Qianfeng Dialogue with Zhao Xianlie of Panfeng Intelligence: The Paradigm of Cross-Border E-commerce Shifts from 'Finding People for Goods' to 'Content as Channel'

【Ebrun Original】Recently, during the 2026 Qianfeng Award interview, Ebrun, Ebrun Think Tank, and Zhao Xianlie, co-founder and COO of Hangzhou Panfeng Intelligent Technology Co., Ltd., held an in-depth discussion around 'The Paradigm Evolution Brought by AI and Going Global.'

Ebrun: This year's Qianfeng Award takes 'The Paradigm Evolution Brought by AI and Going Global' as its core theme. In the context of AI cross-border e-commerce, how do you understand this round of industrial paradigm evolution driven jointly by AI and globalization? What is Moras's positioning and business model?

Zhao Xianlie: This round of 'AI globalization' brings not a simple efficiency improvement, but a 'paradigm evolution.' In the past, going global followed the logic of 'finding people for goods'—stocking up, investing in traffic, buying volume, competing on capital and channels; after AI emerged, content itself became the channel. Two fundamental changes: on the supply side, previously a 10-person team served dozens of influencers, now AI makes content production capacity no longer the bottleneck, and 'people' change from producers to commanders of AI production capacity; on the globalization side, content e-commerce platforms like TikTok Shop have turned 'shelf e-commerce' into 'content stream e-commerce,' and going global is no longer about copying the domestic model abroad, but using AI to drive the cost of localized content down to near zero.

Moras's positioning is an AI-driven cross-border content e-commerce platform, which can also be understood as 'an AI money-making agent for influencers.' The business model is bilateral: on the supply side, we use AI (video generation, intelligent product selection, material fusion) to reduce the production cost of a sales video to a fraction of the original; on the demand side, we connect with brand merchants in the US TikTok Shop region who have growth needs. Influencers publish videos and share commissions after transactions, and the platform earns returns from incremental GMV. We don't sell goods, don't hold inventory, we sell 'AI content production capacity' and earn money from results.

Ebrun: Which core business links is AI currently implemented in?

Zhao Xianlie: Four core links: First, AI video generation—product links or images go in, sales videos come out; Second, intelligent product selection—algorithms pick out products with potential in the US region from a pool of tens of millions or even hundreds of millions of products; Third, fused videos—intelligent mixing and reuse of multiple materials to amplify the lifecycle of hit content; Fourth, LUI—a Personal Agent assigned to each influencer to help them gain insight into industry hotspots, do topic selection, write scripts, make videos, manage publishing, and provide data closed-loop feedback, equivalent to productizing operational experience. Vertically, AI connects 'product selection—creation—publishing—conversion' into an AI workflow; horizontally, it integrates 'people'—turning scattered creators around the world into a schedulable content production capacity network.

Ebrun: Could you share a most representative 'AI value chain' case?

Zhao Xianlie: The most representative is 'AI batch sales videos.' The industry's common pain point is clear: in the content e-commerce era, the biggest bottleneck for merchants and influencers is not traffic, but content production capacity—a sales video that can generate orders used to require a whole set of manual work: product selection, shooting, editing, publishing, which ordinary people simply couldn't afford. Our approach is to AI-ify this chain: influencers select products, AI generates multiple sales videos within minutes, and publishes them directly. The result is that content production capacity goes from 'day-level' to 'minute-level'; for merchants, it means using the cost of one original video to get the exposure and conversion opportunities of dozens of videos.

Ebrun: How do you calculate Token and computing power costs?

Zhao Xianlie: Internally, we don't calculate by 'how many Tokens were called,' but by 'cost per effectively published video'—this is the only honest metric, because only videos that are published and can participate in conversion are effective production capacity. Video generation is the big part; our approach is model routing, off-peak scheduling, and reuse of hit materials to continuously push down the per-video cost.

Ebrun: Regarding data competitiveness, you said that in this system data is not a 'supporting role,' it is the product itself. How should that be understood?

