Advisor Labs Finds 75% of Enterprise AI Fails to Generate Returns

What Works with AI and What to Avoid: Our Lessons Learned Distilled to 20 Mins With Advisor Labs — Ai4

Ai4

Corporate AI initiatives fall short when companies chase software novelty over operational friction, leaving outsized financial returns solely to organizations that build predictive models on proprietary internal data.

AI-assisted summary of What Works with AI and What to Avoid: Our Lessons Learned Distilled to 20 Mins With Advisor Labs — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by Ai4 on August 28, 2025, this presentation features Chris Weidemann, CAIO and Co-Founder of Advisor Labs. The session outlines practical enterprise strategies to help organizations overcome the 75% failure rate of corporate AI initiatives and build iterative roadmaps that generate measurable financial returns.

## TL;DR Chris Weidemann examines the primary friction points causing most enterprise AI projects to stall, including ambiguous ROI metrics, legacy system integration bottlenecks, and novelty-driven tool adoption. He recommends a two-tiered implementation strategy that combines immediate back-office process automation with long-term predictive modeling built around unique proprietary data. To systematically evaluate opportunities, Weidemann presents a 90-day framework focused on cross-functional alignment, dynamic backlog management, and rapid, low-cost proof-of-concept validation. The talk concludes by highlighting that realizing transformative AI value requires top-down executive leadership to actively manage cascading operational changes across interconnected departments.

## Atmosphere & style The presentation has the feel of a pragmatic, fast-paced executive briefing that cuts through speculative generative AI hype with clear-headed engineering realism. Weidemann delivers his insights with steady, consultative authority, anchoring high-level corporate anxieties in concrete historical analogies and relatable enterprise pitfalls. The delivery is structured and solution-oriented, leaving listeners with a sense of methodical clarity rather than technical overwhelm.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=4s) Predictive behavioral modeling, exemplified by Disney using RFID MagicBands to dynamically route foot traffic and boost spending, illustrates how AI creates value by optimizing core business dynamics. - [03:02](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=182s) Only about 25% of enterprise AI initiatives generate returns that exceed their development costs, and merely 10% produce significant financial value. - [03:02](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=182s) The primary friction points stalling enterprise AI are ambiguous ROI metrics, conflicting IT stack priorities, difficult legacy integrations, and misaligned model outputs. - [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) The unprecedented adoption speed of generative AI (ChatGPT reaching 100 million users in two months) severely outpaces typical enterprise readiness and change management cycles. - [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) Widespread organizational deficits hinder AI progress, with only 15% to 16% of executives having deep AI knowledge and over 60% of enterprises lacking dedicated AI personnel. - [06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) Organizations must avoid three common anti-patterns: pursuing tech for novelty without a clear problem, assuming adoption will happen without human-centered workflow design, and trying to 'boil the ocean' all at once. - [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) Successful implementations begin by anchoring initiatives to clear business outcomes (such as cost reduction, risk mitigation, or compliance) and directly surveying frontline staff about operational pain points. - [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) Initial low-hanging fruit and fast efficiency wins are typically found by integrating AI into existing back-office RPA, BPM, and intelligent automation workflows. - [10:41](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=641s) Entering the top 10% of financial value creators requires auditing proprietary data assets that competitors cannot duplicate and training custom predictive models on core business differentiators. - [11:50](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=710s) A structured 90-day framework should establish an impact-versus-feasibility scoring system and align cross-functional stakeholders across business, sales, HR, and operations into an active 1-to-2-year backlog. - [11:50](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=710s) Prior to heavy custom software development, teams should run rapid, low-cost proof-of-concept experiments using accessible, off-the-shelf tooling to validate feasibility. - [14:49](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=889s) Executive leadership is crucial for anticipating and managing cascading downstream organizational ripples across interconnected departments whenever an AI implementation alters a workflow.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=4s) The Disney MagicBand Revelation and the Genesis of Advisor Labs

Chris Weidemann opens with an anecdote tracing his initial fascination with machine learning back to around 2010 during a visit to Walt Disney World. Observing the ubiquitous RFID wristbands guests wore, he initially recognized them as convenient tools for unlocking hotel rooms, boarding cruise ships, and making purchases. However, he soon realized they were actively tracking guest movements throughout the park. In 2012, a technical white paper confirmed his suspicions: Disney had engineered predictive models to dynamically manage park traffic. When a specific ride, restaurant, or retail shop experienced low foot traffic, the system nudged guest flow toward those underutilized assets to drive spending. Fascinated by this strategic manipulation of data and behavioral technology, Weidemann pivoted into machine learning, exploring Twitter data pipelines before spending the next 12 to 13 years immersed in artificial intelligence.

For over 18 years, Weidemann specialized in architecting bespoke, custom enterprise software for organizations that had outgrown off-the-shelf platforms. In 2022, he co-founded Advisor Labs to dedicate all consulting and engineering efforts strictly to enterprise AI implementation. Having delivered solutions across financial services, construction, healthcare, and higher education, the firm observed distinct patterns separating companies succeeding with AI from those stumbling into costly dead ends.

- **2010 Disney Park Insight**: RFID wristbands utilized beyond access control to track real-time human movement. - **2012 Disney White Paper**: Formal confirmation of algorithmic crowd routing models designed to boost retail and attraction revenue. - **Industry Footprint**: Custom enterprise software delivery across financial services, healthcare, construction, and higher education.

---

[03:02](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=182s) The Value Gap: Why Most AI Projects Fail to Deliver ROI

Distilling industry surveys and client feedback, Weidemann highlights a stark reality regarding corporate AI adoption: approximately 25% of enterprise AI initiatives generate more value than their cost of development, but only 10% achieve significant financial value. When organizations evaluate why their investments fall short or stall out, four universal friction points emerge:

```mermaid flowchart TD accTitle: Four Major Friction Points in Enterprise AI Implementation subgraph Hurdles["Enterprise AI Bottlenecks"] A["Unclear ROI Expectations"] B["Stack Prioritization Conflict"] C["Legacy System Integration"] D["Model Output Disconnect"] end Hurdles --> Outcome["75% of Projects Fail to Return Value"] ```

1. **Unclear ROI Measurement**: Because companies are attempting these deployments for the first time, baseline financial expectations and return horizons remain ambiguous. 2. **Prioritization Paralysis**: Leadership struggles to balance speculative AI roadmaps against mandatory day-to-day IT stack maintenance. 3. **Legacy Integration Friction**: While connecting AI to legacy core systems offers immense potential, technical integration into antiquated infrastructure remains exceptionally difficult. 4. **Model Output Mismatch**: Teams invest substantial capital without understanding model mechanics, resulting in production outputs that fail business requirements.

---

[04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) Unprecedented Velocity and Organizational Constraints

The core catalyst behind widespread corporate anxiety is the unprecedented speed of AI adoption compared to prior technology shifts. Historical technology transformations allowed organizations years to adapt, whereas generative AI reached mass market saturation almost instantaneously.

| Technology | Time to Reach User Milestones | |:--- |:--- | | **Telephone** | 50 years to reach 50 million users | | **Personal Computer** | 14 years to reach 50 million users | | **Internet** | 7 years to reach 50 million users | | **ChatGPT** | 2 months to reach 100 million users |

This velocity magnifies internal operational deficits. Only 15% to 16% of executive leaders report possessing AI knowledge beyond what is reported in mainstream local news. Furthermore, over 60% of enterprises report an inability to hire dedicated AI talent or lack even a single employee focused on the discipline. Compounding this challenge, the vendor landscape has devolved into chaotic rebranding: legacy software providers and traditional CRM vendors have suddenly repositioned themselves as "AI platforms," leaving buyers disoriented.

---

[06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) Three Anti-Patterns to Avoid: What Not to Do

When Advisor Labs steps in to remediate stalled enterprise projects, the underlying failure modes invariably map to three fundamental mistakes:

```mermaid flowchart LR accTitle: Three Major Anti-Patterns in AI Adoption subgraph AntiPatterns["Anti-Patterns to Avoid"] direction TB P1["1. Solution Looking for a Problem<br/>'Buying tech for novelty'"] P2["2. Ignoring the Human Element<br/>'Build it and they will come'"] P3["3. Boiling the Ocean<br/>'Trying to fix everything at once'"] end ```

- **Chasing Tech for Tech's Sake**: Purchasing tools purely out of novelty—such as demanding ChatGPT integration without defining a specific business problem or measurable outcome. - **Ignoring the Human Side**: Relying on the flawed assumption that "if you build it, they will come." If the tool does not integrate smoothly into frontline workflows or address user needs, adoption collapses regardless of technical sophistication. - **Boiling the Ocean**: Feeling behind the adoption curve and attempting massive, all-encompassing overhauls simultaneously rather than systematically executing high-impact, prioritized opportunities.

