这两年,"AI转型"几乎成了传统企业老板的标配焦虑。开会要提,年度规划要写,见客户要讲。可你真去看落地,大多数公司干的是同一件事:找几个人,学一学,做几个小工具出来。
然后就没有然后了。
先把最扎心的一句放在前面:做工具,不等于AI转型——就像当年买几台电脑,不等于互联网化。
01大部分企业的"AI转型",停在了工具层
给某个岗位配一个自动化脚本、一个RPA机器人,一天帮他省下两三个小时。老板很满意,觉得转型开始了。
冷静想一下:业务的内核变了吗?没有。还是原来那套流程、那套逻辑、那批人在跑,AI只是钻进来替某一步打了个工。说白了,大多数公司的AI转型,只是给旧流程雇了个更快的临时工:干完这一步,打卡走人,业务的内核一寸没动。
更要命的是,很多人把"降本增效"当成了转型的成绩单。省两个小时,裁一个工时,报表上很好看。但降本增效是工具层的天花板,不是转型的起点——省下来的那点时间,改变不了你赚钱的方式。你还是靠原来那套模式赚钱,只是成本略低了一点,这叫优化,不叫转型。
这跟当年的互联网转型是一回事。不是买了几台电脑、建了个官网,就叫互联网化。真正的转型,是整个组织的运作方式被重做了一遍。
一个关键区分,很多人一辈子没绕过来:
- 工具上的跑——是AI当一个更快的员工,帮你把某个动作做得更快更省。
- 业务生态的迭代——是AI当新的地基,整条业务链围绕它重新长出来。
绝大多数企业以为自己在做第二件事,其实一直在第一件事里打转。在工具层里跑得再快,也跑不到业务层去——这不是速度问题,是方向问题。
02能不能转,先看谁在推
我的判断很直接:不懂业务的人,迭代不了业务。
很多公司把"推AI"交给了IT、行政HR、或一个临时拼起来的"AI小组"。这些人工具玩得溜,但不在业务核心里,不知道这条线真正的痛点、卡点在哪,钱到底怎么进来。
一个不背业务指标的人,你让他去重构业务,他重构不了——他手里只有工具这一把锤子,看什么都像钉子。让不背业务指标的人去主导转型,你只能收获工具,收获不了转型。这不怪他们,是位置决定的。
真正能做业务生态迭代的,只有那种既懂业务、又愿意亲手把自己熟悉的老办法推倒重来的人。这种人本来就稀缺,愿意革自己命的更稀缺。但没有这种人牵头,转型就是空话——再好的工具,也只是给了一个不知道往哪开的人一辆好车。
03最要命的一环:推AI的人,有没有决策权
这是我认为国内传统企业AI转型最深、也最少被讲透的死结。
假设你真组了一个AI团队,也真做出一套能改造业务的方案。接下来会发生什么?
大概率是:团队跑去跟业务部门说"我给你们做了套新东西,建议用一下",业务部门抬头看一眼,"哦,知道了",然后该怎么干还怎么干。
因为这个AI团队没有决策权。他们只能建议,不能拍板。而没有决策权的方案,本质上是一份写得很漂亮的建议书;建议书的归宿,是抽屉。
东西可用可不用,结果一定是"不用"。人有惯性,谁愿意在最忙的时候去改一套自己闭着眼都能跑的流程?你不给强制力,他就一定选舒适区。到这一步,投进去的人力全白费。
所以我一直坚持:一个没有决策权的AI团队,做得越多,浪费越大。
它必须能进到决策层。方案要在项目会上跟业务、运营负责人当面过,过了就一致执行、强制落地——不是"欢迎使用",是"从今天起就这么干"。转型靠自觉是推不动的,指望员工自觉改掉顺手的旧习惯,是管理者最贵的偷懒。转型这事,从来不靠人心,靠机制。
04对人:这是一场没得商量的迁移
AI这波浪,想把所有人都拖上船,不可能。总有人学得飞快,有人怎么推都不动。
我的态度比较硬:想学的人不用逼,他自己就冲上去了;不想学的人,你逼一步他走一步,逼不动就是逼不动。这不是道德问题,是意愿问题。企业没那么多时间,去等每一个人慢慢想通。
这也不冷血。浪来了,上船的往前走,不上船的留在岸上——企业不是慈善机构,没义务等每个人想通。与其粉饰太平,不如早点讲清楚,让愿意上船的人先上、先拿到位置和资源。对真正想学的人,这反而最公平。
05别谈信仰,谈验证
最后一点,也是不让这件事跑偏的刹车:一切都要能验证,而且要很快能验证。
AI最容易让人上头。方案听起来都很美,PPT做得比谁都漂亮,但落地有没有用,得拿数字说话。一条内容用AI做出来到底爆没爆?投放接入AI后ROI到底涨没涨?不该靠感觉,该靠明确的检查点。
别把AI供起来当信仰。工具的尊严,恰恰在于它随时能被验证,也随时能被证伪。一个不敢被验证的AI方案,多半是在掩盖它根本没用。
好在AI这类事验证周期通常很短,做了很快就知道行不行。别怕试错,怕的是既不敢试、又不肯认。
结尾
困局的根,是"既要又要"
串起来看,国内传统企业AI转型的困局,根子往往在最上面。
很多老板的真实状态是"既要又要":既想要转型的结果,又不肯动组织架构,不肯给AI团队决策权,不肯真正配人配资源——只想让下面的人"跑一跑、做点小工具",就把转型给转了。
说难听点,这就像囤了一柜子止痛药,然后宣布自己在做手术。工具是止痛药,转型是动刀子。药吃再多,病灶还在那儿。
工具解决不了转型。转型要的是四样:一次自上而下的组织重构、一个有决策权的推进机制、一批真懂业务的人牵头,外加一套能快速验证的闭环。缺哪一样,最后都只剩一地"可用可不用"的小工具,和一句"我们也在做AI"的自我安慰。
既要又要,是这场转型里最贵的病。想清楚这件事,比多买十个工具都值钱。
For two years, "AI transformation" has been the default anxiety of every traditional-company boss. It goes on the agenda, into the annual plan, onto every pitch. But look at what actually ships, and it's the same thing every time: a few people get trained, a few small tools get built.
