Source | Extraordinary Research

Editor | Tang Jing

Reviewer | Qiu Ping

Image Source: Generated by Unbounded AI

Background

Recently, Amazon Web Services and Extraordinary Capital jointly held the "AI Globalization Special Acceleration Program" offline event in Shanghai, where multiple domestic and foreign leading AI startup companies interacted with over a hundred AI entrepreneurs on site.

This issue is compiled from the roundtable discussions at the event.

Guests

Zhu Tiebin, Founder of Deep Thinking Circle

Tao Feng, Co-founder & President of Gonex

Yan Qidong, Co-founder of Lightstream

Lang Yu, Founder of Maze.Guru

Q: How should different team backgrounds judge whether they are suitable for developing products for overseas markets?

Zhu Tiebin @Deep Thinking Circle: Regardless of whether the founder or the team has overseas backgrounds, going global is a path that many SaaS or AI companies have to choose at present. I have cooperated with many domestic AI product companies, including some independent developers, to help them go global. My experience is that the core and key factor in determining whether a team can truly succeed in the overseas market lies in the team's ability to directly communicate and interact with overseas users.

Many successful AI application products overseas are characterized by a focus on user needs rather than starting from technology. Many companies venturing into the global market for the first time, even those with overseas backgrounds, may have teams with backgrounds in overseas engineering. However, when developing products, they have very little communication and feedback with users. This is very detrimental to creating a truly successful global product.

If a product iteration mechanism based on direct user interaction and feedback is not established, and the product is developed based solely on the team's own ideas or technology, you will quickly encounter problems. For example, you may diligently update tweets and post content related to product features on Twitter and LinkedIn every day, but the views are scarce, and the engagement and conversion are even lower. After releasing on Product Hunt, there may be a small surge in growth, but you still don't know what to do next. This is a problem that almost all global companies will encounter after in-depth communication. The core to solving the question "what to do next" is the need to communicate with users, understand the problems they are trying to solve with the product, and find their pain points and the best solutions.

Lang Yu @Maze.Guru: In Hangzhou, there are actually many apps or applications in potential overseas market-leading fields, with annual revenues reaching the billion level, but many people may not know that they are launched by domestic companies.

When going global, the following two points need to be considered:

First, clarify whether you want to build a brand or create a product. If you want to create a product, the team needs to develop a very vertical product to push into the market. If you want to build a brand, you need to deeply cultivate a product that can provide users with more experiences.

There are many differences between overseas products and domestic products. Overseas apps tend to focus more on vertical application areas, where an app only provides a simple service, and the team spends a very long time cultivating it. On the other hand, domestic products still prefer to create comprehensive products, making users feel like it is an all-in-one product. This model can be quite complex when promoting it overseas.

Second, choose a specific region or market for overseas business, rather than casting a wide net. Many times, people tend to think that going global is simply a division between China and overseas, but in fact, overseas markets are divided into many different regions. There are markets in Japan and South Korea, North America + Brazil, the Middle East, Southeast Asia, and so on. The operational strategies for different markets are completely different. Therefore, it is important to choose the market and determine which market you will invest in the most in the future. It is not advisable to invest in a very broad manner, as it can lead to operating costs and funding being wasted without bringing about good results.

Q: What are the differences in the promotion methods for the same product targeting domestic and overseas markets?

Lightstream @Yan Qidong: In 2016, there were relatively high expectations for the overall development of SaaS in China; however, after this year, most SaaS companies and practitioners have become relatively more pessimistic. A major turning point this year is going global to target a broader market, which is indeed a path that Chinese SaaS companies can consider.

The specific differences are as follows: First, there are significant differences in compliance and privacy issues (especially in the European and American markets).

Domestic customers often have an issue of being unwilling to spend money. Therefore, their expectations for the efficiency, service, and underlying data security of the product are not very high. Additionally, due to the lack of many advantages, there is not as much demand for some cloud services. However, in mature markets such as North America, the standards and conditions that products must meet are very strict.

Second, market differences (each region overseas needs to be individually focused on).

Going global is not a simple distinction between China and overseas, but rather each region needs to be individually focused on. For example, Southeast Asia and Europe and America are two completely different markets. In the process of choosing to go global, the SaaS industry should expand into markets with stronger consumer purchasing power.

Third, localized services and support systems are very important.

Especially for some top customers, it will be very difficult to develop the business without a localized support system. Currently, Lightstream's strategy is to find partners in various regions who can take our products global. These partners with localized and regional industry characteristics play a key role in business expansion in their local areas.