---

[08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) Strategic Execution: Outcome-First, People, and Process Automation

To achieve sustainable returns, organizations must flip their implementation strategy by anchoring every project to defined business outcomes before selecting tools. Teams must isolate the operational symptoms affecting specific departments and quantify the intended business impact—whether that means reducing overhead costs, mitigating operational risk, enhancing regulatory compliance, or lowering staff turnover.

```mermaid flowchart TD accTitle: Value Capture Strategy for AI Implementation subgraph Strategy["Two-Tiered Value Strategy"] direction TB subgraph Tier1["Low-Hanging Fruit: Process Layer"] P["Back-Office Processes"] P --> A1["RPA & BPM Integration"] P --> A2["Intelligent Automation"] end subgraph Tier2["High-Value Transformation: Core Layer"] D["Unique Proprietary Data"] D --> B1["Predictive Modeling"] D --> B2["Competitive Moats"] end end ```

Simultaneously, leadership must prioritize employee engagement. Frontline staff should be surveyed directly about their current operational friction, existing "shadow AI" productivity wins, and areas where tooling could assist them.

For initial implementations, low-hanging fruit almost always resides within back-office business processes. Integrating modern AI capabilities into existing investments in Robotic Process Automation (RPA), Business Process Management (BPM), and intelligent automation workflows delivers predictable, immediate efficiency.

---

[10:41](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=641s) Unlocking Significant Financial ROI via Unique Business Problems

While back-office automation secures quick procedural wins, breaking into the elite 10% of enterprises capturing major financial value requires tackling proprietary, high-leverage business problems. Returning to the Disney example, their outsized ROI emerged because they focused modeling directly on their unique core business dynamic: maximizing guest monetization within physical park boundaries.

To emulate this success, companies must audit their proprietary data assets—information competitors cannot replicate—and explore predictive modeling around their core business differentiators. Significant financial returns stem from utilizing custom machine learning models to extract intelligence, automate key domain-specific decisions, and monetize internal data reserves.

---

[11:50](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=710s) The 90-Day Implementation Framework and Low-Cost Validation

Weidemann outlines an actionable 90-day framework designed to systematically structure an enterprise AI roadmap:

```mermaid flowchart TD accTitle: 90-Day Enterprise AI Implementation Roadmap subgraph Phase1["Step 1: Ideation & Scoring"] A["Draft Initial Opportunity List"] --> B["Score: Impact vs. Feasibility"] end subgraph Phase2["Step 2: Cross-Functional Alignment"] B --> C["Convene Cross-Functional Workshop<br/>(HR, Sales, Operations, Business)"] C --> D["Synthesize Dynamic 1-2 Year Backlog"] end subgraph Phase3["Step 3: Validation & Core Strategy"] D --> E["Run Rapid No/Low-Cost Experiments<br/>(Top 5-10 Opportunities)"] D --> F["Isolate Proprietary Data Opportunities"] end ```

1. **Ideation and Initial Prioritization**: A small leadership group drafts an initial opportunity list, scoring each initiative on business impact versus technical feasibility. 2. **Cross-Functional Collaboration**: Bring together a diverse stakeholder group across business, sales, human resources, and operations. Review the initial list, gather missing pain points, and synthesize the findings into an active 1-to-2-year backlog that is continuously pruned. 3. **Rapid, Low-Cost Experimentation**: Select the top 5 to 10 backlog items and run validation proofs-of-concept before allocating heavy engineering budgets. - *Case in Point*: An educational client wanted AI-generated lesson plans. Rather than building an expensive custom prototype immediately, Advisor Labs guided them to test the workflow using Google's NotebookLM with their source curriculum—validating feasibility for virtually zero software cost. 4. **Identify Proprietary Moats**: Curate a dedicated stream of unique business opportunities centered on proprietary internal data that competitors cannot easily duplicate.

---

[14:49](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=889s) Leadership Governance and Managing Downstream Organizational Ripples

Weidemann concludes by emphasizing the necessity of executive leadership in navigating enterprise AI adoption, referencing a key industry principle:

> "Harnessing machine learning can be transformational, but for it to be successful, enterprise needs leadership from the top. This means understanding that when machine learning changes one part of the business, the other parts must change."

Introducing AI tools creates cascading downstream ripples across interconnected departments, operational backlogs, and individual employee workflows. Leaders must remain proactive in anticipating these systemic shifts, ensuring that tooling enhancements in one domain do not create friction in another. Weidemann closes by inviting attendees to connect with the Advisor Labs team at booth 357 in the exhibit hall.

## Notable quotes > [03:02](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=182s) approximately 25% of enterprise AI initiatives generate more value than their cost of development, but only 10% achieve significant financial value. — Chris Weidemann

> [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) Only 15% to 16% of executive leaders report possessing AI knowledge beyond what is reported in mainstream local news. — Chris Weidemann

> [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) over 60% of enterprises report an inability to hire dedicated AI talent or lack even a single employee focused on the discipline. — Chris Weidemann

> [06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) if you build it, they will come. — Chris Weidemann

> [06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) Boiling the Ocean: Feeling behind the adoption curve and attempting massive, all-encompassing overhauls simultaneously rather than systematically executing high-impact, prioritized opportunities. — Chris Weidemann

> [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) For initial implementations, low-hanging fruit almost always resides within back-office business processes. — Chris Weidemann

> [10:41](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=641s) breaking into the elite 10% of enterprises capturing major financial value requires tackling proprietary, high-leverage business problems. — Chris Weidemann

> [14:49](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=889s) Harnessing machine learning can be transformational, but for it to be successful, enterprise needs leadership from the top. This means understanding that when machine learning changes one part of the business, the other parts must change. — Chris Weidemann

## Mentioned Chris Weidemann, Advisor Labs, Walt Disney World, Disney MagicBand, Twitter, ChatGPT, Google NotebookLM, Ai4

## Video description Chris Weidemann, CAIO & Co-Founder, Advisor Labs

Abstracts

Are you part of the 75% of businesses chasing AI without seeing any real returns? It's a shared struggle across companies: constant pushes to jump on AI without clear direction, tough legacy integrations, and chasing that slippery ROI. We have compiled our hard won lessons learned so you can nail it post conference. In just 20 minutes, you'll accelerate through these lessons and pick up the essentials for learning which projects are really worth it, what to avoid, and iterating roadmaps that deliver value. Attend to walk away ready to join the top 25% and start seeing true AI value back at work.

