I have always believed there's a simple way to determine if a problem is worth solving: put a price tag on it. For instance, using AI. Especially when you try to integrate AI into your business. You say AI can take meeting minutes for you. How much is that worth? You say AI can schedule your trips and draft your emails. How much is that worth? Therefore, recently, I accompanied 17 entrepreneurs from a private advisory board to do one thing: identify a "problem worth $280,000" within their companies and then attempt to solve it using AI.
Why $280,000? A number too small isn't worth the effort for a company. Too large, and it easily becomes a pipe dream. $280,000 is not a precise figure; it merely serves as a reminder: don't waste a tool capable of changing business outcomes on problems that are too trivial.
So, how do you specifically find such a problem? Let me share three stories.
01 The Key to Education Growth: Mass Replication of Top Sales Performers
When it comes to using AI, a common pitfall we fall into is: starting from AI itself to think about what it can do. Don't first think about what AI can do. First think, what is your most valuable problem? For example, an entrepreneur running a chain of educational institutions. Growth in education inevitably involves student recruitment. No matter how good the courses are, if parents don't enroll and pay, it's all for naught. And the core of recruitment is sales. The problem is, sales heavily rely on experience. "I'll think about it" – is there genuinely no need, or have they just not been persuaded yet? When the other party hesitates, should you push forward or step back? You can only close a deal if you get most of this series of judgments right. Therefore, with the same course, some salespeople consistently close deals while others can't sell at all.
This entrepreneur's company has 300 sales consultants. About half of them hover around 200 new student enrollments. Only 15% can achieve over 400 enrollments. Hearing this, I suddenly realized: This is that valuable problem. So I asked: If the average new enrollments could increase from just over 200 to over 300, or even 400, how much additional revenue would that bring? He immediately started calculating: an average increase of over 100 enrollments per salesperson, each salesperson generating an extra $21,000 in performance annually. For 300 people, that's $6.3 million a year. If this capability can be consistently replicated, over three years, it amounts to tens of millions. Although the numbers are estimates, everyone immediately grasped the significance. This is the meaning of "pricing." Often, businesses aren't directionless. It's just that the most valuable problem hasn't been seen.
So, how to improve sales proficiency? The problem with traditional training is: what makes top performers excellent is often the hardest to summarize into scripts. When a parent says "I'll think about it," is it truly a lack of funds, a polite refusal, or needing to go home for their spouse's approval? This is a highly contextual skill, honed through thousands of practical experiences. But with AI, shortening this process becomes possible. How to shorten it? First, you need to find a way to collect sales call recordings. For example, equipping staff with recordable electronic badges. Then, extract patterns from top performers and provide daily improvement suggestions to average salespeople. For instance, if someone closed over a dozen deals in a day, what was their opening line? When a customer hesitated, how did they respond? Deconstruct these high-conversion communications, extract the methods, and replicate them for everyone. Simultaneously, identify where unsuccessful salespeople get stuck. Was the value proposition not fully explained? Or was the push too mechanical? Reasoning to this point, the entrepreneur immediately agreed. Because this is precisely the management action the company has always wanted to implement but couldn't. When the good get better and the weaker ones take an extra step, performance is very likely to increase from an average of just over 200 enrollments to over 300. Scaling the replication of experience inherent to a few individuals – this is the value of AI.
02 Cross-Border E-commerce Shipping Plans: A Mathematical Optimization Problem
An entrepreneur running an Amazon cross-border e-commerce business spoke up next: We have a problem that's driving four colleagues almost crazy. What problem? Shipping. They have over 200 SKUs, many stored in U.S. warehouses. The more inventory you hold and the longer you hold it, the greater the cash flow tied up and the higher the costs. So, you always want to hold less inventory. But once you run out of stock, customers can't buy, leading not only to a significant loss of sales but also a drop in product rankings. Holding less saves money but carries stock-out risks. Holding more is safer but costs more. What to do? They have a dedicated team of four people whose daily work largely involves deciding how much to ship next and whether to use sea or air freight. This is also a typical high-value problem. A shipping plan involves warehousing costs, logistics costs, cash flow efficiency, and sales losses from stockouts. Accurate judgment means the company profits. Wrong judgment means either money is dead in inventory or orders are lost.
