董事会期望从财务AI战略中获得什么：Joyce Li

如果您想通过收听节目获得CPE学分，请访问earmarkcpe.com/fpa。

If you would like to earn CPE credit for listening to the show, visit earmarkcpe.com/fpa.

下载应用程序，完成简短测验，即可获取您的CPE证书。

Download the app, take a short quiz, and get your CPE certificate.

最后，如果您喜欢今天的FPNA节目，请前往您常用的播客平台，点击订阅按钮，并为节目评分和留言。

Finally, if you enjoy listening to FPNA today, please go to your podcast platform of choice, click the subscribe button, and leave a rating and review of the show.

现在，节目正式开始。

And now, onto the show.

这里是数据轨道，为您带来今天的FPNA节目。

From data rails, this is FPNA today.

欢迎收听今天的FPNA节目。

Welcome to FPNA today.

我是主持人格伦·霍珀。

I'm your host, Glenn Hopper.

今天在FP and A Today节目中，我们邀请到了埃瓦伦达合伙公司的CEO兼首席AI战略师乔伊斯·李。

Today on FP and A Today, we're joined by Joyce Lee, CEO and chief AI strategist at Evarenda Partners.

Joyce为我们的对话带来了罕见的专业组合。

Joyce brings a rare combination of expertise to our conversation.

她既是CFA持证人，又拥有计算机科学学士学位及沃顿商学院的MBA学位。

She's both a CFA charter holder and a computer science graduate with an MBA from Wharton.

在她的职业生涯中，她曾共同领导过数十亿美元的投资策略，为全球金融机构提供咨询，如今致力于为董事会和高管层提供AI战略、治理及负责任应用方面的指导。

Over her career, she's co led multi billion dollar investment strategies, advised global financial institutions, and now works with boards and c suites on AI strategy, governance, and responsible adoption.

她还担任OpenBB的顾问委员会成员，并合著了雅典娜联盟AI治理手册，帮助财务领导者应对快速变化的AI世界。

She also serves on the advisory board of OpenBB and coauthored the Athena Alliance AI Governance Playbook, helping finance leaders navigate the fast changing world of AI.

Joyce，今天能邀请到您真是我们的荣幸。

Joyce, it's a real pleasure to have you with us today.

欢迎来到我们的节目。

Welcome to the show.

很高兴能与你交谈，Gwen。

It's great to speak with you, Gwen.

这几个月来认识你真是非常愉快。

It's been it's been great getting to know you over the months.

我知道自从我们第一次交谈已经过去几个月了，每次我们谈话时，我都必须说我非常欣赏你的背景和专注点，既是CFA持证人，又是计算机科学毕业生，还拥有沃顿商学院的MBA学位。

I know it's been a few months since we first spoke, and I just every time we talk, I just I have to say I absolutely love your background and focus as being both a CFA charter holder and a computer science graduate and that MBA from Wharton.

我觉得你拥有近乎完美的资质组合。

I feel like you have the kind of the perfect package.

我想知道，凭借这种方法和洞察力，你会如何描述这种独特组合如何塑造了你的职业道路？

And I'm wondering with that approach and insight, how would you say that unique combination has kinda shaped your career path?

你能大致为我们介绍一下你的教育背景和职业经历吗？

Can you sort of walk us through both your educational and professional background?

好的。

Yeah.

当然。

For sure.

我最近一直在思考这个问题。

I've been thinking about that a lot.

我职业生涯至今的主线是什么？

What's the through line, of my career so far?

因为，显然，如果从外部看，它们有三条发展轨迹。

Because, clearly, if you look outside in, they, it has three tractors.

第一条是计算机科学、工程学，为金融机构做分析工作。

The first one is computer science, engineering, doing analytics work for financial institutions.

那是我早期的职业生涯。

That's my early career.

然后我转换赛道，成为一名分析师，之后成为投资组合经理，管理包括不同领域和各类公司（如对冲基金、纯贷款基金、共同基金、ETF等）的投资。

And then I switched lane and then became a, analyst and then portfolio manager managing investments, including in different, verticals, and also type of firms such as hedge funds, loan only, mutual funds, ETFs, you name it.

而现在，我几乎又回到了技术领域，从事人工智能、金融与治理交叉领域的工作。

And then now I'm doing almost like the back to a little bit more technology and doing the intersection work, between AI, finance, and governance.

我总结出的贯穿线是：我始终保持着好奇心，以及工程师那种'总能做到'的态度。

And the through line I came up with is I just have this curiosity and engineer sort of can do attitude.

我觉得每个足够有趣的挑战，总有一个解决方案在等着我们去发现。

I feel there's always a solution to a challenge that's interesting enough waiting for us to solve.

在我职业生涯早期，那个挑战可能是如何将数据引入金融系统，帮助人们做出决策。

That challenge, early part of my career, maybe how to bring the data into, you know, some of the finance suites, help people to make decisions.

中期阶段则是思考如何为股东和投资者创造价值，帮助他们发现优秀的公司？

In the middle would be, like, how do I create value for our shareholders, our investors to discover great companies?

当这些公司的市值增长时，投资者也能从中获益。

And when they grow in their, market value, the investors benefit as well.

而现在更像是搭建桥梁——帮助人们理解技术，并将其转化为解决实际问题的方案，无论是商业模式的创新、释放员工潜能的策略，还是重组投资团队架构的决策。

And now it's more like a bridging between, how people should look at technology and how they can translate that into the problem they are solving, which may be business model, may be, how to unlock the potential of their their, labor force, or maybe, just simply investment decisions on how to restructure their investment team structure as well.

我觉得这条贯穿始终的主线既带给我无限乐趣，也塑造了我这种'迎难而上'的态度。

So I feel like that is just a through line that gives the give me a lot of fun and that can do attitude.

你也可以说这是一种天真的态度——让我对所有机会都说'好'，并乐在其中。

You can also say very naive attitude for me to just say yes to a of these opportunities and have a lot of fun with it.

是啊。

Yeah.

得益于生成式AI以前所未有的深度融入我们所有领域，现在确实是技术发展的黄金时代。

And it is such an interesting time with technology, thanks to generative AI being so integrated into everything we do at a level that it it never has been before.

因为过去真正运用技术做实事的主要门槛，确实在于编程能力。

Because, truthfully, in the past, that the barrier to access of being able to do real things with technology was the ability to code.