Zhao Xianlie: General large models are affordable to everyone; the real barrier is vertical know-how: what kind of product, paired with what kind of video, finding what kind of influencer, can generate orders in the US region—this knowledge can only grow from full-link behavioral data, cannot be bought, cannot be copied quickly. We place more emphasis on deep cultivation of our own industry data: the full-link data of product selection, generation, publishing, and conversion within the platform is the core asset; cross-entity, cross-industry data collaboration is auxiliary. There are two core challenges: first, attribution—orders must be precisely attributed to videos, and the longer the chain, the more chaotic the metrics, so we built our own video-level attribution and metrics system; second, unification of metrics across multiple languages and markets—the behavioral patterns of English and Spanish content in the US region are different, and data cannot be lumped together.

Ebrun: You mentioned assigning a Personal Agent (LUI) to each influencer and also building your own video-level attribution system. What roles do these two things play in the 'AI content production capacity × alliance commission network'?

Zhao Xianlie: LUI productizes operational experience, giving each influencer an Agent that understands topic selection, scripts, and publishing, lowering the threshold for influencers to create content; the video-level attribution system makes it clear 'which video brought which order,' providing the basis for commission sharing and performance evaluation. The former determines the upper limit of production capacity, the latter determines whether the commission network can operate at scale. Only when the two mesh can the flywheel turn.

Ebrun: How do you view industrial agents rewriting B2B? Is there a long-term necessity for vertical large models? Where is the AI moat?

Zhao Xianlie: Industrial agents rewriting B2B is a high-probability event. The essence of B2B is trust plus efficiency; in the past, intermediaries earned from information asymmetry and execution gaps. Agents turn 'finding goods, comparing prices, following orders, reconciling accounts' into automated execution, links that survive on information asymmetry will be compressed, and value will settle in links that truly create trust—product selection capability, fulfillment capability, after-sales capability. Vertical large models are necessary long-term, but the form is not training a base model from scratch—that's a game for giants. The correct posture for verticals is: general model plus vertical SFT, plus vertical data flywheel, plus vertical toolchain. The model itself will quickly become commoditized; the moat is the product of 'vertical data × workflow × distribution network.'

Ebrun: Regarding overseas layout and digital-intelligence competitiveness, how does Moras do it?

Zhao Xianlie: Currently the core market in the first stage is the US, with business anchored to the US TikTok Shop region. Localization has been done at three levels: content localization—the behavioral patterns of English and Spanish content differ greatly, so AI generation and product selection run in parallel bilingually; people localization—cooperating with local influencer networks rather than transplanting domestic influencers; compliance localization—platform rules, content norms, and taxes all follow local requirements. Digital-intelligence competitiveness has three specific practices: multilingual AI generation, automation and cross-time-zone operations, and data-driven influencer operations. Essentially, it turns cross-border from 'labor-intensive' into 'computing-power-intensive,' and computing power can be scaled.

Ebrun: How do you judge the profit model and future competitive landscape?

Zhao Xianlie: Our profit model has been results-oriented from day one: the core is commission sharing based on actual sales transactions, plus value-added services around AI production capacity. We don't earn money from information asymmetry, we earn incremental money; if influencers don't generate orders, we don't earn either. My judgment on the evolution of industrial internet profit models is: from 'earning channel fees' to 'earning results fees,' then to 'earning AI production capacity fees.' Token payment becoming an industry standard depends not on technology but on pricing method: when Token costs can be precisely attributed to a single business output, customers can calculate ROI clearly, and it naturally changes from a 'cost center' to an 'accountable means of production.' In the next two years, content e-commerce will move from the 'influencer dividend' to the 'AI production capacity dividend,' and the key competitive variable will shift from 'who signed more influencers' to 'whose AI content production capacity costs less and converts better.'

As an annual bellwether in the TO B field, the '2026 8th Ebrun Industrial Internet Annual Conference and First Industrial Agent Conference' will be held in Beijing on November 20, 2026. The theme is set as 'The Paradigm Evolution Brought by AI and Going Global,' focusing on in-depth deliberation of frontier topics such as industrial agent implementation, AI supply chain value reconstruction, data elements and TOKEN economy, and new ecosystem of industrial going global. The conference will gather academic authorities, central state-owned enterprise benchmarks, industry leaders, investment institutions, and service ecosystem partners, pooling core industry forces, deconstructing industrial digital-intelligence transformation paths, and sharing frontier practical experience. In parallel, the annual industrial internet report and benchmark cases will be released, and decision-maker closed-door meetings and the 'Qianfeng Award' ceremony will be held.

This article was first published on the Ebrun official website.