Manchi | 晚点聊 LateTalk Summary

Episode 162 | 晚点聊 LateTalk | Guest: Charles Shen (沈晨岗), Founder & CEO of MeetSocial (飞书深诺)

1. Key Themes

China's Overseas Expansion Is Shifting from Volume-Based Selling to Brand Building

Charles describes a fundamental evolution in how Chinese companies approach overseas markets — from commodities to products to brands to full enterprise globalization. The first decade (2013–2023) was dominated by commodity selling driven by traffic arbitrage. Starting in 2025, the dominant mode is shifting toward product differentiation and brand building.

"If the first ten years were about commodity to product, now it's product to brand — this proportion will grow much larger. Before, people focused entirely on traffic: as long as I have traffic, I can sell goods and recoup my investment the same day or within a week." [00:07:15]

The shift is being forced by rising traffic costs, intensifying commoditization, and media platforms raising compliance/quality bars that block the "race to the bottom" strategies.

"Now media platforms are also closing off those paths more and more tightly — you can no longer compete by lowering standards further. Once those routes are blocked, many companies have no choice but to do the hard but right things." [00:12:07]

Digitalization Is the Multiplicative Force Behind All Other Competitive Advantages

Charles introduces the DMES framework (Digital Competence, Mental Dominance, Product Evolution, Sales Performance) where D is placed at the front because it is a multiplier across all other dimensions — not just an operational tool.

"D, we believe it is a core competitive force that runs through all previous capabilities. It needs to be digitalized in the sales process, in product R&D, in influencing user mindshare, and across enterprise operations. It is a multiplier — a coefficient that amplifies every dimension." [00:18:15]

L'Oréal is cited as a non-obvious example: well-known for its channels, products, and brand — but quietly investing massively in digital transformation behind the scenes.

"What people haven't seen is the enormous capital L'Oréal has invested in its digital transformation across its entire business process over the last three to five years — including all the digital capabilities of its online and offline stores." [00:18:43]

China Has a Structural Advantage Over the West in AI Adoption — Due to Lighter Historical Baggage

Charles makes a counterintuitive argument: Chinese companies, despite being "latecomers," are actually faster at AI adoption because they don't have entrenched legacy systems to dismantle. Western incumbents, including the world's largest ad holding groups, have data fragmented across subsidiaries that cannot talk to each other.

"For many established Western enterprises, they need to first shatter themselves and reorganize before they can use AI well. But for them, this is nearly an impossible task — that's my view." [00:40:52]

He uses Meituan's restaurant management system as a concrete illustration: when they tried to bring their digitalized restaurant OS to the US, they found American restaurants so deeply embedded in existing systems that switching costs were prohibitive even when the new system was demonstrably better.

"China, because of the massive pace of change and competition, has a much higher motivation to change than American companies of this type. So you'll find these application scenarios pioneered in China, and then potentially spreading globally." [00:37:56]

2. Contrarian Perspectives

The Real Scarcity in an AI World Is Not Creativity Volume — It's Creative Quality Frameworks That Humans Must Define

Most people assume AI will democratize creative production and commoditize creative services. Charles agrees with the democratization but argues the truly scarce resource is the knowledge framework — the structured understanding of why a creative works commercially — which AI cannot self-generate.

"If you don't have a method, logic, and process to describe how to make a good creative, AI itself cannot evolve or learn that capability on its own. It requires humans to first abstract the knowledge and framework, and only then can AI iterate within the framework you've given it." [00:51:02]

"AI can do a lot of computation, its processing power is enormous, but it cannot create a new algorithm on its own. An algorithm must always be abstracted and organized by humans first, then handed to AI... It cannot break through the theoretical framework of your cognition." [00:51:32]

AI Agents Are Largely Marketing Hype — Most Cannot Be Tied to Measurable Commercial Outcomes

While the market is flooded with "AI agent" products, Charles takes a skeptical stance: most agents currently cannot be linked to a well-defined, quantifiable business result, making them commercially meaningless as standalone offerings.

"Most of what's called 'agent' in the market today makes it very hard to map to a specific work objective and business outcome. Or they can only define a very vague, abstract result... The precision and quantification of agent output quality is far from sufficient." [00:43:17]

MeetSocial's response is to keep agents internal, paired with human experts, and only deliver the final validated output to clients.

