Interview with Lu Cong of Shenzhen Topology: AI and Data-Driven Strategies Reshape New Paradigms for Cross-Border Independent Site Operations

[Ebrun Original] At the intersection of globalization and digitalization, a group of cross-border service providers driven by AI and data are reshaping the fundamental logic of Chinese brands going overseas. The story of Shenzhen Topology began in 2016—starting with Google and Facebook marketing services, venturing into independent site self-operated business in 2020 to achieve rapid growth, and now evolving into a full-chain managed service provider for independent sites serving over 300 overseas enterprises. Recently, we had an in-depth conversation with Lu Cong, founder of Shenzhen Topology, exploring how AI is reconstructing the competitive edge of independent site operations, the industry trend where GEO is replacing traditional SEO, and the AI-driven upgrade path for managed service models.

"Making successful overseas expansion move from chance to certainty." Lu Cong got straight to the point. Behind this slogan lies Shenzhen Topology's deep belief in data elements and AI technology—transforming independent site operations, which previously relied on individual experience, into a scientific, quantifiable, replicable system with a high success rate.

Ebrun Think Tank: Starting from the slogan 'Making successful overseas expansion move from chance to certainty,' how does Shenzhen Topology utilize 'data elements' and AI technology to transform independent site operations, which previously relied on individual experience (such as product selection, ad placement), into a scientific, quantifiable, replicable system with a high success rate?

Lu Cong: For many sellers, transitioning from platforms to independent sites is a huge leap. On platforms, people look at very few data metrics—traffic, add-to-cart rate, CPC, ad spend, ROAS, about ten or so. But on Facebook, there are a total of five hundred data metrics. How many metrics you can view determines the level of your advertising capability. We have worked with many clients, and one reason their independent sites are unsuccessful is that they lack even the basic knowledge of data indicator systems, let alone application.

Today's data sources are not just first-party data—sales, add-to-cart, checkout, user registrations from your own website. There is also a large amount of second-party, third-party data, and even industry big data. For example, SimilarWeb, SEMrush. For any category we work on, the first step is to obtain industry big data through these tools—understanding industry size, growth rate, how competitors operate, traffic sources, and which countries. This provides excellent preliminary guidance for clients, avoiding detours.

In the past, these distinctions required extensive insights and adjustments from one's own first-party data. But now, a large amount of big data can tell us how industry benchmarks operate before we even start—'copying homework' is taking a shortcut.

Ebrun Think Tank: You mentioned that nearly half of your company's new clients currently come from recommendations by AI tools like DeepSeek and ChatGPT. Can this be seen as a landmark case of 'GEO (Generative Engine Optimization)' replacing traditional SEO?

Lu Cong: Currently, about half of our company's business, in terms of user sources, comes from traffic inquiries recommended by AI tools like DeepSeek, Doubao, and ChatGPT. This is indeed a landmark change.

GEO is currently the most underestimated growth opportunity in the cross-border circle. Drawing on our past SEO experience (I started in SEO in Singapore in 2012), we have developed a set of GEO methodologies. Combined with practical implementation for managed service clients, we have achieved good results. We ourselves have clients who have achieved a hundredfold growth in GEO traffic within a year.

Ebrun Think Tank: Shenzhen Topology serves over 300 enterprises, covering factories, platform sellers, and brand owners. What are the initial gaps in data application capabilities among these three types of clients?

Lu Cong: The gap is quite significant. Some foreign trade factories and OEM enterprises can be said to have no basic data awareness; traditional trade shows and B2B platforms are their main ways of acquiring customers. Amazon sellers are slightly better, but their data perspective is basically limited to the dozen or so metrics within the platform. Brand owners are relatively better, but there are not many who can truly use data effectively.

The first thing we often do for clients is not to place ads, but to help them establish a data indicator system—telling them what to look at, how to look at it, and how to use it. The vast majority of cross-border e-commerce sellers lack this data application capability; they don't understand these tools, can't collect data, and even if they see it, they don't know how to integrate and analyze it.

Ebrun Think Tank: From product research, site building, traffic acquisition to customer service, in which specific links of the 'Seven-Step Service System' has AI technology penetrated?

