Medtronic, EY Say AI Amplifies Flawed Logic Without Process Redesign

AI is not your strategy – it's your mirror and your reset button With Medtronic And EY — Ai4

Ai4

Layering generative artificial intelligence onto legacy workflows merely accelerates bad processes and amplifies flawed decision logic, requiring enterprises to scrap existing steps and rebuild operations around AI natively.

AI-assisted summary of AI is not your strategy – it's your mirror and your reset button With Medtronic And EY — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 29, 2025, this presentation features Stephanie Franklin-Thomas (VP/Global CISO at Medtronic) and Nana Ahwoi (Americas Consumer and Health Cyber Industry Group Leader at EY). According to the video description, the session explores how artificial intelligence exposes flawed decision-making, poor data governance, and organizational unreadiness, asserting that successful AI initiatives require fundamentally redesigning core business processes rather than using technology to patch broken strategies.

## TL;DR Speakers Nana Ahwoi and Stephanie Franklin-Thomas discuss why organizations cannot treat artificial intelligence as a plug-and-play strategic fix, demonstrating that layering AI over convoluted legacy workflows merely accelerates inefficiencies. Drawing from Medtronic's rollout of internal LLMs and EY's 'client zero' lessons, they explain that AI amplifies flawed decision logic and falters without high-quality governed data. Successful enterprise adoption requires addressing human skepticism and ego, co-designing processes alongside end users, and wiping the slate clean to build streamlined, AI-native workflows from the ground up.

## Atmosphere & style The presentation is candid, collaborative, and conversational, balancing pragmatic enterprise realities with lighthearted, relatable humor. The dynamic between Ahwoi and Franklin-Thomas feels like a transparent peer-to-peer exchange, enlivened by audience polling, candid personal anecdotes—such as a chatbot bluntly advising a divorce—and playful banter about elevator music. Rather than delivering an academic lecture, the session creates an engaging, reflective workshop vibe that encourages leaders to confront uncomfortable operational truths without defensiveness.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=tBibN48Zdsw&t=4s) Rapid generative AI adoption—such as ChatGPT reaching 700 million weekly active users by August 2025—does not automatically resolve underlying organizational and operational inefficiencies. - [02:07](https://www.youtube.com/watch?v=tBibN48Zdsw&t=127s) Deploying internal AI tools to automate high-volume tasks (such as vendor security questionnaires) quickly exposes obstacles in prompt variability, employee skepticism, and data hygiene. - [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) Layering AI over convoluted, manual workflows merely creates an expensive, technologically enabled broken process rather than fixing core inefficiencies. - [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) Achieving true AI transformation requires blank-sheet redesign, collapsing complex multi-step legacy workflows into native, autonomous 1-to-2 step outcomes. - [06:14](https://www.youtube.com/watch?v=tBibN48Zdsw&t=374s) Cultural resistance and pre-engineering bias emerge when enterprise teams attempt to isolate AI into predetermined silos rather than reimagining the end-to-end workflow. - [08:26](https://www.youtube.com/watch?v=tBibN48Zdsw&t=506s) AI acts as a logic amplifier: if underlying business rules or reasoning are flawed, AI automation will only magnify and accelerate those flaws. - [08:26](https://www.youtube.com/watch?v=tBibN48Zdsw&t=506s) Successful model implementation strictly depends on teachable, uniform decision logic and well-governed source data before deploying automation. - [10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) Human ego and fear of intellectual displacement hinder adoption, making it necessary to co-design AI solutions directly with end users to establish trust. - [10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) According to the World Economic Forum, 60% of employers anticipate substantial transformation via digital access and AI, paired with significant employee anxiety over evolving job roles. - [12:30](https://www.youtube.com/watch?v=tBibN48Zdsw&t=750s) Driving operational value requires leaders to identify specific processes in need of a reset by challenging foundational assumptions and eliminating legacy waste. - [14:18](https://www.youtube.com/watch?v=tBibN48Zdsw&t=858s) Overcoming enterprise inertia requires empathetic leadership that helps veteran employees separate their personal identity and pride from obsolete legacy workflows. - [14:18](https://www.youtube.com/watch?v=tBibN48Zdsw&t=858s) Breaking through institutional groupthink requires leaders to actively solicit unconventional, unconditioned perspectives from outside traditional leadership circles, such as intern cohorts. - [14:18](https://www.youtube.com/watch?v=tBibN48Zdsw&t=858s) AI cannot replace business strategy; it functions as a mirror that reveals operational realities and a reset button to redesign core operating models along the critical path.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=tBibN48Zdsw&t=4s) Introduction and the AI Adoption Paradox

Nana Ahwoi, EY Americas Consumer and Health Cyber Industry Group Leader and consulting partner, opens the session alongside Stephanie Franklin-Thomas, Vice President and Global Chief Information Security Officer (CISO) at Medtronic. Setting the stage for their talk, "AI is not your strategy, it's your reset button and your mirror," they note that the session is intended to explore why artificial intelligence cannot simply be treated as a plug-and-play strategic fix, but rather as a catalyst that demands deep process redesign. Stephanie adds a quick disclaimer that their remarks reflect personal perspectives rather than official firm stances, while playfully encouraging attendees to check out her solo MedTech security session later in the afternoon.

Nana highlights the sheer velocity of generative AI adoption to frame the scale of the disruption. When ChatGPT first launched, it captured one million users within just five days; by August 2025, that figure surged to an astonishing 700 million weekly active users. This explosive uptake illustrates how deeply embedded AI has become in modern digital culture. However, Nana poses the question that dominates corporate boardrooms: does this rapid adoption mean that AI has fully arrived, that organizational processes will automatically run smoothly, and that business operations will instantly become "exceptionally awesome"? Stephanie's straightforward answer: "Not exactly."

- **Speakers**: Nana Ahwoi (Partner & Cyber Leader, EY) and Stephanie Franklin-Thomas (VP & CISO, Medtronic). - **Scale of Adoption**: ChatGPT reached 1 million users in 5 days; active user volume reached 700 million weekly users by August 2025. - **Core Premise**: Rapid AI adoption does not automatically translate into optimized business operations without foundational process changes.

---

[02:07](https://www.youtube.com/watch?v=tBibN48Zdsw&t=127s) The Vendor Questionnaire Challenge and Medtronic GPT

Stephanie outlines how the operational reality of AI implementation collided with legacy enterprise demands inside Medtronic's cybersecurity organization. Driven by the demands of cyber insurance underwriters and prospective enterprise customers, corporate security teams are bombarded with hundreds, if not thousands, of third-party security questionnaires inquiring about organizational security posture, patching cadences, and foundational vulnerability management. Historically, this meant dedicating valuable human analysts to answer the exact same questions manually across countless spreadsheets and portals.

```mermaid flowchart LR accTitle: Legacy Questionnaire Process vs AI Bot Strategy subgraph Legacy["Traditional Method"] Q1["Third-Party Security Questionnaires"] --> M1["Security Analysts (Manual Repetitive Work)"] end subgraph AttemptedAI["Initial AI Automation Attempt"] Q2["Inconsistent Vendor Formulations"] --> M2["Medtronic GPT / Custom LLM"] D1["Inconsistent Internal Data"] --> M2 M2 --> R1["Inaccurate Outputs & Human Distrust"] end ```

To solve this operational bottleneck, Medtronic deployed an internal instance called Medtronic GPT. The team trained models on internal security documentation with the expectation that the system would instantly generate accurate, automated answers to incoming third-party assessments. The rollout quickly encountered three major obstacles: 1. **Question Phrasing Variability**: External third parties frame and phrase security requirements in vastly different, non-standardized ways. 2. **The Human Element**: Staff members harbored hesitation, distrust, and rigid expectations regarding how and where AI should be used. 3. **Data Integrity**: In line with the classic rule of "garbage in, garbage out," incomplete or unstructured source data produced inaccurate, unreliable answers.

---

[03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) 'Client Zero' Lessons: Why AI Cannot Fix Broken Workflows

Interacting with the audience, Nana asks how many participants have launched an AI project only to face false data, frustrated stakeholders, and poor adoption. Several hands go up, reinforcing that early disillusionment is a widespread phenomenon. Nana then polls the room on whether AI can fix a process riddled with manual workarounds and convoluted steps simply by pushing a button. Only a couple of hands rise, underscoring a hard-won realization across industries: technology cannot magically resolve fundamentally broken workflows.

Nana explains that EY acts as its own "client zero," having invested heavily in deploying AI internally before advising enterprise clients. Experience shows that applying cutting-edge technology directly on top of non-standard, manual workflows does nothing more than create an expensive, "technologically enabled process that is really bad." Instead of wrapping an existing ten-step workflow in automation, organizations must design the system with a blank slate as if "AI was in the room today."

```mermaid flowchart TD accTitle: Broken Automation vs AI-First Redesign subgraph Antipattern["Bolting AI on Broken Processes"] A1["10-Step Manual Process"] --> A2["Apply AI Automation Layer"] A2 --> A3["High-Tech Broken Process"] end subgraph CleanDesign["AI-Native Process Redesign"] B1["Examine Business Intent"] --> B2["Redesign Workflow with AI Native"] B2 --> B3["Streamlined 1-to-2 Step Autonomous Outcome"] end ```

- **Client Zero Concept**: EY tests and scales AI capabilities internally to uncover failure modes before guiding client transformations. - **The Process Trap**: Layering AI over manual steps and unstandardized workarounds merely accelerates inefficiencies. - **AI-First Redesign**: Radical simplification—collapsing 10-step legacy workflows into one or two seamless, automated steps.

