王峰 AI × 经营 约一次诊断

总目 / 文章 / WF-0380

文章 · 存档

深度解读德勤《AI时代的抉择》AI治理已经迫在眉睫

德勤最新发布的一份报告《AI时代的抉择:以信筑基,行稳智远》中称,亚太地区只有不到一成的企业AI 治理能算得上“成熟”,也就是说超过 90% 的公司都还在边摸索边上车。

但另外一头,基层员工,尤其是年轻人,早就自己在用各种生成式 AI 了,有的写方案,有的改代码,有的直接拿来改合同、写邮件。

管理层没完全想清楚怎么玩,底下的人已经在飞了。

这就是今天多数企业的真实状态:AI 已经进来了,但“制度、规则、边界、责任”全都没跟上。

你表面上看是生产效率、内容产出、模型应用的故事;但你往里一戳,全是安全、隐私、合规、声誉的雷。

德勤在调研里问高管最担心什么,排在前面的不是“效果不好”,而是“安全漏洞”“监控泛滥和隐私”“法律和版权风险”。

说白了,大家不是不想用 AI,而是不想用完之后背锅。

谁都知道一旦出事,罚款是小事,招股书、招人、客户信任、品牌口碑,统统会被翻出来问一遍:“你当初是怎么管这个东西的?”

这里有一个很有意思的反差。

很多老板自评时会觉得“我们公司对 AI 的风险是有意识的,也有一些规定,有会议、有汇报、有流程”,但德勤用一套 5 大支柱、12 个指标去测 AI 治理成熟度之后,发现真正在制度层面扎实、可执行、能落地的,只剩不到一成。

这就像很多企业说自己有“全面预算管理”,结果一问预算怎么做的,最后都是“财务部帮各部门做了个数上去”。AI 这件事现在也在重复一遍。

那到底什么叫“信任内功”?

德勤的框架说得很完整:透明、可解释、公平、中立、稳健、可靠、保护隐私、安全、承担责任、可问责。

听上去挺学术,你可以简单理解成一句话:这个系统出事的时候,谁知道发生了什么、谁负责、怎么补救、下次怎么避免。

没有这些,所有“AI 战略”“AI 规划”,最后都会变成一堆 PPT。

真正有意思的是报告还做了一个计量分析,把“AI 治理成熟度”和“AI 使用范围”“员工使用程度”“收入增长”连在一起看。

结论其实很朴素:治理越成熟的企业用 AI 的部门更多,员工用得更深收入增速更高。甚至在控制了“用 AI 的多少”这个变量之后,治理成熟度每往上抬一点,收入增长还能多抬出一截。换成老板能听懂的话就是:治理不是刹车是发动机,它帮你把风险挡在前面,把价值放大在后面。

那到底什么地方最拖后腿?

报告里有一个细节我觉得特别值得中国企业警惕。

五大支柱里,大家在“组织架构”和“监控与评估”上,反而做得还行,甚至很多公司已经设了 AI 委员会、指定了某个 CXO 负责、也会半年或季度评估一次系统。但在“政策与原则”和“程序与控制”上初级水平的企业比例最高。翻译一下,就是:人已经站好了制度还没写清楚流程也没固化。

德勤提到一点我非常赞同:AI 治理归根到底是人的问题,而不是技术问题。调研显示,哪怕在做得比较好的企业,真正具备“能以合规、合伦理方式使用 AI”的员工比例平均也就五六成。提供了培训、有指引、有咨询渠道、鼓励在职学习的公司,员工的胜任程度明显更高;而只喊口号、不教用法、不说边界的公司,员工要么瞎用,要么不敢用。

所以真正要做的不是纠结“我们要不要上一个大模型系统”“选哪家供应商更牛”,而是静下心来问三个问题:

第一,你到底愿意把多少权力交给 AI?让它辅助还是让它决定?哪些业务场景可以“机器先算,人来审”;哪些场景必须“人先判断AI 只给建议”?这个边界不清楚,所有后面的规则都是空谈。

第二,你是否准备好把“AI 使用”当成一种“制度性行为”,而不是个人玩具?有没有一个最小可行的 AI 使用政策让所有人一眼就能看懂,知道什么严禁、什么慎用、什么鼓励?有没有一套简明的“出事处理流程”,哪个部门牵头、怎么隔离问题、怎么对外澄清?

第三,你愿不愿意承认:AI 治理不是 IT 部门的事,而是经营层核心议题?当你在开经营会、预算会、风控会的时候,AI 是不是已经被摆在了桌面上,作为一个长期要管理的“新型生产要素”和“新型风险来源”,而不是“技术试点项目”?