Lu Cong: Our service process can generally be summarized into seven links: research, site building, marketing/advertising, operations, payment, logistics, and customer service. In current work, AI is basically involved in every link. For example, in the research phase, we obtain industry big data through tools like SimilarWeb, then use AI for data integration and analysis. In the site-building phase, mainstream platforms like Shopify themselves have good AI tools to help merchants with analysis and automated site building. In the advertising/marketing phase, Google and Meta's own AI helps us with optimization; there are many tools for automated budget and bid adjustments. In the operations phase, AI helps us with A/B testing, heatmap analysis, and conversion rate optimization. And so on.

The customer service link has seen the biggest change. Previously, when we operated our own brands, we used a Philippine customer service team. Now, AI can solve about eighty percent of customer service issues, including pre-sales inquiries, after-sales queries, and complaints. We use AI tools to achieve 7×24 hour intelligent customer service.

Ebrun Think Tank: The company has a diverse team both domestically and internationally. In terms of AI-assisted content localization and user behavior analysis, how does this global data network collaborate?

Lu Cong: My career started with an international team—working in Singapore after graduation, with many Filipinos, Indians, and Malaysians. What we are doing now is leveraging the advantages of the Chinese supply chain to go overseas. In the future, international enterprises should make good use of global human resource advantages.

The customer service team is in the Philippines, about six people, with low costs, shift capabilities, and language proficiency. For technical matters, such as site building and speed optimization, we have collaborated with an Indian team for over four years; they have low costs and strong professional capabilities. In the future, to serve local American enterprises, we will also hire Americans for creative and design work.

Regarding internal cultivation, each of our colleagues first researches local festivals, looks at local social media, understands what communication forms and advertising presentation styles they identify with, and then applies these to their own advertisements. The core methodology is universal, but the tactics need to be adapted to different countries.

Ebrun Think Tank: As a service provider, how does Shenzhen Topology ensure client data security while utilizing cross-client industry data?

Lu Cong: We take this issue very seriously. First, we are only the managers and users of the data; the clients are the owners. This boundary is very clear. We are just stewards; we cannot take the owner's money or furniture.

Second, when used for public courses or case presentations, data is anonymized—brand information or raw data screenshots will not appear; data is blurred.

Third, when serving clients, we directly promise: for the same industry, the same product, in one country, we only sign one client. This avoids industry competition and data misuse. We have no incentive to use the data elsewhere.

There are considerations in systems and culture, and technical requirements as well. For example, for exporting users and data from the website backend, we have norms—many accounts do not have export permissions; you can use but not export.

Ebrun Think Tank: You mentioned that GEO is the most underestimated growth opportunity in the cross-border circle. Could you illustrate the essential difference between GEO strategy and traditional SEO with a specific case?

Lu Cong: Take NihaoJewelry as an example. This jewelry independent site, through our managed service, saw monthly sales grow from $18,000 to $840,000 in 12 months, with the comprehensive ROAS increasing from less than 4x to over 12x.

Traditional SEO focuses on keyword rankings, targeting the search engine results page. But GEO focuses on making AI recommendation engines—like DeepSeek, ChatGPT, Doubao—recommend your brand when answering user questions. This completely changes the content strategy: you are no longer just optimizing keyword density, but building a content system that can be understood and recommended by AI.

This trend has just begun.

Ebrun Think Tank: In the case mentioned, 'comprehensive advertising ROAS increased from 1.5 to 4 in 3 months,' how much of a role did AI technology play? In the future, will AI completely replace manual optimizers?

Lu Cong: Take WinBridge (megaphone, US market) as an example. Comprehensive advertising ROAS increased from 1.5 to 4 in 3 months, and the natural conversion rate increased by 85%. AI played a very significant role here—automated bidding, dynamic creative, audience prediction—these are all built into Google and Meta's AI systems.

But will AI completely replace manual optimizers? I don't think so. AI greatly lowers the barrier to entry, but how to use AI's capabilities well still has certain requirements. In the future, AI will definitely be a powerful tool for enhancing the data application capabilities of many companies. Many companies don't have this awareness yet; they are not using it. Those who use it might catch up with a decade of others' accumulation in just one or two years; it's entirely possible. But this isn't about how great an individual is; it's about the power of 'human + AI' collaboration.