---

[06:14](https://www.youtube.com/watch?v=tBibN48Zdsw&t=374s) The Human Element and Designing End-to-End

Delving deeper into Medtronic's third-party assessment journey, Stephanie emphasizes that human dynamics often prove far more challenging than technological integration. When Medtronic first introduced AI into the questionnaire pipeline, team members automatically adopted a defensive posture: *"this is our process and we think AI fits here."* By trying to isolate AI to a tiny, predetermined silo within the legacy workflow, the team attempted to pre-engineer the solution rather than stepping back to evaluate the end-to-end outcome.

To illustrate human skepticism toward algorithmic advice, Stephanie shares a humorous personal anecdote from the first time she and her husband experimented with ChatGPT. Acknowledging that they are polar opposites who rarely agree on anything, they asked the model for advice on their marriage—and the chatbot blunt-force advised that they should get divorced. To this day, her husband remains deeply suspicious of anything generated by ChatGPT. Stephanie links this back to her enterprise team: when practitioners are inherently skeptical of AI prompts and outputs, they resist adoption and try to restrict the tool's scope, ultimately undermining the full transformational value of the technology.

- **Pre-Engineering Bias**: Teams often force AI into isolated steps of an existing process instead of reimagining the workflow from end to end. - **Skepticism and Distrust**: Personal and professional biases lead users to question model outputs, creating cultural resistance to adoption. - **The Marriage Anecdote**: A humorous query yielding advice to "get divorced" turned Stephanie's husband into a lifelong AI skeptic—mirroring how a single flawed output can breed lasting enterprise mistrust.

---

[08:26](https://www.youtube.com/watch?v=tBibN48Zdsw&t=506s) Amplifying Flaws: Decision Clarity and Data Quality

Nana reinforces that artificial intelligence does not solve logical failures; rather, "if your logic is flawed, all that AI does is amplify that flawed logic." While many leaders acknowledge this concept theoretically, organizations consistently fail to operationalize it when designing AI initiatives. To achieve tangible value, organizations must focus on decision clarity, teachability, and rigorous data hygiene before deploying models.

```mermaid flowchart TD accTitle: Foundational Pillars for Effective AI Implementation subgraph Foundations["Core Foundations for AI Success"] P1["Teachable & Consistent Decision Logic"] P2["High-Quality, Governed Source Data"] P3["Blank-Sheet Process Redesign"] end Foundations --> Outcome["Reliable, Scalable Enterprise AI Value"] ```

To create resilient AI capabilities, Nana and Stephanie pinpoint three non-negotiable imperatives: - **Teachable Decision Logic**: Business decisions must be uniform and clearly defined. If a poll of team members yields conflicting decisions for the same scenario, training an AI model on those inputs will inevitably create erratic, contradictory results. - **Data Quality Prioritization**: Relying on the adage "garbage in, garbage out," models will reproduce the biases, gaps, and inaccuracies of their training corpora. Reliable outputs demand structured, clean, and governed enterprise data. - **Blank-Sheet Redesign**: Rather than retrofitting automation onto legacy practices, organizations must rethink processes from scratch with a blank sheet of paper, asking what the workflow would look like if built natively for an AI-enabled environment.

[10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) Shifting Mindsets and Building Trust in AI Adoption

Driving meaningful improvement with AI requires organizations to step back from placing rigid constraints on tools and instead reformulate prompts to let the technology highlight unseen opportunities. Stephanie explains that within her own team, adoption was not simply about decreeing that "AI is our coworker," but rather about fostering conversations to actively shift their collective frame of mind. The most substantial hurdle in AI adoption remains the human ego: because humans view themselves at the top of the intellectual food chain, taking a step back to absorb, digest, and learn from new machine-driven insights can feel counterintuitive and humbling.

Nana emphasizes that fixing underlying processes must be paired with addressing the human element, noting that "people are not going to adopt what they don't trust." Instead of building solutions in isolation and delivering them to stakeholders under untested assumptions, technical teams must co-design systems directly alongside end users. Engaging stakeholders directly gives them an active role in reshaping workflows and helps dismantle operational resistance.

- **WEF Future of Jobs Data:** According to the World Economic Forum's *Future of Jobs Report*, 60% of employers expect their organizations to undergo significant transformation through expanded digital access and technologies such as AI. - **Workforce Anxiety:** The report also highlights pervasive reluctance and fear among employees concerning what these systemic shifts mean for their individual roles and day-to-day work.

```mermaid flowchart TD accTitle: Human-Centered Process Reset for AI Adoption subgraph Mindset["1. Mindset Shift"] A["Acknowledge Cognitive Bias / Ego"] --> B["Open Prompts & Uncover Blind Spots"] end subgraph Process["2. Co-Design & Trust"] B --> C["Engage Stakeholders Directly"] C --> D["Audit & Simplify Legacy Workflow"] end subgraph Implementation["3. AI Integration"] D --> E["Embed AI on the Critical Path"] end ```

---

[12:30](https://www.youtube.com/watch?v=tBibN48Zdsw&t=750s) Interactive Exercise: Identifying Processes Ready for a Reset

To ground the discussion in practical action, the speakers invite the audience to participate in an interactive reflection exercise. Stephanie jokes about generational habits—observing that attendees in her age bracket will likely reach for physical paper and pen, while younger cohorts like her son will pull open notes apps on their phones—noting that the act of writing down an idea creates real psychological commitment.

Nana frames the challenge by prompting participants to examine a business process they personally touch rather than deflecting attention onto other departments. Before attempting to automate or patch a workflow with artificial intelligence, leaders must pinpoint the exact facets of the process that need a fundamental reset.

- **Challenge Assumptions:** What foundational, taken-for-granted assumptions about this workflow need to be contested? - **Eliminate Waste:** What legacy steps or handoffs can be completely removed rather than digitized? - **Clarify Logic:** What business rules or decision logic must be clearly defined before introducing model automation?

While the audience reflects, the speakers exchange lighthearted banter about the lack of background elevator music, with Stephanie joking that she could sing into her microphone to fill the silence.

---

[14:18](https://www.youtube.com/watch?v=tBibN48Zdsw&t=858s) Leadership Traits for Overcoming Legacy Inertia and Final Reflections

An audience member asks what core leadership capabilities are required to move an established enterprise with entrenched legacy workflows forward. Stephanie highlights the need for an agile mindset and deep empathy. When leaders attempt to overhaul longstanding processes, veteran employees who originally designed those systems often feel personally attacked, reacting as if "you're calling my baby ugly." To defuse this defensiveness, Stephanie uses analogies—such as reminding teams that humanity once created fire with flint but evolved past it—to demonstrate that abandoning obsolete methods is a normal, positive progression.

Nana adds that leaders must actively guard against groupthink and familiarity traps. Instead of continually consulting the same inner circle, effective leaders seek out overlooked "corners of excellence" across the organization and remain genuinely open to unconventional input. Stephanie shares an ongoing leadership practice of pitching raw process problems to incoming intern cohorts; despite occasional initial skepticism, these fresh, unconditioned perspectives frequently provide breakthrough nuggets of insight that seasoned leaders overlook.

An audience contributor points to organizational change management (OCM) as a critical focus area, underscoring that transformative initiatives cannot succeed in isolation without keeping the broader organization informed and aligned. In closing, Nana and Stephanie leave the audience with a final charge: AI acts as both an unsparing mirror and a reset button. To capture its true value, enterprises must redesign their underlying business models around AI along the critical path, challenge the status quo, and have the courage to press that reset button.