这些问题,如果你愿意认真回答,企业的 AI 之路自然会清晰很多。

再说一句现实的。很多老板会本能地觉得治理就是添麻烦,就是减速,就是给业务上锁。

德勤的数据恰好反过来印证,治理做得越好的企业,AI 部署越快、使用越广、创收越明显。原因也不复杂:有规则,大家才敢真用。

未来三到五年,企业之间会形成一道新的“经营分水岭”——不是谁先用 AI,而是谁先把“AI+治理”这套内功打扎实。

有的企业会走出一条道路:AI 深度嵌入日常经营,财务、业务、风控、人力都围绕一套数据和规则来协同,AI 帮你放大正确的经营决策;另外一批企业,会陷在反复试点、反复踩坑、反复公关危机、反复被监管叫去喝茶的循环里。

我给一个落地得不能再落地的建议,其实很简单,从明天起可以就动手:先把公司里所有正在使用、准备使用、员工私下使用的 AI 工具,拉一个完整清单出来;然后挑两个你最在意的关键业务场景——比如对外报价、合同文本、对客户的回复邮件——先写一版“AI 使用最低标准”,白纸黑字、两页 A4 解决,不要写成论文。

接着选一批骨干员工给他们一场半天的工作坊,不是教他们“什么是 AI”,而是教他们“在你这个岗位,AI 可以帮你做什么,不能碰什么”,同时把他们的实际疑问都收回来,倒推完善你的规则。这种小步快跑、边用边治理的方式,远远比等一个“完美 AI 战略”靠谱。

这正是我近一年开发的《AI财务落地课》的初衷,我可以让任何一名财务人员用2天的时间就能够完全掌握AI来实现自己真实工作场景的自动化,是任何一名,并且所有人都能真正学会写代码、写复杂的公式、让复杂的逻辑轻而易举的实现自动化完成。课程里也会专门把“AI 治理”“可信 AI”做一个通俗又能落地的模块,也会开放共享 CFO 顾问服务,把这一套帮助企业做成自己的“AI 内控体系”。你如果觉得这些东西对你的公司有价值,那就不是“帮我”,而是帮你自己给企业加上一层长期的护城河。

AI 时代,看上去大家都在比模型、比算力、比落地案例,其实真正决定生死的是一句话:客户、员工、监管、股东相不相信你这家公司用 AI 是有边界、有责任、有底线的。

能做到这一点的企业,不用追风口,它自然会活得很久。

In-depth Breakdown of Deloitte's Report

In-depth Breakdown of Deloitte's Report “Choices in the AI Era”: AI Governance Can't Wait Anymore

Deloitte just released a new report called “Choices in the AI Era: Build on Trust, Go Far with Wisdom.” It says that in the Asia-Pacific region, less than 10% of companies have “mature” AI governance. That means over 90% of companies are still figuring things out while jumping on board.

But on the other side, regular employees—especially young ones—are already using all kinds of generative AI on their own. Some use it to write plans, some to fix code, some even to edit contracts or write emails.

The bosses haven't fully decided how to play this game, but the people below them are already flying.

This is the real situation in most companies today: AI is already inside the door, but rules, boundaries, duties, and responsibilities haven't caught up at all.

On the surface it looks like a story about higher productivity, faster content creation, and cool model use. But once you dig a little, it's full of hidden bombs: security, privacy, legal compliance, and reputation risks.

When Deloitte asked executives what worries them most, the top answers were not “it doesn't work well,” but “security holes,” “too much monitoring and privacy issues,” and “legal and copyright risks.”

To put it simply: everyone wants to use AI, but nobody wants to take the blame if something goes wrong.

Everyone knows that if something blows up, fines are the small problem. The big problem is that your IPO documents, hiring, customer trust, and brand reputation will all be dragged out and questioned: “How exactly did you manage this thing?”

Here's a funny contrast.

Many bosses think “Our company is aware of AI risks. We have some rules, meetings, reports, and processes.” But when Deloitte tested real maturity using 5 big pillars and 12 measures, only less than 10% of companies actually have solid, workable, real rules on paper.

It's just like when companies say “We have full budget management,” but when you ask how the budget was made, it turns out the finance team just made up numbers for every department. The same thing is happening again with AI.

So what exactly is “trust as core strength”?

Deloitte's framework lists everything clearly: transparent, explainable, fair, neutral, robust, reliable, protects privacy, safe, takes responsibility, and can be held accountable.

It sounds very academic, but you can boil it down to one simple sentence: When this system messes up, who knows what happened, who is responsible, how to fix it, and how to stop it next time?

