Christine Livingston > [02:55](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=175s) Purposeful Integrated Solutions — Christine Livingston > [05:18](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=318s) 自研（Build）、外购（Buy）、借用/微调（Borrow/Fine-tune） — Christine Livingston > [09:23](https://www.youtube.com/watch?v=KJEkRxRjEo4&t=563s) Center of Excellence, COE — Christine Livingston ## Mentioned Christine Livingston, Protiviti, Inc, Ai4, IBM Watson, McKinsey (麦肯锡), Microsoft 365 Copilot, Google Gemini, ServiceNow, Jira, Salesforce, SAP, GenAI (生成式 AI), Agentic AI (智能体 AI), MCP servers (模型上下文协议服务器) ## Video description Christine Livingston, Managing Director - Global AI Leader, Protiviti, Inc Abstracts Have you ever wondered why your AI investments aren't delivering the expected results? This session will explore how to maximize AI investments by taking a value-driven approach. Many companies have implemented AI solutions without fully understanding the expected value or how to measure it, leading to mismatched expectations and results. This talk will cover how to select the right use cases, align them with business goals, and address common challenges like ROI qualification before experimentation. You'll learn how to develop AI strategies to ensure your AI initiatives deliver real business impact. [Source](https://www.youtube.com/watch?v=KJEkRxRjEo4)

Source quotes

[0:04] Christine Livingston: 感觉我们落后了，没有跟上别人的步伐 [5:18] Christine Livingston: 在两周的原型开发中你所学到的东西，远比进行两个月的纸上分析要多得多 [9:23] Christine Livingston: 创意是去思考新事物，而创新则是真正去践行新事物（Creativity is thinking about new things. Innovation comes when we actually do new things） [0:04] Christine Livingston: AI 是万能的魔法 [0:04] Christine Livingston: Data hungry [2:55] Christine Livingston: Purposeful Integrated Solutions [5:18] Christine Livingston: 自研（Build）、外购（Buy）、借用/微调（Borrow/Fine-tune） [9:23] Christine Livingston: Center of Excellence, COE

Source quotes

[0:04] Christine Livingston: we think we're behind; we're not as far along as everybody else. [1:18] Christine Livingston: magic [1:18] Christine Livingston: transformative [1:18] Christine Livingston: data hungry [4:16] Christine Livingston: You will learn more in two weeks of prototyping than you will in two months of analysis. [4:16] Christine Livingston: build, buy, or borrow [11:02] Christine Livingston: creativity is merely thinking about new things, whereas innovation requires actually executing them [12:29] Christine Livingston: no advice

Original transcript

[0:04] who's heard a lot about AI over the last [0:06] couple of years, the whole room, right? [0:08] I personally have been in this space for [0:11] about 12 years. Started with the first [0:13] IBM Watson platform, so seen quite a few [0:16] um deployments, some successful, some [0:18] not, right? Over the last 12 years or [0:20] so. And I think one of the things that I [0:22] want to level set with everybody right [0:24] now, the number one thing I hear is like [0:25] I think we're we think we're behind. [0:27] we're not as far along as everybody [0:29] else. And the reality is we're still [0:31] really early in the Genai adoption [0:34] curve. So this is a study from McKenzie [0:37] done um almost a year ago now. You can [0:40] see only 1% of organizations said they [0:43] were actually mature in this space with [0:45] a huge number saying we're going to go [0:47] do a lot more. Um we did our own survey [0:50] fairly recently. This just came out a [0:53] month ago and we lost the stats on this [0:57] slide so it's a little less helpful. But [0:59] when we look at the progression of [1:00] organizations, our study that we just [1:03] released showed only 8% of organizations [1:05] were in that stage of transformation [1:08] with the vast majority of organizations [1:10] still in like initial and [1:12] experimentation phases. So we're still [1:14] very very early in the deployment cycle. [1:18] And one of the things we're here to talk [1:19] about today is how do we help bridge [1:21] this expectation gap? There's an [1:22] expectation that AI is magic, that it [1:24] works wonders and it does it quickly and [1:27] returns huge value. Um, one of the [1:29] things we started to look at was does [1:32] your maturity impact your return on [1:35] investment. And what was really [1:37] interesting to see was overall we had [1:40] about 85% of organizations saying that [1:42] AI was either meeting or exceeding its [1:45] expected value. When you get all the way [1:47] to the transformative stage, 96% of [1:51] organizations are saying AI is either [1:53] meeting or exceeding our expectations [1:55] for it. So really interesting to see. [1:58] You can see a pretty big jump from [2:00] experimentation into definition from [2:03] organizations saying we're actually [2:04] starting to see the value that we expect [2:06] from these solutions. Um and