[Source](https://www.youtube.com/watch?v=DlRsWRu5Pwg)

中文简报

Advisor Labs 发现 75% 的企业 AI 未能产生回报 当企业追逐软件的新奇感而非解决运营阻力时，其 AI 项目往往难达预期；超额的财务回报仅属于那些基于专有内部数据构建预测模型的机构。 Chris Weidemann 深入探讨了导致多数企业 AI 项目停滞的核心阻力，包括模糊的 ROI 指标、传统系统集成的瓶颈，以及盲目追求工具新奇感的使用倾向。他建议采用分层的两步实施策略，将即期的后台流程自动化与依托独特专有数据构建的长期预测模型相结合。为了系统化评估落地机会，Weidemann 提出了一套 90 天框架，重点关注跨职能协同、动态积压任务管理以及快速、低成本的概念验证。他在演讲结尾强调，要实现具变革性的 AI 价值，需要高管团队自上而下发挥领导作用，积极统筹和管理各互联部门之间连锁发生的运营变革。 约25%的企业AI项目能收回开发成本，而仅有10%能创造显著的财务价值；导致AI项目停滞的主要阻碍是ROI指标模糊、IT优先级冲突、传统系统集成困难以及模型输出脱节。 (3:02) 生成式AI的爆发速度极大地超出了企业的变革准备能力，目前仅有15%到16%的高管具备深入的AI知识，超过60%的企业缺少专职AI人才。 (4:47) 推进AI时必须避开三大反模式：为了尝鲜技术而脱离具体业务问题、忽视以人为本的工作流设计、以及试图一次性全盘改造所有业务。 (6:35) 项目落地应以降低成本、控制风险或合规等明确业务目标为锚点，早期收益通常来自于将AI与现有的RPA、BPM及智能自动化后台流程相结合。 (8:15) 企业可通过90天框架梳理AI路线图，利用低成本或零成本工具快速验证前5到10个高潜力场景，并通过挖掘自身专有数据构建核心竞争壁垒。 (11:50)

Reading summary

Drawing on enterprise AI deployments across financial services, construction, healthcare, and higher education, Advisor Labs co-founder Chris Weidemann argues that most corporate AI projects fail because they prioritize novelty over defined business outcomes. To join the 10% of organizations generating significant financial returns, enterprises must combine disciplined back-office automation with predictive modeling built on proprietary data. The ROI Gap and Adoption Velocity (3:02) According to industry surveys and client feedback cited by Weidemann, roughly 75% of enterprise AI initiatives fail to generate enough value to cover their development costs, while only 10% deliver significant financial returns. He attributes these stalled investments to four recurring bottlenecks: ambiguous ROI baselines for first-time deployments, difficulty balancing speculative AI work against daily IT maintenance, the technical complexity of integrating AI into legacy core systems, and a lack of understanding of model mechanics that leads to unusable outputs. These technical hurdles are compounded by the speed of generative AI adoption. While the telephone, personal computer, and internet took 50, 14, and 7 years respectively to reach 50 million users, ChatGPT reached 100 million users in just two months. This rapid saturation has outpaced internal readiness: Weidemann notes that only 15% to 16% of executives report AI knowledge beyond mainstream news coverage, and over 60% of enterprises lack dedicated AI talent. Meanwhile, traditional software and CRM vendors have aggressively rebranded as "AI platforms," further confusing corporate buyers. Three Implementation Anti-Patterns to Avoid (6:35) When remediating stalled projects, Advisor Labs consistently encounters three failure modes. First, organizations purchase technology for its novelty—such as demanding ChatGPT integration—without first defining a measurable problem. Second, teams neglect the human element, assuming employees will automatically adopt sophisticated tools even if they disrupt frontline workflows. Third, companies that feel behind the curve attempt to "boil the ocean" with sweeping overhauls rather than executing prioritized opportunities. To counter these mistakes, Weidemann advises anchoring every project to quantifiable business impacts, such as reducing overhead, mitigating operational risk, improving regulatory compliance, or lowering staff turnover. Leaders should also survey frontline employees directly to uncover operational friction and identify existing "shadow AI" productivity practices before selecting tools. Capturing Value: Process Automation vs. Proprietary Moats (8:15) Weidemann outlines a two-tiered strategy for capturing returns. For initial deployments, the most reliable low-hanging fruit lies in back-office processes, where modern AI can enhance existing Robotic Process Automation (RPA) and Business Process Management (BPM) workflows for immediate efficiency gains. However, breaking into the top 10% of financial performers requires applying custom machine learning to proprietary data and core business differentiators. Weidemann illustrates this with Walt Disney World's RFID MagicBands, which he observed in 2010 and saw confirmed in a 2012 technical white paper. Beyond serving as room keys and payment tools, the wristbands tracked real-time guest movements, feeding predictive models that dynamically routed crowds toward underutilized rides, restaurants, and shops. By modeling a core business driver—guest monetization inside physical parks—using data competitors could not replicate, Disney achieved substantial returns. A 90-Day Framework for Low-Cost Validation (11:50) To operationalize these lessons, Weidemann proposes a 90-day roadmap. First, a small leadership group drafts an initial opportunity list and scores each idea on business impact versus technical feasibility. Second, a cross-functional workshop spanning sales, HR, operations, and business units refines these pain points into a dynamic one-to-two-year backlog. Third, before committing heavy engineering budgets, teams validate the top five to ten backlog items through rapid, low-cost experiments—while maintaining a parallel track for proprietary data initiatives. For example, when an educational client wanted custom AI software to generate lesson plans, Advisor Labs first had them test the workflow using their existing curriculum inside Google's NotebookLM, proving feasibility at virtually zero software cost. Ultimately, Weidemann stresses that technical validation must be paired with top-down executive governance. Because introducing machine learning into one department inevitably creates downstream ripples across connected workflows and backlogs, leadership must proactively manage organizational change so that efficiency gains in one area do not create bottlenecks in another.