- **Empathy in Process Redesign:** Acknowledge the historical value of legacy workflows while helping long-tenured employees separate their personal identity from legacy systems. - **Diversifying Input:** Solicit ideas beyond the typical leadership circle by engaging interns and cross-functional teams to break through institutional echo chambers. - **The Core Mandate:** AI does not substitute for strategic transformation; it requires leaders to look in the mirror, evaluate operational realities, and fundamentally rethink core operating models.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=tBibN48Zdsw&t=4s) AI is not your strategy, it's your reset button and your mirror — Nana Ahwoi & Stephanie Franklin-Thomas

> [00:04](https://www.youtube.com/watch?v=tBibN48Zdsw&t=4s) Not exactly. — Stephanie Franklin-Thomas

> [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) technologically enabled process that is really bad — Nana Ahwoi

> [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) AI was in the room today — Nana Ahwoi

> [06:14](https://www.youtube.com/watch?v=tBibN48Zdsw&t=374s) this is our process and we think AI fits here. — Stephanie Franklin-Thomas

> [08:26](https://www.youtube.com/watch?v=tBibN48Zdsw&t=506s) if your logic is flawed, all that AI does is amplify that flawed logic. — Nana Ahwoi

> [08:26](https://www.youtube.com/watch?v=tBibN48Zdsw&t=506s) garbage in, garbage out — Nana Ahwoi & Stephanie Franklin-Thomas

> [10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) AI is our coworker — Stephanie Franklin-Thomas

> [10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) people are not going to adopt what they don't trust. — Nana Ahwoi

> [14:18](https://www.youtube.com/watch?v=tBibN48Zdsw&t=858s) you're calling my baby ugly. — Stephanie Franklin-Thomas

> [14:18](https://www.youtube.com/watch?v=tBibN48Zdsw&t=858s) corners of excellence — Nana Ahwoi

## Mentioned Stephanie Franklin-Thomas, Nana Ahwoi, Medtronic, EY, ChatGPT, Medtronic GPT, World Economic Forum, Future of Jobs Report, Ai4

## Video description Stephanie Franklin-Thomas, VP/Global Chief Information Security Officer, Medtronic Nana Ahwoi, EY Americas Consumer and Health Cyber Industry Group Leader, EY

Abstract

AI reveals what your business is (and is not) ready for. Broken processes produce broken automation – Start designing for what AI can do. Many organizations think AI will change their business. What it can do is reveal the flaws in how decisions are made, how data is governed and whether people are ready to change. AI doesn't fix strategy – it exposes it. Successful AI initiatives require redesigning core processes to align with how AI actually works. The real win is transformation. Join us to hear lessons from the field while reflecting on how YOU can challenge how things have always been to capitalize on how things CAN be in this era of AI.

[Source](https://www.youtube.com/watch?v=tBibN48Zdsw)

中文简报

美敦力与安永：若不重塑业务流程，AI将放大错误逻辑 将生成式人工智能直接套用于传统工作流只会加速不良流程并放大错误决策逻辑，企业须打破现有步骤并围绕AI原生重构运营。 演讲嘉宾 Nana Ahwoi 与 Stephanie Franklin-Thomas 探讨了为何企业不能将人工智能视为即插即用的战略补丁，指出将 AI 叠加在复杂的传统工作流程之上只会加剧低效。结合 Medtronic 推广内部 LLM 以及 EY 作为“零号客户”的经验教训，他们阐述了 AI 会放大错误的决策逻辑，且在缺乏高质量治理数据时难以发挥作用。企业若要成功实现 AI 落地，必须化解人员的疑虑与心理抗拒，与最终用户协同设计流程，并彻底打破陈规，从零开始构建精简的原生 AI 工作流程。 生成式 AI 的迅速普及并不等同于企业运营效率的自动提升，将 AI 强行套用在混乱的手动流程上，只会制造出昂贵且低效的错误自动化。 (0:04) 美敦力（Medtronic）在部署内部 Medtronic GPT 处理第三方安全问卷时，遭遇了提问表述不一、员工抵触怀疑以及底层数据治理不佳等核心瓶颈。 (2:07) AI 本质上是逻辑放大器，若底层业务逻辑存在缺陷，自动化只会放大错误；企业必须在白纸上重构端到端流程，将繁琐的传统步骤精简为原生的一至两步自治结果。 (3:52) 克服员工心理防线与智力替代恐惧是落地关键，技术团队需与终端用户共同设计解决方案建立信任，并通过引入不同视角的意见打破组织固有惯性。 (10:06)

Reading summary

Artificial intelligence cannot serve as a plug-and-play corporate strategy; instead, it acts as an operational mirror and reset button that exposes broken workflows and forces organizations to redesign processes from scratch. The AI Adoption Paradox (0:04) Consumer adoption of generative AI expanded at unprecedented speed, with ChatGPT acquiring one million users in five days and growing to 700 million weekly active users by August 2025. According to Nana Ahwoi of EY and Stephanie Franklin-Thomas of Medtronic, this consumer enthusiasm creates an executive illusion that deploying AI will instantly modernize corporate operations. In practice, rapid technological adoption does not automatically fix business operations without foundational restructuring. Automating Broken Workflows Amplifies Flaws (2:07) Applying automation directly onto fragmented legacy processes merely accelerates inefficiencies, producing an expensive technological layer over bad workflows. Drawing on EY's internal "client zero" tests, Ahwoi noted that AI does not correct flawed logic but amplifies it. When Medtronic deployed "Medtronic GPT" to automate third-party security questionnaires, the initiative encountered major obstacles: external partners phrased identical requirements inconsistently, and unstandardized, incomplete internal data triggered inaccurate outputs. To avoid this pitfall, Ahwoi and Franklin-Thomas argued that organizations must abandon ten-step legacy workarounds and adopt blank-sheet redesigns that build workflows natively around AI. Data Hygiene and Teachable Decision Rules (8:26) Reliable AI implementations require structured data governance and clear, teachable business logic before any model deployment. In accordance with the principle of "garbage in, garbage out," systems trained on unstructured or contradictory inputs yield erratic results. Ahwoi emphasized that if members of a team cannot agree on a single, standardized decision rule for a given scenario, an algorithm cannot be expected to generate consistent outcomes. Enterprise readiness therefore depends on harmonizing rules and cleaning source data rather than expecting models to infer missing standards. Human Skepticism and the Challenge of Trust (10:06) Human dynamics and organizational culture present steeper barriers to AI adoption than technical integration. Franklin-Thomas observed that practitioners often exhibit a defensive posture, pre-engineering solutions to isolate AI into rigid silos of an existing routine rather than allowing end-to-end transformation. Citing the World Economic Forum's Future of Jobs Report, the speakers noted that while 60% of employers expect significant transformation through digital access and AI, widespread employee anxiety persists regarding role displacement. Ahwoi stressed that end users will not adopt tools they distrust, making direct co-design essential. Overcoming Institutional Inertia and Legacy Bias (14:18) Dismantling legacy processes requires empathetic leadership to counter the defensive reaction of employees who built those systems. Franklin-Thomas advised separating professional identity from outdated practices, reminding teams that abandoning obsolete tools is a normal progression. To counter groupthink and institutional echo chambers, leaders must broaden their sources of input. At Medtronic, this includes presenting process bottlenecks to unconditioned intern cohorts to discover unconventional solutions that seasoned teams overlook. Ahwoi and Franklin-Thomas conclude that AI delivers value only when enterprise leaders have the institutional courage to confront their operational realities and place redesign along the critical path. However, the session leaves open how organizations should quantify the financial return of discarding functional, albeit inefficient, legacy systems in favor of long-term workflow overhauls.

精炼阅读摘要

安永（EY）合伙人Nana Ahwoi与美敦力（Medtronic）全球首席信息安全官Stephanie Franklin-Thomas指出，人工智能不能被视作即插即用的万能战略工具，其本质是一面反映企业底层逻辑缺陷的“镜子”与迫使业务流程重构的“重置按钮”；若无端到端的零基流程再造与组织信任，单纯叠加AI只会放大原有业务的低效。 采纳狂潮与企业落地现实的脱节 (0:04) Ahwoi指出，生成式AI的用户增长速度极为惊人，ChatGPT上线仅5天即突破100万用户，到2025年8月周活跃用户已达7亿；然而许多企业董事会误认为技术的普及意味着企业流程能自动运行顺畅并创造卓越价值，这一设想与实际脱节。Franklin-Thomas以美敦力网络安全团队的实践说明了该落差：面对来自网络保险商及企业客户数以千计的第三方安全评估问卷，团队原先不得不投入大量安全分析师在电子表格中重复手动应答；为突破瓶颈，美敦力基于内部安全文档部署了Medtronic GPT，期望实现自动化答复。 但在实际推行中，自动化尝试迅速遭遇三大阻碍：一是外部供应商对安全要求的提问句式千差万别且缺乏标准化；二是员工对AI输出心存疑虑、缺乏信任，且对工具的应用场景抱有僵化预期；三是源数据不够结构化且存在缺漏。实践表明，“垃圾进、垃圾出”（garbage in, garbage out）的规律在AI时代依然适用，不完整或未治理的数据直接导致模型输出了不准确、不可靠的结果。 叠加AI只会放大缺陷流程与错误逻辑 (3:52) Ahwoi基于安永作为“零号客户”（Client Zero）的内部试验指出，将前沿AI技术直接覆盖在原本充满非标手工绕行和繁琐步骤的流程之上，最终只会制造出一个“由技术赋能但极其糟糕的高成本流程”。Ahwoi强调，如果底层业务逻辑本身存在缺陷，AI所做的仅仅是放大这种错误逻辑。企业不能简单将既有的10步传统操作包裹在自动化外衣下，而必须采取“白纸设计法”（blank-sheet redesign），以AI原生视角重新审视业务目标，将其精简压缩为一到两个流畅的自动化步骤。 为确立稳健的AI能力，两位讲者提出了三项非协商性的基础要求：首先是确立“可传授的决策逻辑”，确保同一业务场景在团队内部具有一致的判断规则，否则相互冲突的决策输入必然导致模型输出混乱；其次是高标准的数据治理与质量把控，避免模型复刻语料中的偏差与数据漏洞；最后是从零重构流程，在原生AI环境中自底向上重新定义工作流。 破解组织心理防御与技术信任缺失 (6:14) Franklin-Thomas强调，相比技术整合，人性层面的阻力往往更难化解。美敦力在推行问卷AI工具初期，团队员工本能地表现出防御姿态，试图将AI禁锢在既有流程的极小孤岛中（即预先预设“AI只适合放在这里”），从而阻碍了端到端的整体评估。此外，从业者常因对算法结论心存戒备而抵制技术采纳；Franklin-Thomas分享了个人生活趣事作为类比——她与丈夫曾向ChatGPT咨询婚姻建议，模型竟直接建议他们离婚，导致其丈夫自此对AI输出产生长期怀疑，这与企业员工因单次错误输出而丧失信任的心态如出一辙。 人类长久以来习惯于处于认知食物链顶端，向机器洞见妥协并退后学习往往会引发自尊心防御。Ahwoi援引世界经济论坛《未来就业报告》数据指出，60%的雇主预计数字化普及与AI将推动组织发生重大变革，这在员工中引发了对个人岗位前景的广泛焦虑。Ahwoi指出，“人们不会采纳自己不信任的东西”，技术团队绝不能在孤岛中闭门造车，而必须邀请最终用户参与系统协同设计（co-design），让业务人员在重塑流程中掌握主动权以化解阻力。 突破路径依赖的同理心与多元领导力 (14:18) 针对如何推动深陷历史惯性的大型企业转型，Franklin-Thomas指出领导者必须具备敏捷心态与深刻的同理心。当管理层试图革新老旧流程时，那些最初亲手设计这些系统的资深员工往往会产生情感防御，如同面对“有人说自己的孩子丑陋”一般；领导者需要运用类比——例如人类曾用燧石取火但终会演进迭代——帮助资深员工将个人专业身份与过时系统解耦，理解告别陈旧方式是正常且积极的进步。 Ahwoi与Franklin-Thomas进一步指出，领导者必须主动打破群体思维与舒适区陷阱，避免仅在固定的核心圈子里征求意见。有效做法包括发掘组织内被忽视的“卓越角落”，以及向不受既定思维束缚的群体征集视角；例如美敦力管理层会将未加工的原生业务痛点抛给新入职的实习生群体，这些未受体制驯化的新视角往往能提供资深人士忽视的突破性见解。结合全面的组织变革管理（OCM），企业才能在保持全员认知一致的前提下推进转型。 Ahwoi与Franklin-Thomas总结认为，AI既是一面毫不留情的镜子，映射出企业长期掩盖的数据混乱与逻辑盲区，也是一个不可或缺的重置按钮。技术的最终价值不取决于算法模型本身，而取决于管理层是否具备勇气打破既得利益与流程惯性，在核心业务关键路径上践行端到端的零基再造与全员协同变革。