we also see [2:10] that as you go through different stages [2:11] of maturity, your challenges change. Um, [2:15] so what's really interesting is to see [2:17] how data changes from stage one to stage [2:20] five. So that first curve. So presumably [2:22] as organizations are becoming more [2:24] mature with their AI solutions, AI is [2:28] data hungry. It's looking for more and [2:30] more information to consume and to [2:32] leverage. So they start to see more [2:34] challenges with data as they progress [2:36] through maturity and develop and adopt [2:38] more advanced solutions. Um, you'll also [2:41] see that integration is a pretty [2:44] consistent challenge with it being kind [2:46] of the biggest in that definition and [2:48] optimization stage, which makes sense as [2:50] when we're trying to actually integrate [2:52] AI at scale into these systems. [2:55] So, I want you to think about a couple [2:57] different categories, so to speak, of [3:00] ways that you should think about value [3:02] and genai adoption in your enterprise. [3:05] So, you're going to have individual [3:07] productivity solutions. Let me say these [3:08] are not any vendor recommendations, just [3:11] illustrative icons, right? Where the [3:13] first category are really solutions that [3:15] are just going to help people work [3:17] faster, more efficiently, more [3:18] effectively as an individual. They are [3:20] by design your personal assistant. This [3:23] is going to be things like C-pilot, [3:25] Gemini, these built-in productivity [3:27] tools um who are really going to help [3:29] you again work more efficiently as an [3:31] individual. Every single software [3:34] platform I think on the face of the [3:36] earth has now incorporated Gen AI, [3:37] right? You've seen it in everything [3:39] from, you know, Service Now to Jira to [3:41] Salesforce to SAP. Every core enterprise [3:44] platform now has AI capabilities. You've [3:47] got to decide when do I turn those on? [3:49] Which ones are going to add value? What [3:50] does it even mean to turn it on? And [3:52] then the last category um is where we [3:55] typically see the most measurable value, [3:59] which is categories that are really [4:01] looking for purposeful integrated [4:02] solutions. Not necessarily saying build [4:05] your own AI, but integrate AI in a [4:08] meaningful way into your organization. [4:11] So again, how do we do this in a way [4:14] that's really impactful and meaningful? [4:16] Um so I love these humorous takes on [4:18] this, right? So many organizations are [4:21] chasing AI for the sake of chasing AI [4:23] and doing something interesting and fun. [4:25] And of course, it's interesting and fun [4:27] and exciting to be in this space. But we [4:29] really believe you need to focus on [4:32] value definition to your organization. [4:34] And that also means how are you going to [4:35] measure that value. So when you start to [4:38] think about how are we going to adopt [4:41] and embrace AI and what are we going to [4:42] do in a meaningful way you know again [4:45] similar to the last conversation in here [4:47] think about what are we actually trying [4:48] to accomplish why are we using AI what [4:50] outcomes are we hoping to achieve how [4:52] does it align with our mission and then [4:54] you have to start to think about you [4:56] know where are we going to get data from [4:57] that for that solution what systems are [4:59] we going to need to integrate with do [5:01] the models need to be explainable and [5:02] transparent and then of course most [5:05] importantly what value are we going to [5:06] derive out of How are we going to [5:08] measure that value? What technology do [5:10] we need to deploy it? What skills? And [5:13] how are we going to manage the change [5:15] within our organization? [5:18] So, we've kind of worked through a [5:19] pretty simple successful framework for a [5:21] lot of organizations to say, you know, [5:24] envision the potential and establish [5:25] your plan. Think about where you want to [5:27] go, what you want to do, and let's build [5:29] a plan to get there based really on the [5:31] first step of identify the [5:33] opportunities. And this is what are [5:35] people doing inside your industry? What [5:37] are people doing outside of your [5:38] industry? But really importantly, you [5:40] know, what are the actionable ideas that [5:43] are going to drive value for your [5:45] organization? So you try to identify all [5:47] these opportunities. You've got to think [5:49] through, you know, some industries more [5:50] than others, responsible AI practices, [5:53] regulatory guidelines, etc. Once you've [5:56] got this great list of ideas, now let's [5:59] go prioritize them. And this is really [6:01] where you're starting to look at value [6:02] and complexity. And you want to find an [6:05] interesting mix, right, of high value, [6:07] typically low to medium complexity to [6:10] get started. But thinking through, you [6:13] know, what are we going to do to measure [6:15] value of these solutions? We've got some [6:16] ideas we'll share in the next couple of [6:18] slides and really think about how [6:21] organizationally ready are we for these [6:24] solutions? Do we have the right [6:25] stakeholders? Are they going to champion [6:26] this? Are they going to move it forward? [6:28] And then the key is that align [6:30] technology right so are we going to [6:32] build are we gonna buy are we going to [6:34] borrow right are we going to do some [6:35] like maybe fine-tuning some slight [6:37] adaptation of models to integrate that [6:39] into our network and let's build our [6:41] roadmap for action with the goal being [6:45] get to action and experimentation as [6:47] quickly as possible. I would say like [6:49] you will learn more in two weeks of [6:52] prototyping than you will in two months [6:53] of analysis. So try to get to that [6:55] action where you're working on something [6:57] very quickly. You're seeing it on your [6:58] data in your environment and you can [7:00] actually see it moving the needle again. [7:02] So ways to think about this are think [7:05] about what are we trying to accomplish? [7:07] What are the goals? What are some [7:08] creative ways to think about [7:09] accomplishing this? We found this really [7:11] great to happen getting everybody in a [7:13] room together sometimes digitally, [7:15] sometimes physically. Um but you're [7:17] really trying to get to, you know, just [7:19] this list of ideas. What are the things [7:22] that we might want to do? And typically [7:24] you'll start to see some patterns and [7:26] some themes emerge from that list of [7:29] ideas. And as you start to get really [7:32] interesting ideas and valuable ones, you [7:35] might want to conceptualize it, right? [7:37] So not just talk about it, but in the [7:40] bottom right you'll see what might this [7:41] actually look like, right? If we were to [7:43] realize this capability, um who would [7:46] use it? How would they interact with it? [7:48] What value would we create? And you can [7:50] use those ideas to get really to um a [7:54] priority view very quickly. So this is [7:56] again looking at what technical [7:57] capabilities do we need? How do they [7:59] stack up across multiple use cases? Um [8:02] here's again some representative KPIs, [8:05] not exhaustive, but this is always the [8:07] magic list people want, right? What are [8:08] the KPIs I can actually measure and use [8:10] to impact? Um, and you take again those [8:14] decomposition activities, the value [8:16] complexity, and you'll pretty quickly [8:18] see that we get to a short list of ideas [8:22] that say these are the top like three to [8:23] five high value, low to medium [8:26] complexity ideas that we can start to [8:28] build off of. And you need to understand [8:30] the value that they're going to create. [8:33] That will help you build that road map, [8:35] right? How do we actually get to [8:36] something actionable and executable? [8:39] What value are we going to deliver along [8:40] the way? what um technology [8:42] prerequisites do we need? Um and you [8:45] know, I'm going to move super quick. I [8:46] said we were going to move fast. I think [8:48] I've got two minutes left. [8:51] Um oh, seven. I got more time. Um so, [8:54] you know, again, along those lines, one [8:56] of the biggest challenges that we [8:59] typically see and is consistent with our [9:01] survey results is a lot of organizations [9:03] can find great ideas. You might have [9:05] already deployed one or two. The next [9:08] challenge is how do we start to do that [9:10] consistently, repeatably and at scale, [9:13] right? How do we deliver this value to [9:15] our organization [9:17] um in a consistent scalable repeatable [9:19] way where we can measure and demonstrate [9:21] that value? [9:23] So I know that some of you might have an [9:25] allergic reaction to the term center of [9:26] excellence like call it a steer co a [9:29] committee a COE whatever you want to [9:31] call it working group right but you need [9:33] to start to think about these [9:35] capabilities to really scale AI in a [9:39] meaningful way throughout your [9:40] organization right and so again this is [9:42] going to be like what are we going to do [9:44] what functions do we need um you know [9:46] how are we going to actually operate a [9:49] big question I hear I don't know if [9:50] anyone here's struggling with this how [9:52] are we going to take in these ideas at [9:54] scale and evaluate them quickly. I've [9:56] heard upwards of a thousand use cases in [9:59] an AI backlog for evaluation and [10:02] prioritization. Um, so