精炼阅读摘要

Advisor Labs 联合创始人 Chris Weidemann 指出，企业在推进人工智能落地时普遍面临投资回报率低和组织能力匮乏的挑战；企业唯有摒弃盲目追求工具新奇感的误区，以可量化的业务成果为导向，结合低成本概念验证与专有数据资产，才能从低阶的后台流程优化跃升至高价值的核心商业突破。 投资回报差距与落地的四大核心阻力 (3:02) Advisor Labs 汇总行业调查与客户实践指出，企业在人工智能投入上存在显著的“价值鸿沟”：仅有约 25% 的企业 AI 项目创造的实际价值能超过其研发成本，而实现重大财务回报的比例仅占 10%。这一数据表明，绝大多数企业的 AI 投入未能转化成切实可观的利润增量。 Weidemann 将阻碍项目推进并拉低投资回报率的原因归纳为四大摩擦点：一是投资回报率预期模糊，由于多数企业属于首次尝试此类技术落地，缺乏清晰的财务基准与回报周期设定；二是技术栈优先级冲突，管理层难以在具有投机色彩的长期 AI 规划与必须持续维护的日常 IT 系统之间取得平衡；三是与陈旧老旧系统集成的技术阻力极其严重，尽管连接核心数据资产潜力巨大，但打通底层陈旧架构难度高昂；四是模型产出与业务需求脱节，团队在未充分理解模型运作机理的情况下投入重金，导致上线后的实际输出无法满足真实业务场景。 技术爆发速度与组织能力的结构性错配 (4:47) 生成式 AI 的爆发速度打破了以往技术变革的演进规律，成为加剧企业集体焦虑的催化剂。在用户普及速度的对比中，电话耗时 50 年触达 5000 万用户，个人电脑耗费 14 年，互联网耗时 7 年，而 ChatGPT 仅用 2 个月便突破了 1 亿用户大关。这种瞬时饱和的渗透节奏剥夺了企业按部就班吸收消化新技术的时间窗口。 这种速度反衬出企业内部的严重能力短板。据调查，仅有 15% 至 16% 的企业高管表示自己对 AI 的了解超越了主流大众新闻的范畴。同时，超过 60% 的企业无法招募到专职 AI 人才，甚至没有任何一名员工专注于该领域。此外，供应商生态的概念炒作进一步扰乱了决策视野，大量传统软件与 CRM 厂商纷纷换壳改称为“AI 平台”，使本就缺乏辨别能力的企业选型过程愈发混乱。 规避三大实施反模式与成果优先原则 (6:35) Advisor Labs 在协助企业修复陷入停滞的 AI 项目时，发现失败案例高度集中于三类反模式：第一是“技术至上而脱离场景”，即单纯出于新奇感采购工具，例如在未界定明确业务痛点与衡量标准的情况下执意接入 ChatGPT；第二是“忽视人的要素”，盲目假设只要系统搭建完毕员工就会主动使用，若工具无法无缝融入一线现有工作流，无论技术多先进最终都会被弃用；第三是“妄图煮沸整片海洋”，由于担心落后而盲目铺开全盘重构，缺乏分阶段针对高影响机会的精准聚焦。 为纠正这些偏差，企业必须执行“成果优先”的策略转向。在选定工具前，团队必须精准定位特定业务部门的痛点症状，并严格量化预期的业务成效，如削减管理间接成本、降低运营与合规风险，或减少人员流动率。同时，管理层必须主动开展员工调研，摸排一线实际阻力与已有的非正式 AI 生产力习惯，确保技术服务于人而非增加使用负担。 从后台自动化到专有数据的双层策略 (8:15) 为了在平衡风险的同时捕获价值，企业应采取分层的推进机制。第一层也是最稳妥的起步切入点在于后台业务流程。企业可以通过将现代 AI 能力无缝嫁接到既有的机器人流程自动化（RPA）、业务流程管理（BPM）和智能自动化系统之中，以较低风险迅速取得可预测且立竿见影的日常运营效率提升。 然而，要突破瓶颈跻身获取巨额财务回报的前 10% 企业阵营，关键在于攻克高杠杆、不可替代的核心业务命题。正如 Weidemann 回顾 Disney 的案例所示，Disney 在 2010 年观察到并于 2012 年白皮书确认利用 RFID 智能手环数据构建预测模型，动态调节园区人流以最大化提升餐饮与零售收益。企业应当全面审计竞争对手无法复制的专有内部数据资产，针对自身核心差异化优势构建定制化的机器学习预测模型，以此构筑难以逾越的竞争护城河。 90天敏捷实施框架与高管跨部门协同 (11:50) Weidemann 提出了一套标准化的 90 天实施框架以系统化落地 AI 规划：第一步由核心领导团队梳理初期机会，按业务影响与技术可行性两个维度进行量化评分；第二步召集销售、运营、人力资源和业务跨职能研讨会，补齐遗漏痛点，形成动态维护的 1 至 2 年项目储备池；第三步针对评分前 5 至 10 个项目开展快速且极低成本的概念验证，例如某教育客户在开发专属教案生成系统前，先使用 Google NotebookLM 结合自有教材课标进行闭环验证，在零额外软件投入下先行验证了技术可行性。 这一框架的成功离不开最高管理层的系统性全局治理。引入机器学习会引发跨部门的链式级联反应，直接波及上下游业务流程与员工操作习惯。领导层必须深刻认识到，当 AI 优化了业务的某一节点时，其余协同环节也必须作出配套变革，唯有主动预判并化解系统性震荡，才能确保工具的局部效能不演化为整体组织的摩擦内耗。 Advisor Labs 的核心论点在于，企业在 AI 浪潮中突围的前提是将注意力从追逐工具热度转移到深挖专有业务问题上。若缺乏清晰的财务度量、跨部门协同治理以及对自有独特数据的深度运用，单点技术的狂热投入将极难跨越从流程自动化到重大财务价值创造之间的鸿沟。