Without these, all your “AI strategy” and “AI plans” will just end up as a bunch of PowerPoint slides.

The really interesting part is that the report did a numbers analysis. They linked “AI governance maturity” with “how widely AI is used,” “how deeply employees use it,” and “revenue growth.”

The conclusion is actually very simple: Companies with better governance use AI in more departments, employees use it more deeply, and revenue grows faster. Even after removing the effect of “how much AI they use,” every step up in governance maturity still pushes revenue growth higher.

In words bosses understand: Governance is not the brake—it's the engine. It blocks risks in front and makes value bigger in the back.

So which part is holding companies back the most?

There's one detail in the report that I think Chinese companies really need to watch out for.

Among the five big pillars, companies are actually doing okay on “organizational structure” and “monitoring & evaluation.” Many already have an AI committee, named a CXO to be in charge, and do reviews every six months or every quarter.

But the highest number of companies are still at beginner level on “policies & principles” and “procedures & controls.”

In plain words: The people are already in position, but the policies are not written clearly and the processes are not locked in.

Deloitte said something I totally agree with: At the end of the day, AI governance is a people problem, not a technology problem.

The survey shows that even in the better companies, only about 50–60% of employees really know how to use AI in a legal and ethical way.

Companies that offer training, clear guidelines, help channels, and encourage learning on the job—their employees are clearly much better.

Companies that only shout slogans, never teach how to use it, and never say where the lines are—employees either use it blindly or are too scared to touch it.

So the real thing to do is not to keep asking “Should we buy a big model?” or “Which vendor is better?”

Instead, calm down and answer three questions:

First: How much power are you really willing to give to AI? Just help, or actually decide? In which business situations can we do “machine calculates first, human checks”? In which situations must we do “human decides first, AI only gives suggestions”? If this line is not clear, everything after is just talk.

Second: Are you ready to treat “using AI” as official company behavior, not a personal toy? Do you have a short, clear, minimum AI policy that everyone can understand at a glance—what is totally forbidden, what needs extra care, what is encouraged? Do you have a simple “what to do if something goes wrong” process—who leads, how to isolate the problem, how to explain to the outside?

Third: Are you willing to admit that AI governance is not just the IT department's job—it's a core topic for top management? When you have management meetings, budget meetings, or risk meetings, is AI already on the table as a new production factor and a new source of risk that you have to manage long-term—not just a “tech pilot project”?

If you are willing to answer these questions seriously, your company's AI road will become much clearer.

One more realistic thing: Many bosses feel deep down that governance = extra trouble = slowing down = locking the business.

Deloitte's data proves exactly the opposite: Companies with better governance roll out AI faster, use it more widely, and make more money. The reason is simple: When there are clear rules, people dare to really use it.

In the next 3–5 years, a new dividing line will appear between companies—not who uses AI first, but who builds solid “AI + governance” basics first.

Some companies will find a good path: AI deeply built into daily operations, finance, business, risk control, and HR all working together around one set of data and rules. AI helps you make the right business decisions bigger.

Other companies will stay stuck in endless pilots, endless mistakes, endless PR crises, and endless trips to talk to regulators.

Here's the most practical advice you can start tomorrow: Make a full list of every AI tool your company is using, plans to use, or that employees are using secretly.

Then pick two key business scenes you care about most—for example, pricing quotes to customers, contract text, or reply emails to clients—and write a one-page (max two A4 pages) “Minimum AI Use Standard.” Keep it simple, black and white, no long papers.

Then pick a group of key employees and run a half-day workshop—not to teach “what is AI,” but to teach “in your job, what can AI help with and what you must never touch.” Collect all their real questions and use them to improve your rules.

This small-step, fast-run, govern-while-using way is way more reliable than waiting for a “perfect AI strategy.”

This is exactly why I created my “AI for Finance Real-World Course” in the past year. I can help any finance person master AI in just 2 days and automate their real daily work. Anyone—no matter who—can truly learn to write code, write complex formulas, and make complicated logic automatic and easy.

The course also has a simple, practical module on “AI Governance” and “Trustworthy AI.” I'll also open my CFO advisory service to help companies build their own “AI internal control system.”

If you think this is valuable for your company, it's not “helping me”—it's helping yourself add a long-term moat to your business.

In the AI era, everyone looks like they're racing on models, computing power, and success stories. But what really decides who lives or dies is one sentence: Do customers, employees, regulators, and shareholders believe your company uses AI with clear boundaries, clear responsibility, and a solid bottom line?

Companies that can do that don't need to chase trends—they will naturally live long and strong.