Detailed notes — zh

## Background 本内容由 Ai4 于 2025 年 8 月 29 日发布在 YouTube 平台上。视频记录了美敦力（Medtronic）副总裁兼全球首席信息安全官 Stephanie Franklin-Thomas 与安永（EY）美洲区消费与健康网络安全行业负责人 Nana Ahwoi 的联合演讲，核心探讨 AI 如何暴露企业决策与数据治理的缺陷，并强调 AI 的本质并非企业战略本身，而是推动底层核心业务流程重构的契机与镜子。 ## TL;DR 两位嘉宾通过美敦力早期尝试利用内部 GPT 自动化处理海量网络安全问卷时遭遇的数据与信任瓶颈，指出技术无法自动修复存在缺陷的流程，生搬硬套 AI 只会放大低效。演讲提出了基于“决策可教化”、“数据质量优先”以及“白纸重构思维”的核心原则，强调企业应与最终用户共同构建系统以建立信任，而非闭门造车。最后，针对拥有沉重历史包袱的传统组织，演讲呼吁领导者破除小圈子思维并保持敏捷开放，敢于按下业务重置键，从根本上革新端到端业务路径。 ## Atmosphere & style 整场演讲充满务实、风趣且极具现场共鸣的氛围。Nana 与 Stephanie 配合默契，两人既有行业大咖的高屋建瓴，又频繁抛出如“ChatGPT 劝人离婚”以及“现场打趣献唱”等生活化笑料，引得观众频频会心一笑。互动环节引导全场掏出纸笔反思自身业务，节奏张弛有度，既有对自动化落地痛点的辛辣直言，也有对组织文化变革充满激情的鼓舞。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=tBibN48Zdsw&t=4s) AI 本质上不是企业战略的核心终点，而是一面暴露组织流程与治理缺陷的“镜子”，以及重构业务底层逻辑的“重置键”。 - [00:04](https://www.youtube.com/watch?v=tBibN48Zdsw&t=4s) 尽管 ChatGPT 等 AI 工具用户规模飞速突破数亿，但技术的普及并不意味着企业原有的业务流程已经自动变得高效顺畅。 - [02:07](https://www.youtube.com/watch?v=tBibN48Zdsw&t=127s) 美敦力在自动化安全合规问卷的实践表明，单纯接入 AI 会迅速遭遇提问表述非标、员工信任缺失以及底层数据质量欠佳三大落地瓶颈。 - [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) 技术无法自动修复本质上有缺陷的流程；在充满人工变通方案的糟糕流程上叠加 AI，只会产生一个更昂贵、被先进技术武装的糟糕流程。 - [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) 企业应摒弃“贴膏药”式的局部修补，以“假设 AI 就在房间里”的白纸思维从头设计端到端流程，大幅削减冗余步骤。 - [06:14](https://www.youtube.com/watch?v=tBibN48Zdsw&t=374s) 组织变革中最大的阻碍往往来自“人的因素”——员工因不信任而产生心理抵触，或因惯性思维将 AI 局限在孤立狭窄的环节。 - [08:26](https://www.youtube.com/watch?v=tBibN48Zdsw&t=506s) 业务决策逻辑必须具备标准化与“可教化性”（Teachable Decisions）；若决策规则内部充满矛盾，AI 只会放大缺陷并输出不一致的结论。 - [08:26](https://www.youtube.com/watch?v=tBibN48Zdsw&t=506s) 数据质量是 AI 赋能的绝对前置条件，必须严格治理并杜绝“垃圾进，垃圾出”（Garbage in, garbage out）导致的错误扩散。 - [10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) “人们绝不会采用自己不信任的技术”，推进 AI 项目必须与业务最终用户及利益相关者共同构建（Co-build），而非闭门假设。 - [12:31](https://www.youtube.com/watch?v=tBibN48Zdsw&t=751s) 在将 AI 套用于业务之前，组织与个人需要主动自省，识别并挑战既定假设、果断剔除无用工序并厘清底层逻辑。 - [14:19](https://www.youtube.com/watch?v=tBibN48Zdsw&t=859s) 重构传统遗留流程需要领导者具备敏捷思维与开放心态，包容历史演进并有效化解老员工因既有流程被否定而产生的抵触情绪。 - [14:19](https://www.youtube.com/watch?v=tBibN48Zdsw&t=859s) 警惕小圈子决策引发的“群体思维”，领导者可通过引入没有既定流程包袱的新人（如实习生）进行逆向压力测试，发掘流程优化盲点。 - [17:05](https://www.youtube.com/watch?v=tBibN48Zdsw&t=1025s) AI 时代的深层业务重塑必须依托组织变革管理（OCM），唯有敢于按下重置键并将技术融入核心业务路径，才能实现真正的转型胜局。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=tBibN48Zdsw&t=4s) 开场介绍：AI 不是战略本身，而是重置键与镜子 安永（EY）美洲区消费与健康网络安全行业负责人兼咨询合伙人 Nana Ahwoi 与美敦力（Medtronic）副总裁兼全球首席信息安全官（CISO）Stephanie Franklin-Thomas 共同登台，开启了题为《AI 不是你的战略——它是你的重置键与镜子》（AI is not your strategy – it's your reset button and your mirror）的演讲分享。在简短介绍个人背景及阐明本次演讲仅代表个人观点而非公司立场的免责声明后，Stephanie 还幽默地预告了自己当天下午在三楼的另一场个人演讲。Nana 指出，本次分享的核心主旨在于阐明 AI 并非企业战略的核心终点，而是一次重新审视并重构底层业务流程（Process Redesign）的契机。 为了说明 AI 席卷全行业的惊人速度与普及度，Nana 回顾了 ChatGPT 在上线初期仅用 5 天就突破 100 万用户的历史，并分享了一组最新数据：截至 2025 年 8 月，ChatGPT 的周活跃用户数已达 7 亿（700 million weekly users）。然而，Nana 随即抛出一个带有调侃意味的反问：面对如此庞大的数据和 AI 的全面到来，这是否意味着一切都已完美无缺、企业流程已经变得无比高效顺畅了？ - **演讲嘉宾**：Nana Ahwoi（EY 美洲消费与健康网络安全行业主管合伙人）与 Stephanie Franklin-Thomas（Medtronic 副总裁兼全球 CISO）。 - **用户规模**：ChatGPT 上线 5 天突破 100 万用户；截至 2025 年 8 月，周活跃用户数已达 7 亿。 --- ### [02:07](https://www.youtube.com/watch?v=tBibN48Zdsw&t=127s) 美敦力的安全问卷困境与 AI 初体验的碰壁 针对 Nana 的反问，Stephanie 给出了断然否定的回答：“并非如此（Not exactly）”。作为全球医疗科技巨头美敦力的 CISO，Stephanie 负责管理公司全方位的网络安全事务。随着近年来网络安全保险（Cyber Insurance）的普及以及企业客户对第三方供应商网络安全态势审查的日益严格，美敦力安全团队每天都要应对海量涌入的安全合规问卷——数量高达成百上千份。过去，这项工作依赖大量人工逐条回复，而问卷中充斥着极其重复的基础问题，例如“你们是否打补丁？”、“如何处理系统漏洞？”等。 为了从繁重的问卷回复中解脱出来，团队决定利用 AI 技术（包括美敦力内部开发的 Medtronic GPT）来建立专属问答模型，期望只要把知识库灌入模型，就能瞬间实现自动应答。然而，这一初步尝试很快遭遇了三大现实困境： 1. **提问方式千差万别**：不同客户在提问同一项安全举措时，用词、句式与逻辑架构各不相同，导致模型识别与匹配存在困难； 2. **被低估的“人的因素”（Human Element）**：团队对 AI 输出的信任度与接受度不足，流程协同出现脱节； 3. **数据质量陷阱（Garbage in, Garbage out）**：如果底层输入的数据不够严谨准确，AI 模型生成出来的答案同样漏洞百出。 ```mermaid flowchart TD accTitle: 美敦力安全问卷自动化早期尝试中的三大困境 subgraph Input["输入端瓶颈"] A["外部客户/保险问卷"] --> B["提问形式多样化<br/>（句法/术语不统一）"] C["历史底层数据"] --> D["数据质量不一<br/>（垃圾进，垃圾出）"] end subgraph Model["AI 处理与应用"] B --> E["Medtronic GPT / AI 模型"] D --> E E --> F["自动生成回答"] end subgraph People["落地与信任障碍"] F --> G["员工对输出持怀疑态度"] G --> H["未能实现预期自动化收益"] end ``` --- ### [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) 拒绝“贴膏药”式赋能：修补坏流程只会产生更昂贵的坏流程 Nana 与现场观众展开了互动，询问有多少人在企业或个人尝试使用 AI 时因虚假数据或无人使用而感到无比沮丧，现场不少观众纷纷点头认同。Nana 接着向全场提问：“如果一个流程本身充满了人工变通方案（Manual Workarounds）、过度人工干预以及繁琐复杂的工序，有谁认为只要按下 AI 按钮，AI 就能自动把这个流程修好？”全场仅有一两只手犹豫地举起。 Nana 结合安永作为“零号客户”（Client Zero，即安永自身投入重金全面探索 AI 应用并赋能客户）的经验指出，**技术无法自动修复一个本质上存在缺陷的流程**。如果在充满非标步骤和人工补丁的流程上强行叠加 AI，得到的不过是一个“被先进技术武装起来的糟糕流程”（Technologically enabled process that is still bad）。与其在原有的 10 步流程中让 AI 在后台被动打转，企业更应该以“假设 AI 今天就在房间里”的全新视角重构逻辑——去思考这个流程是否原本可以简化为两步甚至一步，从根本上打破既有惯性。 --- ### [06:14](https://www.youtube.com/watch?v=tBibN48Zdsw&t=374s) 组织文化与“人的因素”：从离婚建议趣事谈对 AI 的认知阻碍 深入复盘美敦力的历程时，Stephanie 强调组织中“人的要素”往往是最大的绊脚石。在最初推进项目时，团队的惯性思维往往是先画定既有流程，然后武断地下结论说“AI 应该放在流程的这个环节”，这实际上是用既有框架去生搬硬套 AI，而不是从端到端（End-to-end）的整体视角去革新。 为了说明人对 AI 的不信任与认知偏差，Stephanie 分享了一段生活中的真实趣事：她第一次与丈夫一起使用 ChatGPT 时，两人性格迥异，几乎在任何事情上都观点相左。他们向 ChatGPT 询问关于两人婚姻的建议，结果 ChatGPT 竟然直接建议他们“应该去离婚”。直到今天，她的丈夫都对 ChatGPT 极度排斥和怀疑，输入任何提示词都先预设得到的结果不可靠。Stephanie 指出，企业团队中同样存在这种心态：员工要么对 AI 结论抱有强烈的怀疑与抵触，要么过度保守地将 AI 严苛限定在极其狭窄的孤立环节，从而极大地削弱了 AI 本应带来的变革价值。 > “如果数据不可靠，输出就不可靠；如果你的逻辑本身存在缺陷，AI 所做的仅仅是把这种缺陷逻辑放大而已。” --- ### [08:26](https://www.youtube.com/watch?v=tBibN48Zdsw&t=506s) 流程可教化与白纸重构原则 围绕如何让 AI 真正赋能业务，Nana 与 Stephanie 提炼出了几个关键的核心原则。首先是**流程决策的清晰度与“可教化性”（Teachable Decisions）**。如果让同一间会议室里的每个人去回答一个业务判定，大家给出的规则五花八门，那么用这些充满矛盾的判定去训练模型，产出的结果必然充斥着不一致性；因此，决策逻辑必须具备明确、结构化且可传授的标准。 其次，必须将**数据质量**置于最高优先级，严格杜绝“垃圾进，垃圾出”（Garbage in, garbage out）。