Q: What insights does Gonex have on the trend of going global?

Gonex @Nick: The first dimension is data security issues.

Data security issues are more complex in terms of human resources, including a lot of sensitive information, such as employee personal files, SMS mobile data, and so on. Data security is a major concern when it comes to cross-border services, whether serving Chinese or international companies, and it requires more time in this dimension.

The second dimension is how to truly get close to users.

In different markets with different levels of maturity, user needs are completely different. Currently, the majority of Chinese companies going global are in a state of relatively low maturity, especially in terms of human resources, where many things overseas are still in the stage of being educated. Therefore, Gonex has launched the following products: The first product is somewhat like the "Ten Thousand Whys" of overseas HR policies, as a lead-in product, its usage rate is particularly high. We use AI technology to turn it into a vertical small application. Another product addresses some blind spots and pain points. For example, how to evaluate candidates from different countries? When it comes to cross-border and cross-domain, customer pain points become numerous. Therefore, we are constantly iterating and developing products.

The third dimension is user stickiness.

Chinese customers, due to their cross-border management in the early stages, are very reliant on service providers, so their stickiness is very good. After starting to go global, Chinese characteristics also need to gradually adapt to international trends, accept international rules, and gradually begin to accept the practices of European and American companies. In this process, the research and development efficiency and adaptability of domestic enterprise products will continue to improve.

Q: Can our existing experience in going global with products be directly replicated to AI products going global?

Zhu Tiebin @Deep Thinking Circle: For Chinese teams developing products for overseas markets, the current promising direction seems to be productivity tools or SaaS products, and adopting the PLG (Product Led Growth) growth model in the early stages. One advantage of PLG is that it doesn't require physical presence overseas in the early stages; instead, it can use online social media, content marketing, and SEO to generate traffic and achieve early growth for the product. For example, Twitter is a popular channel for promoting AI products, and many AI products initially gain high traffic and quality attention through Twitter and other online social media, successfully achieving a cold start.

Additionally, Product Hunt is also a great promotional channel. Although some people may think that the traffic from Product Hunt may only be a short-term surge and not worth much investment, Product Hunt actually has a unique advantage. Once a product enters the TOP5 or TOP10, it will receive a lot of attention and sharing from newsletters and KOLs. Newsletters are equivalent to public accounts overseas. Product Hunt is a source of material for many newsletters, including many well-known KOLs on Twitter, whose daily tweet material also comes from products on Product Hunt.

Furthermore, AI directories are also a great promotional channel for this wave of AI products. Although many overseas directories are paid, if you can achieve a good ranking on Product Hunt, you can directly email the directory administrators and basically get listed for free, and possibly even appear on the homepage. Some directories may even actively include your product. I know that the automated inclusion rules of some directories involve crawling the top 10 products on Product Hunt for inclusion.

In addition, the pricing strategy for AI products has undergone some new changes compared to traditional PLG SaaS. Traditional PLG SaaS products usually adopt a strategy such as "Freemium," where users can use the product directly without needing to provide credit card information, attracting seed users. However, current AI products tend to prefer binding cards first and then offering refunds within a certain period. Why this change? First, the barriers to entry for AI products are not as strong, so when a product is good, there may suddenly be many imitators, so most of them tend to collect money first. Additionally, most AI products are based on large model development, and any usage requires payment for the large model. Therefore, I have observed an overall shift in pricing strategy.

My personal experience in the past year with AI applications and middleware products is that the capabilities of underlying models have not yet reached the stage of paradigm shift that everyone expects, and can only support the step of SaaS + AI. This also means that the current capabilities of underlying models cannot support the creation of a truly AI-native product, but only the capabilities of SaaS + AI. Faced with this situation, many startups may encounter the "downward pressure" from overseas SaaS giants. For example, a startup creating an AI writing plugin will face significant disadvantages compared to integrating AI writing features directly into tools like Notion and Craft, because the existing workflow has not undergone a truly revolutionary change. Additionally, many previously popular agent building platforms may directly face pressure from OpenAI, and the boundaries of underlying large model companies are not yet very clear, which is also a risk that AI startups face, as they cannot establish their own true barriers, and their important features may be directly included in a single update of the large model.

