Fidelity AI VP Warns Moonshot Trap Drives 85% Project Failure Rate

Low Hanging Fruit, and Building Your AI Program Ground Up With Fidelity Investments — Ai4

Ai4

Most corporate AI initiatives collapse because leaders chase superficial demos instead of messy operational bottlenecks, whereas targeting existing manual workflows with human-in-the-loop augmentation yields a 3.7-fold return on investment.

AI-assisted summary of Low Hanging Fruit, and Building Your AI Program Ground Up With Fidelity Investments — original source. Check the original for full context.

Source published: 2025-08-24

Detailed notes

## Background Published on YouTube by Ai4 on August 24, 2025, this presentation features Sachin Solkhan, Vice President at Fidelity Investments. The session outlines a practical framework for building an enterprise AI program ground up, focusing on identifying high-ROI 'low-hanging fruit,' avoiding common execution pitfalls, and demonstrating measurable business value to stakeholders.

## TL;DR Sachin Solkhan delivers a pragmatic guide for enterprise AI implementation, contrasting the 85% failure rate of tech-first 'moonshot' projects with the 3.7x ROI realized by disciplined, workflow-focused initiatives. He outlines clear criteria for selecting early low-hanging fruit—prioritizing manual, unstructured data bottlenecks with humans in the loop—and details how to navigate risks like poor data quality and change management failure. The talk culminates in a structured 90-day action plan supported by a three-tiered value framework spanning operational, strategic, and institutional capability metrics.

## Atmosphere & style The session has the focused, authoritative feel of an executive strategy briefing, cutting through industry buzzwords with sober realism and practical corporate wisdom. Solkhan infuses the talk with relatable humor—such as comparing AI to an uninvited meeting attendee—while methodically framing AI adoption as an urgent organizational change challenge. The delivery is structured, candid, and empowering, designed to give leaders clarity and operational discipline amid pervasive market hype.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) The enterprise conversation has transitioned from speculative experimentation to an operational mandate, separating the 80% of companies adopting AI merely as table stakes from the top 20% strategically profiting from it. - [03:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=186s) "Moonshot syndrome" and chasing flashy tech-first demos lead to an 85% project failure rate, while disciplined focus on well-defined operational bottlenecks achieves 3.7x ROI. - [05:05](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=305s) Generic LLM prompts are insufficient for production; enterprise utility requires models securely grounded in proprietary data assets such as transcripts, customer interactions, and financial records. - [05:05](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=305s) Enterprise AI success hinges on change management, governance, and user trust rather than just underlying technology and infrastructure. - [07:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=426s) True "low-hanging fruit" opportunities target existing painful manual workflows, focus on unstructured data, and deliberately keep humans in the loop to augment productivity. - [09:15](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=555s) The four most common pitfalls derailing enterprise deployments are unclear objectives, unrealistic timelines, poor data quality, and neglecting change management. - [11:01](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=661s) Deployments should follow a phased three-step execution framework: Assess & Plan, Pilot & Learn, and Scale & Govern. - [11:01](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=661s) Measuring AI success requires a Three-Tiered ROI framework combining traditional operational/financial metrics, strategic competitive advantage, and long-term capability maturity. - [13:28](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=808s) Sustainable scaling requires alignment across four foundational pillars: People, Process, Platform, and Future Readiness. - [13:28](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=808s) A disciplined 90-day implementation plan should progress from readiness audits (Weeks 1–2) to value/effort prioritization (Month 1) and MVP pilot deployment (Months 2–3). - [15:42](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=942s) Battle-tested enterprise GenAI use cases that consistently deliver value today include developer coding assistants, enterprise search/knowledge copilots, and marketing content generation.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) Cutting Through the AI Hype and the AI Mandate

The conversation begins with an acknowledgment of the overwhelming AI hype cycle dominating the technology landscape over the past few years, with headlines oscillating between "utopian visions" and "doomsday predictions." Sachin Solkhan, Vice President of AI Product Engineering at Fidelity Investments, jokes that AI often feels like "that colleague of yours who shows up for all your meetings but are never invited." Cutting through this noise, Solkhan sets the stage for a pragmatic discussion aimed at moving the needle within organizations from pure hype to measurable enterprise value backed by real-world implementation insights rather than abstract theory. After providing the standard disclaimer that his views represent his own perspective and not those of his employer, he highlights how the enterprise conversation around AI has fundamentally shifted from a tentative *"maybe we should"* to an urgent *"we can't afford not to."*

The transition from exploratory R&D pilots and speculative model testing to core operational capability is happening rapidly across every sector. Enterprise adoption is no longer just about procuring off-the-shelf software tools or debating which large language model is superior, but rather building the muscle to deploy AI effectively to remain competitive. Solkhan poses a critical strategic question: will an organization end up in the 80% that deploy AI merely as table stakes to conduct day-to-day business, or join the top 20% who strategically profit from it?

- Gartner data shows approximately three-quarters (75%) of organizations are actively spending on software with baked-in AI or GenAI features. - Enterprise AI adoption has surged to 78%, up from 55%. - 92% of organizations report increasing their overall AI investments.

---

[03:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=186s) The ROI Paradox and "Moonshot Syndrome"

Despite massive capital allocation and executive mandates, a dangerous paradox is unfolding in enterprise AI execution. A vast majority of enterprise AI initiatives are stalling or getting scrapped altogether at rates far exceeding traditional IT failures. Solkhan attributes this high casualty rate to what he terms the "moonshot syndrome"—a pervasive trap where leadership teams chase flashy, superficial demos and adopt a "spray-and-pray" or technology-first approach rather than addressing real, messy, high-value business bottlenecks right in front of them.

True implementation success requires turning away from searching for problems to solve with the newest models and instead surgically applying targeted AI solutions to thoroughly understood operational workflows. Organizations that avoid scattershot experimentation and maintain disciplined execution around well-defined use cases achieve significantly higher multiples on their technology investments. "A disciplined approach beats hype every time," Solkhan emphasizes, underscoring that business problem clarity must always precede tool selection.

```mermaid flowchart TD accTitle: Moonshot Syndrome vs Disciplined AI Execution subgraph Antipattern["The Failure Pattern: Moonshot Syndrome"] A["Tech-First Mindset / Shiny Demos"] --> B["Scattershot Problem Search"] B --> C["85% Project Failure Rate"] end subgraph BestPractice["The Disciplined Pattern: Value-First"] D["Messy, High-Value Business Workflow"] --> E["Surgical AI Application & Governance"] E --> F["3.7x Return on Investment"] end ```

- 85% of enterprise AI projects are failing—double the failure rate seen in traditional IT projects. - 42% of organizations have had to scrap AI initiatives entirely. - 40% of agentic AI projects are currently getting canceled before reaching production. - Enterprises executing on tightly scoped, well-defined use cases realize an average of 3.7x ROI over unfocused competitors.

---

[05:05](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=305s) Enterprise Realities: Grounding in Data and Change Management

Examining modern enterprise patterns reveals that generative AI is moving away from standalone sandboxes and becoming deeply embedded directly within existing business workflows, such as customer support and sales enablement. In these critical workflows, generic model responses generated purely from out-of-the-box LLM prompts are an absolute non-starter. To provide actual enterprise utility, systems must be securely grounded in proprietary enterprise data—including historical sales calls, customer support transcripts, and audited financial reporting records.

Simultaneously, AI oversight has evolved into a mandatory board-level concern rather than a secondary technical consideration. Solkhan shares an anecdote about a board member friend who noted that the board chairperson spent more time interrogating the company's AI governance policies than reviewing the financial results. Building on an insight from a previous session speaker, Solkhan stresses that AI is fundamentally a change management challenge: if employees do not trust the outputs, if existing business processes cannot absorb the new workflows, and if underlying platforms cannot support the workload, any AI program is destined to fail.

---

[07:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=426s) The Blueprint for "Low-Hanging Fruit"

When deciding where an organization should launch its first AI initiative, selecting the right initial project is essential to building internal momentum and avoiding organizational friction. Solkhan outlines a specific set of criteria that define a genuine "low-hanging fruit" project—one with the highest return-on-effort ratio. The single most important formula for an early AI win is finding an existing manual process that is notoriously painful, repetitive, and time-consuming, centers on unstructured data, and keeps humans firmly in the loop to augment human skill rather than replace it entirely.