Detailed notes — zh

## Background 该内容由 Ai4 官方频道于 2025 年 8 月 28 日在 YouTube 平台发布，主讲人为 Advisor Labs 联合创始人兼首席 AI 官（CAIO）Chris Weidemann。视频简介指出，当前有 75% 的企业在缺乏明确方向和受困于旧系统集成的背景下追逐 AI 却难以见效，本场演讲旨在分享一线实战经验，帮助企业辨别真正值得投入的项目以跻身实现真实回报的行列。 ## TL;DR 演讲从迪士尼利用智能手环调度客流的启发切入，揭示了仅有 10% 的企业能通过 AI 创造重大财务价值的残酷现实，指出量化 ROI 难、旧系统集成受阻和认知断层是主要瓶颈。Weidemann 明确指出了盲目跟风炫技、无视员工体验和试图“煮干大海”三大雷区，主张以业务结果为先、聚焦人员与后台流程并深挖独有数据壁垒。最后，他提出了包含初筛排序、跨部门 Backlog 评审、低成本原型验证及专有优势深挖的 90 天四步实操框架，强调需以顶层领导力应对 AI 引发的组织连锁反应。 ## Atmosphere & style 整场分享节奏干练、务实且充满实操针对性，透露出讲者多年深耕企业定制软件研发与一线排雷的深厚经验。讲者以生动的叙事开篇，迅速剥离 AI 行业的炒作浮沫，用直击痛点的行业数据引发极强的现实共鸣。整体基调专业而清醒，通过清晰的逻辑框架和接地气的原型验证案例，将宏大的技术焦虑转化为可按部就班执行的落地路径。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=4s) Chris Weidemann 受迪士尼 MagicBand 手环通过实时位置追踪与数据模型调度游客分流的启发进入机器学习领域，并创立 Advisor Labs 专注于企业定制化 AI 落地。 - [03:02](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=182s) 当前企业应用 AI 面临严峻的 ROI 困局：仅有 25% 的企业项目收益能覆盖构建成本，仅 10% 能创造出重大财务价值。 - [03:02](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=182s) 阻碍企业 AI 成功的四大核心瓶颈在于：难以量化 ROI、缺乏优先级排序机制、传统旧系统集成困难，以及对模型能力理解不足导致产出不达预期。 - [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) 技术扩散速度呈现史无前例的指数级压缩，ChatGPT 仅用 2 个月即突破 1 亿用户，远快于历史上的电话、PC 和互联网。 - [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) 企业在 AI 爆发期普遍面临内生能力断层：仅约 15% 领导者具备充足认知，超 60% 企业缺乏全职 AI 专业人才，且面临鱼龙混杂的供应商生态干扰。 - [06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) 企业落地 AI 必须规避的三大雷区：盲目跟风追逐酷炫技术（Tech for Tech's Sake）、漠视终端员工习惯与体验、试图激进全盘押注而战线过长（Boiling the Ocean）。 - [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) 成功推进 AI 的首要原则是结果导向，必须先梳理降低成本、控制风险、提升合规或降低流失率等明确业务指标，再寻找技术匹配点。 - [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) 寻找 AI 落地的低垂果实应聚焦人员与后台既有流程，结合原有的 RPA 与 BPM 自动化管线以最低风险实现快速见效。 - [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) 若要迈向顶级 10% 的高额财务回报，企业必须盘点竞争对手无法获取的专有数据，围绕独特业务逻辑构建算法护城河。 - [11:50](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=710s) 企业应建立 90 天四步实操计划：从初筛可行性打分、跨部门评审动态待办池（Backlog），到低成本原型验证与深挖专有优势清单。 - [11:50](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=710s) 在投入重金自研前，应通过现有现成工具（例如利用 NotebookLM 进行课程教案生成验证）开展零成本或极低成本的概念验证。 - [14:49](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=889s) AI 落地并非孤立的技术替换，必须依赖顶层领导力管理其引发的下游系统级连锁反应（Ripple Effects）与全流程变革。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=4s) 迪士尼手环的启示与顾问实验室的创立背景 Chris Weidemann 讲述了自己踏入机器学习领域的戏剧性起点。大约在 2010 年，他在迪士尼世界度假区游玩时，注意到了几乎人手佩戴的 MagicBand 智能手环。经过调研，他发现这些手环不仅是用来刷开酒店房门、登上邮轮或进行免密支付的便捷工具，更是一套极其精密的追踪系统，实时监控着游客在园区内的移动轨迹。到了 2012 年，迪士尼发布了一篇相关白皮书，证实了他们开发的数据模型不仅是在记录轨迹，更是在主动“调度和分流人群”——当某个游乐项目、餐厅或商店客流不足时，模型便会通过各种机制引导游客涌入这些区域消费。这一将数据与技术转化为直接商业操纵的范式令他深感震撼，并促使他转向机器学习。 在经历迪士尼的启发以及随后在 Twitter 对机器学习算法的深入实践后，Weidemann 在过去十余年里深度参与了 AI/ML 的浪潮。他在过去的 18 年中一直专注于为大型企业开发定制化（Bespoke）软件，帮助那些无法仅靠通用 CRM 等现成工具满足需求的企业构建专属系统。2022 年左右，他与合伙人创立了 Advisor Labs，将多年定制开发的经验与 AI 技术全面融合，专注于协助金融、建筑、医疗保健及高等教育等各行业落地实施 AI 解决方案。 --- ### [03:02](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=182s) 企业 AI 落地的残酷现实：ROI 迷局与四大核心阻碍 根据对行业调查和客户反馈的提炼，当前企业在应用 AI 时普遍面临极低的回报转化率： - **25% 的企业**：其实施的 AI 项目所产生的价值能够覆盖其构建成本。 - **仅有 10% 的企业**：能够通过 AI 创造出“重大财务价值（Significant Financial Value）”。 在探究阻碍企业成功的核心瓶颈时，受访企业普遍指出了四大核心痛点： 1. **难以量化与理解 ROI**：由于大部分企业此前从未涉足过此类项目，完全不知道合理的投资回报基准是什么，更无从预估收益。 2. **缺乏优先级排序机制**：在日常 IT 基础设施和既有繁重业务的夹击下，团队不知如何将 AI 新技术合理融入现有的技术栈中。 3. **传统旧系统（Legacy Systems）集成极为艰难**：虽然 AI 最具潜力的场景之一正是与存量系统打通，但复杂的系统架构使集成成本和难度居高不下。 4. **模型理解不足与输出不达预期**：许多企业盲目投资，事前未花时间透彻理解所采购或构建的模型能力，最终产出的结果与业务预期严重脱节，导致前期的重金投入打了水漂。 --- ### [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) 史无前例的技术扩散速度与企业内生能力断层 AI 之所以带来如此强烈的焦虑与混乱，其根本原因在于技术渗透与普及的速度彻底颠覆了历史常态。从用户规模增长到 5000 万人所耗费的时间维度来看，技术演进呈现出指数级压缩： | 技术媒介 | 达到 5000 万用户所需时间 | 突破 1 亿用户所需时间 | |:--- |:--- |:--- | | **电话 (Telephone)** | 50 年 | — | | **个人电脑 (Computers)** | 14 年 | — | | **互联网 (Internet)** | 7 年 | — | | **ChatGPT** | — | 仅用时 2 个月（1 亿用户） | 这种超高速的变革给组织带来了极大的压力与挑战，而企业的内部准备度却严重滞后： - **管理层认知匮乏**：仅有 15% 至 16% 的企业领导者表示自己对 AI 的认知超过了“地方新闻播报的普及水平”。 - **专业人才极度短缺**：超过 60% 的企业表示无法招募到合适的人才，甚至组织内连一名专门负责 AI 的全职人员都没有。 - **供应商生态鱼龙混杂**：一夜之间所有软件服务商都自称是“AI 供应商”，原本的 CRM 供应商突然摇身一变成为 AI 架构顾问，给企业的技术选型制造了巨大干扰。 --- ### [06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) 企业落地 AI 必须规避的三大雷区（Don'ts） 基于 Advisor Labs 进场为大量遭遇失败的企业“救火”的经历，Weidemann 总结了三个最具代表性的错误做法： 1. **盲目追逐技术本身（Tech for Tech's Sake）**：许多企业决策者会说“我就是想要 ChatGPT”或“我觉得这个技术很酷炫”，完全脱离了“发现问题并提供解决方案”的商业本质，纯粹为了跟风而采购。 2. **漠视人性与使用者体验（Ignoring the Human Side）**：采购了顶级技术却不花心思去理解终端员工的实际工作流、习惯与真实诉求。“如果你把它造出来，人们自然会来用（If you build it, they will come）”的传统软件开发思维在 AI 领域完全失效；如果员工抗拒或用不顺手，再高价值的工具也会被束之高阁。 3. **试图“煮干整片大海”（Boiling the Ocean）**：许多企业因为落后焦虑，急于迎头赶上，不再按部就班寻找低垂的果实（Low-hanging fruit），而是采取“全盘押注、多线齐发”的激进做法，最终导致项目战线过长而全盘崩溃。 --- ### [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) 成功落地 AI 的三大核心原则（Do's） 为了跻身真正创造重大价值的 10% 顶尖企业行列，必须回归商业逻辑本身： ```mermaid flowchart TD accTitle: AI 实施的核心成功路径 subgraph S1["1. 