当然，凭借你的计算机科学背景，你早已具备这种能力。

And of course, with your computer science background, you already had that.

但现在，那些以前不会编程的人也能通过可视化编程接触Python等语言的强大功能，并将其应用于日常工作中。

But now, people who couldn't code before, they they can get into vibe coding and access the the power of Python and whatever other languages they're using in their everyday job.

不过我在想，显然我们接下来会重点讨论人工智能。

I'm wondering though, and we're gonna talk a lot about AI, obviously.

但你在管理投资组合或从事计算机科学以外的其他工作时，是否曾想过可以将其自动化，或者建立模型、进行投资组合平衡之类的操作？

But were you tempted or did you ever, when you were managing investment portfolios and and doing your other work outside of computer science, did you ever think, I can maybe automate this or do some modeling or some portfolio balancing or whatever it is?

你在进行投资组合管理时，是否考虑过编写程序或应用计算机科学技术？

Did you ever think about writing programs or did you apply the computer science when you were doing portfolio management?

噢，当然经常这么想。

Oh, definitely all the time.

虽然我可能并不总能做好，但这始终是我会向编程能力更强的同事提出的问题，尤其是在处理海量数据或复杂建模技术时，总认为有更好的技术手段可以利用。

I may not always been doing it well, but this is always one of the questions I will ask for my colleagues or who who are much more skills in programming, especially in the past, like dealing with big amount of data or highly complicated modeling techniques, are always been assuming there's a better way to leverage technology to whatever we're doing.

我给你举个例子。

So I'll give you an example.

我们投资团队过去研究的一个问题是如何获取非结构化数据。

One of the things that we looked at as a investment team in the past was how do you sort of get the unstructured data?

当然，那时候还没有生成式AI。

Of course, at that time, there's no Gen AI.

所以很多时候，我们会去参加贸易展会或查阅政府备案数据库，获取各种非标准数据源。

So a lot of times, we went into trade shows or we went into government, sort of filing, databases and get all these, nonstandard data sources.

有趣的是，正是这些地方，只需一点技术手段就能挖掘出许多有价值的洞察。

And interestingly, that's where a little bit of technology can do a lot of mileage to uncover some of the interesting insights.

当我执行多空策略时，我们实际上通过这种方式发现了许多虚报财务数据或可疑的商业行为。

And when I was doing long short strategy, we actually discover a lot of sort of inflated financial claims or, just maybe some of the, you know, questionable business practices by doing that.

可以说这种能力——始终追问'能否用现有技术以不同方式处理问题'——对我的职业生涯大有裨益。

So I would say that benefited my, career a lot, the ability of always asking, even just asking the question, can we do something differently with the technology available to us?

我真心希望能说服或鼓励其他人也这样思考。

And I I really look forward to convince or encourage everyone else to think about that.

事实上，这或许也是你经常在做的事——鼓励人们思考：当今技术能如何放大我的能力，或帮我发现未知的新事物？

And in fact, that's maybe one of the things that you also do a lot is, like, encourage people to think about what can I do with technology these days that can either multiply my ability to do things or maybe discover something that I didn't know before?

是的。

Yeah.

完全同意。

Absolutely.

我想我明白我们是在与高层领导以及财务和会计专业人士对话。

And I think I understand we're talking to senior leadership and and finance and accounting professionals.

我并不是说那些在其他领域积累了专业经验的人需要转行成为机器学习工程师或重新获得计算机科学学位。

And I'm not saying that anyone who's carved out a career with domain expertise in another area that they I'm not saying they need to go become a machine learning engineer or get a new degree in computer science.

但我认为重要的是，如果我们要使用人工智能，至少要在某种程度上理解其底层原理。

But it isn't I think it's important that we understand at some level what's going on under the hood if we're going to use AI.

我想到你现在所处的位置，我在思考——别误会我的意思。

And I I think about where you are right now, and I'm wondering and don't get me wrong.

我很高兴看到现在有这么多财务领导者，实际上所有行业和职业的人，都在积极拥抱生成式AI并越来越擅长使用它。

I love there are so many finance leaders right now, and just across all professions and all industries, are really leaning into generative AI and are getting very good at using it.

但很多时候，他们并没有花时间去深入了解原理，比如说他们已经摸索出了一套有效的提示策略。

But a lot of times, they don't take the time to look under the hood and say, you know, they've figured out a good prompting strategy.

他们已经找到了可以交给AI处理的好任务，但并不理解其底层运作机制。

They've figured out good things that they can offload to AI but they don't understand what's happening underneath the hood.

凭借工程师的思维方式和计算机科学背景，你的理解确实会更深入些，或者说要深入得多。

And with that engineers mindset and computer science background, you do have a little bit better of a read or a significantly better read, would say.

因为你多少了解其中的工程原理。

Because you understand sort of the engineering that's happening.

当你与领导层、董事会或任何对部署AI感兴趣的人交流时，你如何划定这条界限？对于我们需要掌握多少技术知识而非仅仅成为优秀用户，你有什么看法？

When when you're talking to whether it's leadership or boards or anyone that's interested in rolling out AI, where do you draw that line and what do you what are your thoughts on how much on the technical side we need to know versus just being good users?

是的。

Yeah.

这确实很有趣。

It's really interesting.

我们一直在讨论生成式AI这个话题。

We've been talking to about, you know, Gen AI.

为什么它与过去的技术进步不同？

Why is it different from past technology advancements?

我知道你对这个问题也有很强烈的看法。

And then I know you have a strong opinion on this as well.

我认为生成式AI对企业领导者或任何人来说都非常容易上手。

And I believe Gen AI is really easy for a business leader or anyone to get get onto.

比如，他们可以立即开始使用。

Like, they can start using it.

他们可以开始编写提示词，甚至可以利用他人创建的提示词库，已经能做出令人惊叹的事情。

They can start prompting, and then they can even create this or leverage on the prompt library other people create and really be doing amazing things already.

然而，学习曲线就在这里停滞了。

However, the curve stopped there.

如果你想创造真正能释放商业战略价值的思考，就必须再深入一步。

So if you want to create, you know, truly value unlocking business strategy type of thinking, you have to go a step deeper.

你必须持续使用并不断思考：好吧...

You have to keep using it and keep thinking, oh, okay.

其他人还在用它做什么其他事情？

What are the other things that other people are using it for?