Ebrun Think Tank: The company once self-operated B2C business and maintained high growth. What unique capabilities were accumulated during this experience of 'being both a coach and an athlete'?

Lu Cong: During the pandemic in 2020, we launched cross-border e-commerce independent site B2C business, maintaining over 20% average monthly growth in two product categories. This experience allowed us to accumulate comprehensive independent site operational capabilities—from product selection, site building, and advertising to logistics and customer service; we have personally run every link.

More importantly, we accumulated a complete set of data assets—user behavior models, product selection algorithms, advertising placement models. These capabilities feed back into the services we provide to clients, allowing us to not just 'talk the talk' but truly know the pain points and solutions at each stage.

Ebrun Think Tank: What is the biggest bottleneck currently constraining independent site sellers from unleashing the value of data? How does Shenzhen Topology help clients overcome these constraints?

Lu Cong: The internal bottleneck is primarily team cognition—as mentioned earlier, many sellers don't even have knowledge of data indicator systems; this is the main internal bottleneck. External obstacles are mainly platform data barriers and privacy policy changes.

Our solutions are two-fold. The first is to help enterprises enhance their internal capabilities—conducting training, public courses, and live streams. We have held over 100 sessions on independent sites, Facebook, etc., in more than 20 cities across the country, training over 3,000 participants cumulatively. The second is to recommend enterprises to introduce external capabilities—managed services, leveraging external professional expertise.

Ebrun Think Tank: Looking to the future, what are Shenzhen Topology's new strategic plans regarding 'AI + Data'?

Lu Cong: This is a new proposition. We are still practicing and haven't yet elevated it to a very clear strategy. But the vague direction is clear—the future is about AI, but more importantly, data.

First, GEO is a direction we highly value. Second, make good use of Shopify's own AI system for automated site building and data analysis. Third, leverage top foreign AI models to help us with site building, analysis, and strategy formulation.

We have hired some fresh graduates; after extensive exposure to AI, they quickly learn what people with three, five, or ten years of experience know. So AI is helpful for employment—it is creating things that humans couldn't do originally.

We have been thinking whether to design AI capabilities into a product in the future—for example, using AI to automatically diagnose websites and perform automated optimization. I meet with five to ten clients almost every week and find that the vast majority of companies don't know where their problems lie. If we can internalize diagnostic experience into a tool for use by a large number of independent site sellers, it would be very valuable.

Ebrun Think Tank: If you were to give one core piece of advice to those just starting out, eager to build their own data competitiveness for going overseas, what would it be?

Lu Cong: In one sentence—actively embrace AI, data, and branding. You cannot operate like in the past cross-border e-commerce platforms, opening numerous stores, links, and listing products; this model is becoming increasingly unviable. Once you value AI, data, and branding, you open a completely new chapter. Use AI to build data processing capabilities—just like the data element competitiveness model you proposed, start from the four dimensions of 'acquisition, governance, utilization, security,' and AI can help you in every link.

In the future, for running an independent site, all you need to know is how to ask questions and state requirements. AI can help you with advertising creatives, placement plans, page building and optimization, automated process design, data acquisition and analysis. These are things that traditionally required people with three to five years of experience, but AI can do them.

Of course, prompt engineering capability is crucial. AI prompts are a conversational process, not one-way questioning. Being good at expression, providing context, requirements, cases, and related information to AI, it will be used very well.

Conclusion: 'Making successful overseas expansion move from chance to certainty.'

Shenzhen Topology's practice shows that Chinese brands going overseas are shifting from 'traffic-driven' to 'data and AI dual-driven.' From product research to advertising placement, from website operations to customer service systems, AI is reconstructing every link of independent site operations. And GEO, this underestimated growth opportunity, is opening new growth space for enterprises willing to embrace change.

"Making successful overseas expansion move from chance to certainty"—this is not only Shenzhen Topology's mission statement but also a promise of certainty given by the AI and data era to all enterprises going overseas. In today's deep integration of technology and globalization, the exploration by Shenzhen Topology may provide a new paradigm for Chinese brand independent sites going overseas. Ebrun Think Tank will continue to focus on the competitiveness enhancement of data elements in related industries and enterprises, and report on new developments, achievements, and cases.

Contact email: huangbin@ebrun.com