```mermaid flowchart LR accTitle: Low-Hanging Fruit Selection Criteria A["Well-Defined Business Problem"] --> B["Existing Human Workflow ('Good' Exists)"] B --> C["Unstructured Enterprise Data"] C --> D["Human-in-the-Loop Augmentation"] D --> E["High-ROI Quick Win"] ```

- **Problem Definition:** Must address a clearly understood operational bottleneck with quantifiable metrics. - **Pre-existing Workflow:** A baseline human-executed process must already exist so standard performance and quality benchmarks are already established. - **Unstructured Data Sweet Spot:** Projects should leverage text, transcripts, or documents where GenAI excels naturally over rigid rule-based systems. - **Human-in-the-Loop:** Early initiatives should deliberately augment knowledge workers, ensuring safety while driving user buy-in.

---

[09:15](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=555s) Avoiding Common Implementation Pitfalls

Identifying a promising project is only half the challenge; navigating the common operational landmines is what separates sustainable programs from scrapped pilots. Solkhan outlines the four primary failure modes that frequently derail well-intentioned enterprise deployments:

Solkhan reiterates that "technology is the easy part here," noting that successful execution ultimately lives or dies based on adoption, governance, and clarity of purpose.

---

[11:01](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=661s) Phased Execution and the Three-Tiered ROI Framework