有时候它并不需要与你的业务功能直接相关。

And sometimes it doesn't have to be directly related to your business function.

有时候仅仅是在生活中使用它，可能会激发你在工作中应用的灵感。

Sometimes it's just the idea that, you know, you use it at life could be sparking a idea that you can use it at work.

这更像是一种思维方式，当你每天使用它时，你会逐渐扩展自己的思维边界——虽然我不确定这个比喻是否恰当。

It's more like that mindset that once you use it every day, you sort of you expand your sphere of of maybe I don't know the right analogy.

但基本上，想象你处于一个球体的中心，你用得越多，这个球体的表面积就会扩大。

But, basically, if you think about you are sitting in the middle of a sphere, the more you use it, your sphere surface is gonna expand.

你会产生更多有趣的想法，同时也会培养出这种品味。

You're gonna have more interesting ideas, and, also, you create you create this taste.

我知道这听起来有点抽象，但请耐心听我说完。

I know it's a little bit, like, fluffy, but bear with me.

我确实认为对于商业领袖而言，很多时候我们的第二层思维是基于第一层思维发展而来的。

I do think for business leaders, a lot of times, we develop that second level thinking based on our first level thinking.

所以对于传统商业领域里什么是好主意、什么不是好主意，我们已经形成了这种判断品味。

So that taste of what is a good idea, what's not a good idea in traditional business domains, we're so used to it.

但在生成式AI领域，如果你能同样运用这种思维方式，就能更有信心地判断哪些AI项目值得公司考虑，哪些噪音其实与业务核心竞争力无关而应该放弃。

But on Gen AI, if you can think about it similarly, you're gonna be able to be much more confident on determining what is the right AI initiative that your companies should consider, what are some of the noise that is not really related to your true competitive edge of this business, therefore, you should pass.

所以我确实认为这种判断力需要通过日常不断实践来培养，尽管你无法预知哪次使用会带来百分百的天才创意。

So I do think that taste has to be developed by just grinding through all these daily usage even though you don't know which one will give you that, you know, that a 100% genius idea.

嗯。

Yeah.

另一方面我们经常讨论，既然能在生成式AI中进行各种计算预测而非Excel，Excel会被淘汰吗？

And the other part of it so we talk a lot about if you can do, you know, all these calculations and forecast in generative AI, not in Excel, is Excel going anywhere?

我个人认为不会，我们经常讨论这个话题。

And I don't I don't think it is, and we talk about it all the time.

无论你是财务、数据科学、商业智能还是其他领域从业者。

I don't care if you're in finance or or data science or or BI or whatever.

它确实是进行数据分析的完美工具。

I mean, is just it's a perfect format to to do data analysis in.

回想我职业生涯早期，可以说是Excel高手出身，当时为自己能编写各种公式感到非常自豪。

And I think about my early career, I kinda came up, you know, being an Excel warrior and really proud of all the formulas I could make and all that.

但要知道，我已经担任首席财务官二十多年了，现在不怎么用Excel了。

And but it's, know, I was I was been a CFO for a couple of decades now and I don't do as much in Excel anymore.

我所有这些话的重点是，如果你懂得如何操作数据，无论格式是Excel、R还是其他平台，你就能培养出更工程化的思维方式。

But and and my point with all this is, if you understand ways to manipulate data, if the if the format is Excel or or if it's R or whatever platform you're in, then you have more of an engineering mindset around it.

你会思考什么是可能的，无论是否在编写公式，你都知道能得到什么结果。

And you think you think about what's possible and whether you're writing the formula or not, you know the outcomes you can get.

我觉得现在这种氛围编程很酷，只要对着生成式AI说话就能让它帮你开发应用什么的。

And I sort of think all this vibe coding right now, it's really cool if you can just talk to generative AI and have it build an app for you or whatever.

但要考虑一个生产级应用需要理解什么才能更智能地给出提示——即使你不是优秀程序员，只要知道结构，明白条件循环原理，清楚自己想要什么样的整体架构，就能更好地引导提示词。

But to think about a production ready app and what you need to understand about it, to be able to prompt more intelligently, even if you're not a great coder, if you know constructs and you know this is a conditional loop and this is how it works and this is the way that I'm gonna do this, and this is the, you know, sort of the the over overall architecture I want, then you can guide the prompts better.

所以我认为在我们的职业生涯中，首先仍需要保持专业领域的专长。

So I think for us in our careers, yes, we still need to have that domain expertise, and that's thing one.

如果AI能让工作更轻松，那当然很好。

And if it gets easier through AI to do our job, that's great.

但我们仍需知道该问什么问题，以及如何构建和引导提示词——无论是思维链还是其他方式。

But we still need to know the questions to ask and how to structure and guide the prompts if it's a chain of thought and all that.

所以我不确定。

So I don't know.

我想说的是，你现在有在尝试氛围编程吗？

It's it's gotta be I guess, all that to say, are you doing any vibe coding right now?

你有在让AI帮你编写代码构建什么项目吗？

Are you building anything that you're having AI write write code for you?

有的。

Yes.

确实。

Yes.

就像我说的，如果我不亲自尝试，怎么知道该问什么问题呢？

Again, like, if I don't do it, how do I know what questions to ask?

有趣的是，我想先延伸一下你关于Excel建模的观点，然后再谈谈氛围编程。

And interestingly, I actually I I wanna extend your your comments on the Excel modeling a little bit, and then I go into the vibe coding.

我以前经常研究很多上市前的公司。

So I used to look at a lot of pre IPO companies.

当他们进行IPO时，有一个步骤你可能很熟悉，但为听众解释一下，投行分析师卖方会建立他们的模型。

And when they come IPO and there's a step you probably are familiar with, but just for your audience, the the sell side of investment banking analysts will have their model build up.

而对我们来说，在IPO前，我们也会基于与管理层的沟通加上财务申报文件，建立我们自己的模型。

And then for us, before the IPO, we're also based on the communication with management plus the financial filings, we also build up our own model.

所以总会有一场会议让我们比较各自的假设，我会质疑或我们团队成员会质疑他们的假设。

So there's always this meeting where we'll compare our assumptions, and I'll question or our team members will question their assumptions.

他们可能会试图为自己的假设辩护，这种来回讨论会让双方各自做出决定。

They may try to defend their assumptions, and that back and forth will will make each side have their own decisions.

这种提出关键问题的能力会影响模型的结果。

That ability to ask what's key questions that will influence that model.