"We have large numbers of these applications across various environments internally, but we cannot say these AI agents can be handed directly to clients to use and deliver value on their own. That would be irresponsible." [00:44:43]

High-Value Commercial Decisions Cannot Be Delegated to AI — The Accountability Gap Is Fatal

Charles draws a parallel to autonomous driving: even if AI makes fewer errors than humans, a single high-stakes mistake is catastrophic, and there is no accountability mechanism.

"Who bears responsibility for the commercial outcome? AI finds it very difficult to be accountable for high-value commercial decisions or high-value commercial products. If from our perspective the AI's decision error tolerance in financial/commercial decisions is small, you simply cannot deploy it there." [00:27:18]

Chinese Companies' Speed Advantage in AI Application Comes From Having More Consumer Scenarios, Not Better Technology

Most observers credit Silicon Valley with AI leadership. Charles inverts this by pointing out that application velocity — not model capability — is where China leads, driven by denser consumer and B2B scenarios that create faster feedback loops.

"In business application speed and iteration depth, I think China is actually faster. Because China has more application scenarios — more consumer scenarios, 2C consumption scenarios, or 2B scenarios — we can apply and iterate products faster." [00:35:01]

3. Companies Identified

Anker (安克) Consumer electronics brand (chargers, charging cables, audio, etc.) that expanded globally from Amazon. Cited as the model example of a company that consciously invested in product R&D and brand building while still in the commodity-selling phase — long-term thinking from day one.

"Anker invested more of its early revenues into R&D — how to make charging cables better, how to improve charging efficiency, even how to improve the craftsmanship of connector heads... At the same time, they built their brand so consumers would genuinely like them, remember them, and gradually love them." [00:13:07]

MeetSocial / 飞书深诺 (including MeetExperience / 蜜月科技) China's leading overseas marketing services group, founded 12 years ago. Serves 100,000+ clients including Anker, SHEIN, Temu, Alibaba, miHoYo, Lilith Games, ByteDance, Meitu. Highlighted for: (1) data infrastructure built since 2015 giving it a structural AI advantage; (2) processing $20M+ in ad spend daily through its systems; (3) building an agentic creative pipeline that delivered 6,000 video creatives in one week for a short-drama client, scaling that client from thousands to $100K+ daily ad spend.

"We have over $20 million in daily spend data accumulating and learning within our system. This is something no single enterprise can match — neither in learning speed nor learning quality. It makes no sense for them to compete with us on this." [00:58:13]

L'Oréal Global cosmetics giant. Cited as a non-obvious example of a company excelling across all four DMES dimensions, with particular emphasis on its largely invisible but massive investment in digital transformation over the past 3–5 years.

"What people haven't seen is the enormous capital L'Oréal has invested in its digital transformation across its entire business process." [00:18:43]

Meituan (美团) Chinese food delivery and local services platform. Cited as an example of advanced digital operations — real-time algorithmic dispatch, incentive design, and delivery routing — and also as a case study of how mature Chinese digital infrastructure struggles to penetrate the US market due to legacy system lock-in.

"Meituan wanted to apply their restaurant management capabilities to the US market, but found that American restaurant settlement and management systems were already deeply embedded, and changing them — even knowing the new system was better — carried enormous switching costs." [00:37:00]

Anthropic (Astropec / Anthropic) AI research company behind Claude and Claude Code. Singled out as producing the most revolutionary near-term AI capability shift — AI coding agents that can replace the output of 100 engineers.

"Anthropic's AI coding — Claude Code — should be a truly revolutionary change. It represents a leap in capability along the AI development trajectory. It can AI-ify the programming capabilities of our engineers... what used to require 100 engineers writing code, the AI coding agent can now accomplish to a very high level in a very short time." [00:29:44]

4. People Identified

Charles Shen / 沈晨岗 (CEO, MeetSocial / 飞书深诺) Founder and CEO of China's leading overseas marketing services company, 12 years operating, serving 100,000+ clients. Former early builder of data infrastructure for performance marketing at scale. Notable for: disciplined thinking about AI limitations, refusal to overpromise agent capabilities to clients, and having invested ~1 million RMB in a data warehouse as early as 2015 when it was considered "a huge amount."