To systematically minimize operational and political risks while accelerating value delivery, organizations should structure their AI journey across a clear three-step phased framework:

```mermaid flowchart LR accTitle: Three-Step AI Implementation Framework subgraph Step1["1. Assess & Plan"] A["Audit Skills & Data"] --> B["Evaluate System Readiness"] end subgraph Step2["2. Pilot & Learn"] C["Launch Focused Pilot"] --> D["Track Concrete Metrics"] end subgraph Step3["3. Scale & Govern"] E["Embed AI Governance"] --> F["Expand Capabilities"] end Step1 --> Step2 Step2 --> Step3 ```

Because "what gets measured always gets funded," tracking progress requires expanding beyond standard operational scorecards. While traditional financial metrics (cost savings, revenue growth) and operational indicators (cycle times, throughput) remain essential, Solkhan proposes a more holistic Three-Tiered Value Framework:

1. **Traditional Operational & Financial ROI:** Measuring baseline cost reductions, efficiency improvements, workforce productivity, and direct customer satisfaction metrics. 2. **Strategic ROI:** Gauging the broader competitive differentiation and agility the enterprise gains over industry rivals. 3. **Capability ROI:** Evaluating the organization's evolving AI maturity—a critical dimension that tracks institutional learning, platform maturity, and the team's ability to repeatedly deploy AI solutions faster over time.

---

[13:28](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=808s) The Four Scaling Pillars and the 90-Day Action Plan

Achieving durable, enterprise-wide scale requires harmonizing capabilities across four foundational pillars: **People** (upskilling, continuous training, and empathetic change management), **Process** (governance structures, explicit KPIs, and structured feedback loops), **Platform** (robust data pipelines, infrastructure, tooling, and developer systems), and **Future Readiness** (preparing for agentic AI architectures and evolving compliance/ethical frameworks).

To operationalize these pillars immediately, Solkhan presents a concrete 90-day execution roadmap structured to secure an early win while building long-term institutional momentum:

- **Weeks 1–2 (Readiness Assessment):** Audit organizational skills, identify technical and talent gaps, evaluate existing infrastructure, and review data readiness. - **Month 1 (Use Case Identification & Prioritization):** Map operational workflows on a prioritization matrix to pinpoint the sweet spot of low effort and high business value with clear projected ROI. - **Months 2–3 (Pilot Launch):** Build the Minimum Viable Product (MVP), deploy it to initial users, establish tracking for performance metrics, and conduct end-user training.

---

[15:42](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=942s) Proven Enterprise Use Cases and Path Forward

Rather than chasing over-hyped, experimental applications, Solkhan points to three battle-tested GenAI use cases that consistently deliver reliable enterprise value today:

1. **Developer Coding Assistants:** A proven productivity enhancer that speeds up software development and engineering workflows. 2. **Enterprise Information & Search Copilots:** Real-time knowledge retrieval assistants that equip front-line employees with precise answers and documentation while navigating live customer service or sales calls. 3. **Marketing Content Generation:** Accelerating the authoring and variation testing of enterprise marketing assets and collateral.

Concluding his address, Solkhan reminds the audience that the enterprise AI mandate is undeniably here. The rewards for getting it right will be massive, just as the penalties for missteps will be steep. Entering the top 20% of organizations that strategically win with AI requires replacing frantic experimentation with a series of deliberate, well-governed, and tightly executed steps.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) that colleague of yours who shows up for all your meetings but are never invited. — Sachin Solkhan

> [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) maybe we should — Sachin Solkhan

> [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) we can't afford not to. — Sachin Solkhan

> [03:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=186s) moonshot syndrome — Sachin Solkhan

> [03:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=186s) A disciplined approach beats hype every time — Sachin Solkhan

> [07:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=426s) low-hanging fruit — Sachin Solkhan

> [09:15](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=555s) technology is the easy part here — Sachin Solkhan

> [11:01](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=661s) what gets measured always gets funded — Sachin Solkhan

## Mentioned Sachin Solkhan, Fidelity Investments, Gartner, Ai4

## Video description Sachin Solkhan, Vice President, AI Search, Fidelity Investments

Abstracts

A practical session that guides organizations through establishing an effective AI program, cutting through the hype to deliver actionable strategies. The speaker will share real-world case studies demonstrating how companies have successfully implemented AI solutions that deliver measurable value without excessive investment. A framework will be shared for identifying your organization's true "low-hanging fruit" - those opportunities with the highest return-on-effort ratio. The session will address common pitfalls to avoid when launching AI initiatives and provide a metrics framework to demonstrate tangible value to stakeholders.

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

中文简报

富达AI副总裁警告：“登月计划”陷阱导致85%的项目失败率大多数企业的AI项目之所以折戟，是因为领导层盲目追求表面光鲜的演示Demo，而非解决繁琐的实际运营瓶颈；相反，针对现有手动工作流进行人机协同赋能，可带来3.7倍的投资回报率。 Sachin Solkhan为企业实施AI提供了一份务实指南，将技术优先的“登月”项目高达85%的失败率，与严谨且专注于工作流的举措所实现的3.7倍投资回报率进行了对比。他列出了挑选早期“低垂果实”的明确标准——优先解决需要人工介入的非结构化数据瓶颈，并详细说明了如何应对数据质量差和变革管理失败等风险。演讲最后提出了一个结构化的90天行动计划，并由涵盖运营、战略和机构能力指标的三层价值框架提供支持。企业 AI 已从探索性试点转变为战略刚需，但盲目追求技术演示与宏大设想的“登月综合征”（moonshot syndrome）导致 85% 的项目失败，远高于传统 IT 项目。 (3:06) 通用的 LLM 提示词无法直接用于生产环境，企业级 AI 必须深度结合内部专有数据，并将变革管理与用户信任作为核心支柱，防止技术成果被业务端排斥。 (5:05) 真正的“低垂果实”应锁定既有的高痛点手动流程，专注非结构化数据并坚持人机协同（Human-in-the-loop）模式，聚焦明确业务瓶颈的企业平均可实现 3.7 倍的投资回报率。 (7:06) 企业推进 AI 应遵循“评估规划、试点学习、规模化与治理”三步走路径，并依托涵盖运营财务、战略竞争优势以及能力成熟度的三层 ROI 框架进行全面衡量。 (11:01)

Reading summary