Therefore, I believe that the current opportunity lies in identifying a certain aspect of real user demand and setting reasonable expectations. Currently, I have collaborated with many independent developers to go global, and it is already very good for each product to expect to generate monthly revenue of $5,000 to $20,000. In this area, there are already some successful product examples overseas, such as Chatbase, PDF.ai, and SiteGPT, among others. In this process, it is more about waiting for the right opportunity to appear, rather than excessively pursuing the perfect AI-native product.

Gonex @Nick: There are many software companies and developers with strong technical capabilities overseas. Our strategy focuses on leveraging information asymmetry. Taking HR SaaS as an example, although there are many excellent software companies overseas, the human resources departments of even top companies in the United States may only be familiar with their own applications and may not know much about other key information. This information asymmetry provides us with opportunities.

I believe that although American companies may be very strong in certain fields, once they enter China, Southeast Asia, or Europe, they will still face many problems and need to address many challenges that they cannot cope with. In this regard, we have the opportunity to provide unique solutions. Therefore, our strategy may be different from other companies.

Lang Yu @Maze.Guru: First, PLG and SLG are not mutually exclusive, but rather a phased relationship. Many times, the applicability of PLG and SLG varies in different situations. For example, in cases where funds are not abundant, we are more willing to choose PLG.

For us, the first thing we consider is this: AI has not matured to the point where it can deeply intervene in B-end applications. AI projects on Twitter or YouTube are still more about entertainment operations, curiosity, and personal enjoyment. Products in this direction are very suitable for the PLG model and can achieve very good results from the start. Some KOLs and major influencers online are willing to share your product, and you may even be able to establish a project library on Product Hunt, which is very helpful for the early growth of the project and can receive some good feedback.

Second, after the product project goes live, what should be the next step? AI is a major trend in today's era. When Chat GPT releases new content, its user traffic and social attention grow very quickly, and it is essential to invest in related operations at this time. In my opinion, YouTube has the best quality of AI user sources and the highest willingness to pay. Although there are more users on Twitter, overall, people tend to try products rather than subscribe and pay for them. Therefore, investing some operational funds in KOLs on YouTube can achieve relatively good results.

Project promotion in China is usually relatively simple, such as posting information on Xiaohongshu, doing film and television advertisements, or on platforms with concentrated traffic like Weibo and Douyin. But overseas, traffic is very dispersed, and you need to invest corresponding operational funds in different platforms. However, this situation is not suitable for everyone, and timing and financial preparation are also very important. At the same time, you need to pass a test in your own mind while preparing funds, because overseas promotion is very expensive. If the timing is indeed right, then investing in promotion is definitely worthwhile.

Third, localized business is a necessary condition for scaling up the product. We chose Web3 not only because it is more in line with the original field, but also because the development of the product cannot be separated from localized business. When doing AI products, there will be some B-end customers, and it is difficult to deeply enter the B-end business of a specific country without a background in Web3. Based on Web3, we can find similar people or some large B-end customers.

Q: Does Lightstream currently have a certain integration of AI in its products?

Lightstream @Yan Qidong: First of all, based on the actual situation, we are targeting the market for rapid application development. One rule is that IT costs are usually very high for enterprises, but the satisfaction on the business side is often very low. The main reason for this is that traditional software development efficiency is too low, including the entire mode of development and collaboration.

We have been trying to reduce the labor cost in the middle of this process in recent years. Empowering business personnel to develop applications autonomously is the most efficient approach because they have a clear understanding of their needs and can have a lower threshold. Through Lightstream's practice, in the past few years, low-code and no-code have significantly reduced the cost of IT within enterprises, basically achieving a level of 2% to 1% of the original, and customer satisfaction has also improved.

With AI being so popular this year, we have also been forced to think: what will be the endgame of this market? We made a basic assumption: in the future, the work of software development or productivity will definitely be gradually replaced by AI.

This year, we made an attempt to achieve a grafting method using GPT's open-source capabilities. Specifically, users input a requirement description on our platform, such as "I need a CRM system, including which features?" in about 50 words. We pass this requirement to GPT, and it can return structured data, calling our product's interface. Based on this logic, the entire process takes less than 30 seconds to build a basic CRM prototype.

We plan to release this application in July and conduct internal testing in August. At that time, many customers showed strong interest in this feature, and many people signed up to use it. However, in fact, about a month later, almost no one mentioned or inquired about this feature. The reason is simple: although we have implemented this feature and it can optimize the initial work of the application, the value provided to users by saving an hour of initial development work is relatively limited. Although it looks cool, to truly scale up user applications, it will still require a long road. We are still observing, this is an attempt on our development side, and we have been continuously improving this feature.