hopefully that [10:04] sounds like a big number to a lot of [10:06] you. Um, but how are we going to do that [10:08] really at scale? And then starting to [10:10] think through, you know, again the [10:11] governance, who's making decisions, what [10:13] challenges are we going to have, who do [10:14] we need to involve? [10:16] Lots of questions around legal, privacy, [10:20] compliance concerns. um you know what [10:22] people are we going to incorporate to [10:25] again as our as our strategy to build [10:28] buy borrow these capabilities. We've got [10:30] to have the talent that aligns with [10:32] that. How are we going to engage and [10:34] interact with the organization as a [10:36] whole? Again, how are we going to take [10:37] in these ideas? How are we going to help [10:39] them execute on them? How are we going [10:40] to help them realize them? What tools do [10:42] we need? And of course, the value. How [10:45] are we going to measure effectiveness? [10:47] How are we going to demonstrate that [10:49] these solutions are working? um which is [10:51] why I will encourage you think about [10:53] value really really really early on [10:56] right it's step two in this process how [10:58] are we going to measure value and [11:00] demonstrate that very quickly [11:02] not again not a secret organizations are [11:05] already seeing tremendous value from [11:08] generative AI um you'll see you know [11:10] again that individual productivity [11:12] that's typically introducing about a 40% [11:15] performance boost so pretty impactful [11:17] for those employees who figure out how [11:19] to use AI and how to use it. Well, um, I [11:22] love the stats there around ROI, right? [11:24] For every dollar, you're starting to see [11:26] $3.50 in return. Number one thing, it [11:30] takes less than 12 months. I've been in [11:32] the space for a really long time. Used [11:34] to take us 3 to four months to even [11:36] prototype something. You can do that in [11:37] 3 to 4 days, right? So, these [11:39] deployments are not as extensive. [11:41] They're not as lengthy. They're probably [11:43] not as costly as you may think. And a [11:46] lot of them are actually done in less [11:47] than 6 months. Um, you know, so I was I [11:50] like this quote to kind of think to end [11:53] on, right? Creativity is thinking about [11:54] new things. You know, the innovation [11:56] comes when we actually do new things. So [11:58] it encourage you think about the value [12:00] you're going to create, how you're going [12:01] to measure that for your organization, [12:03] and then actually move to doing that [12:06] will help you sort through the the [12:08] whatifs, what might be, what value could [12:10] we realize, and will this work very very [12:12] quickly. Um, again, so encourage you to [12:15] do new things. Um, and I think this is [12:18] our code for our booth if you want to [12:21] stop by and learn a little bit more [12:23] about turning some of those AI ideas um [12:26] into value. [12:29] And [12:29] >> wow, that was amazing. A lot of great [12:32] information. Please, a round of applause [12:33] for Miss Christine Livingston. [12:39] Now, we do still have three minutes left [12:41] in her presentation time. If there are [12:43] any questions, feel free to raise your [12:44] hand and step into the aisle if you have [12:46] a question for our speaker. Yes, sir. [12:48] Can I hear [12:56] a quick question? Thank you for the [12:58] presentation. When when it comes to [13:00] legal and security concerns, uh what is [13:03] your advice on how to address it? [13:07] >> I have no advice. No, I'm just kidding. [13:10] Um it's interesting. I often see now um [13:13] CIOS, CTO's like ultimately sisos a lot [13:17] of things are are ending in legal office [13:20] for just you know discernment on which [13:22] way they want to go. the security side [13:25] is really like platform by platform [13:27] especially as we start to get into [13:28] agentic [13:30] um solutions MCP servers if any of you [13:32] are really on the the techy side that [13:34] gets really complicated really quickly [13:36] um you know but I would say shortly it's [13:38] thinking through you know are we [13:40] purchasing tools that we're already [13:41] comfortable with the security of do we [13:42] have a specific security platform that [13:44] we need to put in place to secure them [13:47] um and often that connects with your [13:49] data governance platforms as well so if [13:51] you've already got something that's kind [13:52] of monitoring your data pip pipelines, [13:53] they'll often plug into that. Um, and [13:55] then legal is, I would say, evolving and [13:59] complicated and typically takes takes [14:01] the the chief legal teams um and general [14:04] counsel's input and advice really to [14:06] determine where they land on things.