Sachin Solkhan, Vice President of AI Product Engineering at Fidelity Investments, contends that while enterprise AI adoption has rapidly shifted from speculative experimentation into an urgent operational mandate, organizations face high failure rates unless they abandon superficial 'moonshot' projects in favor of disciplined, data-grounded workflows. The Enterprise Adoption Boom and Moonshot Pitfalls (3:06) Enterprise commitment to artificial intelligence has accelerated dramatically, with Gartner data indicating that enterprise adoption jumped from 55% to 78%, 92% of organizations are increasing AI investments, and 75% are actively spending on software with embedded AI capabilities. However, Solkhan highlights a pervasive operational paradox: despite extensive capital deployment, 85% of enterprise AI projects fail—double the historical failure rate of traditional IT initiatives. In addition, 42% of organizations have had to scrap AI programs entirely, and 40% of agentic AI deployments are currently canceled prior to reaching production. Solkhan attributes this attrition rate to 'moonshot syndrome,' an antipattern where executive teams pursue flashy demonstrations and a technology-first 'spray-and-pray' model rather than solving concrete business bottlenecks. In contrast, enterprises executing on disciplined, tightly scoped, and well-defined use cases achieve an average 3.7x return on investment over unfocused rivals, effectively dividing the market into the 80% of organizations using AI merely as standard operational table stakes and the top 20% who strategically profit from it. Blueprint for High-ROI Low-Hanging Fruit (7:06) To build organizational momentum and avoid operational friction, Solkhan recommends that leadership teams focus their initial efforts on 'low-hanging fruit' projects that maximize the return-on-effort ratio. An ideal early initiative must target an understood operational bottleneck with quantifiable metrics, center on a pre-existing manual process where established performance benchmarks already exist, and leverage unstructured enterprise data—such as text, documents, or transcripts—where generative AI models naturally outperform rigid rule-based systems. Crucially, initial deployments must maintain humans in the loop to augment knowledge workers rather than attempt complete workforce replacement, ensuring both operational safety and internal user buy-in. Solkhan emphasizes that relying on generic out-of-the-box model prompts is unacceptable for core workflows like customer support or sales enablement; tools must be securely grounded in proprietary enterprise assets, such as audited financial reporting records and historical support logs, to provide measurable business utility. Overcoming Governance and Change Management Friction (9:15) Solkhan asserts that technology is typically the straightforward component of enterprise AI, warning that initiatives predominantly falter due to structural and human friction. He identifies four central failure modes: addressing unclear objectives that lack business stakeholder interest, managing unrealistic expectations by viewing AI as an instant fix, deploying on poor-quality or unvetted underlying data, and neglecting change management. Because technical efficacy fails if end users distrust outputs or find tools unergonomic, enterprises must prioritize explainability, trust-building, and user workflow integration. Governance has concurrently elevated into a critical board-level responsibility, with corporate directors scrutinizing compliance and risk mitigation policies as rigorously as financial performance. If employees do not trust model results or organizational platforms cannot handle integrated workloads, even well-engineered deployments will fail. Phased Delivery and Multi-Tiered Performance Measurement (11:01) To systematically manage delivery risks, Solkhan outlines a phased execution framework structured across a 90-day plan: auditing internal skills, infrastructure, and data readiness in weeks 1–2; prioritizing high-value, low-effort workflows in month 1; and launching a measured minimum viable product with end-user training in months 2–3. Sustainable long-term scaling requires aligning four operational pillars: People (upskilling and change management), Process (governance and metrics), Platform (data pipelines and developer tools), and Future Readiness (preparing for agentic systems and evolving regulatory standards). To ensure initiatives remain funded, Solkhan introduces a Three-Tiered Value Framework that evaluates more than immediate cost reductions and cycle times. In addition to traditional operational and financial ROI, organizations must track Strategic ROI (gaining competitive differentiation and agility) and Capability ROI (developing institutional learning, platform maturity, and the capacity to deploy subsequent AI capabilities faster). For immediate traction, he highlights three proven applications delivering reliable returns today: developer coding assistants, real-time enterprise search copilots, and marketing content generation. Solkhan concludes that while an enterprise AI mandate is inevitable, sustained competitive advantage requires organizations to replace unguided experimentation with deliberate, governed execution; however, the framework notes ongoing uncertainty regarding how enterprises will navigate rising compliance requirements and maintain stability as they prepare for emergent agentic architectures.

精炼阅读摘要

富达投资（Fidelity Investments）人工智能产品工程副总裁 Sachin Solkhan 指出，企业 AI 已从实验性探索转变为生死攸关的战略刚需，但盲目追求技术演示的“登月综合征”导致高达 85% 的项目夭折。他主张企业必须以明确业务痛点为先导，立足私有数据、锁定“低垂果实”、坚持人机协同并辅以严格的分阶段实施，方能跻身依靠 AI 获取战略利润的前 20% 领先梯队。投资热潮与高失败率的ROI悖论 (0:04) Solkhan 强调，当前企业对 AI 的态度已从迟疑的“或许应该尝试”转变为紧迫的“承受不起缺席的代价”。根据 Gartner 的数据，约 75% 的企业正在积极为内置 AI 或生成式 AI 功能的软件付费，企业 AI 整体采用率已由 55% 攀升至 78%，更有 92% 的组织正在追加 AI 投资。然而，Solkhan 提出企业正面临战略分水岭：未来只有 20% 的组织能够凭借 AI 获得实质战略利润，而其余 80% 仅能将其作为维持基本运营的筹码。大规模资本投入的背后普遍存在“登月综合征”（Moonshot Syndrome）与投资回报率悖论。统计显示，高达 85% 的企业 AI 项目最终失败，这一比例是传统 IT 项目失败率的两倍；42% 的企业不得不彻底废弃已启动的 AI 规划，40% 的智能体（Agentic AI）项目在投产前被取消。Solkhan 认为，这主要是因为管理层往往执迷于光鲜的演示效果，采取漫无目的的“技术先导”尝试，而非解决核心业务瓶颈。相比之下，针对清晰业务用例进行精准施策的企业，其平均技术投资回报率可达未聚焦同行的 3.7 倍。落地根基：私有数据与组织变革管理 (5:05) 生成式 AI 正从孤立的实验沙盒全面融入客户支持与销售赋能等核心业务流程。在这些关键场景中，直接调用未经微调的通用大语言模型提示词无法满足业务要求。Solkhan 指出，AI 系统必须基于企业专属的私有数据进行深度锚定（Grounding），包括历史销售通话记录、客服工单转录文本以及经过审计的财务报告数据，唯有如此才能输出可靠的业务价值。与此同时，AI 治理已上升为董事会层面的核心议题，其审议深度甚至超越了对常规财务业绩的关注。Solkhan 强调，AI 实施本质上是一场深刻的组织变革管理：技术的实现相对容易，真正的阻碍在于组织内部的吸收能力与信任度。若员工对模型输出存疑、现有业务流程无法承载新工作流或基础设施不足，项目必将搁浅。企业须重点规避四类致命陷阱：业务目标模糊、将 AI 视为速效药的脱节预期、缺乏清洗导致输出失真的劣质数据资产，以及因忽视人机工学和透明度而引发的终端用户抵制。低垂果实筛选法则与三大成熟用例 (7:06) 为了在组织内部建立推进信心并积累早期动能，企业选定首批项目的标准至关重要。Solkhan 提出了筛选“低垂果实”（Low-Hanging Fruit）的方法论，即寻找具备最高“努力-回报比”的切入点。该模式的核心在于定位那些公认繁琐痛苦、重复耗时且围绕非结构化数据展开的既有手工流程。具体筛选标准包括：明确且量化的业务痛点指标、具备可供比对质量基线的既有成熟流程、契合生成式 AI 文本处理优势的非结构化数据，以及坚持“人机协同”（Human-in-the-Loop）以赋能而非替代人力，确保输出安全并争取员工认可。在实践中，Solkhan 建议企业优先采用三类已被市场充分验证的生成式 AI 用例：一是显著提升工程效能的开发者代码编写辅助工具（Coding Assistants）；二是在客服与销售场景中实时调取精准文档的企业级知识与搜索副驾驶（Copilots）；三是能够大幅缩短制作周期并支持多版本测试的市场营销内容生成工具。分阶段实施路径与三层ROI评估框架 (11:01) 为有效化解运营与组织政治风险，企业应推行“评估规划、试点学习、规模化治理”的三步走框架。Solkhan 提出了一个可落地的 90 天行动路线图：前两周完成组织技能、技术缺口、基础设施与数据就绪度的全面审计；首月通过优先级矩阵锁定低投入、高商业价值的甜蜜点业务场景；第二至第三个月构建最小可行产品（MVP），推向初始用户群进行培训并跟踪成效指标。实现持久的规模化扩展需要协同支撑四项基础支柱：人员（技能提升与同理心变革管理）、流程（治理架构与明确 KPI）、平台（稳健的数据管道与工具链）以及面向未来架构（如智能体架构与合规伦理标准）的准备。在价值度量方面，Solkhan 倡导建立“三层价值评估模型”：除了度量成本削减和效率提升的传统运营与财务 ROI 之外，还须考量企业相对于同行建立的敏捷度与差异化优势（战略 ROI），以及衡量组织 AI 成熟度、沉淀学习经验与重复部署效率的基础能力 ROI。 Solkhan 总结指出，企业部署 AI 的外部要求已不可逆转，伴随高额潜在回报的是激增的执行风险。然而，企业在拥抱这一浪潮时面临的核心挑战，依然在于如何平衡长远技术演进与眼前的务实收益；若无法克服跨部门协作摩擦、数据基建割裂以及员工信任缺失等现实约束，盲目的技术投入极易沦为沉没成本。唯有摒弃不切实际的技术幻想，采取严谨治理、精细规划与价值先导的务实路线，企业才有可能真正突破高失败率的困局。