你不会问那些会影响某个单元格进而影响另一个公式的微小细节。

You wouldn't ask, like, what's the, you know, tiny little detail that will affect the cell that will lead to another formula.

你想问的是那些问题。

You wanna ask that.

你要问最关键的控制因素。

You ask the most important levers.

但你如何发现这些关键问题呢？

But how do you discover that?

你如何发现所有这些杠杆因素？

How do you discover all these levers?

它们如何协同运作？

How do they work together?

当然，通过Excel公式的形式呈现，但为什么一个因素会引发另一个因素，以及为什么假设差异会导致巨大不同，这是需要时间积累才能理解的。

Of course, in the form of Excel formulas, but why one will lead to another and why the assumption difference will make a huge difference here is developed over time.

我认为这种工程思维——或者你称之为分析师思维——确实需要比实际建模更高层次的认知能力。

And I I do think that engineering mindset or whatever you call it, analyst mindset, that requires a a little bit, like, more literacy than, you know, maybe not actually building the model.

其实我也很久没有亲自建模了，但我清楚该问什么问题。

I haven't actually building the model also for a long time, but I know what to ask.

我只需看十分钟模型，就能知道该从哪些漏洞入手质疑。

I I know within ten minutes of looking at model, I know what are some of the holes I will poke on.

对吧？

Right?

所以进入氛围编程吧。

So and go into the vibe coding.

氛围编程重新开始，上手超级简单。

Vibe coding to get start again, getting started is super easy.

几分钟内你就能发现些真正惊人的东西，而且你尤其可以向孩子炫耀你绝对站在技术前沿。

Within minutes, you'll find something really amazing, and and you can especially show off to your kid that you are you are absolutely in on the cutting edge.

这非常简单。

That's very easy.

然而，一旦你到达那一步，该如何测试呢？

However, once you get there, how do you test?

又该如何评估呢？

How do you evaluate?

这就是为什么我认为如今人们倾向于说评估是关键，我认为对董事会成员或高管层来说尤其如此。

That's why I think nowadays people tend to say eval is the key, I think, for board members or for c level executives.

这才是关键标准。

That's the key criteria.

但除非你理解或至少具备对人工智能达到一定认知水平的好奇心，了解AI能做什么，以及AI可能破坏什么或制造什么问题，否则你将无法提出这些直接影响决策的关键问题。

But unless you understand or at least have that curiosity to learn to a certain level of literacy around AI, what AI can do, and what AI might be able to destroy or create problems, you wouldn't be able to ask these very targeted questions that will influence your decision making.

因此，董事会成员和高管们持续学习人工智能的能力与潜在风险之间的关联至关重要。

So that linkage between the capabilities, the potential risk is very critical for board members and and executives to keep learning.

今天的FP&A内容由DataRails为您呈现，全球排名第一的FP&A解决方案。

FP and A today is brought to you by DataRails, the world's number one FP and A solution.

DataRails是专为Excel用户打造的人工智能驱动的财务规划与分析平台。

DataRails is the artificial intelligence powered financial planning and analysis platform built for Excel users.

没错。

That's right.

你可以继续使用Excel。

You can stay in Excel.

但不同于每次预算、月末结算或预测时都要面对地狱般的折磨，你可以享受数据整合、高级可视化、报告和AI功能的天堂，外加改变游戏规则的洞察力，即时获得答案并在几秒内生成你的故事。

But instead of facing hell for every budget, month end close, or forecast, you can enjoy a paradise of data consolidation, advanced visualization, reporting, and AI capabilities, plus game changing insights giving you instant answers and your story created in seconds.

了解为何超过一千个财务团队使用数据轨道来揭示他们公司的真实故事。

Find out why more than a thousand finance teams use data rails to uncover their company's real story.

不要取代Excel。

Don't replace Excel.

拥抱Excel。

Embrace Excel.

了解更多信息，请访问datarails.com。

Learn more at datarails.com.

所以我对你目前在市场上观察到的情况非常好奇。

So this I'll be very curious to hear what you are seeing in the market right now.

去年大家都在谈论AI，但没人有预算。

So last year, everybody was talking about AI, but nobody had budget.

今年我们仍处于Gartner炒作周期的高峰期，围绕生成式AI。

This year, we're still we're flying off the edge of the Gartner hype cycle on on generative AI.

不过我认为，我们开始看到一些迹象表明它正滑向幻灭的低谷。

And I think though, we're starting to see a little bit of that slipping into the trough of disillusionment.

关于这一点最有趣的是，我认为这并不是因为技术上的任何不足。

And the super interesting thing around this, I don't think it's because of any letdown in the technology.

技术发展快如闪电，而且每天都在进步，前沿供应商之间的军备竞赛不断推出新功能，一切进展之快令人难以置信。

The technology is moving lightning fast and, you know, it's it's better every day and there's new features for the arms race between the the frontier providers is is insane how how quick everything's happening.

但已有研究表明——这时候我就特别羡慕那些大型播客，他们有现场制作人可以随时查询数据，而我既没人帮忙也没提前准备。

But there have been studies and this is one where is a time where I get jealous of the really big podcast that they have like a producer in the room and they can ask for the stats and get all that since I don't have anyone and I didn't prepare ahead.

你知道的，我们讨论过这个问题，我也撰文分析过。

You know, we've talked about it and and I've written about it.

目前媒体上充斥着人工智能项目失败的负面报道。

There's a lot of noise in the media right now about AI projects failing.

但你我心知肚明，绝大多数时候——我敢说绝大部分情况——问题并不在于技术本身不行。

And you and I know the reasons for those and a lot of a lot of times it's I would say the bulk of the time, it's not because the technology it's bad.

而是因为他们设定的目标脱离了当前技术发展水平，根本不切实际。

It's because maybe the desired outcome, what they were going out for was unrealistic given where technology is today.

不过我想知道你目前在市场上观察到什么现象，因为我感觉...从我掌握的信息来看，这种推动力依然存在。

But I'm wondering what you're seeing in the market right now because I there's this I think there's still that push from what I'm saying.

自上而下的'必须搞人工智能'压力仍然存在，尽管连'搞AI'具体指什么都不明确。

There's still that push from the top down of we have to do AI with nebulous unclear what do AI means.