结果与目标导向"] A["明确业务痛点与症状"] --> B["确立核心指标<br>(降本 / 控险 / 合规 / 减员流失)"] end subgraph S2["2. 聚焦人员与流程"] C["倾听一线员工诉求与经验"] --> D["自动化存量后台流程<br>(RPA / BPM / 智能自动化)"] end subgraph S3["3. 攻克专属核心难题"] E["盘点独有专有数据"] --> F["构建预测与业务洞察模型<br>(迈向顶级 10% 财务回报)"] end S1 --> S2 --> S3 ``` - **目标与结果先行（Think About Your Outcome First）**：首先坐下来与团队梳理每天遇到的真实业务问题，明确预期产出是降低成本、降低运营风险、提升合规性，还是降低内部部门的人员流失率，然后再去寻找技术与 AI 的匹配点。 - **聚焦人员与既有流程（Focus on People and Processes）**： - 先去向基层员工提问：“你们现在最需要什么？你们自己已经用 AI 取得了哪些小成果？你们觉得可以如何借助这些工具？”员工往往能提供极具价值的落地方向。 - 业务中阻力最小的“低垂果实”通常隐藏在后台流程（Back-office processes）中。结合过去投资的 RPA（机器人流程自动化）和 BPM（业务流程管理），将 AI 融入现有自动化管线，能以最低风险实现快速见效。 - **挖掘自身独有的业务问题与数据壁垒（Unique Business Problems）**：优化通用流程只能拿到基础收益，若要创造非凡的财务价值，必须像迪士尼那样，认清自身业务的独特运行逻辑，盘点企业拥有而竞争对手完全没有的专有数据，围绕这部分数据迭代算法，进行深度预测与价值挖掘。 --- ### [11:50](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=710s) 90 天落地实操行动计划与低成本验证 Weidemann 为企业提供了一套可在 90 天内推进落地的四步法框架： ```mermaid flowchart LR accTitle: 企业 AI 落地 90 天四步实操框架 Step1["第 1 步: 初筛清单<br>基于影响与可行性排序"] --> Step2["第 2 步: 跨部门协作<br>动态清洗与完善 Backlog"] Step2 --> Step3["第 3 步: 低成本验证<br>挑选 Top 5-10 快速实验"] Step3 --> Step4["第 4 步: 独立专有清单<br>聚焦专有数据与独占机遇"] ``` 1. **拟定初筛清单与可行性排序**：由一到两名核心人员发起，列出 AI 可以在企业内创造价值的所有潜在场景，并依据“业务影响力（Impact）”和“当前实施能力/可行性（Ability to perform/build）”进行优先级打分。 2. **开展跨部门评审并建立动态待办池（Backlog）**：召集销售、HR、业务等跨职能团队，让他们审视该清单，收集一线反馈，剔除伪需求，并增补新想法。由此形成的 Backlog 将成为企业未来 1 至 2 年内动态维护、持续修剪的技术路线图（Roadmap）——明确哪些直接采购现有软件，哪些自研定制，哪些根本无需引入 AI。 3. **低成本/零成本原型实验（Experimentation & Validation）**：挑出 Backlog 中排名前 5 到 10 个用例开展小规模概念验证。Weidemann 分享了一个教育行业客户的真实案例：该机构希望利用 AI 自动生成课程教案，Advisor Labs 并未直接为其定制高昂的独立系统，而是建议他们直接将现有教学大纲和教案导入 Google 的 NotebookLM 进行交互提问，以此验证大模型是否具备按其逻辑生成高质量教案的能力。这种近乎零成本的快速实验能在投入大笔资金前精准验证可行性。 4. **单列并深挖独占性机遇（Unique Opportunities）**：在通用需求之外，专门建立一份“企业专有优势清单”，全面梳理竞品无法获取的私域数据和行业隐性知识，将其转化为决定性的竞争护城河。 --- ### [14:49](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=889s) 领导力推动与系统级连锁反应（Ripple Effects） 在总结分享时，Weidemann 引用了一段关于机器学习实施的核心洞察： > “利用机器学习可以带来颠覆性的变革，但要想取得成功，企业必须拥有来自顶层的领导力。这意味着必须深刻认识到：当机器学习改变了业务的一个环节时，其他所有环节也必须随之改变。” 构建 AI 系统绝非孤立的技术替代，Backlog 中每一个项目的落地都会在组织内部引发广泛的“下游涟漪效应（Downstream Ripples）”。企业领导者在推进技术落地的同时，必须充分预判并管理这种连锁反应对组织架构、上下游流程以及人员协作方式所带来的全方位冲击。最后，Weidemann 邀请现场参会者前往展厅的 357 号展位，与 Advisor Labs 团队进一步交流探讨。 ## Notable quotes > [03:02](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=182s) 25% 的企业其实施的 AI 项目所产生的价值能够覆盖其构建成本，仅有 10% 的企业能够通过 AI 创造出重大财务价值。 — Chris Weidemann > [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) 仅有 15% 至 16% 的企业领导者表示自己对 AI 的认知超过了地方新闻播报的普及水平。 — Chris Weidemann > [04:47](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=287s) 超过 60% 的企业表示无法招募到合适的人才，甚至组织内连一名专门负责 AI 的全职人员都没有。 — Chris Weidemann > [06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) 盲目追逐技术本身（Tech for Tech's Sake）：许多企业决策者完全脱离了‘发现问题并提供解决方案’的商业本质，纯粹为了跟风而采购。 — Chris Weidemann > [06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) ‘如果你把它造出来，人们自然会来用（If you build it, they will come）’的传统软件开发思维在 AI 领域完全失效；如果员工抗拒或用不顺手，再高价值的工具也会被束之高阁。 — Chris Weidemann > [06:35](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=395s) 试图‘煮干整片大海’（Boiling the Ocean）：许多企业因为落后焦虑，急于迎头赶上，不再按部就班寻找低垂的果实，而是采取‘全盘押注、多线齐发’的激进做法，最终导致项目战线过长而全盘崩溃。 — Chris Weidemann > [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) 目标与结果先行（Think About Your Outcome First）：首先坐下来与团队梳理每天遇到的真实业务问题，明确预期产出是降低成本、降低运营风险、提升合规性，还是降低内部部门的人员流失率，然后再去寻找技术与 AI 的匹配点。 — Chris Weidemann > [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) 业务中阻力最小的‘低垂果实’通常隐藏在后台流程中。结合过去投资的 RPA 和 BPM，将 AI 融入现有自动化管线，能以最低风险实现快速见效。 — Chris Weidemann > [08:15](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=495s) 若要创造非凡的财务价值，必须认清自身业务的独特运行逻辑，盘点企业拥有而竞争对手完全没有的专有数据，围绕这部分数据迭代算法，进行深度预测与价值挖掘。 — Chris Weidemann > [14:49](https://www.youtube.com/watch?v=DlRsWRu5Pwg&t=889s) 利用机器学习可以带来颠覆性的变革，但要想取得成功，企业必须拥有来自顶层的领导力。这意味着必须深刻认识到：当机器学习改变了业务的一个环节时，其他所有环节也必须随之改变。 — Chris Weidemann ## Mentioned Chris Weidemann, Advisor Labs, 迪士尼 (Disney), MagicBand, Twitter, ChatGPT, Google NotebookLM, RPA (机器人流程自动化), BPM (业务流程管理), Ai4 ## Video description Chris Weidemann, CAIO & Co-Founder, Advisor Labs Abstracts Are you part of the 75% of businesses chasing AI without seeing any real returns? It's a shared struggle across companies: constant pushes to jump on AI without clear direction, tough legacy integrations, and chasing that slippery ROI. We have compiled our hard won lessons learned so you can nail it post conference. In just 20 minutes, you'll accelerate through these lessons and pick up the essentials for learning which projects are really worth it, what to avoid, and iterating roadmaps that deliver value. Attend to walk away ready to join the top 25% and start seeing true AI value back at work. [Source](https://www.youtube.com/watch?v=DlRsWRu5Pwg)

