Unique Research · Interview / Roundtable · November 22, 2023 · Guests: Zhu Tiebin (朱铁斌), Founder, Deepthink Circle (深思圈); Tao Feng / Nick (陶峰), Co-founder & President, Gonex; Yan Qidong (严琦东), Co-founder, QingFlow (轻流); Lang Yu (朗豫), Founder, Maze.Guru · From AWS × Unique Capital “AI Globalization Accelerator” Shanghai offline
Q: How should teams of different backgrounds judge fitness for go-global products?
Zhu (Deepthink): With or without overseas backgrounds, go-global is often unavoidable for SaaS/AI companies today. Zhu helps domestic AI products and indie hackers go abroad. The core: ability to talk and iterate directly with overseas users. Successful overseas AI apps start from user needs, not tech. Even overseas-engineer-background teams often under-communicate with users — fatal for a real go-global product. Without a user-feedback loop you get Twitter/LinkedIn posts with tiny reach, a Product Hunt bump, then “what next?” The fix is talking to users about the real problem and best solution.
Lang (Maze.Guru): Hangzhou hides many apps earning hundreds of millions RMB annually that people don’t realize are Chinese-built. Two checks: (1) Brand vs. product — vertical product for market push vs. brand that deepens experience. Overseas apps stay vertical (one simple service, long cultivation); domestic teams like all-in-one, which is harder abroad. (2) Pick a concrete region/market — not “China vs. overseas.” Japan/Korea, North America+Brazil, Middle East, SEA each need different ops; spray-and-pray sinks budget.
Q: How do domestic vs. overseas promotion differ for the same product?
Yan (QingFlow): 2016 optimism on China SaaS faded; many practitioners are more bearish after this year — go-global to a larger market is a real path. Differences: (1) Compliance/privacy gaps (esp. Europe/US) — domestic buyers often reluctant to pay and have lower expectations on efficiency, service, and data security, weaker cloud appetite; mature North American SaaS markets are strict. (2) Every region needs separate attention — SEA ≠ West; chase higher purchasing-power markets. (3) Localized service/support — critical for top accounts; QingFlow finds regional partners to carry the product.
Q: Gonex insights on go-global trends?
Nick (Gonex): (1) Data security — especially complex in HR (personal files, SMS/phone data); cross-border makes it hotter for Chinese and international clients alike. (2) Get close to users — maturity differs by market; most Chinese firms are early on overseas HR literacy. Gonex shipped an overseas HR-policy “100,000 Whys” AI vertical as a high-usage acquisition product, plus products for cross-border candidate evaluation pain. (3) Stickiness — Chinese clients early in cross-border management lean on vendors; as they adapt to international rules and Western playbooks, domestic product R&D efficiency and fit keep rising.
Q: Can prior go-global playbooks copy onto AI go-global?
Zhu: Promising lanes for Chinese teams: productivity tools / SaaS with early PLG — social, content, SEO without being physically abroad. Twitter is a hot AI launch channel; Product Hunt’s lasting value is TOP5/TOP10 → newsletter and KOL pickup (newsletters ≈ overseas WeChat Official Accounts). AI directories: free listing often follows a strong PH rank (some crawl PH Top10 daily). Pricing shift vs. classic Freemium: many AI products require card-on-file with refund windows — weak moats invite copycats, and every token costs money. Zhu’s take after a year of AI apps/middleware: foundation models have not hit the paradigm shift for true AI-native products — only SaaS+AI — so startups face Big SaaS embedding AI (Notion/Craft writing) and Agent platforms risk OpenAI encroachment when model-vendor boundaries blur. Deterministic opportunity: nail one real user need with sane expectations — indie products at $5K–$20K MRR are already solid (Chatbase, PDF.ai, SiteGPT). Wait for the right opening rather than force perfect AI-native.
Nick: Play information asymmetry. Overseas has strong software firms, but even top Chinese HR orgs may only know their own stack — that gap is opportunity. U.S. leaders still hit frictions in China/SEA/Europe that local teams can solve.
Lang: PLG and SLG are stages, not opposites — PLG when capital is tight. AI for deep B-side is immature; Twitter/YouTube still skew entertainment/curiosity — good for PLG cold start with KOLs and PH. Post-launch: when ChatGPT news spikes attention, spend is necessary. YouTube users convert better; Twitter tries more than pays — put ops budget into YouTube KOLs. Overseas traffic is fragmented and expensive vs. China’s concentrated Xiaohongshu/Weibo/Douyin — timing and capital readiness matter. Localization is required to scale; Maze chose Web3 partly for fit and for reaching same-tribe / large B clients that pure AI without Web3 background struggles to penetrate country-by-country.
Q: How is QingFlow combining AI in product?
Yan: Target: rapid application development. IT is costly yet business satisfaction low — classic software efficiency from mode to collaboration is too low. Driving business users to build themselves is most efficient. Low/no-code at QingFlow cut IT cost to roughly 1%–2% of before with higher satisfaction. Endgame assumption: AI will gradually replace software-development labor. Experiment: ~50-character NL CRM requirement → GPT → structured data → QingFlow APIs → basic CRM shell in <30 seconds. Announced for July release, August beta — many signed up; a month later almost no one asked again. Why: AI cut the first hour of scaffolding to 30 seconds, but production-ready CRM still needs 20+ hours — cool demo, limited scale value. Still iterating.
Longer thesis: AI will structurally change enterprise management logic. Since the 1960s enterprise software’s interaction/function form barely changed (DB-upper apps). Future automation moves from execution to decision and sensing — not only “send this email” but “decide when to send and send it.”
Q: Will Agents shock no/low-code vendors like QingFlow?
Yan: Yes eventually — Agents will replace low-value repetitive production work inside enterprises. As infra matures, software engineering becomes more commodity labor; programmers writing thin glue over middleware are replaceable. Two jobs in software: understand my problem (high value, implicit context, high ToB bar — theoretically AI-solvable but not soon mid-term) vs. solve it (many tools). First Agent cut-in: programmers; PMs and business people still needed mid-term. Near-term “assets” are customer-need understanding and industry-case commercial strategy — not tech moats.
Q: Is the AI wave also an opportunity for no/low-code?
Yan: Highly related — lower thresholds speed diffusion. QingFlow’s vision: everyone-usable no-code for business users, not only China’s ~7M IT workers — empower ~200M business people to develop. Second challenge: demand–supply connection; as a startup they plan to partner DingTalk/WeChat/Feishu for reach while they catch demand and drive adoption.
Q: AI × Web3?
Lang: Data decentralization — e.g. Stable Diffusion open so the public trains personal styles. Paid models needed for value circulation vs. hobby-only open communities; Web3 on-chain rights (LLM-as-NFT transfer/usage/ownership) can close the loop with transparent incentives — turning amateur painting/model markets into more professional ones with clearer data rights (usage only for those who use it on-chain).
— END —