模型训练完全基于喂给它的数据，若基础数据未经过治理与净化，自动化只会加速错误的蔓延。最后，Nana 引导 Stephanie 探讨“当 AI 就在房间内时的重新设计”（Redesigning with AI in the room）——这意味着企业需要放弃修修补补的改良主义，像面对一张白纸一样，重新推导整个业务流程的构建逻辑。 - **决策可教化（Teachable Decisions）**：业务决策逻辑必须标准化、无内部矛盾，才能作为有效数据用于模型训练。 - **数据质量优先**：AI 会直接放大底层数据的缺陷，确保数据底座的纯净度是成功的前置条件。 - **白纸重构思维（Blank Sheet Redesign）**：摆脱历史包袱，以 AI 的原生能力为基准，从零重新设计端到端流程。 ### [10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) 人与信任：与利益相关者共同构建而非闭门造车 推进 AI 落地的过程中，最艰难的挑战往往不在于算法或技术本身，而在于“人”的因素。Stephanie Franklin-Thomas 指出，在美敦力（Medtronic）推进业务改善并取得成功的过程中，团队经历了一个关键的认知转变：不再对 AI 施加过多的预设限制，而是转变提问和提示词（prompts）的构建方式，让 AI 自主指出它所看到的优化机遇。面对团队时，她并没有生硬地宣布“AI 就是我们的新同事，工作模式必须变成这样”，而是引导大家消化 AI 给出的见解，共同探讨如何重塑思维认知。人类习惯于自居在“食物链顶端”，总认为自己是最聪明的，但正因为这种“自以为什么都懂”的心态，当面对全新视角时，往往需要停下脚步深入消化，才能真正让新技术拓展思维的边界。 Nana Ahwoi 强调，在修复流程与训练模型之后，“人”的维度再怎么强调都不为过，因为“人们绝不会采用自己不信任的东西”（"People are not going to adopt what they don't trust"）。企业推进 AI 项目时，切忌关起门来为某个业务群体做假设性设计，而必须“与他们共同构建”（build it with them），让利益相关者深入参与全流程，从而打破既有思维定式。世界经济论坛（WEF）发布的《未来就业报告》（*Future of Jobs Report*）显示，60% 的雇主认为随着 AI 等技术的普及，数字化进程将从根本上重塑其组织架构。然而报告同时指出，员工对技术给个人岗位与日常工作带来的冲击普遍存在恐惧与抵触心理。要真正推动落地采用，就必须在整个旅程中持续吸纳最终用户与核心利益相关者的参与。 ```mermaid flowchart TD accTitle: 业务流程重置与 AI 协同构建机制 subgraph S1["流程重塑阶段"] A["审视核心业务流程"] --> B["破除既定假设 & 剔除冗余"] B --> C["厘清业务逻辑"] end subgraph S2["以人为本构建阶段"] C --> D["与利益相关者深度共创"] D --> E["建立信任消除抵触"] E --> F["将 AI 嵌入关键业务路径"] end ``` ### [12:31](https://www.youtube.com/watch?v=tBibN48Zdsw&t=751s) 互动重置练习：审视你自己的业务流程 会议现场进入了随堂互动环节。Stephanie 观察现场打趣道，通常年龄和她相当或更成熟的听众习惯掏出纸笔手写答案，而像她儿子那代年轻受众则会立刻拿出手机打开便签 App。她呼吁全场观众无论采用哪种载体，都务必把思考亲笔写下来，因为“只要写下来，就代表你认真对待这件事了”（"if you write it down, then you're serious about it"）。 Nana 随之抛出了互动核心命题：人们在审视业务时，往往倾向于挑剔别人的流程有问题，而认定自己的流程很完美。他要求台下所有人反思一个自己亲自参与或密切相关的真实业务流程，并在盲目套用 AI 之前，先确定该流程中哪个环节最需要彻底“重置”。在思考过程中，需要深入回答以下三个核心问题： - 在当前流程中，有哪些习以为常的既定假设是必须去挑战的？ - 流程中有哪些步骤其实完全可以被彻底剔除？ - 流程内部有哪些底层的业务逻辑亟待理顺与厘清？ 在给观众预留思考时间的间隙，现场还出现了一段幽默互动。Nana 笑称自己手头没有准备商用电梯背景音乐供大家思考时聆听，Stephanie 则立刻接茬笑称自己手里有麦克风可以现场献唱，Nana 连忙打趣回应“千万别”，引得全场欢笑。 ### [14:19](https://www.youtube.com/watch?v=tBibN48Zdsw&t=859s) 传统业务重塑：领导者必须具备的思维与技能 现场观众就传统机构转型提出了一个尖锐问题：面对拥有大量老旧遗留流程（legacy processes）的成熟组织，引领业务前行的领导者需要具备哪些核心技能？Stephanie 率先回应，领导者首要具备的是“敏捷思维”（agile mindset）以及对新事物的接纳度。老流程往往是由组织内资深的老员工一手建立起来的，当新管理者提出推倒重来时，这些员工的第一反应往往会觉得“你在说我的孩子长得丑”（"you're calling my baby ugly"）。面对这种抵触，Stephanie 常用生火的演变打比方：人类祖先曾用燧石取火，但除非在荒野求生节目中，今天没人会再这么做。领导者需要帮助团队理解变革在历史上一直带来益处，进而培养愿意测试、乐于倾听的敏捷心态。 Nana 对此表示高度赞同，并补充了“极致开放”的重要性。如果领导者固守己见，就无法构想未来的可能性。此外，敏捷思维还要求领导者审视自己所依赖的核心圈子，警惕是否总是在向同一批人征求意见，避免陷入“群体思维”（group think）与熟人偏见；组织内部往往存在尚未被发掘的“卓越角落”（corner of excellence），领导者应打破壁垒让更多元的角色参与进来。 Stephanie 随后分享了自己多年职业生涯中的一个私藏技巧——向实习生群体借力。每当迎来一批新实习生，她会把棘手的流程问题抛给他们。虽然有时实习生的想法天马行空，甚至让人初听之下觉得“这绝不可能发生，他们根本不懂业务”，但在许多情况下，正是这些未经职场条条框框约束的年轻大脑，能够直击本质，提供极具突破性的见解与启示。 - **核心领导技能维度**：开放包容、敏捷求变、倾听与测试意愿、破除经验主义偏见。 - **团队多元化策略**：避免长期依赖单一智囊小圈子，主动挖掘边缘业务部门的潜力人才。 - **逆向思维启发机制**：利用无既定流程包袱的新人（如实习生）进行压力测试，发掘流程优化的盲点。 ### [17:05](https://www.youtube.com/watch?v=tBibN48Zdsw&t=1025s) 组织变革管理与总结：按下业务重置键 在现场分享环节，一名观众指出“组织变革管理”（Organizational Change Management, OCM）是传统流程改造的核心支柱，必须确保全公司上下对正在推进的变革有清晰一致的认知，因为深层次的根本性变革绝不可能在孤立的真空环境中完成。 Nana 总结指出，技术演进的脚步不会停歇，未来与人类并肩作战的“数字同事”将不仅停留在工位旁，更会常驻在每台电脑与终端里。想要借助 AI 实现真正的脱胎换骨，企业就必须将 AI 融入业务的关键路径，从根本上重构商业模式。演讲最后留给全场一个深刻的自省命题：当你们以 AI 为镜审视自身时，面对那个近在眼前的“重置按钮”，你是选择维持现状、抱残守缺，还是敢于挑战自我、打破既定常规？“去大胆地颠覆旧习，重塑一切吧。” ## Notable quotes > [00:04](https://www.youtube.com/watch?v=tBibN48Zdsw&t=4s) AI 不是你的战略——它是你的重置键与镜子 — Nana Ahwoi > [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) 技术无法自动修复一个本质上存在缺陷的流程。 — Nana Ahwoi > [03:52](https://www.youtube.com/watch?v=tBibN48Zdsw&t=232s) Technologically enabled process that is still bad（被先进技术武装起来的糟糕流程） — Nana Ahwoi > [06:14](https://www.youtube.com/watch?v=tBibN48Zdsw&t=374s) 如果数据不可靠，输出就不可靠；如果你的逻辑本身存在缺陷，AI 所做的仅仅是把这种缺陷逻辑放大而已。 — Stephanie Franklin-Thomas > [10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) People are not going to adopt what they don't trust（人们绝不会采用自己不信任的东西） — Nana Ahwoi > [10:06](https://www.youtube.com/watch?v=tBibN48Zdsw&t=606s) build it with them（与他们共同构建） — Nana Ahwoi > [12:31](https://www.youtube.com/watch?v=tBibN48Zdsw&t=751s) if you write it down, then you're serious about it（只要写下来，就代表你认真对待这件事了） — Stephanie Franklin-Thomas > [14:19](https://www.youtube.com/watch?v=tBibN48Zdsw&t=859s) you're calling my baby ugly（你在说我的孩子长得丑） — Stephanie Franklin-Thomas > [17:05](https://www.youtube.com/watch?v=tBibN48Zdsw&t=1025s) 去大胆地颠覆旧习，重塑一切吧。 — Nana Ahwoi ## Mentioned Stephanie Franklin-Thomas, Nana Ahwoi, 美敦力 (Medtronic), 安永 (EY), ChatGPT, Medtronic GPT, 世界经济论坛 (World Economic Forum), 《未来就业报告》 (Future of Jobs Report), Ai4 ## Video description Stephanie Franklin-Thomas, VP/Global Chief Information Security Officer, Medtronic Nana Ahwoi, EY Americas Consumer and Health Cyber Industry Group Leader, EY Abstract AI reveals what your business is (and is not) ready for. Broken processes produce broken automation – Start designing for what AI can do. Many organizations think AI will change their business. What it can do is reveal the flaws in how decisions are made, how data is governed and whether people are ready to change. AI doesn't fix strategy – it exposes it. Successful AI initiatives require redesigning core processes to align with how AI actually works. The real win is transformation. Join us to hear lessons from the field while reflecting on how YOU can challenge how things have always been to capitalize on how things CAN be in this era of AI. [Source](https://www.youtube.com/watch?v=tBibN48Zdsw)