Source quotes

[3:02] Chris Weidemann: 25% 的企业其实施的 AI 项目所产生的价值能够覆盖其构建成本，仅有 10% 的企业能够通过 AI 创造出重大财务价值。 [4:47] Chris Weidemann: 仅有 15% 至 16% 的企业领导者表示自己对 AI 的认知超过了地方新闻播报的普及水平。 [4:47] Chris Weidemann: 超过 60% 的企业表示无法招募到合适的人才，甚至组织内连一名专门负责 AI 的全职人员都没有。 [6:35] Chris Weidemann: 盲目追逐技术本身（Tech for Tech's Sake）：许多企业决策者完全脱离了‘发现问题并提供解决方案’的商业本质，纯粹为了跟风而采购。 [6:35] Chris Weidemann: ‘如果你把它造出来，人们自然会来用（If you build it, they will come）’的传统软件开发思维在 AI 领域完全失效；如果员工抗拒或用不顺手，再高价值的工具也会被束之高阁。 [6:35] Chris Weidemann: 试图‘煮干整片大海’（Boiling the Ocean）：许多企业因为落后焦虑，急于迎头赶上，不再按部就班寻找低垂的果实，而是采取‘全盘押注、多线齐发’的激进做法，最终导致项目战线过长而全盘崩溃。 [8:15] Chris Weidemann: 目标与结果先行（Think About Your Outcome First）：首先坐下来与团队梳理每天遇到的真实业务问题，明确预期产出是降低成本、降低运营风险、提升合规性，还是降低内部部门的人员流失率，然后再去寻找技术与 AI 的匹配点。 [8:15] Chris Weidemann: 业务中阻力最小的‘低垂果实’通常隐藏在后台流程中。结合过去投资的 RPA 和 BPM，将 AI 融入现有自动化管线，能以最低风险实现快速见效。 [8:15] Chris Weidemann: 若要创造非凡的财务价值，必须认清自身业务的独特运行逻辑，盘点企业拥有而竞争对手完全没有的专有数据，围绕这部分数据迭代算法，进行深度预测与价值挖掘。 [14:49] Chris Weidemann: 利用机器学习可以带来颠覆性的变革，但要想取得成功，企业必须拥有来自顶层的领导力。这意味着必须深刻认识到：当机器学习改变了业务的一个环节时，其他所有环节也必须随之改变。

Source quotes

[3:02] Chris Weidemann: approximately 25% of enterprise AI initiatives generate more value than their cost of development, but only 10% achieve significant financial value. [4:47] Chris Weidemann: Only 15% to 16% of executive leaders report possessing AI knowledge beyond what is reported in mainstream local news. [4:47] Chris Weidemann: over 60% of enterprises report an inability to hire dedicated AI talent or lack even a single employee focused on the discipline. [6:35] Chris Weidemann: if you build it, they will come. [6:35] Chris Weidemann: Boiling the Ocean: Feeling behind the adoption curve and attempting massive, all-encompassing overhauls simultaneously rather than systematically executing high-impact, prioritized opportunities. [8:15] Chris Weidemann: For initial implementations, low-hanging fruit almost always resides within back-office business processes. [10:41] Chris Weidemann: breaking into the elite 10% of enterprises capturing major financial value requires tackling proprietary, high-leverage business problems. [14:49] Chris Weidemann: Harnessing machine learning can be transformational, but for it to be successful, enterprise needs leadership from the top. This means understanding that when machine learning changes one part of the business, the other parts must change.