因此，投资者、董事会和高层管理人员将这种压力传导给团队，要求他们进行人工智能项目，却没有明确的目标和预期成果。

So that, you know, investors, boards, senior management is pushing that down to their teams, telling them to do AI without clear goals and outcomes.

而我们这些夹在中间的人就很困惑，你到底要我拿这个做什么？

And then kind of those of us stuck in the middle are, well, what do you want me to what do you want me to do with this?

我可以试着在这里做点什么，随便吧。

I'm I can try to do it here, whatever.

所以这类大型资本投资项目正在停滞、放缓、无法取得成果，我认为原因正是我刚才谈到的——人们既不了解技术，也没有围绕它设定清晰目标。

So these sort of big capital investment projects are stalling, slowing down, not coming to fruition because of what I I think, because of what I was just talking about where people don't understand the technology and don't have clear goals around it.

但我们确实在前线员工使用工具的场景中看到了更多成功案例和效率提升，有些是企业明确许可的，有些则是员工私下用个人账户进行的影子AI项目。

But where we are seeing more success and more efficiencies happening at the front row or at the front lines rather, where employees are using the tools, sometimes with explicit permission from their companies and sometimes they're just doing shadow AI off on their own and and using their own personal account.

不过现在我要把话题拆解成几个具体问题了，铺垫已经够多了。

But I'll break this into a couple of questions now that I've laid out all that exposition.

第一个问题是：当董事会现在来找你时，你是否察觉到他们对AI项目有降温或担忧的迹象？比如可能要踩刹车？

So the first question is, when boards come to you right now, are you seeing a cooling off of or a fear around, okay, maybe we're gonna pump the brakes on AI projects.

他们目前都在提出哪些类型的人工智能问题？

What what kind of AI questions are they asking right now?

你感觉目前人们对尝试新AI项目的情绪和意愿如何？

What's what's your sense of people's mood and appetite for trying new AI projects right now?

是的。

Yeah.

我认为投资回报率问题仍是董事会最关心的。

I think ROI question is number one, still in boards' mind.

我不会说董事会要求暂停并重新评估。

I wouldn't say, boards would say, let's pause and review.

事实上，我认为董事会现在仍然保持着'我们必须解决这个问题'的紧迫感。

In fact, I do think, if anything, boards now still have that urgency of let's figure this out.

也许我们之前所做的给了我们一些宝贵的经验教训。

Maybe what we have done give it give us valuable lessons.

试点项目的成功或失败都是为了让你能够从中获得经验来指导决策。

Successes or failures of pilot projects are meant to give you insights that you can guide your decisions.

所以我认为投资回报率仍然是最关键的问题。

So I I would say ROI is still very much the big ask.

但我认为我看到的重大变化是，人们意识到这些AI计划与业务目标之间需要更紧密的关联性。

But, also, the I think the big change I I'm seeing is the realization of closer attribution of these AI initiatives to business goals.

我的意思是，过去可能是'让我们想想如何将生产力提高15%'，因为15%似乎是人们常挂在嘴边的神奇数字。

So what I mean by that is maybe in the past, it would be, oh, let's let's think about how to, you know, improve our productivity by 15% because that seems to be the magic number people storing around.

但现在情况可能是这样：好吧，

But now maybe it's like, okay.

如果你是首席营收官，你的增长目标是什么？

If you are, you know, the chief revenue officer, what are your growth goals?

AI在这方面能如何提供帮助呢？

How can AI help there, if anything?

可能并不是AI在提供帮助。

It may not be AI helping.

可能是AI帮助追踪营销活动，更敏锐地判断哪些营销活动效果良好等等。

It may be AI helping to track the marketing campaign and really be a little bit more vigilant on which marketing campaigns are doing the right things and all that.

但无论应用在哪个领域，它都与业务负责人密切相关。

But regardless of where it is, it's linked to that business head as well.

因此，无论是业务部门负责人还是职能部门主管，都需要签署确认特定的AI目标，而AI团队的八个关键绩效指标也将与该目标挂钩。

So the both the business division head function head would be a would be able would be asked to sign off on a certain AI goal, and the eight KPIs of the AI team will be linked to that goal as well.

所以我发现，相比六个月或十二个月前，现在这种讨论要多得多——那时候更多是自上而下的指令模式。

So I see that being talked about a lot more than, I would say, six months and twelve months ago when, at that time, it was a little bit more, like, top down.

让我们想办法提高生产力。

Let's figure out how to improve productivity.

回到你的观点，我认为理解AI能做什么、不能做什么，以及AI风险可能出现在哪里，确实是推动这种变革非常重要的输入因素。

And I do think, going back to your point, understanding what AI can do and cannot do, where the AI risk might happen, really is part of very important input to drive that change as well.

因为现在你不能轻易糊弄过关了。

Because now you can say it's it's not that easy to just, you know, bluff.

对吧？

Right?

这更像是'如何被衡量就会如何被执行'。

It's it's more like how do you get measured will will get done.

另外，我确实感受到这种中层管理困境的存在。

The other thing is I do feel, the there's of this middle management dilemma.

我对那部分人才群体深表同情，因为很多时候，取决于部门负责人是否清楚了解AI能做什么不能做什么，有时很难沟通如何从目标落实到实施层面的现实情况？

I I have a lot of sympathy, on on on that group of, talent because a lot of times, depending on the whether the the head of the division have a clear sense of what AI can do or cannot do, sometimes it's a little, like, difficult to communicate how do you go from that goal to what are the, what are the implementation reality is?

特别是，我很想听听您对数据观点的反馈。

And, especially, I think one of the area I would love to hear your your feedback on that is the views on data.

我们准备好了吗？

Are we ready?

这是第一个问题。

That's the first question.

您的部门负责人可能对什么有着截然不同的看法？

Your head of division may have very different view to the what?

作为中层管理者。

As the middle management.

第二件事是即使我们在数据层面准备好了，对我们核心能力或数据宝藏的预期真的准确吗？

The second thing is even if we're ready on data, is that expectation of our core competence or our this treasure from our data, is that really true?

我们五年后的竞争优势或核心能力，真的会来自这些数据吗？

Our competitive edge or our core competence five years down the road, is it really coming from the this data?

第三点，我也觉得存在很大差异，特别是在财务部门的FPA职能需要计算批量返利时，这种理解至关重要：我们是在做自动化，还是真正在利用人工智能？

And the third thing, I I also feel like there's a lot of difference, especially in the finance folks who FPA functions have to calculate the the volume payback, is this understanding of, are we doing automation, or are we truly leveraging AI?