Source quotes

[0:04] Nana Ahwoi: AI 不是你的战略——它是你的重置键与镜子 [3:52] Nana Ahwoi: 技术无法自动修复一个本质上存在缺陷的流程。 [3:52] Nana Ahwoi: Technologically enabled process that is still bad（被先进技术武装起来的糟糕流程） [6:14] Stephanie Franklin-Thomas: 如果数据不可靠，输出就不可靠；如果你的逻辑本身存在缺陷，AI 所做的仅仅是把这种缺陷逻辑放大而已。 [10:06] Nana Ahwoi: People are not going to adopt what they don't trust（人们绝不会采用自己不信任的东西） [10:06] Nana Ahwoi: build it with them（与他们共同构建） [12:31] Stephanie Franklin-Thomas: if you write it down, then you're serious about it（只要写下来，就代表你认真对待这件事了） [14:19] Stephanie Franklin-Thomas: you're calling my baby ugly（你在说我的孩子长得丑） [17:05] Nana Ahwoi: 去大胆地颠覆旧习，重塑一切吧。

Source quotes

[0:04] Nana Ahwoi & Stephanie Franklin-Thomas: AI is not your strategy, it's your reset button and your mirror [0:04] Stephanie Franklin-Thomas: Not exactly. [3:52] Nana Ahwoi: technologically enabled process that is really bad [3:52] Nana Ahwoi: AI was in the room today [6:14] Stephanie Franklin-Thomas: this is our process and we think AI fits here. [8:26] Nana Ahwoi: if your logic is flawed, all that AI does is amplify that flawed logic. [8:26] Nana Ahwoi & Stephanie Franklin-Thomas: garbage in, garbage out [10:06] Stephanie Franklin-Thomas: AI is our coworker [10:06] Nana Ahwoi: people are not going to adopt what they don't trust. [14:18] Stephanie Franklin-Thomas: you're calling my baby ugly. [14:18] Nana Ahwoi: corners of excellence