Original transcript

[0:04] A little bit about me, kind of why you'd [0:07] even want to sit and listen to me today. [0:10] Somewhere around 2010, [0:13] I was I was at a Disney park. I was in [0:15] Disney World. And I saw those wristbands [0:19] that everybody had on. And I started to [0:22] think about that and I started to do a [0:25] little bit of research and found out [0:27] that not only are those really nifty [0:29] ways for us to get into our hotel rooms [0:31] and cruise ships and pay for things, but [0:35] they were also tracking us. They were [0:38] watching us move through the park. And I [0:41] started to think about that a little bit [0:42] more and it got me interested in why [0:45] might they be doing that? [0:47] 2012 rolls around and a white paper [0:50] comes out about this very thing. And I [0:53] read that white paper and I get really [0:55] interested and come to find out they've [0:57] built models to move us through the [1:00] park. So when they had an attraction or [1:03] a restaurant or a store that wasn't [1:07] getting enough traffic, they had built a [1:09] model to find ways to push us into those [1:13] attractions, restaurants, or effectively [1:15] to spend our money. I thought this was a [1:17] really unique way to use information, to [1:20] use technology, and it started me down [1:23] the path of machine learning. After [1:27] Disney World and figuring out how they [1:29] can manipulate us, I started to look at [1:31] Twitter and that's where I first started [1:33] to get into this crazy world of machine [1:35] learning. And I've spent the last 12 13 [1:37] years now enjoying the path that we're [1:40] all on that's been accelerated over the [1:42] last few years of of AI and ML. [1:46] A little bit more about me. So I've been [1:48] building software for a long time. [1:51] Really last 18 years has been bespoke [1:54] software. So custom software for [1:56] businesses that are are big enough that [1:59] they need their own software. Maybe it's [2:01] not just go out and buy your own CRM or [2:03] something off the shelf, but it's it's [2:05] time to build your own. So I've been [2:07] helping companies do that for a long [2:09] time. And about three years ago in 2022, [2:13] uh we started Advisor Labs. We thought, [2:15] you know what, this consulting business [2:17] that we're in is great. We enjoy [2:18] building software, but let's take what [2:21] we know in in AI and let's really focus [2:24] all of our efforts on helping companies [2:26] implement and build AI. So, we've been [2:29] doing that. We have a lot of experience [2:31] in a few different industries, the [2:33] financial industry, in construction, [2:36] healthcare, higher education, lots of [2:38] great companies in there that that we've [2:40] helped. But throughout that process, we [2:43] have found some lessons learned, some [2:45] things that we believe a lot of [2:47] companies are doing right or doing [2:50] wrong. So my my goal today is to really [2:54] share with you, accelerate you through [2:56] those lessons learned that we have [2:57] already picked up over the last handful [2:59] of years. [3:02] So here's what we know. Like if we [3:04] distill all of the surveys, all of the [3:06] feedback down into kind of one slide, we [3:11] know that about 25% [3:14] of all AI projects are creating value, [3:18] more value than it costs to build. [3:23] But we know that only 10% are creating [3:27] significant financial value. There's a [3:30] lot of reasons behind that. There's [3:33] plenty of because and why not and and [3:37] technical reasons and non-technical [3:39] reasons. But when you ask companies, why [3:43] is this happening or what is the thing [3:45] that's holding you back? They'll tell [3:47] you these four things. They'll say that [3:50] it's really hard to understand the ROI [3:54] of these types of projects. We've never [3:56] done it before, so we don't know what [3:57] this looks like. We don't know what we [3:59] should expect for a return. [4:02] We don't know how to prioritize these [4:04] projects. We have everything else that [4:05] we're doing every day. How do we take [4:08] this new AI thing and fit it into the [4:10] rest of of our of our IT stack? How do [4:14] we find integrations into our existing [4:17] systems? AI is great. Some of the best [4:20] use cases out there are building [4:22] integrations into our legacy systems. [4:24] And that's that's hard to do. [4:26] Integrations are hard. [4:28] And then at at the end, a lot of people [4:31] invest, they haven't taken the time to [4:33] really understand the models that [4:34] they're investing in. And at the end, [4:36] they don't get the outputs that they [4:38] were looking for. So they've spent all [4:40] this time, they've made all these [4:41] investments, and ultimately things get [4:44] in the way. [4:47] Now, why why is this happening? Again, I [4:50] said there's a lot of wise. I think this [4:51] is probably the biggest why. This is [4:54] happening so fast. [4:57] Many of us have lived through some of [4:58] these. If you look at the amount of time [5:01] it took the telephone to get from one [5:03] user to 50 million users, it's a long [5:06] time, 50 years. If you look at the same [5:08] thing with computers to 50 million [5:11] users, it took 14 years. If we look at [5:15] the internet, it took seven years. [5:18] Chat GPT was 100 million users in two [5:21] months. So this is much faster than any [5:25] transformation we've ever dealt with. [5:28] That brings a lot of anxiety. It brings [5:30] a lot of stress. It brings challenges. [5:33] It a lot of fun stuff. But ultimately [5:35] this is moving much faster than any of [5:37] us are used to. [5:40] So not only does this move faster, but [5:42] then we have these other constraints or [5:44] risks that we're all dealing with. [5:47] There's only 15 or 16% of leaders out [5:50] there that say they know more about AI [5:54] than what you can learn from your local [5:56] news. That's a problem. [5:59] We know that it's really hard to hire [6:01] staff. Over 60% of companies can't hire [6:06] or say that they don't even have an [6:08] individual that's dedicated to AI. It's [6:10] hard to implement AI when you don't have [6:12] the team or don't have a person to help [6:14] you do it. [6:15] We also have this issue with vendors [6:17] where now every vendor is an AI vendor [6:21] and how do we work through that whole [6:22] mess of well this company used to be our [6:25] CRM company but now they're our AI [6:27] company. All of these things become [6:30] obstacles in the way of us implementing [6:33] AI. [6:35] If we distill all of that down I think [6:37] we can classify the work that we do. So [6:40] when we come into companies, a lot of [6:42] times they've tried to do some of these [6:44] things and ultimately they've failed. [6:46] They've decided, oh, we need to bring in [6:48] someone from the outside to help us. And [6:50] ultimately we can distill those down [6:53] into one of these three categories. [6:56] So don't do these things. First one, we [7:00] hear so many times a company say, "Well, [7:03] I just needed chat GPT or I really [7:06] wanted this tech. It wasn't about a [7:09] problem and a solution, but instead it [7:12] was I want to buy this thing because it [7:15] sounds like fun. Don't do that. [7:18] The other piece, number two, is ignoring [7:21] the human side. You can buy the best [7:24] piece of technology, but if you haven't [7:27] taken the time to understand how people [7:29] use that software, how they use that [7:32] tool, if it's impactful to them, it [7:35] doesn't matter the value it can bring to [7:37] them. If they don't want it, they won't [7:39] use it. That old analogy of if you build [7:41] it, they will come does not work with [7:43] AI. Don't try it. [7:46] Boiling the ocean. This is probably the [7:48] biggest problem. When we talk to [7:50] organizations, companies feel like [7:52] they're behind. They need to do [7:54] something and they want to catch up is [7:57] really the the key. And so instead of [8:00] finding lowhanging fruit and [8:02] prioritizing and doing it in a very [8:05] strategic way, instead it's well, let's [8:08] go all in. Let's do everything at once. [8:11] Let's boil that ocean. Don't do that. [8:15] So what should we do? Here's where you [8:18] know if if you're going to get anything [8:20] today, next three slides [8:24] here. [8:26] Think about your outcome first. Decide [8:29] what the problem is that you need to [8:30] solve. This sounds very simple. I'm sure [8:34] all of us are going, "Well, yeah, of [8:35] course." But it's not something we all [8:38] do. [8:40] If you sit down with your team and think [8:42] about the problems you all experience [8:44] day in day out, hey, this department is [8:47] is suffering from this symptom. [8:51] Most of the time we can find [8:52] technological solutions to those [8:54] problems. And now all of us are [8:57] searching for AI and how AI can solve [9:00] those problems. Think about your outcome [9:02] first. Find the problem, then think [9:04] about the solution. [9:07] Also, it's important to think about what [9:09] you're trying to get out of that [9:11] solution. Do I want to save money? Do I [9:14] want to re reduce risk? Do I want to [9:17] increase compliance? [9:19] Do I want to reduce turnover with inside [9:22] of my departments? What is it that I'm [9:24] trying to get out of this investment? [9:28] Number two, this is really two [9:30] individual pieces, but we'll talk about [9:31] it one slide. focus on people and [9:35] processes. So people first. Like we were [9:38] saying in the don'ts, people go out and [9:41] buy software and they don't think about [9:42] how that impacts people. First thing you [9:45] should do is think about people. Go ask [9:46] the people, what is it that you need? [9:49] What what wins have you had by using AI [9:52] already? What are the ways that you [9:54] think we can leverage these tools? [9:56] They're going to tell you some great [9:58] ideas. They're going to tell you what [10:00] they've already been doing and and [10:02] you'll get a lot of value out of that. [10:05] The other piece is I guarantee you your [10:07] lowhanging fruit is in the processes [10:10] inside of your business. Probably the [10:12] back office processes. Now, I'm not [10:15] saying that's your biggest impact, but [10:18] that's probably your lowhanging fruit. [10:21] So think about the things that we've [10:22] invested in from an RPA perspective many [10:25] years ago. Business process management [10:27] that we've been doing for a while. Now [10:29] intelligent automation, find ways to [10:32] integrate AI with those existing [10:34] processes. That's a lowhanging fruit. [10:41] The other one, so let's go back to [10:43] Disney. [10:44] They have this unique business. They [10:47] have people come into a park. They've [10:50] already had those people spend money, [10:52] but they need to get more money out of [10:54] them. That is their business model, at [10:56] least for the parks. And I enjoy going [10:58] to Disney, but they thought about that [11:00] problem as a unique business. If you [11:03] want to be the 10% of businesses that [11:07] are driving significant financial value [11:09] from implementing AI, you have to find [11:12] your ne unique business problems. [11:15] Certainly, I said in the last slide, [11:17] your processes will drive lowhanging [11:20] fruit value. If you want big big wins, [11:24] it's in your unique problems. Think [11:27] about what your business does, where [11:29] your data is, what unique data you have [11:32] that your competitors don't have. Think [11:34] about all of those things and then start [11:36] iterating through ideas. How do I [11:38] generate value from that? How can I [11:40] learn from that? How can I make [11:42] predictions based off of that data? [11:45] That's where your significant value [11:47] comes. [11:50] So, if you leave here today, here's a [11:52] 90-day program for you. Go home, do [11:55] these three, there's really four, but do [11:58] these things. Spend the next 90 days [12:01] doing this. [12:03] So, first sit down, you maybe one other [12:07] person, and create a list of things you [12:10] think AI can produce value in in your [12:13] business. [12:14] Then prioritize those based off of the [12:17] impact you believe those will have in [12:19] your business and the ability you have [12:21] to perform or build that thing. So, [12:25] create that list. [12:27] Then, you're going to go out and [12:29] collaborate with teams. So after you've [12:31] generated that list, get a team together [12:33] of crossf functional team members. Grab [12:36] people that are not like you from other [12:38] departments, business, sales, HR. Bring [12:42] them all in a room and ask them about [12:44] that list of things that you created. [12:47] Ask them if they if they think it's [12:49] valuable. Ask them if they think it [12:51] provides impact. Ask them what they [12:54] would add to this list. Ultimately, [12:57] you're going to get a backlog that comes [13:00] out of these conversations, [13:02] which is a roadmap that over the next [13:04] year, two years, you're going to [13:06] dynamically work through and prune. [13:08] You're going to find opportunities. Oh, [13:10] there's this software out here that will [13:12] solve item number three. You know what? [13:14] Item number two is so important to us. [13:16] We're going to go out and build that. [13:18] Item number four, we've found another [13:20] solution for that. We don't need need to [13:22] use AI. But this becomes the backlog [13:25] that you'll work off of over the next [13:27] year to two years. And then finally, [13:31] you've taken this list, maybe look at [13:33] the top five or 10 and start [13:35] experimenting. Validate that you can [13:38] actually do these things using AI. Great [13:41] example, we had a customer that's in the [13:43] education space. They were trying to see [13:46] if they could they could have AI build [13:48] lesson plans for them. And certainly we [13:51] could build a prototype to do that. But [13:53] one of the conversations we had, we [13:54] said, "Hey, why don't you go out and [13:56] just use Google's notebook LM, put your [13:59] lesson plans into there, and start [14:00] asking it questions. Can it produce a [14:03] lesson plan that would make sense to [14:04] you?" That's a nocost experiment for [14:07] them. maybe a few dollars if they don't [14:09] already have the license. But those are [14:11] the types of validations that will help [14:13] you work through that list, that [14:15] backlog, and help you prioritize [14:18] ultimately what that backlog looks like. [14:20] And then finally, that fourth item, [14:22] don't forget the unique stuff. So after [14:24] you've created that big long list of all [14:26] these other things, then create another [14:29] list that is the unique opportunities [14:31] your business has. Again, think about [14:34] the data that you have. Think about what [14:36] you have that your competitors don't. [14:37] Think about what you're doing or what [14:39] you know in your business that no one [14:41] else knows. Those are the unique [14:43] opportunities that will drive [14:45] significant value for you. [14:49] You don't have to take my word for it. [14:50] There are plenty of people that have [14:52] said similar things to this. Ultimately, [14:56] I I think these quotes are great, but I [14:59] like the one on the bottom the best. [15:02] It reads, "Harning machine learning can [15:05] be transformational, but for it to be [15:07] successful, enterprise needs leadership [15:10] from the top." This means understanding [15:13] that when machine learning changes one [15:15] part of the business, the other parts [15:18] must change. Also, don't forget about [15:21] the downstream ripples of this backlog [15:23] that you built. Think about the ways [15:25] that it impacts others. Thinks about [15:27] Think about the ways it impacts you. [15:29] Think about the ways that it impacts [15:31] your backlog. There is a lot of [15:34] downstream impact with these systems and [15:36] tools that we build. [15:40] Ultimately, thank you for your time. We [15:42] have a booth in the exhibit hall. We're [15:44] in booth 357. Look forward to hope [15:47] hopefully you come visit us today. But I [15:48] appreciate your time. Thank you.