Detailed notes — zh

## Background 本内容由 Ai4 频道于 2025 年 8 月 24 日发布在 YouTube 平台上。视频记录了富达投资（Fidelity Investments）AI 搜索与产品工程副总裁 Sachin Solkhan 的主题演讲，主要探讨企业如何穿透 AI 炒作迷雾、锁定“低垂果实”，并从零开始构建高回报的企业级 AI 落地项目与评估框架。 ## TL;DR 演讲指出当前 85% 的 AI 项目因“登月综合征”与脱离业务流而失败，企业必须从技术导向转向务实执行。成功的 AI 战略依赖于扎根私有数据、人机协同以及以变革管理为核心的推进模式。针对实施路径，演讲提出了寻找高价值且流程成熟的“低垂果实”筛选法则、涵盖人/流程/平台/规模的四支柱体系、三层价值衡量框架以及清晰可操作的 90 天落地路线图，助力企业跻身最终获利的 20% 领跑梯队。 ## Atmosphere & style 整体基调务实稳健、直击要害，洋溢着资深一线工程管理者的清醒与务实。演讲者语言风趣犀利，开场将 AI 比作“不请自来的会议客”迅速拉近与观众的距离，随后层层剖析惨淡的失败数据与落地雷区，形成严谨沉稳的叙事节奏。聆听过程如同参与一场面向企业管理层的实战复盘会，既有直面行业炒作泡沫的冷峻警示，又给出了条理分明、鼓舞人心的系统化行动指南。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) 企业对 AI 的态度已从边缘试验转向核心战略刚需，管理者面临的关键抉择是沦为日常维持运营的 80% 平庸企业，还是跻身通过 AI 赢取战略利润的 20% 领跑梯队。 - [03:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=186s) 警惕 AI 的“ROI 悖论”与“登月综合征”：高达 85% 的 AI 项目因盲目追求光鲜 Demo 和广撒网而失败，而深挖高价值业务痛点、务实精准落地的企业能斩获 3.7 倍的投资回报率。 - [05:05](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=305s) 企业级 AI 必须牢牢扎根于自有私域数据，且其实质是一场变革管理（Change Management），依赖人员、流程与平台三者的协同进化。 - [07:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=426s) 锁定首个“低垂果实”（速赢项目）的核心准则：优先选择已有成熟流程、耗时重复、重度依赖非结构化数据且坚持人机协同（Human-in-the-Loop）的业务场景。 - [09:15](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=555s) 规避目标模糊、数据质量低下与期望不实等落地雷区，遵循“评估与规划、试点与学习、规模化与治理”实战三步走框架推进。 - [11:58](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=718s) 超越单一财务指标，构建融合“基础运营回报、战略竞争优势、组织能力成熟度”的三层价值衡量框架以确保持续的资源与预算支持。 - [13:54](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=834s) 围绕人员技能、治理流程、技术平台与规模化融入四大支柱，系统性构建支持 Agentic AI 等未来技术的企业级基础设施。 - [14:40](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=880s) 推行 90 天快速行动路线图：前两周完成就绪评估，第 1 个月圈定高价值甜点区，第 2-3 个月上线 MVP 试点并跟踪成效。 - [16:09](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=969s) 拒绝概念炒作，优先布局代码编写助手、企业信息搜索助手以及营销内容生成等已被行业充分验证的成熟场景。 - [17:01](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=1021s) AI 时代的商业指令已不可逆转，唯有依靠深思熟虑的战略规划与坚决务实的执行力，企业才能真正释放 AI 的商业价值。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) 破除炒作迷雾：从“不请自来的会议客”到落地企业价值过去几年中，所有人都在被铺天盖地的 AI 炒作周期所淹没，头条新闻充斥着从“乌托邦式的美好愿景”到“世界末日式的灾难预言”等各种极端言论。富达投资（Fidelity Investments）AI 搜索与产品工程副总裁 Sachin Solkhan 开场便以风趣的比喻切入：如今的 AI 有时就像“那位每场会议都会不请自来、准时现身的同事”。本次演讲的核心目的，正是要拨开这些噪音，探讨企业如何切实从 AI 中获取真正价值，而不是任由 AI 在各个业务场景中盲目“蹭场子”。他强调，本次分享基于真实的落地实践经验而非空洞的理论，旨在提供一套经过验证的实用步骤，推动企业 AI 从概念炒作迈向可衡量的业务成果。随着市场认知发生转变，企业对待 AI 的态度已经从“或许我们该试试”变成了“我们承担不起落后的代价”。AI 已不再是局限于实验室里的原型研发、单纯采购现成工具或争论哪款 LLM 模型最优，而是演变成了企业必须构建的组织级部署能力与核心竞争力： - **预算与采纳激增**：根据 Gartner 的统计，近 75%（3/4）的企业正在采购内置 AI 或生成式 AI 功能的企业级软件； - **使用率攀升**：企业使用 AI 的比例已从早期的 55% 攀升至 78%； - **投资持续加码**：高达 92% 的组织正在加大对 AI 领域的资金与资源投入； - **战略分水岭**：企业管理者如今面临的关键抉择是——究竟是要沦为那 80% 仅将 AI 当作日常维持运营门槛的企业，还是要跻身于那 20% 能真正通过 AI 获得战略性商业利润的领头羊。 --- ### [03:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=186s) AI 的 ROI 悖论与“登月综合征” 当前业界正显现出一个极其危险的“ROI 悖论”：尽管大笔资金与战略指令源源不断地涌入，但落地的实际成效却极为惨淡。高达 85% 的 AI 项目最终走向失败，这一失败率是传统 IT 项目的两倍；多达 42% 的企业正在直接废弃现有的 AI 倡议，更有 40% 的智能体（Agentic AI）项目惨遭叫停。导致这种大面积溃败的核心诱因在于领导层普遍患上了“登月综合征”（Moonshot Syndrome）——决策者沉迷于光鲜亮丽的技术 Demo，盲目追求“广撒网、碰运气”（spray and pray）的技术导向路径，却忽视了眼前那些虽然繁琐复杂但具有高商业价值的真实业务流。事实证明，以技术为先、拿锤子找钉子的做法是公认的失败根源。真正的成功并非拿着最先进的模型去四处物色问题，而是深入业务一线痛点，进行外科手术式的精准赋能。那些聚焦于高价值、定义清晰场景并严格执行的企业，获得了近 3.7 倍的投资回报率（3.7x ROI），表现远超盲目试水者。“在任何时候，严谨务实的执行力都能击败盲目的炒作。” ```mermaid flowchart LR accTitle: AI 项目落地路径对比 subgraph Bad["失败模式：登月综合征（85% 失败率）"] direction TB B1["盲目追逐光鲜 Demo"] --> B2["技术导向 / 广撒网"] B2 --> B3["脱离复杂业务流"] end subgraph Good["成功模式：严谨务实执行（3.7x ROI）"] direction TB G1["深挖高价值业务痛点"] --> G2["精准匹配 AI 解决方案"] G2 --> G3["聚焦清晰场景与人机协同"] end ``` --- ### [05:05](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=305s) 企业级 AI 落地模式：数据根基、治理与变革管理当前企业级 AI 正处于从边缘实验全面融入核心业务流的关键转折期，例如深度嵌入客户支持与销售赋能等高频场景中。在这一过程中，单纯调用通用大模型生成的泛化结果根本行不通，AI 系统必须牢牢扎根于企业自有的私域数据之上——包括客户通话录音文本、销售沟通纪要以及内部财务报告等结构化与非结构化数据。与此同时，AI 治理（Governance）已经从单纯的技术合规演变为主流董事会的核心议题。Sachin 分享了一个生动的事实：一位担任董事会成员的朋友透露，在最近一次董事会上，董事长针对 AI 治理与风险控制提出的质询，甚至超过了对常规财务状况的关注。从根本上说，“AI 本质上是一个变革管理（Change Management）问题”。成功的落地依赖于“人员（People）、流程（Process）与平台（Platform）”三者的协同进化：如果一线员工不信任它、业务流程无法包容它、底层技术系统无法支撑它，那么任何 AI 倡议都注定会以失败告终。 --- ### [07:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=426s) 寻找“低垂果实”：首战告捷的核心筛选标准对于刚起步的企业而言，能否精准锁定并摘取“低垂的果实”（Low-hanging Fruit）是决定 AI 战略成败的关键第一步。首个 AI 赢局必须建立在具备明确成功衡量标准的场景之上，其最佳筛选标准可提炼为： - **已有成熟范式**：优先选择已有成熟人工流程的场景（即“良好基础已存在”），避免在完全未知的流程中引入新技术； - **非结构化数据密集**：大量涉及非结构化文本、音频或文档，这正是生成式 AI 最擅长释放威力的核心甜蜜点； - **坚持人机协同（Human-in-the-Loop）**：始终保留人类专家的审核与干预机制，确保输出可靠与风险可控。 Sachin 强调，如果整场演讲中听众只能记住一句话，那就是这句项目筛选准则：“**企业斩获首个 AI 胜果的最佳切入点，必然是一个令人痛苦、高度重复、耗时费力、重度依赖非结构化数据，且 AI 能有效增强人类技能的业务流程。**” 这样一个可见度极高、能迅速落地的“速赢”（Quick Win）项目，能够迅速为组织建立信心、激发内部需求，并有效化解员工的抵触心理。 --- ### [09:15](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=555s) 避开落地雷区与三步走实施框架在锁定了潜在项目之后，识别并规避可能导致项目脱轨的“隐形地雷”构成了另一半战场： - **业务目标模糊**：在无人关心的边缘问题上空耗资源；必须坚持业务优先，确立明确清晰的 ROI 目标； - **设定不切实际的期望**：AI 落地不是短跑冲刺，而是一场需要长期投入与清晰沟通的“马拉松”，必须从小而精、价值明确的狭窄场景切入； - **数据质量低下**：技术领域“垃圾进，垃圾出”（Garbage In, Garbage Out）的铁律依然适用； - **忽视变革管理**：技术往往是最容易解决的部分，实施的成败最终取决于终端用户的采纳度、组织的治理规范以及初心的清晰度。为系统化推动落地，Sachin 提出了实战型三步走框架： ```mermaid flowchart LR accTitle: 企业 AI 落地三步走框架 step1["1. 评估与规划 (Assess & Plan) 技能审计 / 数据完备度评估"] --> step2["2. 试点与学习 (Pilot & Learn) 聚焦试点 / 收集核心关键指标"] step2 --> step3["3. 规模化与治理 (Scale & Govern) 嵌入合规治理 / 沉淀组织专长"] ``` 这种循序渐进的阶段性推进方式，不仅能显著加速成果产出，还能大幅降低企业在推进新技术时面临的运营与组织内部政治风险。 --- ### [11:58](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=718s) 三层价值衡量框架：超越单纯的财务 ROI 在项目推进过程中，“凡是能量化的指标，才最终能获得预算支持”（What gets measured always gets funded）。传统企业往往将评估局限于财务成本、运营效率、客户满意度或人力节省等常规维度，但要实现 AI 在企业内的长远扎根，必须构建一个立体的“三层价值衡量框架”（Three-tiered Value Framework）： | 价值层级 | 评估核心与内涵 | |:--- |:--- | | **基础价值 (Financial & Operational ROI)** | 关注直接的财务节省、成本削减、流程吞吐量与运营效率的量化改善。 | | **战略价值 (Strategic ROI)** | 评估 AI 对企业核心竞争壁垒的强化程度，以及在市场中的差异化竞争优势。 | | **能力价值 (Capability ROI)** | 衡量组织整体 AI 成熟度、技术演进速度以及跨部门数据与创新文化的深层跃迁。 | 唯有将硬性的财务回报与软性的战略、文化影响力深度融合，企业才能建立起持久的 AI 能力，并在人员、流程与平台这几大核心支柱之间形成合力，支撑企业级 AI 的全面规模化扩展。 ### [13:54](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=834s) 企业级 AI 落地的四支柱框架构建企业级 AI 能力必须依靠系统化的支撑体系。规模化落地（Scale）并非一蹴而就，而是需要通过“人、流程、平台与规模”这四大支柱，将人工智能真正融入企业的组织肌理与业务脉络中。 ```mermaid flowchart TD accTitle: Enterprise AI Four Pillars subgraph P1["人员 (People)"] A["技能培训"] --> B["变革管理"] end subgraph P2["流程 (Process)"] C["治理框架"] --> D["明确 KPI 与反馈循环"] end subgraph P3["平台 (Platform)"] E["基础设施与工具管线"] --> F["未来就绪性 (智能体/合规伦理)"] end subgraph P4["规模化 (Scale)"] G["全面融入企业组织肌理"] end P1 --> G P2 --> G P3 --> G ``` - **人员（People）**：核心在于系统性的技能培训与变革管理，确保团队能够适应 AI 驱动的新工作模式。 - **流程（Process）**：关键在于建立严格的治理框架，制定清晰可量化的关键绩效指标（KPI），并构建及时的反馈闭环。 - **平台（Platform）**：涵盖底层基础设施、工具链与数据管线的建设，同时兼顾未来的技术就绪度（Future Readiness），包括面向智能体 AI（Agentic AI）的准备工作以及合规与伦理框架的搭建。 - **规模化（Scale）**：依托上述三大支柱，将 AI 技术无缝嵌入企业日常运营的各个环节。 --- ### [14:40](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=880s) 从就绪评估到首个试点：90 天行动路线图针对希望立即启动 AI 项目的企业，演讲者提出了一个兼顾效率与稳健性的 90 天实施路线图。该方案的设计初衷在于“既足够短，能够快速锁定早期收益（Early Wins）；又足够长，足以形成持续推进的组织势能”。 ```mermaid flowchart LR accTitle: 90-Day AI Action Plan subgraph W12["第 1-2 周: 就绪评估"] A["技能与差距分析"] --> B["数据与基础设施审查"] end subgraph M1["第 1 个月: 场景优先级排序"] C["用例矩阵梳理"] --> D["锁定低投入/高价值 Sweet Spot"] end subgraph M23["第 2-3 个月: 启动首个试点"] E["MVP 最小可行产品开发"] --> F["指标跟踪与用户培训"] end W12 --> M1 M1 --> M23 ``` - **第 1–2 周（就绪评估，Readiness Assessment）**：对企业内部的技能储备进行全面摸底与差距分析，审查现有的数据资产质量、系统基础设施及各项报告。 - **第 1 个月（用例识别与优先级排序，Use Case Identification & Prioritization）**：构建用例矩阵，计算各项方案的投资回报率（ROI），精准锁定“低投入、高价值”的甜点区（Sweet Spot），筛选出真正具有实操价值的“低垂果实”。 - **第 2–3 个月（启动首个试点项目，Launching Pilot Project）**：推进最小可行产品（MVP）的开发与上线，建立关键性能监控指标，并同步开展针对终端用户的操作培训。 --- ### [16:09](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=969s) 拒绝炒作：三类已被验证的企业级落地场景在评估具体应用时，企业应当跳出过度炒作的概念陷阱，优先选择那些已在众多大型企业中被反复验证、具备明确商业成效的成熟场景： - **代码编写助手（Coding Assistants）**：在软件工程团队中部署成熟的代码补全与生成工具，这是目前行业内验证最为充分、回报最明确的落地场景之一。 - **企业信息搜索助手（Enterprise Search Assistants）**：在客服座席接听客户电话或响应客户咨询时，通过智能搜索助手快速检索企业内部知识库，提供精准信息支撑与实时应答辅助。 - **营销内容生成（Marketing Content Generation）**：赋能市场与运营团队，规模化自动化生成各类营销文案与推广物料，大幅缩短内容产出周期。 --- ### [17:01](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=1021s) 把握 AI 时代机遇：通过审慎执行进入领跑梯队 “AI 的时代指令已经明确到来（The AI mandate is here）。”这一趋势带来的潜在回报是巨大的，但如果执行出现偏差，面临的惩罚和代价也同样沉重。根据行业规律，未来只有约 20% 的企业能够从这一波技术浪潮中获取绝大部分的商业价值。要跨入这 20% 的领先梯队并释放真正的 AI 潜力，企业不能盲目跟风，而必须采取“一系列深思熟虑且执行得当的坚实步骤（a series of deliberate and well-executed steps）”。现阶段战略框架与实施路径均已清晰，关键在于迅速行动与坚决执行。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) 那位每场会议都会不请自来、准时现身的同事 — Sachin Solkhan > [00:04](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=4s) 我们承担不起落后的代价 — Sachin Solkhan > [03:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=186s) 在任何时候，严谨务实的执行力都能击败盲目的炒作。 — Sachin Solkhan > [05:05](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=305s) AI 本质上是一个变革管理（Change Management）问题 — Sachin Solkhan > [07:06](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=426s) 企业斩获首个 AI 胜果的最佳切入点，必然是一个令人痛苦、高度重复、耗时费力、重度依赖非结构化数据，且 AI 能有效增强人类技能的业务流程。 — Sachin Solkhan > [09:15](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=555s) 垃圾进，垃圾出（Garbage In, Garbage Out） — Sachin Solkhan > [11:58](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=718s) 凡是能量化的指标，才最终能获得预算支持（What gets measured always gets funded） — Sachin Solkhan > [14:40](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=880s) 既足够短，能够快速锁定早期收益（Early Wins）；又足够长，足以形成持续推进的组织势能 — Sachin Solkhan > [17:01](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=1021s) AI 的时代指令已经明确到来（The AI mandate is here）。 — Sachin Solkhan > [17:01](https://www.youtube.com/watch?v=jsJsaDCSxmA&t=1021s) 一系列深思熟虑且执行得当的坚实步骤（a series of deliberate and well-executed steps） — Sachin Solkhan ## Mentioned Sachin Solkhan, 富达投资 (Fidelity Investments), Ai4, Gartner ## Video description Sachin Solkhan, Vice President, AI Search, Fidelity Investments Abstracts A practical session that guides organizations through establishing an effective AI program, cutting through the hype to deliver actionable strategies. The speaker will share real-world case studies demonstrating how companies have successfully implemented AI solutions that deliver measurable value without excessive investment. A framework will be shared for identifying your organization's true "low-hanging fruit" - those opportunities with the highest return-on-effort ratio. The session will address common pitfalls to avoid when launching AI initiatives and provide a metrics framework to demonstrate tangible value to stakeholders. [Source](https://www.youtube.com/watch?v=jsJsaDCSxmA)

Source quotes

[0:04] Sachin Solkhan: 那位每场会议都会不请自来、准时现身的同事 [0:04] Sachin Solkhan: 我们承担不起落后的代价 [3:06] Sachin Solkhan: 在任何时候，严谨务实的执行力都能击败盲目的炒作。 [5:05] Sachin Solkhan: AI 本质上是一个变革管理（Change Management）问题 [7:06] Sachin Solkhan: 企业斩获首个 AI 胜果的最佳切入点，必然是一个令人痛苦、高度重复、耗时费力、重度依赖非结构化数据，且 AI 能有效增强人类技能的业务流程。 [9:15] Sachin Solkhan: 垃圾进，垃圾出（Garbage In, Garbage Out） [11:58] Sachin Solkhan: 凡是能量化的指标，才最终能获得预算支持（What gets measured always gets funded） [14:40] Sachin Solkhan: 既足够短，能够快速锁定早期收益（Early Wins）；又足够长，足以形成持续推进的组织势能 [17:01] Sachin Solkhan: AI 的时代指令已经明确到来（The AI mandate is here）。 [17:01] Sachin Solkhan: 一系列深思熟虑且执行得当的坚实步骤（a series of deliberate and well-executed steps）

Source quotes

[0:04] Sachin Solkhan: that colleague of yours who shows up for all your meetings but are never invited. [0:04] Sachin Solkhan: maybe we should [0:04] Sachin Solkhan: we can't afford not to. [3:06] Sachin Solkhan: moonshot syndrome [3:06] Sachin Solkhan: A disciplined approach beats hype every time [7:06] Sachin Solkhan: low-hanging fruit [9:15] Sachin Solkhan: technology is the easy part here [11:01] Sachin Solkhan: what gets measured always gets funded

Original transcript

[0:04] So again good afternoon everyone. Um let [0:07] me start by acknowledging that all of us [0:11] have been inundated by uh the whole AI [0:14] hype cycle over the past couple of [0:16] years. Uh lot of headlines about AI u [0:20] from utopian visions uh to doomsday [0:23] predictions. But today I want to cut [0:25] through the noise. [0:27] uh AI sometimes feels like that uh [0:32] colleague of yours who uh shows up for [0:35] all your meetings but are never invited. [0:38] Um [0:40] I I want to talk uh in the next 15 20 [0:43] minutes about how uh you can get value [0:46] out of AI and not AI just crashing the [0:50] party. Uh so in the next 15 minutes I [0:53] will share a set of uh practical uh but [0:56] proven steps uh to move AI needle in [0:59] your organization uh get you from hype [1:02] to measurable enterprise value uh backed [1:06] by real world insights um and not just [1:09] theory. So my name is Sachin Sulken. I'm [1:11] a VP of AI product engineering at [1:13] Fidelity Investments. Uh before I begin, [1:16] okay, first a disclaimer uh that uh all [1:19] the views are my own and uh not [1:21] reflected uh of my company uh uh not [1:25] their thoughts uh of my employer. So [1:27] let's uh let's get to the heart of the [1:30] matter. Uh so the AI mandate is here. Uh [1:34] let's start by looking at why the AI [1:37] conversation has moved from maybe we [1:40] should to we can't afford not to. [1:44] uh some metrics to talk about almost uh [1:48] as per Gartner 3/4 of the organizations [1:52] um are uh going to be [1:56] spending on AI related software some [1:59] kind of AI or genai baked into it. uh [2:02] 78% of organizations are using AI up [2:05] from 55% [2:07] and uh 92% organizations are increasing [2:10] investment uh in AI. [2:14] Uh so AI is no longer u about pilot R&D [2:19] projects uh or simply buying AI tools or [2:23] figuring out which is the best LLM model [2:25] to use. uh but really it is building [2:28] about building your organization [2:30] capability to deploy them effectively. [2:33] Uh this is becoming table stakes for [2:35] competitiveness. Uh the critical [2:38] question that we all should be asking [2:40] here is do we really want to be part of [2:43] that 80% of the organizations who in the [2:46] future will employ just to do business [2:49] or do we want to be part of the 20% of [2:52] the organizations who are going to [2:54] strategically profit from it. uh because [2:57] we are seeing some interesting paradox [2:59] uh that dangerous paradox that is [3:01] emerging uh if we get this wrong [3:06] the ROI paradox um so the AI mandate is [3:10] clear the investments are flowing in but [3:14] uh there is a flip side to all this [3:15] mandate [3:17] almost 85% of AI projects are failing [3:20] twice the rate of uh traditional IT [3:22] projects [3:27] uh 42% organizations are u just [3:30] scrapping AI initiatives. Uh 40% of [3:34] agentic AI projects are getting [3:36] cancelled. [3:38] uh [3:39] the one reason I think uh for this uh is [3:43] what I call the moonshot syndrome right [3:46] so this is really leaders [3:48] uh chasing shiny [3:51] demos instead of really looking at their [3:55] business flows high value business flows [3:57] messy but high value business flows with [4:01] problems which are right in front of [4:03] them so this pre or and prey technique [4:07] or a tech first approach is a as we know [4:09] is a proven recipe for failure. So, so [4:13] what is the reality here? [4:16] Uh success of AI implementation is not [4:19] about just searching for a problem u to [4:23] solve using this latest and greatest [4:25] models. It really comes from deeply [4:28] understanding the business problem and [4:30] surgically applying the right AI [4:32] solution to it. [4:34] Uh that's why enterprises that execute [4:37] well uh on [4:40] well-chosen well-defined use cases are [4:43] seeing almost 3.7x return on investment [4:47] and consistently outerforming uh [4:49] outperforming those who are following [4:52] like a scattershot approach. So key [4:54] takeaway is that a disciplined approach [5:00] beats hype every time. [5:02] every single time. [5:05] So how do we operationalize this [5:07] disciplined execution? [5:10] It starts with understanding the [5:12] tangible [5:14] um shapes that are right that are [5:18] shaping enterprise AI today. [5:23] So the patterns are clear across [5:26] enterprises. [5:28] AI is moving from experimentation to [5:30] being embedded within business [5:32] workflows. [5:33] So think of like customer support um or [5:37] sales en enablement uh kind of use cases [5:41] all grounded in [5:43] uh enterprise data as generic results [5:46] are just a non-starter. Generic results [5:49] that you get from just calling LLM [5:52] models are just a non-starter. So think [5:54] of data like your customer call [5:57] transcripts, your sales call [5:59] transcripts, your financial reporting [6:02] data. [6:04] AI governance is now a board level [6:08] discussion, not just a tech issue. Uh a [6:11] friend of mine who is on the board [6:12] recently told me uh about his recent [6:15] board meeting where the chairperson of [6:16] the board was asking more AI governance [6:19] questions than uh than about um uh [6:23] financials [6:25] and successful AI implementation [6:28] requires um [6:31] uh people process and platform [6:33] capabilities. [6:35] uh as the previous speaker at the end [6:37] was saying and I it was it's a good uh [6:40] start for my conversation and my topic [6:42] also is that as she said AI is [6:45] fundamentally a change management [6:47] problem if people don't trust it if [6:51] processes cannot accommodate it and and [6:54] your platforms your technology systems [6:57] cannot support it then any AI initiative [7:01] is doomed to be a failure Yeah. [7:06] So, how do you um where do you start? Uh [7:11] as the previous speaker was saying at [7:12] the end, identify the lowhanging [7:15] projects. Uh how how do you choose the [7:17] projects, right? How do you identify [7:20] those lowhanging fruits? [7:22] I I want to propose a a criteria for it [7:26] starting with [7:29] uh start with a well- definfined problem [7:33] uh success where um which followed by [7:43] uh followed by those kind of workflows [7:47] where good already exists. What I mean [7:50] is already there is a human executed [7:52] manual workflows that exist [7:55] u followed by uh where there is [7:58] unstructured data involved uh which is [8:01] kind of the sweet part spot for any of [8:03] the genai capabilities and always always [8:07] with human in the loop. [8:09] uh I think this is a good criteria for [8:12] starting with your like first um kind of [8:16] AI project. [8:18] If there is nothing else you remember [8:21] from my talk today um this is one thing [8:25] and what this slide says is what I want [8:27] you to remember. your your best bet for [8:30] your first AI win will be a process that [8:34] is [8:35] uh painful, that is repetitive, that is [8:39] time consuming, [8:41] involves unstructured data and where AI [8:44] is augmenting human skill. [8:48] Uh such a visible kind of quick win will [8:51] can build the momentum, can build the [8:53] demand and and avoid resistance. um uh [8:58] which is again the change management [8:59] problem. Now finding such kind of [9:03] projects is just half uh the battle. The [9:06] other is avoiding some pitfalls, some [9:08] land line landmines that could uh derail [9:12] many of your well-meaning initiatives. [9:15] So what are those? [9:21] So the first one u and it's some of [9:23] these uh I'm u it reiterating [9:27] um some of the already mentioned uh [9:30] unclear business objectives like working [9:33] on problems that nobody cares [9:36] uh and and uh also providing a uh I'm [9:39] providing a solution for it right [9:40] obviously business first approach rather [9:42] than a uh technology first approach is [9:45] good clearly defined ROI targets uh [9:48] second [9:50] uh setting unrealistic expectations [9:53] implementing AI is is a marathon. [9:58] Um it's it's it requires sustained [10:01] commitment. It requires clear [10:03] communication. Uh so start with well the [10:06] solutions start with well definfined [10:08] narrow use cases