When your boss uses AI to review your proposal, how should the CIO cope with this situation?
At the end of the year, are many CIOs currently busy drafting various plans for the coming year, while also preparing themselves for queries and criticism?
A couple of days ago, a CIO friend working in the manufacturing industry messaged Lao Yang, saying: "Lao Yang, I pulled four all-nighters to finish the 2027 digital transformation plan, integrating all the long-standing messy accounts of the company, unavoidable pitfalls in business operations, and technical trade-offs into it. But right after I submitted the plan, my boss forwarded it directly to an AI tool. Five minutes later, he called me into his office, held up his phone and read the result to me: The AI says your technical route is too conservative, and there is far too much padding in the budget. Is it that you lack competence, or that your mind is too rigid?"
He tried to explain for a long time, but his boss finally dropped a line: "AI will never lie to me!"
He said he was totally speechless at that moment, and knew he could not make it clear to the boss in a short time.
Have you ever encountered a similar scenario?
Will AI Never Lie to the Boss?
"AI will never lie to me! AI is smarter than humans and considers things more comprehensively!" This is a line many bosses often say after they start using AI tools.
This statement sounds like a compliment to AI, but in essence it is defining the nature of the budget matter. When a boss says AI will not lie to him, what is the hidden subtext? You will. When you submit a budget proposal, you must have ulterior motives, and you must be considering your own personal interests.
Now, here comes an AI that is "free from interest entanglements", and it tells the boss exactly what he wants to hear.
To put it bluntly, the boss is not looking for verification, he is looking for endorsement.
Before he opens the AI, he already has the answer in his mind. He wants to cut the budget, and he thinks you are too conservative, but he does not understand technology and cannot refute you from a professional perspective. At this moment, AI just happens to speak out what he wants to say. Speaking through the mouth of AI makes him look far-sighted, objective and impartial, and also leaves you no room to refute.
The fact that AI has no interest entanglement with you does not mean it has no source bias. It reads public data across the entire network, among which white papers from technology vendors account for the majority. It provides a general optimal solution, but when this general optimal solution is applied to a specific company, the cost is always underestimated.
More critically, the boss using AI to cut your budget is a manifestation of his pre-set stance. He is by no means a neutral referee. He is a party to this game, but he just borrows the mouth of AI, the so-called referee.
The Company AI Describes Is Not Yours
Why does AI say there are loopholes in your plan, and it sounds so convincing? Because what it says is theoretically correct. It is describing an ideal enterprise with clean data, standardized processes and full alignment of all staff, which is extremely rare in reality.
AI does not know that your 10-year-old legacy ERP system has more than 200 stored procedures at the bottom layer that no one dares to modify, otherwise the system will throw errors immediately.
It does not know that the sales director created a bunch of fake data in the system to cover up the declining performance.
It does not know that the finance department is so strict with your budget because an informatization project collapsed unfinished 3 years ago, and the psychological shadow has not faded away to this day.
The suggestions AI most likes to put forward are breaking down data silos, introducing agile development, and reconstructing the technical architecture. These terms sound very pleasant and vivid. But breaking down one data silo may require several business departments to argue for 3 months. Cleaning up a batch of historical dirty data may cost dozens of man-months. Adding one approval node may mean readjusting the relationship between two vice presidents.
In other words, the conclusions drawn by AI are basically based on ideal states, but the real work scenarios of enterprises are far from ideal.
As a CIO of a certain enterprise put it: What AI masters is the average value of human public knowledge, while what we master is the exact value of this company's long-standing messy accounts. No matter how beautiful the average value is, it cannot solve problems that require accurate figures.
What Should You Compete With AI For?
You will never win if you compete with AI for intelligence.
Once you figure this out, the CIO's idea of fighting back becomes clear.
When it comes to refuting AI, many CIOs initially want to defeat AI, find its loopholes and make it look bad. But in reality, even if you really defeat AI, the boss will not think you are excellent, he will think he has been fooled. The consequences are well known and will not be elaborated here.
So never confront head-on when encountering such situations. If you say to your boss "AI knows nothing at all", that is the stupidest move. The boss has just been convinced by AI, and your words are equivalent to calling him brainless to his face.
A smart CIO will not prove that AI is wrong. AI is not wrong, it just does not know the actual situation of your company. What you need to prove is another thing: The company AI describes is not our company.
Translate Technical Accounts Into Management Accounts
When your boss says your budget is too high based on AI's output, do not rush to explain technical details.
The problem is that the more you explain the architecture, the more he thinks you are beating around the bush. What he wants to hear is not technology, but money.