Original transcript

[0:04] Thank you for joining us for our talk [0:05] this morning on AI is not your strategy, [0:08] it's your reset button and your mirror. [0:11] I am thrilled to be presenting alongside [0:14] Stephanie Franklin Thomas. Um I'm Nana [0:16] Aoy. I'm a consulting partner. I focus [0:18] on cyber security and I also serve as [0:21] our America's consumer and health [0:23] industry group leader. [0:25] >> And I Good morning everyone. I am [0:27] Stephanie Franklin Thomas. I am the vice [0:29] president and chief information security [0:31] officer for Medronic which is a medtec [0:33] company you might be familiar with might [0:36] not but I am speaking on a solo [0:38] engagement late this afternoon so if you [0:39] want more information on that be on the [0:41] third floor it'll be very interesting so [0:43] highly recommend that uh quick [0:45] disclaimer our views are those of ours [0:47] alone and doesn't necessarily represent [0:49] the views of our entire firm whether [0:52] it's EY or Metronic um what we're going [0:54] to and you you just found out who we [0:56] were talking about. So, a couple things [0:58] that we're going to do today is just [1:00] talk through um this concept of of AI um [1:04] not being core to your strategy per se, [1:06] but really requiring process redesign. [1:08] We'll share some stories via Stephanie [1:11] on what she's experienced along the way [1:13] and then leave you with some with some [1:15] thoughts as well. We want this to be [1:17] engaging and so as we ask questions, [1:19] please continue to raise your hand. If [1:21] something comes to mind, let us know as [1:23] well. Um and so you know when we think [1:26] about the topic of AI and we think [1:30] through the introduction of chat GPT [1:33] release what spurred it all on right we [1:35] started talking about how a million [1:37] users jumped on it right away within 5 [1:39] days of it being introduced and I found [1:41] out fun fact that as of August 2025 so [1:44] as of right now there are 700 million [1:47] weekly users of chat GPT and that is [1:49] just an indication of how prevalent this [1:52] has become for for us and and that is [1:54] just a microcosm of what AI can do and [1:56] where it's headed. So Stephanie, when we [1:59] think about those stats, that means AI [2:02] has arrived. Everything's perfect. [2:04] Processes are going to be exceptionally [2:06] awesome, right? [2:07] >> Not exactly. Um so um in the role that [2:11] I'm I'm in at Metronic, you know, I have [2:13] responsibility for all of cyber [2:16] security. And as you guys know, probably [2:18] in the last few years, what tends to [2:20] happen is with cyber insurance and every [2:23] company wants to know about another [2:24] vendor or customer's cyber security [2:26] profile or posture. And so my [2:29] organization gets hundreds if not [2:32] thousands of questionnaires around what [2:35] do we do around security, right? And so [2:38] you know when when Natada talked to me [2:40] about this and you know she is a great [2:42] supporter of Metronic but when she [2:44] talked to me about this you know I was [2:46] telling her about our journey in terms [2:48] of how do we address all of these [2:50] questionnaires that we get in because [2:51] you know originally we had a bunch of [2:53] people that would just respond to all [2:54] the questionnaires and it's the same [2:56] questions over and over. Do you patch? [3:00] Do you address your vulnerabilities? All [3:02] the foundational um security stuff. So [3:04] what we thought using chat GPT or [3:07] metronic we have a metronic GPT we have [3:09] our own um uh AI but what we thought was [3:12] okay we're going to use AI to maybe [3:14] answer some of these questions. So we'll [3:16] train a model we'll get all the [3:18] information in there and instantly you [3:20] were going to be able to answer all the [3:21] questions that someone asked. [3:23] Unfortunately there were a few [3:24] downfalls. Um not everybody asks the [3:27] questions in the same way. Um, and then [3:30] there's this whole human element that I [3:33] don't think we fully accounted for. Um, [3:36] and I think we're going to talk a little [3:37] bit about that. So, I'll kind of defer [3:38] on that, but I would say that um, and [3:40] then data, you know, if garbage in, [3:43] garbage out. So, if your data is not [3:44] good, the data you get back, the answers [3:46] to the questions may not be accurate as [3:49] well. So, a little bit of a struggle [3:51] when we first started. [3:52] >> Yeah. So I want to see show of hands. [3:54] How many of you either within your [3:56] organizations or in a personal capacity [3:58] attempted to do something with AI? You [4:00] thought it was going to be brilliant and [4:02] then you got incredibly frustrated [4:04] because that it gave you false data. [4:06] Nobody wanted to use it. Yeah, I'm [4:08] seeing some head nods. Y [4:09] >> excellent. So it is it is not the end [4:11] all be all. And to Stephanie's point, [4:14] sometimes the outcomes that we're [4:15] looking for are not the first outcomes [4:16] that we're going to get. Um and so [4:19] there's a question that we wanted to ask [4:20] of the group. So if you have a process [4:23] that h that is full of manual [4:25] workarounds, um lots of manual [4:27] intervention, complicated workflows, how [4:30] many of you show of hands think that AI [4:32] can just fix it? You just push the [4:35] button and it [4:36] >> fixes it. [4:38] >> I think I saw like one hand. Oh, couple [4:40] hands. Okay. [4:41] >> Yeah. [4:43] >> Couple hands. Well, what what we have um [4:47] noticed uh in some of the work that [4:49] we've done within the firm as well, we [4:50] call ourselves client zero um because [4:52] we've we've significantly invested in AI [4:55] as a company ourselves and are working [4:58] with clients as they embark on their AI [5:00] journeys as well is when a process [5:02] really is full of work workarounds, [5:06] manual interventions [5:08] um and non-standard steps, [5:11] it's highly unlikely that you're going [5:13] to get the outcomes that you're looking [5:14] for right away. So, what we what we um [5:18] end up having to do is really look at [5:19] the underlying processes and ask [5:22] ourselves what needs to be different [5:25] with that process because and this is [5:28] not a a new concept of looking at the [5:30] underlying process before you slap on [5:32] technology to make it smarter because [5:34] you can have a very awesome technology [5:37] technologically [5:39] enabled process that is really bad. it [5:42] just you know you have technology on top [5:44] of it. Um but if you change the [5:46] fundamental process then you can get to [5:48] the outcomes that you are looking for. [5:50] Um and so really we need to think about [5:53] this in the construct of building it as [5:56] if AI was in the room today. What would [5:58] that look like? Should your 10step [6:00] process be a 10-step process that you [6:02] have AI working all in the background [6:04] for or could it be two or could it be [6:07] one? So you could fundamentally [6:09] challenge the way that things have [6:10] always been done to get to the outcomes [6:12] that you're looking for. So Stephanie [6:14] talked through the journey so far that [6:18] you've you've explored for the third [6:19] party process um and some of these [6:22] similar drivers, right, for why the [6:25] outcomes weren't exactly what you were [6:27] looking for. So what what have you done [6:30] differently? So, you know, and I kind of [6:32] alluded to kind of what our our I guess [6:35] top three issues were, but you know, I'm [6:37] going to I'm going to really kind of [6:38] double click on the people element of it [6:40] because I think what happened when we [6:43] started thinking about when we started [6:45] thinking about how we were going to to [6:47] use AI, the automatic disposition was [6:51] this is our process and we think AI fits [6:53] here. [6:55] The problem with that is we were already [6:59] engineering the AI pieces or we were [7:01] already solving the problem of where we [7:03] think AR AI could be when the reality is [7:06] we needed to look at it from end to end. [7:08] people were not really um I would say [7:11] ready to accept or ready to tap into [7:15] what AI was offering us in terms of the [7:17] the responses you know so I'm going to [7:19] digress for a minute so I'm sorry but [7:22] you know the first time I used um chat [7:24] GPT I was with my husband and the [7:26] questions that we asked we were very [7:27] very different people should we agree [7:29] upon nothing but somehow it worked but [7:32] the first question we asked it was about [7:34] our marriage and it actually came back [7:35] and said we should get divorced [7:38] um to this day he doesn't like to use [7:40] chat GPT right but so out of that I mean [7:44] I relate that to this by saying he's one [7:47] of those that could put something in and [7:50] ask for advice or whatever depending [7:52] upon his prompt he's always suspect [7:54] around what he gets so he is not a [7:56] person that's going to accept and I feel [7:58] like I got some of this from my team and [8:00] and they were very specifically saying [8:02] this is the this is in end this is a [8:04] process we only want to use it here. But [8:06] I think that really kind of detracts [8:08] from the value of what we could drive [8:11] from AI. So that was kind of what the [8:14] thing that we had to go over go through [8:15] first. I think people is are critical to [8:19] understanding how we can best use AI in [8:22] our environment and how we can improve [8:24] our processes. [8:26] >> Absolutely. [8:27] And the moral of the story really is [8:30] right if if the data is not sound your [8:33] output is not going to be sound. If your [8:35] logic is flawed, flawed, all that AI [8:37] does is amplify that flawed logic. And [8:40] again, what we're talking through, you [8:43] might be sitting thinking, well, we all [8:44] we all know that, but it's not [8:46] manifesting itself in the way that we're [8:47] thinking about AI and how we drive value [8:49] for our organizations and even in our [8:52] own individual experiences. Um, so a [8:55] couple of key call outs that we wanted [8:56] to drill down um into. The first is [9:00] around the uh clarity around the process [9:04] right and the decisions that you need to [9:05] you need to make because if your [9:08] strategy um is flawed it's going to be [9:11] difficult for the for the AI capability [9:14] that you're building to really again [9:15] give you the outcomes that you're [9:16] looking for. So your decisions have to [9:18] be teachable. If I pulled everyone in [9:20] this room and we all gave vastly [9:22] different responses and we try to train [9:25] a model on that set of responses, you're [9:27] going to have inconsistencies in in the [9:29] in the um outcomes when you work with [9:32] the with the AI model. Um the second [9:35] really comes into play with what [9:37] Stephanie was saying, prioritizing data [9:38] quality. Again, old