对吧？

Right?

如果选择自动化，计算投资回报率更容易，有时也更具吸引力，我必须承认这一点。

So if you do the automation, it's it's, like, it's easier to calculate ROI, and then sometimes it's much more attractive, I have to admit.

但如果我们假设AI真的会在这个商业模式中发挥巨大作用，那么显然需要更多的沟通、协调和讨论。

But in order to really let's assume that AI really will play a huge difference in this business model, then there's definitely need to have a lot more communication, a lot more alignment, a lot more discussion.

因为如果只做简单的计算比较，你可能会选择流程自动化，从而错过许多其他潜在的有趣机会。

Because just simple calculation comparison, you may easily go with the process automation, which you are missing out a lot of other potentially interesting things.

是的。

Yeah.

这非常有趣，因为存在所有这些压力。

And it's so interesting because there is all this pressure.

有时会让人觉得，如果没有人深入挖掘并弄清楚我们到底想在这里实现什么目标？

And sometimes it can feel like if somebody doesn't dig in and figure it out and get down to like the brass tacks of what what are we trying to do here?

这是一个基于规则的工作流程吗？AI在其中处于什么位置？

And is it a rule based workflow or where where is AI in there?

有时感觉就像盲人给盲人带路。

It can feel like the blind leading the blind.

就像大家互相告诉对方，我们要做AI。

It's like everybody just telling each other, we're gonna do AI.

我们要做AI，却连定义都不明确，甚至不清楚最终目标是什么，除了提高生产力或其他什么。

We're gonna do AI without defining what it means or even necessarily defining what their end goal is other than productivity increase or or whatever the case is.

而我一直在进行这场语义之争，至今不愿放弃。

And I I've had this semantic battle that I'm I'm just not ready to let go of.

但你知道，2025年这一年，人们把所有东西都称为智能体。

But, you know, 2025 this year, everybody calls everything an agent.

这对营销人员来说倒是件好事。

And that's, you know, great for marketers.

他们可以把这种聊天机器人称为智能体。

They can call this chatbot an agent.

他们可以把这种工作流程也称作智能体，诸如此类。

They can call this workflow an agent and all that.

但智能体在人工智能领域是一个具有特定含义的工具。

But an agent is a very specific tool in artificial intelligence.

它是具备自主行动能力的实体。

It is something that has agency.

就像你我都有自主行动能力一样。

Like, you and I have agency.

智能体是你下达指令后就会去执行，并且独立完成任务的存在。

An agent is something that you tell it to go do it, and it goes off and does it.

除非遇到问题，否则不会中途返回。

And unless it has a problem, doesn't come back.

它会完成所有计算工作，直到任务完成才向你汇报。

It does and performs all the calculations and doesn't come back to you till it's done.

所以如果你把聊天机器人或其他非真正智能体的工具称为智能体，就是在稀释这个概念的真正分量。

So if you call your chatbot an agent or if you call, you know, whatever tool you have that's not truly an agent, then you're watering down how significant that is.

而且，你正在扰乱人们的期望，他们从各种媒体听到'哦，我们可以直接让代理去做这件事'。

And also, you're messing with people's expectations where they hear from various media, Oh, we could just have an agent do that.

其实没那么简单。

Well, it's not that simple.

我认为这就是为什么许多项目失败的原因——不切实际的期望以及对人工智能能做什么和不能做什么缺乏理解。

And I think that's why a lot of these projects are failing is just unrealistic expectations and a lack of understanding of what AI can and cannot do.

而我唯一能想到要妥协、允许把那些我们知道不是代理的东西称为代理的理由是...

And and the only reason I would think of going back and like sort of conceding on calling things that I that we know are not agents and and just allowing them to be called an agent.

如果对终端用户来说它看起来像个代理，那么它实际上是不是代理，或者只是一个编排好的工作流程，这重要吗？

If it looks like an agent to the end user, does it matter if it's actually an agent or if it's just an orchestrated workflow?

我不知道。

I don't know.

比如，如果我让电脑去执行某项任务，而我并不知道后台那些复杂的计算过程、决策树和典型自动化流程的具体运作方式，然后它完成时返回结果。

I mean, if I tell my computer to go off and do some task and I don't know the Rube Goldberg machine that's going on in the background that's doing all the calculations and decision trees and typical automation flow, and then it comes back when it's done.

对我来说，它就像个代理，所以我就会称它为代理。

To me, it seems like an agent, so I'm gonna call it an agent.

也许这就是我可以放手的地方。

So maybe that's where I could let it go.

但在董事会层面，当你和他们交谈时，我觉得他们可能不想听这种区分。

But at the board level, when you're talking to them, I feel like they probably don't wanna hear that distinction.

对吧？

Right?

我的意思是，你如何...我想我只是...我无法摆脱我们需要技术理解的需求，同时也需要有人来担任那个解释者并付诸实践。

I mean, how do you, I guess, I I just I can't get past the need for us to have technical understanding, but also someone to be that interpreter and go do it.

但我认为如果你没有那个解释者，没有具备那种深度理解的人，这就是这些项目失败的原因。

But I think if you don't have that interpreter and you don't have someone with that deep level of understanding, that's why these projects are failing.

所以我不知道董事会是否听到了这些，他们是否意识到，是否在意，或者他们对此的看法是什么。

So I don't know if boards are hearing that, if they're aware, if they care, or what their take on it is.

简短的答案是，至少我还没遇到过真正在意这种区分的人。

The short answer is, at least I haven't met someone who really care about that distinction.

当然，当首席技术官或首席信息官在他们面前被问到这类问题时，这个问题就会出现。

Of course, when the CTO or the CIO, went before them, got asked about these type of questions, then it will come up.

但就主动询问而言——你们是否在使用真正自主、真正工具化、真正能从历史决策中学习的AI系统——至少我还没遇到过这种情况。

But in terms of just proactively say, are you doing the Asian with true autonomy, true tool use, and true ability to learn, from the past decisions, I haven't at least I haven't encountered.

这并不意味着不存在这样的案例。

That doesn't mean there's none.

我只是说目前还不常见。

I'm just saying it's not common yet.

但我认为部分原因是，很多董事会成员，特别是在更传统、受监管的行业里，认为这离企业核心业务功能还很遥远。

But then I do think one of the reasons is, a lot of board members, especially in the more traditional, I would say, regulated industries, assume this is far away from the core business functions of this business.

