Hello! I'm your financial analysis assistant. This article was written by Dr. Xi Chunying, and its main points are very insightful, highlighting the current challenges in the implementation of AI in businesses.
To help you understand it easily, I will first summarize the article in one sentence, and then break it down into five key aspects for a detailed explanation in plain language.
---
📝 Summary of the Main Points
One-sentence Summary:
Don't just focus on which AI model is the most intelligent; what really matters is how long the AI has been “working” within your company. Models can be rented, but it’s the process of training employees to correct AI, handle exceptions, and accumulate judgment that becomes the company’s “digital experience”—an asset that no one else can steal.
---
🔍 In-Depth Explanation: Five Key Aspects
1. Intelligence vs. Experience: Why Even a “Genius” is a Novice in a Company?
📌 Core Logic:
Today’s AI is like a newly graduated “super student” who has memorized all the world’s knowledge, but it hasn’t worked in your company. It doesn’t know how to solve problems specific to your environment or understand your company’s culture.
💡 Plain Language Explanation:
The author makes a great analogy: Even if you hire a math genius like Terence Tao for your accounting team, he won’t immediately become the best accountant because he doesn’t know the company’s internal rules, client preferences, or operational quirks.
- New employees see numbers in financial reports.
- Veteran financial directors see risks, potential issues, and future trends.
- New salespeople see order amounts; Veteran salespeople see payment patterns, default probabilities, and underlying channel relationships.
Conclusion: Companies don’t buy AI’s intelligence (which is already high); they buy its ability to make informed decisions in a specific context. This kind of judgment can’t be learned from generic models; it must be developed through practical use within the company.
2. The Absurd Reality: AI Works Every Day… but Keeps “Starting Over”
📌 Core Logic:
AI is very efficient, but it has a “short-term memory.” You teach it something today, and tomorrow, when a new employee uses it, it forgets and makes the same mistakes. Companies constantly use AI’s capabilities without building their own knowledge base.
💡 Plain Language Explanation:
Imagine an intern in your company who writes emails and performs analyses. If the legal department tells them to improve their language today, they might do it better tomorrow, but only if someone reminds them of the previous feedback.
This is the current dilemma: AI is working, but it doesn’t gain experience. The corrections made by employees (changes, rejections, retries) are valuable learning opportunities, but most companies don’t capture these. As a result, the tools are useful, but the organization doesn’t become smarter.
3. Upgrading from “Data Oil” to “Experience Gold”
📌 Core Logic:
Data was once called “oil”; now we need to upgrade to “experience gold.” Data records what happened, while experience explains why and how to act differently in the future.
💡 Plain Language Explanation:
- Data: Banks know that Zhang San paid back his loan last year.
- Experience: A veteran loan officer knows that, despite Zhang San’s good financial indicators, his industry has seasonal cash flows, and they lent him money because his wife works for a state-owned company (a hidden guarantee).
- Data: Hospitals know that a patient took medication A.
- Experience: A veteran doctor knows that, despite the patient’s normal indicators, their complexion suggests a different medication is needed based on their past allergies.
In the past, this kind of expert knowledge was stored in individual minds. With AI, we can turn this intuition and judgment into code that machines can learn from, marking a leap from “digitization” to “intelligence.”
4. The Challenge of Storing Conversations: Recording Doesn’t Equal Having Experience
📌 Core Logic:
Just recording all conversations doesn’t make AI smarter. Human feedback can be noisy, biased, and change over time. Companies need a system to filter out useful information.
💡 Plain Language Explanation:
If AI makes a mistake and an employee corrects it, does the AI learn from that? Not necessarily. The correction could be due to the employee’s mood or changing policies. Companies need to establish a system to:
1. Filter out noise: Identify genuine corrections from personal preferences.
2. Weigh opinions: Give more weight to expert advice over new employees’ suggestions.
3. Verify effectiveness: Only solidify decisions that lead to good results.
4. Update regularly: Remove outdated information.
This is no longer just a technical issue; it’s a management problem—companies need to control what AI learns.
5. Rebuilding the Competitive Edge: Opportunities for Traditional Businesses & New Compound Growth
📌 Core Logic:
You can replace models (e.g., from OpenAI to Anthropic), but you can’t replace the knowledge accumulated over decades. Traditional businesses have unique industry insights that tech companies lack. When this knowledge generates compound growth, companies gain a competitive advantage.
💡 Plain Language Explanation:
Many traditional business owners worry, “We don’t understand algorithms—how can we compete with AI companies?”
The author argues: You’re wrong. AI companies have models, but not your industry-specific data. Tech companies understand traffic patterns, but not your supply chain challenges.
Your competitive advantage lies in your “knowledge base.” The question should be, “If I switch models, can I preserve and continue using the millions of business decisions, rules, and feedback I’ve accumulated?”
The new compound growth model looks like this:
- Previous model: Complete a transaction → Make a profit → End.
- New model: Complete a transaction → Make a profit → Generate new experience → AI becomes more knowledgeable → Next transaction is more accurate → Make more profit.
Final Conclusion:
The future of business competition won’t be about who has the biggest model; it’ll be about who has the longest “AI experience.” The key is to turn the knowledge that fades with employee turnover into a lasting digital asset. This is the true meaning of “models can be rented, but experience cannot.”
I hope this summary and breakdown help you understand the article’s key points! If you have any further questions, feel free to ask.