Summary of Key Points
This article demystifies the glamour surrounding technology, corporate titles (such as those from large companies), and capital narratives by examining cases involving Chen Mian and AI video applications like LibTV and Flova. It emphasizes that, whether for individuals or businesses, the key is to find an irreplaceable niche. For AI applications, this niche lies not in the models themselves but in the “middle layers” of value, such as demand definition, workflows, and data closed loops. For individuals, true capabilities lie outside of organizational platforms. Moreover, real demand must be verified through multiple dimensions, including repeat purchases and natural growth; for hardware, the return rate is a crucial indicator, and business decisions are based on long-term accumulation from actual transactions.
1. AI Applications: Focus on the Middle Layer, Not Just the Models
Many people think that AI applications are merely “models wrapped in a UI” and that they will eventually be swallowed up by large model manufacturers. However, the middle layer will never disappear—as long as users don’t want to research model parameters or deal with complex processes, they will need someone to help them achieve their goals. For example, when creating an ad video, users don’t want to just “generate a video” but to have 20 versions ready for Meta by tomorrow, each with different elements and a unified character design that is easy to modify. Middle-layer applications like LibTV connect various models, materials, and workflows, allowing users to get the final product without worrying about the underlying technology. Large model manufacturers may be skilled in technology but cannot meet the specific needs of every industry; NVIDIA won’t start its own robotics company, and淘宝 doesn’t produce all its products itself. The value of the middle layer lies in its control over users’ workflows, data, and brand perception—Adobe’s platforms handle designers’ projects, materials, and approval processes, and users won’t switch to another software just because of a model update. These are barriers that others cannot overcome.
2. Real Demand: The Test from “One-Time Purchase” to “Repeated Purchases”
Many AI applications boast about high Daily Active Users (DAUs) and revenue, but these may be misleading metrics. Real demand must pass five tests:
- First test: Users actually pay for the product (this is more reliable than surveys but still not enough).
- Second test: They pay at the full price (not through discounts or subsidies).
- Third test: There are no returns, indicating the product is useful.
- Fourth test: There is natural growth—users continue to use the product even without advertising.
- Fifth test: The product performs well during economic downturns and recovers quickly during upswings.
For example, an AI tool that generates 10 million in revenue per month through ads but costs 12 million may be losing money; another that generates 3 million with 70% of sales coming from natural traffic is more sustainable. Real demand means that users are willing to continue paying for the value it provides.
3. Demystifying Corporate Titles: True Abilities Lie Outside of Organizations
Many people place too much faith in corporate titles (e.g., “Head of a certain department”), but these often come with organizational platforms. In large companies, directions are set, budgets are allocated, and resources are readily available. Success may be attributed to the platform rather than individual capabilities. For instance, someone who manages 100 people in a large company might struggle to find funding or recruit new staff after leaving because their achievements relied on the organization’s brand, traffic, and supply chain. True ability lies in being able to start from scratch outside of an organization—defining problems, securing funding, developing products, selling them, and managing cash flows. Entrepreneurs like Zhang Yiming succeeded not because they fit certain labels but because they could achieve things without a platform’s support.
4. The “Second Vote” of Hardware: Return Rates Reveal More Truth Than Sales
Hardware is different from software; users need to open, install, and use it, so return rates provide a more honest reflection of its quality. For example, if a hardware product sells 100,000 units through ads and crowdfunding but has a 30% return rate, it indicates that users found it ineffective or unsatisfactory. Good hardware should attract natural traffic, have low return rates, and gain word-of-mouth from existing users. The real demand for hardware is not about how many units are sold but how many users stay loyal to the product.
5. The Secret of Business Judgment: Learning from Real Transactions
Business judgment is often seen as mysterious, but it’s actually a result of long-term experience with real transactions. For example, salespeople analyze SKU data, inventory turnover, and return rates to understand why certain products don’t sell well. A seasoned consumer won’t just look at sales figures but will also consider factors like whether the revenue comes from discounts or if there are high return rates and if sales continue without advertising. These insights help filter out false prosperity and reveal true value.
In Conclusion
Whether for AI applications or individuals, focus on becoming irreplaceable by solving specific problems that others can’t solve and creating a dependency among users. That’s your niche.
(End of article)