At this moment, you can calculate another set of accounts for him.
"Boss, AI says this step can save 500,000, which is based on two assumptions: 100% data accuracy and full cooperation from all business teams. But our current data accuracy is only 60%. To achieve the effect AI mentioned, we first need to spend 800,000 to clean the data, and then spend 200,000 to confirm data ownership for each business department. AI has not included this 1 million of hidden cost."
"My budget is not too high, it is because I have taken into account the real long-standing messy accounts of our company."
Why does this paragraph work?
Because it translates technical problems into management problems. The boss may not understand what architecture decoupling means, but he definitely understands what hidden cost means. He has been calculating accounts in the business field all his life, as long as you present the accounts in the language he is familiar with, he will figure it out himself.
Let AI Contradict Its Own Previous Conclusion
Take the AI-generated opinion that your boss approves, complete all the preconditions, and feed it back to AI. Do not modify its conclusion, only supplement the actual facts that it did not know before.
You can ask it like this: Suppose you are the person in charge of a traditional manufacturing enterprise, the existing data is extremely chaotic, the general AI literacy of employees is not high, and 60% of the originally planned basic data governance budget has been cut. Please deduce 3 problems that will definitely break out half a year after this plan is forcibly implemented, and estimate the corresponding losses.
The output will most likely make the boss frown.
Then you hand this deduction to him: "Boss, I followed your instructions and asked AI to do an in-depth deduction again, it also reminds that if the basic governance work cannot keep up, the system will most likely crash after going online. What do you think, shall we make sure the underlying foundation is solid first?"
This is not a trick. It is asking the question again after completing all the assumed conditions. AI's answers strongly depend on the preconditions you give. The version the boss asked earlier is based on an ideal enterprise; the version you ask now is based on our real enterprise.
The same AI, with two sets of preconditions, produces two different conclusions. The boss will realize where the truth lies by himself.
No Payment Before Successful Trial Operation
The boss thinks you are too conservative, and says the disruptive innovation proposed by AI is better. At this moment, do not argue, and pass the ball back appropriately.
"Boss, the direction proposed by AI is indeed very good. How about this, we cut 20% from next year's budget and set it up as a separate innovation fund. But we can also set a rule: for all new AI projects, we will first run a 2-week POC. As long as it can operate smoothly in the real business flow, I will immediately add additional budget. If it cannot run smoothly, it means the idea is still in the AI's hallucination. The company's money cannot be used to let AI make trial and error."
The benefits of this proposal are twofold. The boss gets the respect he wants: he is brilliant, he dares to innovate, and he is not fooled by his subordinates. You also get the practical result you want: the risk is locked in that 20% of the fund, no one can bet all the money on a single risky move.
It is worth noting that you also need to be able to translate the boss's grand narrative. When he says he wants to build a full-link intelligent agent, do not argue with him about whether the term is correct. You just reply: "Got it, next week I will first build an assistant that answers common questions in the customer service department, and run it for 2 weeks to track the data." Use a small, implementable product to respond to his grand, uncheckable statement.
Keeping Traces Is Not Underhanded
CIOs need to note that all important decisions should leave traces in forms such as emails. Write down your professional opinions, the risks you put forward, and who finally made the decision. This is not to shirk responsibility afterwards, but to provide a basis for subsequent reviews.
The essence of keeping traces is to store professional judgments and administrative decisions separately. Write down clearly what your judgment is and what the company finally decides. If a problem really occurs one day, this record protects not only you, but also the company. The boss may be unhappy at that moment, but he will thank you for keeping this record when you explain it a year later.
This is professional ethics, not office politics. Keeping traces is to make the facts clear.
Final Summary
AI can write the most beautiful architecture diagram. But it cannot replace IT engineers who get up at 3 a.m. to go to the computer room and restart the crashed server.
In this contest between human beings and machines, the boss seems to hold the sharp weapon of AI. But what he really cannot do without is the people who can implement the work on the ground.
So next time when your boss holds up his phone and reads AI's words to you, what you need to prove is not that AI is wrong. What you need to prove is that the company AI describes is not our company. Remember: the average value can never fill the pit of accurate accounts.
Finally, Lao Yang still wants to ask a question: If the boss of an enterprise does not even have the patience to sit down and listen to the content written by his subordinate who has pulled four all-nighters, and can only rely on AI to make judgments, does this company still have a future?
This article is from the WeChat official account "Xiangjiang Digital Review" (ID: benpaoshuzi), written by Lao Yang, and published with authorization from 36Kr.