所以他们往往认为智能体更多用于市场推广、潜在客户开发营销，或者某些邮件处理场景。

So they tend to assume that agents are used more in, let's say, go to market, you know, lead generation marketing, and maybe some of the email.

在日常生活中，可能就是订机票这类事务。

In casual life, it will be, like, booking a flight ticket or whatever.

但实际上，我们看到越来越多的企业将智能体作为核心能力——至少是工作流程的核心部分。

But in reality, we are seeing more and more the agents becoming a core competence, at least workflow, is in some of these businesses.

观察什么能引起董事会成员的关注将会很有趣。

And it will be interesting to see what will get the board members' attention.

我敢打赌他们会非常担忧与之相关的风险，因为当你提到代理权时，人们拥有自主权固然很好。

My bet would be they will be very worried about the risk associated with it, because when you say agency, it's great to you know, for people to have agency.

但一旦假设机器拥有代理权，风险警报就会立即拉响，谁能阻止一个代理在做出超出防护栏的意外行为前停下呢？

But once you assuming machine has agency, the risk alarm bells will just start ringing, and who will be able to, you know, stop an agent before they do something that's unexpected or, above the guardrail.

所以我们之前已经提到过氛围编程。

So we already mentioned vibe coding.

其实我很想抽空参加这类黑客马拉松活动，那些AI代理初创公司甚至已不能算初创了。

Actually, one thing I really like to see, and I I I would like to attend, when I have time is to go to these type of hackathons where these AI agent startups are not even startups.

它们现在至少都有数千万美元的收入规模，作为开发者工具呈现。

They are already, like, hundreds at least dozens of millions of revenue in size to present, like, them as a tool for developers.

我注意到代理的发展重点正从工作流能力（如代码、低代码流程）转向更关注防护栏机制。

And where they are focusing on, I also noticed for agents are moving from, you know, what agents can do in terms of workflow, like, code, low code workflow, you put together an agent, build an agent, to now very much focus on what are the guardrails?

我们该如何监控？

How do we monitor?

我们如何获得对这些意外行为的AI响应？

How do we get, you know, almost AI response to some of these unexpected behaviors?

我们如何创建审计追踪，这样即使代理可能产生所有这些理想结果，但万一出现问题，我们也能向检查者提供所有这些审计日志。

How can we create audit trails so that even though agents may, you know, create all these desirable outcome, But in case something happens, we can offer all these audit trail logs to anyone who's checking that.

最后我想说，如果你假设不仅你自己拥有代理，对方也拥有代理，那么你的业务立场如何与这类代理对手进行沟通、销售甚至交易？

And lastly, I would just say, also, if you assume not just yourself have agents, right, and then your counterparts have agents as well, how how is your position how is your business position to talk to or sell to or even deal with these type of agent counterparts?

你是真正将其视为机遇而接纳，还是因为风险指标而将它们全部拒之门外？

Are you really accommodating as opportunity, or are you sort of shutting them all out because the risk metrics?

我很希望我们能就此展开更多讨论，但我想说时机尚未成熟。

I would love for us to have more discussion about that, but I would just say not yet.

是的。

Yeah.

我是说，当智能体开始相互交互时，它们能做什么。

I mean, in what agents can do when they start interacting with each other.

我知道，因为谷歌刚推出了一个智能体支付协议，现在围绕它已经建立了很多协议。

I know because I think Google just came out with an agent payment protocol, and there's a lot of protocols being built around it now.

有趣的是，如果你在设计机器人，你会把它们设计成人形，因为你希望它们在一个为人类设计的世界中运作。

And it is funny to think we're, you know, if you're designing robots, you design robots to be human form because you want them to operate in a world that was designed for humans.

所以如果你有一个两英尺高、带轮子、只有一只机械臂的装置，它能做很多事，但无法像人类那样灵活操作。

So if you have a, you know, two foot tall something on wheels with one arm, you know, one mechanical arm or whatever, it can do a lot, but it can't manipulate around the way that that humans can.

这说的是物理世界中的设计理念。

So, you know, that's in the physical world and what you're designing.

但如果你设计的是数字代理，就会觉得围绕网站设计的整个用户界面概念，以及我们浏览应用和互联网的方式都非常以人类为中心，对代理来说效率极低。

But then if you're designing digital agents, it's kind of funny to think that the whole idea of a UI around a website and the way we navigate apps and the internet and everything is very human centric and it's very inefficient for agents.

所以随着网络发展，看到更多类似情况会很有趣——你提到个人用途演示总是很有趣。

So it's gonna be interesting to see what happens on the web as more and more like, think I've it's funny you mentioned that it's always the demo for personal use.

总是预订航班或餐厅预约之类的场景。

It's always booking a flight or booking dinner reservations and all that.

但想想这有多复杂：登录达美航空官网，选择出发地和目的地，查看不同航班选项、舱位等级、座位选择等等。

And it's but if you think about how complicated it is, going to the Delta website and navigating, you know, where to from and then looking at the different options and the different flight class options and then the seat options and all that.

这种浏览方式非常繁琐，而直接通过API查询可用航班和座位数据库会简单得多。

It's a very cumbersome way to navigate when really an API that just went in and looked at the database of available flights and seats and all that would be much easier.

所以如果你现在经营企业，在为这种代理主导的互联网做多少规划？特别是当代理甚至还不能帮你对账信用卡时，你该投入多少精力？

So if if you're in business right now, how much are you planning for this sort of agent run Internet and and how much your focus on it is when you can't even get an agent to reconcile your credit card accounts or whatever.

我确实认为对于财务规划与分析部门的同事来说，有两点是大家越来越需要考虑的。

I do think for FP and A colleagues, there are two things that people are more and more have to consider.

所以一方面可能是从产品角度来看，如果你销售的是AI类产品，智能代理意味着定价策略将变得截然不同。

So one is maybe from a product side of things, if you sell AI type of products, Agents means the pricing strategy will be very, very different.

对吧？

Right?

比如基于结果的、基于行动的，甚至是基于使用量的。

So outcome based or even action based, usage based.

有一点可以确定的是，绝不会再采用按席位收费的模式。

One thing for sure is not gonna be seat based.