How to solve it? In essence, this is a mathematical problem, naturally suited for software processing. My undergraduate degree is from the Mathematics Department of Nanjing University, studying how to find optimal solutions under constraints. For example, a fruit shop had 200 pounds of apples left over from the day before yesterday, 20 pounds remained. Yesterday they stocked 180 pounds, but it wasn't enough. How much should they stock today? Amazon shipping is just a more complex version of the same problem. Faced with such problems, people get fatigued, are inconsistent, and find it difficult to make consistently optimal judgments under the changing constraints of over 200 SKUs daily. So, we must rely on software, on fixed algorithms, to repeatedly obtain the theoretical optimal solution.
How to obtain such software? In the past, you'd need to find algorithms, find developers, easily involving dozens of people over several months. Costs started at the million-dollar level, or even more. Most companies simply couldn't do it. But today, it's different. Because AI has significantly lowered the barrier to software development. So, I opened an AI programming software on the spot and told it all the "constraints": I am an Amazon e-commerce entrepreneur with about 200 SKUs. Now, I need to formulate a daily shipping plan from China to the U.S. based on past sales and forecasts of future sales... How can such a system be implemented? AI quickly provided an answer. As it contained too many mathematical terms, I won't elaborate here. You just need to understand one thing: it basically proved the idea is feasible. The general process is: use a forecasting model to predict future demand. Use an optimization model to factor in warehousing/flow costs and stockout risks. Finally, generate daily shipping suggestions for human approval. Right. It became clear immediately. This entrepreneur was also excited. If this works, it saves not just manpower. The combined savings from warehousing/logistics, reduced capital tied up, minimized stockout risks, and increased sales revenue would far exceed $280,000. Many companies' problems aren't due to insufficient execution, but because problems that should be handed over to mathematics and software are still being shouldered by people.
03 The Dilemma of E-commerce Operations: Too Many Buttons, Not Knowing Which to Press
The third entrepreneur is in e-commerce. She operates dozens of flagship stores. The operations team is busy every day but feels particularly anxious. Why? To use an imperfect analogy: doing e-commerce operations is like a novice sitting in a Boeing cockpit with hundreds of buttons in front of them. But which button actually works? I really don't know. In the morning, should we change the main product image? In the afternoon, should we adjust the price? Should we optimize keywords? Should we add more ad spend? Each action is a "button." So, the operations team sits in the cockpit every day, pressing buttons. Then, they stare at the data on the screen. Sales increased. Which button was pressed correctly? Was it the changed main image or the optimized keywords? Sales dropped. Which button was pressed wrongly? Was the price raised, or was ad spend not keeping up? Everything seems related, but to what extent specifically? No one knows. Causality is already confusing enough, and often you must make quick decisions. A competitor suddenly lowers prices – should we match? Keyword traffic suddenly spikes – should we increase the budget? Which negative review is affecting conversion? These questions can only be answered based on experience. It's exhausting and not necessarily effective. Can we find a way to improve the efficiency of operational actions? You see, this is a problem worth far more than $280,000. For a business worth hundreds of millions, any optimization in decision-making translates to real money. The key issue isn't a lack of data, but the absence of a system that can consistently transform this changing data into executable operational actions.
Let's try together. I continued speaking to the AI programming software: I operate many flagship stores on an e-commerce platform. Now I want to optimize operational actions to drive greater sales. I hope to build a system. This system can analyze the correlation between results and operational actions through data... It can also provide us with operational suggestions, preferably on an hourly basis. As long as the person in charge agrees, it can automatically execute... After listening, AI immediately asked a few questions. How do you obtain real-time e-commerce data? Do you have your own data middle platform? What's the scale – how many stores? How many SKUs? Which operational actions require manual confirmation? Which can be automated? It even specifically reminded: For competitor landed prices, remove member discounts and coupons. It was indeed very thoughtful. After supplementing the information one by one, AI began deconstructing the system implementation approach.