showing clear value. Uh [10:12] the other which is a very well-known [10:14] problem uh is is poor data quality as we [10:17] know garbage in garbage out. Um and I [10:21] think it's a it's a it's a well- [10:22] definfined issue with lot of previous [10:24] technology uh technologies also uh next [10:28] is the change management neglect as we [10:31] talked about um uh users want any [10:35] solution that you develop any any [10:38] approach that you're following any any [10:40] outcome that you want users want to use [10:42] it and adopt it. Uh key takeaway is [10:46] again technology is the easy part here. [10:50] Successful execution uh lives and dies [10:54] based on adoption, governance and [10:56] clarity of purpose. [11:01] I want to propose a [11:04] where where do you start? How how do you [11:06] begin? Um uh I want to propose a [11:10] three-step framework. first assess and [11:13] plan. Uh so this is all about um [11:18] auditing your skills uh basically [11:20] auditing auditing uh your the skills of [11:23] your uh enterprise uh about your data [11:26] your readiness. Next is pilot and learn. [11:30] Uh start with focus pilots, collect [11:33] relevant metrics and and and the third [11:36] is scale and go one. Right? So this is [11:37] really expand by embedding governance uh [11:41] as part of your imple AI implementation [11:44] build the expertise. So such a phased [11:47] step-by-step approach will not only [11:49] accelerate win but also reduce the [11:52] operational and political risk that you [11:55] might face. [11:56] So how how do we define success? How do [11:58] we measure value? [12:02] One of the key things uh that I [12:04] mentioned in the phased approach uh uh [12:06] previously was of measuring value, [12:09] right? What gets measured always gets [12:11] funded. So in in terms of specific [12:13] metrics there are different kinds of [12:15] ones like financial, operational, [12:18] customer related, workforce related and [12:21] some examples u I have given as part of [12:24] that uh and should be should look very [12:26] familiar to all of you but but I want to [12:29] propose that we want to go beyond this [12:31] usual metrics. I want to propose a [12:33] three- tiered value framework uh to [12:36] think about measuring uh the value [12:38] measuring the ROI. uh so one is [12:41] obviously the the metrics that we just [12:44] looked at but along with that I want to [12:46] propose uh a strategic ROI which is [12:49] measuring really your enterprise [12:51] competitiveness and then the most [12:54] important which I feel is often get [12:56] ignored is the capability ROI. So this [12:59] is really assessing the AI maturity of [13:02] your organization as you move forward. [13:04] Uh so it it is really not just about the [13:08] dollar value u and going beyond just [13:11] measuring dollars uh savings but it is [13:14] the really this combination of the [13:16] financial aspects along with the [13:18] strategic cultural impact uh that you [13:22] want to sustain AI within your [13:25] organization. So bringing all these [13:28] pieces together [13:31] uh just summarizing uh scaling for the [13:35] enterprise uh as I mentioned uh what it [13:40] takes to scale is for lasting u a [13:44] lasting enterprise AI capability is it [13:47] is alignment across this four pillars. [13:49] Uh so we talked about people process [13:51] platform uh in in the in the in in in [13:54] the previous slides. Um so people it is [13:57] all about [13:59] uh training change management process is [14:02] all about your governor's frameworks [14:04] having clear KPIs feedback loops [14:07] platform is all about your pipelines [14:10] your infrastructure your tools systems [14:12] and and also the future readiness um uh [14:16] like agent AI preparation your [14:19] compliance ethics framework etc. [14:22] scale is going to come uh by building AI [14:25] into the fabric of the enterprise by [14:27] following this uh fourpillar approach. [14:36] Now, [14:40] uh if you were to start now, here is how [14:43] I would propose a 90-day action plan for [14:46] you. So week one two is uh the readiness [14:50] assessment. [14:54] All right, [14:56] let me talk about that. Um so it is it [14:59] is really as we talked about right [15:01] assessing the skills uh assessing the [15:04] gaps uh in those uh your data analysis [15:08] your infrastructure system report [15:11] review. Uh the next is month one uh [15:14] identifying and prioritizing use cases. [15:16] Uh as I mentioned in the in the previous [15:18] slides I talked about uh some of the uh [15:22] the the ways in which you can identify [15:24] the use cases the low hanging fruits. So [15:27] really creating that use case ma matrix [15:30] and finding that sweet spot between low [15:32] effort high value uh g get finding out [15:36] the ROIs and and figuring out uh ways to [15:39] prioritize those use cases and and then [15:42] month two and three is launching your [15:43] first pilot project. So this is all [15:45] about your MVP development performance [15:48] metrics uh user training etc. [15:54] Th this kind of plan is something that [15:57] you can take away uh as part of this [15:59] session. Uh it is short enough uh to get [16:02] some early wins but also long enough to [16:05] build the momentum. Uh as I talk about [16:09] this uh certain use cases that probably [16:14] many of you might have heard which have [16:16] been successful within the enterprise. [16:17] These are not the the hyped use cases [16:20] but the actual ones that have been [16:21] successful within many of the [16:23] enterprises uh which I can give examples [16:26] are uh you obviously the coding [16:28] assistant is a is a proven one. Uh [16:31] second is uh the uh kind of the the the [16:37] your assistant uh for searching for [16:40] information uh the enterprise [16:42] information uh like while attending a [16:45] customer call or while uh answering a [16:48] query from your customer uh etc. uh the [16:52] that kind of use case and and then the [16:55] third is the marketing generating [16:57] marketing content use case. Uh [17:01] let me leave you with one final thought [17:03] here. The AI mandate is here. Uh the [17:07] rewards uh will be enormous and so are [17:11] the penalties if there are any missteps. [17:15] the path to be part of that 20% of [17:17] enterprise uh enterprises uh who are [17:20] going to immensely benefit uh is to [17:24] unlock the real AI value and and [17:26] unlocking the real AI value is to take [17:31] series of deliberate and wellexecuted [17:34] steps. I hope my talk has given you the [17:37] framework. Uh I hope that it has given [17:40] you the plan and now is the time to [17:42] execute. [17:45] Thank you.