Secondly, we believe that AI will structurally change the logic of internal business management and operational management within the entire enterprise in the future. Looking back, from the 1960s to today, for enterprise-level software, there has not been much difference in the form of software from interaction to functionality, all based on some optimization of upper-layer applications on the database. But this will change greatly in the future, as the interaction between humans and machines, and humans and software, has always been changing.

Now we often encounter automation, where some predefined logic is executed by machines. But we believe that in the future, the entire automation will not only stay at execution but will gradually transition to decision-making and perception. For example, today's automation only helps me send an email, but in the future, it can make decisions for me about when to send an email and successfully send the email. This will also be the trend of structural and formal changes in the future for SaaS.

Q: How will the current popularity of Agents impact low-code/no-code vendors like Lightstream?

Lightstream @Yan Qidong: To be honest, there will definitely be an impact. Under foreseeable trends, the agent model will definitely replace some low-value repetitive production work within enterprises. As the infrastructure becomes more and more sound, software engineering becomes more and more low-value repetitive labor; the more sound the underlying technical infrastructure, the less important the upper-layer application. For example, a programmer writes an application, can call a lot of intermediate interfaces, and quickly write an application using the underlying infrastructure, but much of the code he writes is very low-level and can actually be completely replaced.**

Software engineering is divided into two tasks. The first task is "understanding my problem." For example, if you need to help me solve a problem today, you need to understand my problem, to realize my problem, and understanding my problem is the highest value-added part; if only talking about solving the problem, there are actually many tools; but the threshold for understanding the problem is relatively high, which is also the key to implicit context. Therefore, this problem is relatively high in terms of the threshold for ToB services, and this series of operations can indeed be solved by AI in theory, but not so quickly in the short to medium term.

So in my opinion, the first entry point for agents is programmers, and product managers and business personnel may still be needed, but programmers are indeed at high risk.

In the future, these things will definitely change. At the current stage, what we can call assets is our understanding of customer needs and the skills of these industry cases at the business level. Technically speaking, this will not become a new hard barrier in the short to medium term.

Q: When the AI trend comes, do you think this will also be an opportunity for low-code/no-code?

Lightstream @Yan Qidong: The correlation is very high. Lowering the threshold helps to promote the rapid popularization of technology, and this theory is completely valid in logic. We are trying AI for an important reason, which is to continuously lower the application threshold for enterprises. Lightstream's vision is to create a no-code platform that everyone can use, is able to use, and knows how to use. We serve business personnel rather than IT personnel. We are committed to enabling 200 million business personnel to have development capabilities, not just improving the efficiency of 7 million IT workers in China. From a technical perspective, lowering the threshold is a key point.**

Another challenge is the connection between supply and demand. In this regard, our capabilities are limited because we are still a startup. Therefore, we plan to cooperate with platforms such as DingTalk, WeChat, or Feishu to reach more enterprises. These platforms can do more in market education, and we are committed to meeting market demand and promoting the popularization of these tools.

Q: How should we view the combination of AI and Web 3?

Lang Yu @Maze.Guru: First, it is a manifestation of decentralized data. For example, after the open source of stable diffusion, it is open to the public, and the public can use their own data to train corresponding models, presenting a special style of their own painting, which is a manifestation of decentralization.**

How can decentralized AI tools generate revenue? Or through what means can their value system and ecosystem be put into operation? That is to adopt a paid business model. If no money is spent, it will only become a gathering place for enthusiasts as a kind of open-source community application. In this case, on-chain in Web3 can help execute some rights confirmation work. Large language models can be traded as NFTs, achieving transfer, change of usage rights, or transfer of ownership. Through this model, the two can be combined to generate corresponding economic benefits, establish a circular value system through decentralized data and large language models on-chain, and make the entire ecosystem operational. Web3 can establish an incentive mechanism, and this mechanism is very transparent, and aspects such as chains are very open.

This actually extends the industrial chain, gradually transforming some amateur activities into professional affairs. Compared to the past, many painting markets, model markets, and Bilibili are more dominated by enthusiasts, and the products they make hope to receive user feedback, user love, or train private or commercial models. However, there is a significant risk in this case.

In Web3, large language models actually represent the organization form of data, which is a high-level form. Only those who use it on the chain can obtain the right to use the data, making it easier to confirm rights.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。