With massive data, AI can attempt to sort out causality. For example, a certain sales increase wasn't due to a better main image change, but happened to coincide with a keyword traffic surge and competitor stockouts. Based on these causal relationships, AI can push operational suggestions every half hour. After confirmation, it can automatically execute. Finally, it can continuously monitor and optimize attribution. Interesting. If the system can continuously track data, analyze results, and provide suggestions, then operations are no longer just about gut feelings and snap decisions. Because the system will increasingly understand which actions are effective and which are not. The accuracy of operational actions will continuously improve. In the past, experience was scattered in individual explorations. Now, experience is crystallized within AI systems.
Seeing this, the entrepreneur's speech quickened: Excellent. Recently, I was struggling with how to streamline operational processes and even considered hiring consultants for a special project. But using AI, it was sorted out in half an hour. The best future enterprises may no longer have massive employee teams but will certainly possess a set of their own "digital assets."
04 AI's Greatest Value: Not Doing Your Work, But Helping You Build Tools
Three stories, finished. I wonder, what are your thoughts? Many people always want to use AI to get results directly. Analyze this table for me. Write this email for me. Yes. AI can do these things. But each time it does them, the outcome can be different. Why? Because AI's chain of thought is unstable. The essence of large models is "generation." It doesn't look up standard answers; based on probability, it creates an answer it deems most reasonable. The path of creation can be different each time. So, although large models are brilliant, if you ask it 10 questions, it might give you 10 different answers. This instability is fatal for business production. Can you tolerate your ERP system using algorithm A for reports today and algorithm B tomorrow? Of course not. Can you tolerate your shipping plan saying to use air freight today and changing its mind to sea freight tomorrow? Certainly not either. What businesses need is not a brilliant artist, but a high-precision machine tool. You give the machine tool an instruction, a blueprint, and it can produce a million identical parts with no deviation. It has no emotions, no fluctuations in state. It is always reliable. In the business world, this "high-precision machine tool" is software. Software logic is fixed. Given an input, it inevitably produces a determined output. This determinism is the cornerstone businesses can rely on.
Therefore, we shouldn't let AI, this "artist," go to the workshop day after day tightening screws. We should let it design the "high-precision machine tool" that can tighten screws. Use AI to write software. Then, use the software to solve problems. In the AI era, the real workflow for businesses is not "AI → Result," but "AI → Software → Result."
Looking back at those three cases: Improving sales capability isn't about letting AI listen to recordings daily and directly give suggestions to salespeople. It's about letting AI help write an "automated analysis of materials, providing optimization suggestions" software. Your managers and salespeople use this software daily. Optimizing shipping plans isn't about letting AI recalculate everything daily. It's about letting AI, based on operations research models, write a "software capable of handling 200 SKUs and dynamically calculating the optimal solution." Your supply chain team opens this software daily. Optimizing e-commerce operations isn't about letting AI change prices or adjust main images for you. It's about letting AI sort out the causal relationships between all operational actions and data, writing a "real-time monitoring, intelligent suggestion" software. Your operations team monitors this software daily.
Perhaps in the future, judging whether a company is advanced may no longer depend on how many employees it has, but on how many sets of software assets it possesses – assets built by AI that solve core business problems. AI doesn't directly change the world. It changes the world by changing software.
Final Thoughts
Recently, I heard a story from an executive at a real estate company with tens of billions in scale. Their company has 20 people whose daily job is: using pen and paper, registering employee clock-in/clock-out times at construction sites. Then, inputting them into the system. I was particularly shocked. It's 2026, friends. A company worth tens of billions still has 20 people daily doing a job software could handle in 20 minutes. At that moment, I strongly realized: This world hasn't been thoroughly baptized by mathematics and software. Over the past years, we always said the internet changed the world, that the digital wave swept through. But in many corners beyond our sight, vast amounts of work still rely on human effort and intuition. Inventory forecasting? Rely on gut feelings. Operational attribution? Rely on morning meetings. Packing solutions? Rely on manual Excel calculations. Payment term management? Rely on the boss keeping an eye. Customer service response? Rely on throwing people at the problem. Behind these lie one $280,000 problem after another.
Therefore, we have also launched a new project for this: The $280,000 AI Implementation Camp. The entire project lasts 5 days. My colleagues and I will accompany you throughout to identify that $280,000 problem within your company. Then, develop a real, usable software prototype for that problem. The first session is already full. The second session is scheduled to start on August 3rd. And there are only ten spots available. If you're interested, you can scan the code to inquire.
Comments