那么如何在充满不确定性且缺乏现成参考方案的情况下建立这种模型？

So how do you model that out and especially with a lot of uncertainty and not a lot of existing playbook yet?

这是针对销售此类产品的情况。我认为作为一名财务规划与分析专业人士，如果能思考如何解决这个问题并建立相应框架，将是脱颖而出的绝佳方式。

That's from a if you sell these type of products, I think if you, as a FP and A professional, can think about ways to think through that, create a framework around that, is a great way to stand out and differentiate yourself.

这只是我的个人观点。

Just my opinion.

但如果你作为采购方购买这类产品，在你的成本结构中，这部分支出也会变得越来越重要。

But if you are buying these type of products on the other side, like your cost side of things, also could be more and more important in your in your cost structure as well.

这也是需要考虑的一点。

So that's something also to consider.

我确信格伦在未来的节目中会大量讨论这个话题。

I'm pretty sure that, Glenn, in the future episodes, you will talk about that a lot.

我很期待收听那些内容。

So I look forward to listening to those.

我想多谈谈治理方面的问题。

I wanna talk a little bit more about governance.

不过在开始之前，我要先暂存关于董事会指导方向的问题，以及最新研究显示大量AI项目失败的情况。

But before I do, I wanna put a pin in my question about sort of the direction from boards and and the latest studies that have shown, you know, all these AI projects failing.

我反复看到的很多数据都显示CFO们表示认同。

A lot of the numbers I see over and over are CFO says, yep.

AI具有战略意义。

AI is strategic.

我们已准备好预算。

We have budget.

我们准备推进了。

We're ready to go forward.

然后你问有多少在做试点项目，这个比例是某个百分比。

And then you ask how many are doing pilots and it's it's some percentage.

但如果你问有多少已经超越试点阶段，真正在生产环境中大规模使用AI，这个数字就会急剧下降。

But then if you ask how many have gone beyond pilots and are actually using AI at scale in production, that number drops precipitously.

而且你知道，这还是一项非常新兴的技术，特别是在财务和会计领域。

And I'm you know, and it's also it's a very nascent technology and, you know, finance and accounting.

我们，你知道，我们本就应该规避风险。

We're not, you know, we're we're meant to be risk averse.

所以我们不会冲在最前沿。

So we're not gonna be out on the bleeding edge.

但你认为现在更多是时机问题吗？

But it what do you is it more than timing right now?

那么为什么在AI实施的愿景与执行之间存在如此大的差距呢？

What's why why is there such a gap between, call it, aspiration and execution on AI implementation?

是的。

Yeah.

我知道你自己不会这么说，但我确实觉得你在Substack上关于这个话题的文章非常出色，所以我强烈建议大家去关注你的Substack。

I know you wouldn't say it yourself, but I do feel like your article on this topic, on your substack was excellent, so I would highly recommend people to track down your sub stack.

但我的观点是，除了通常的时间安排、人员管理、变革管理等因素外，我想提出两个我认为相关但较少被讨论的点。

But my opinion, other than the normal, you know, the the timing and all that, people management, change management, I do want to bring in two points that I feel are relevant but less talked about.

首先是第一阶段，即使用AI、将AI作为副驾驶或聊天机器人来采用。

So one is you do have that first phase of, you know, using AI, adopting AI as a Copilot or chatbot.

有时甚至需要内部定制开发以适应内部使用场景。

Sometimes it's even built internally to customize for the internal use case.

这本来很好，直到它不再适用。

And that was good, but until it's not.

聊天机器人作为界面非常僵化。

It's it's the chatbot as the interface is very rigid.

这又创造了一个新的工作流程。

It creates another workflow.

最初，人们或许能从聊天机器人的帮助中受益。

And, initially, maybe people are, you know, benefiting from that chatbot help.

但随着AI能力持续发展，人们更希望将AI嵌入现有工作流程，而不是额外打开新窗口或不断复制粘贴内容。

But with AI's capability continue to develop, people will love to have embedded AI into their existing workflow instead of having another window or have to copy paste and copy paste back.

这会导致使用率快速下降，而技术发展会自然从传统界面（我们称之为传统，实则是两年前的聊天机器人界面）演进为如今更智能的代理或完全原生的嵌入式工具。

So that really fast the utilization rate a lot, and that's also a natural way of progression, I guess, for the technology to go from a traditional UI, which is a we call it traditional, but two year old UI type chatbot to now is, like, agent or embedded completely native type of tools.

我认为这种转型取决于公司所处阶段，有时人们会卡在过渡期。

I do think that transition, depending on your company's stage, sometimes people are stuck in between.

比如如何从结构固定、易于理解的聊天机器人，转向功能更强大但需要一定集成工作的项目类型。

Like, how do I move to from a very canned, very easy to understand type of chatbot to a much more powerful, but also has a little bit more funds through integration work type of projects.

另一个我认为当前讨论不足的是实施方式的问题。

So the other thing that I feel people are now not talking about it enough is the way of implementation.

过去我们可能面临两种选择路径。

So in the past, maybe again, like, we chose to there are two ways.

一种是购买这种通用性强但功能强大的聊天机器人。

One is to buy this very generic use but very powerful chatbot.

另一种方式是我们拥有秘密配方或优质数据，必须自行构建，因此需要聘请大批人才来打造自己的系统。

The other way is we have this secret formula or great data we have to build on our own, and therefore, hire an army of talents and build our our own.

但现在双方都有所欠缺——通用方案有时无法完全整合，也无法充分挖掘业务潜力，更无法让你脱颖而出。

But now both sides have something else to desire because the general use sometimes cannot, again, like, cannot be fully integrated and fully fully harvesting the potential of that business and also doesn't make you differentiate.

对吧？

Right?

所有人都在用同样的东西。

Everyone is using the same thing.

那你凭什么能与众不同呢？

Like, why would you be so different?

而自行构建的那一方，正如MIT文章提到的，失败率甚至更高——因为你的团队要么跟不上技术发展，要么存在预算超支风险，资源分配可能达不到最优成本效益。

And then the other side, building yours on yourself, and it's even has even more, I would say that MIT article also mentioned it has even higher failure rate because your talent either cannot keep up, or, there's also a risk of overspending or budgeting, where your your talent may or your allocation may not be moving into the most cost effective way.

我们都知道，过去两年AI模型训练技术发生了翻天覆地的变化。

And we all know that, you know, the techniques of AI model training has been changing so much over the last two years.