And then nothing.
So let me lead with the line that stings: building tools is not AI transformation, no more than buying a few computers made you an internet company.
01Most companies are stuck at the tool layer
You give someone an automation script or an RPA bot, it saves two or three hours a day, the boss is happy, everyone decides the transformation has begun.
Be honest: did the core of the business change? No. Same process, same logic, same people. AI just crawled in to do one step faster. Most companies' "AI transformation" is just hiring a faster temp for the old process: it does one step, clocks out, and the core of the business never moves an inch.
And here's the trap dressed up as a win: people treat "cost savings" as the scorecard. Two hours saved, one headcount trimmed, looks great on a slide. But cost-cutting is the ceiling of the tool layer, not the starting line of transformation. The hours you save don't change how you make money. You're still making money the old way, just slightly cheaper. That's optimization, not transformation.
It's the internet transition all over again. A website didn't make you an internet company. Real transformation meant the whole organization was rebuilt on a new foundation.
The distinction most people never get past:
- Running on tools: AI is a faster employee doing one action more cheaply.
- Iterating the business: AI is the new foundation, and the whole value chain regrows around it.
Most companies think they're doing the second. They've been circling the first the whole time. You can sprint as fast as you like inside the tool layer and never once cross into the business layer. That's not a speed problem, it's a direction problem.
02Whether you can transform depends on who's driving
I'll be direct: people who don't understand the business can't reinvent the business.
Too many companies hand "push AI" to IT, to HR, to some hastily assembled task force. Great with tools, but not in the core of the business. They don't know where the real pain is, or how the money comes in. Give the wheel to someone who carries no business number, and all they can hand you is more tools. When your only tool is a hammer, everything looks like a nail. That's not their fault; it's their vantage point.
The only people who can iterate the business are the ones who understand it deeply and are willing to tear up their own familiar playbook. Rare to begin with, and rarer still the ones willing to cut against their own habits. But without them leading, transformation is just talk. The best tool in the world is still just a fast car handed to someone with no idea where to drive.
03The killer: does the person pushing AI have any authority?
This is the deepest knot, and the one almost nobody names.
Say you build an AI team, and they build something that could genuinely reshape the business. What happens? The team walks over and says, "we built you this, you should try it." The business unit glances up, says "noted," and keeps doing exactly what it did.
Because the team has no decision-making power. They can only suggest, never decide. And a plan with no authority behind it is just a beautifully written recommendation. Recommendations end up in a drawer.
When something is optional, the answer is always no. Nobody rewrites a workflow they can run in their sleep in the middle of a busy week. Deny them the mandate and they'll pick comfort every time. Every hour of headcount you invested is wasted.
So here's my hard line: an AI team with no authority does more damage the more it produces. It has to sit at the decision table. The plan clears the room with the business and ops leads, and then it's mandatory. Not "you're welcome to use it," but "this is how we work starting today." Transformation never runs on goodwill. Betting on people to voluntarily drop a habit they can run in their sleep is the most expensive form of managerial laziness. It runs on mechanism, not on hope.
04On people: this is a migration, not a vote
You cannot drag everyone onto the boat. Some learn fast; some won't move no matter how hard you push.
My stance is blunt: the ones who want to learn don't need a push; they're already sprinting. The ones who don't, take one step per shove, and when the pushing stops, so do they. Not a morality problem, a will problem. A company doesn't have infinite time to wait for everyone to come around.
And this isn't cruelty. The wave is here: the ones who board move forward, the ones who don't get left on the shore. A company isn't a charity; it owes no one an infinite wait. Better to say it plainly and early, so the people who want on get on, and get the seats and the resources first. For anyone who actually wants to learn, that's the fairest deal there is.
05Skip the faith. Verify.
The brake that keeps this from going off the rails: everything has to be verifiable, and fast.
AI is intoxicating. Every plan sounds beautiful, every deck looks sharp. But whether it works has to be settled with numbers. Did the AI-made content take off? Did ROI move after AI went into the pipeline? Not vibes, but a clear checkpoint. Don't put AI on an altar. The dignity of a tool is precisely that it can be verified, and disproven, at any time. An AI plan that can't survive a test is usually hiding the fact that it doesn't work.
The good news: the verification cycle is short. Ship, and you know fast. Don't fear testing. Fear refusing to test and refusing to admit it.
In closing
The root of the trap: wanting it both ways
String it together, and the root sits right at the top.
The honest state of many bosses is "I want it both ways": they want the result of transformation, but won't touch the org structure, won't give the AI team real authority, won't staff or resource it. They just want the people below to "run around and make a few tools" and call it transformation.
It's like stockpiling a cabinet of painkillers and announcing you're doing surgery. Tools are the painkiller; transformation is the scalpel. Swallow all you want, the tumor's still there.
Tools don't solve transformation. It takes four things: a top-down org rebuild, a decision mechanism with real authority, people who genuinely understand the business leading it, and a fast verification loop. Miss one, and all you're left with is a pile of optional little tools and a comforting line: "we're doing AI too."
Wanting it both ways is the most expensive disease in this whole transformation. Getting clear on that is worth more than buying ten more tools.