adage, garbage in, [9:42] garbage out, right? Whatever you're [9:44] training your your model on, it's going [9:46] to spit out exactly what you train it [9:48] on. So, we have to make sure that the [9:50] underlying data that we're using to [9:52] build um the model and train the model [9:55] is is sound. Um and Stephanie, maybe [9:58] I'll pull you in with the redesigning [10:00] with AI in the room, which gets to the [10:02] concept of thinking about this with a [10:04] almost a blank sheet of paper, [10:06] >> right? And I again I taking that back to [10:08] the people. I think that that's the [10:09] biggest struggle in all of this. So um [10:12] you know in order for us and and [10:13] ultimately we were successful and be [10:15] able to being able to drive improvement [10:17] in the area that we were seeking but it [10:19] really required us to kind of take a [10:21] state back step back and not really give [10:24] AI constraints but really um change our [10:27] prompts and change the way they were [10:29] looking at it so it would tell us where [10:31] it saw the opportunities. Now you know [10:33] for my team I didn't say well because AI [10:36] is our coworker and it's going to look [10:38] like this. um rather we took that [10:40] information in and then we began we [10:42] began to talk about it and think about [10:44] how we could change our frame of mind in [10:46] terms of bringing AI in. So I think that [10:49] is probably the biggest struggle um when [10:51] it comes to the adoption of AI because [10:53] we as humans were the top of the food [10:55] chain. You know we're we we think we're [10:57] the smartest and we are right. There's a [11:00] lot of really smart people, but when you [11:01] hear a new idea, it takes a minute to [11:04] kind of take a step back because we I [11:06] don't know, in my house, we think we [11:07] know it all and we don't, right? We have [11:09] to take that information and and really [11:12] begin to digest to see how it can really [11:14] improve our thinking. Yeah. And and we [11:16] already talked about fixing the process, [11:18] then training the model. But that last [11:20] point on the people part, it cannot be [11:22] emphasized enough, right? People are not [11:25] going to adopt what they don't trust. [11:26] And so instead of building for a [11:28] stakeholder group and assuming that you [11:30] know what they're looking for. We talked [11:32] a bit about that when we were preparing [11:34] for this session, you really want to [11:36] build it with them. So in involve the [11:38] stakeholders in the process because it [11:40] gives them an opportunity to challenge [11:42] their thinking. There was an interesting [11:44] stat in the World Economic Forum. They [11:46] released a future of jobs report. It's a [11:49] it's a nice long read, but one of the [11:52] things that the Roy highlight is that [11:53] 60% of employers believe that their [11:56] organizations are going to be [11:57] significantly transformed through the [11:59] advent of digital access um spurred on [12:02] by technologies like AI, right? And when [12:05] you think about that, that means um the [12:08] personnel that are in these [12:09] organizations are in some way expected [12:11] to come along. And another another fact [12:14] that they called out is there there is a [12:16] bit of reluctance and fear around what [12:18] this means for the individual in their [12:20] process. So again in order to drive [12:23] adoption of what you're building we have [12:26] to make sure we're engaging the end user [12:27] and the stakeholders um along the [12:29] journey we have a bit of um homework. We [12:32] wanted this to be additionally more [12:34] interactive. So we want everybody [12:36] >> well so wait what we're going to do on [12:38] the homework right. So, when I saw the [12:41] homework, I said, "Okay, well, anybody [12:42] my age and and above is probably going [12:44] to pull out a sheet of paper and try to [12:45] write down the response." Um, anybody [12:47] who is like my son's age will pull up [12:50] their phones and get into notes and then [12:52] answer the question. So, I'm going to [12:53] ask you guys to do that. So, whatever [12:55] you feel comfortable with, I'd really [12:56] like for you to pull up whatever device [12:58] or pull out a sheet of paper or a note [13:00] and really begin to write down because I [13:02] I I'm a believer if you write it down, [13:04] then you're serious about it, right? you [13:06] really begin to take on whatever it is [13:09] that you're writing down. So, I will [13:10] turn it over to you to ask the question, [13:12] but I'm encouraging you guys to actually [13:14] pull out your notes and commit to [13:17] something on on digital paper or paper [13:20] and we'll see what comes out from the [13:21] crowd. I think we we'll have a little [13:22] bit of time to get a question or two. [13:26] >> Yes. [13:27] >> Excellent. So if you think about your [13:29] business a business process and when we [13:32] were when we were talking we said we [13:34] often will look at someone else and say [13:35] well your process needs to be fixed not [13:37] mine. So we want you to challenge is [13:39] there a business process that you're [13:40] either a part of or involved with in [13:43] some way and what part of that needs the [13:47] reset the most before you look at [13:48] leveraging AI to fix it. And a couple of [13:51] thoughts and questions as you think [13:52] through that like what assumptions would [13:55] you challenge about the process [13:58] like what steps might you completely [14:00] eliminate in the process? [14:02] And the last piece is what what logic [14:04] would you need to clarify in that [14:06] process? So we'll give you just a couple [14:08] of minutes. I do not have um elevator [14:10] music to play while you think about your [14:13] response. [14:13] >> I do have a mic so I could sing. [14:15] >> Oh no, [14:15] >> no. I Thank you. Thank you. Um we do [14:18] have actually Nana we do have a question [14:19] over here. [14:20] >> Oh excellent. Well the question was when [14:22] you have an established organization [14:24] with legacy processes what are some of [14:27] the skill sets that you'd want to see in [14:29] the leader or leaders who are aligned to [14:32] move the business forward. Right. So [14:35] I'll throw it to you. [14:36] >> Oh okay. [14:37] >> And then um I can add anything else. Um [14:39] so you know when you were asking the [14:41] question the first thing um I was [14:43] thinking was about just having an ind or [14:46] an individual or set of individuals that [14:49] are um I guess agile in their thinking [14:52] right um acceptance of new the new right [14:55] because anytime you have a legacy [14:57] process someone created that process and [14:59] what I run into because I have a lot of [15:01] longterm [15:03] folks that have been in my organization [15:05] and because they created the process and [15:06] if I come in and say I want to change it [15:08] they're like you're calling calling my [15:09] baby ugly and that's not it right so I [15:12] always use the analogy around you know [15:14] we used to create fire with flint we [15:16] don't do that anymore well unless you're [15:18] a survivor type person but you know [15:20] trying to show them where they have [15:22] accepted change um in order to think [15:25] about how you can change something for [15:26] the future so that what I would be [15:28] looking for truly is is creating a [15:31] mindset and having them understand where [15:34] change has been good for them because [15:35] change is not all bad right so it's It's [15:38] having that agile mindset to kind of [15:40] accept what's what's going out there, [15:42] willing to test, willing to listen. [15:43] Those are all really valuable skills in [15:45] my mind. [15:46] >> 100%. I don't know one more to add [15:48] beyond that. Definitely openness, right? [15:51] Because you want to bring the input from [15:53] the team and making sure that you have [15:55] you have the willingness to to challenge [15:57] and change what that looks like. If [15:58] someone's stuck in stuck in their ways, [16:01] if you will, it gets quite challenging [16:02] to imagine what could be different. The [16:05] other thing I would say is um going to [16:07] going along the the lines of like having [16:10] an agile mindset is also looking at the [16:12] team that the leader brings along and [16:14] the leader is going to have to challenge [16:15] am I going to the same set of people all [16:18] the time, right? Because there might be [16:20] group think and familiarity and the [16:21] ideas that come out. So you you might [16:24] have a corner of excellence that is not [16:26] yet engaged in the process. So really [16:28] challenging again the composition of [16:30] who's involved in the process and [16:31] staying open. You know, one tip, I'll [16:34] give you one other tip that I personally [16:35] use is when I'm thinking about a process [16:37] because I've been, this is not my first [16:39] rodeo. I've done this a while. Um, when [16:41] I get an intern group, I just throw [16:43] things in the air and see what they tell [16:46] me. And sometimes they tell me stuff [16:48] that, you know, I walk away, I'm like, [16:49] "Oh, that's never going to happen. They [16:50] don't get it." But it happens and they [16:52] get it. So, I it may just be a little [16:55] nugget that I need to take away from it. [16:56] But I think young minds are incredible. [16:58] >> Absolutely. Um, and so I if I have a [17:01] chance to pitch there, I think that [17:02] that's good for a leader as well. [17:05] >> Thank you. Any other questions or anyone [17:08] want to share maybe one person share [17:11] what you wrote down for the process that [17:13] you think might need a bit of a [17:16] Thank you very much. So she talked about [17:18] organizational change management as one [17:20] of the core facets and making sure that [17:22] the broader organization is being read [17:24] in on what's happening, right? because [17:26] it's not going to be isolated. If you're [17:28] trying to drive fundamental change, [17:29] you're going to need to make some [17:30] adjustments. So, appreciate the [17:33] engagement today. I think that this is a [17:35] topic that we'll continue to see evolve. [17:38] Someday we'll have our co-workers not [17:40] necessarily sitting next to us [17:41] physically, but certainly on our [17:43] computers and laptops. So, um we'll [17:45] leave you with this. To totally [17:47] transform AI, organizations need to [17:49] reimagine their business models, right? [17:51] With AI along the critical path. So the [17:54] question to go back and ask yourselves [17:56] is what are you going to do with that [17:58] reset button when you're looking in the [18:00] mirror? Are you going to leave things as [18:01] they are or are you going to challenge [18:03] yourselves to do things differently? So [18:06] go forth and do everything differently [18:08] and challenge the status quo. Thank you [18:10] very much. [18:10] >> Thank you. Thank you.