"We started building our first-generation data warehouse in 2015, using HP and Tableau systems. At that point I invested about one million RMB — which was a huge sum for me at the time. From day one we were accumulating data digitally, with everything — all media, all creatives, all ad operations — sedimenting into one system." [00:39:23]

5. Operating Insights

Decompose AI Integration Into Nested Sub-Tasks With Defined Output Quality at Each Node

Rather than deploying a single AI agent to handle an entire workflow, Charles's operational approach is to decompose complex tasks into many small, specific nodes — each with its own agent — and keep humans at the orchestration layer. This creates a measurable, improvable system rather than a black box.

"We break [a creative task] into many very small task nodes and mini-agents. There's an agent for understanding client needs, an agent for creative strategy, an agent for image production, a different agent for video, another agent for quality checks. A product/project manager orchestrates these agents. The client sees only the delivered output." [00:46:09]

The key operating principle: every step is digitally logged so the system can learn which steps produced better outcomes, and humans can label good vs. bad outputs to train the agents further.

"The entire flow is an agentic flow. And because the whole process is digitally managed, you can learn — which steps could be done better. That data feeds back to our agents so they can continue learning." [00:47:38]

Fix the Client's Data Collection Infrastructure Before Optimizing Their Marketing Spend

MeetSocial discovered that many clients obsessing over marginal marketing ROI improvements had fundamental data collection errors on their own websites — meaning all downstream optimization was built on corrupted inputs. Fixing foundational digital infrastructure (website tracking, user data collection, CDP) delivered 10–20% performance gains vs. 1–2% from ad optimization.

"Many times clients ask why their ad performance is declining — and it turns out their website data collection is misconfigured. How can you run good ads on broken data? After we helped clients fix this, they might see a 10–20% improvement, whereas before we were working hard to squeeze out 1–2% gains from advertising." [01:00:39]

Prioritize Data Infrastructure From Day One — Retrofitting It Later Multiplies Costs

Directed specifically at founders beginning their overseas expansion journey: the cost of building data infrastructure mid-journey or late-stage is exponentially higher than doing it at inception.

"From the very beginning, you must give sufficient weight to data and technology. If you only discover this problem in the middle or late stages and try to catch up, your costs will be enormous." [01:09:28]

6. Overlooked Insights

The Compounding Flywheel: AI Adoption Creates an Accelerating Data Moat That Becomes Structurally Unassailable

This point was mentioned briefly but not dwelt upon. Charles describes a self-reinforcing loop where AI adoption → more digital data accumulation → faster iteration → lower costs → greater scale → more data. For a platform like MeetSocial processing $20M+ in daily ad spend across 100,000 clients, this flywheel compounds at a rate that no single advertiser — or a competitor starting from scratch — can match. The implication for investors is significant: the moat here is not the AI model itself (which is commoditizing), but the proprietary training corpus built on real commercial performance data at scale.

"The more AI-ified you become, the more digital sediment accumulates, the faster your iteration speed and efficiency, the lower your costs, the greater your scale capacity — this forms a positive feedback loop." [00:48:07]

"We have over $20 million in daily spend data accumulating and learning in our system. A single enterprise simply cannot match us on learning speed or learning quality." [00:58:13]

Second-Generation Founders Are a Quiet but Powerful Force Reshaping Chinese Brand Globalization

This was mentioned only in passing but is actually a significant structural trend. Legacy Chinese manufacturers (Generation 1) have strong production capabilities but commodity-era operational inertia. Their children (Generation 2), often educated abroad and market-native to Western consumer culture, are launching new overseas brands independently — bypassing the parent company's constraints entirely.

"The second generation can better understand what overseas consumers want... They may have received more market-oriented, more cutting-edge marketing education abroad. They'll reorganize products or brands and extend them toward higher dimensions. The founder's child at Metersbonwe is creating a new overseas brand — very differently from how the previous generation did it." [00:14:06]

This is a non-obvious investment signal: look not just at established Chinese exporters, but at the independent ventures being launched by their second-generation heirs who combine manufacturing supply chain access with Western brand-building instincts.