Unique Research · Panel · May 10, 2024
From Unique Capital’s 2024 AIGC Application Development Summit (Shanghai, Jan 24). Moderator: Wang Chaochao, Partner, Unique Capital.
Panel
- Diao Shengxin (Kevin) — Founder & CEO, Fenshi Yuedong (分时跃动)
- Zhang Tianhe — Founder, Shenpu Intelligent (深谱智能)
- Cao Hui — Partner, Huibo Tech (慧博科技)
- Qiu Huihui — Founder & CEO, Feidi Tech (飞笛科技)
Introductions
Diao: 9 years entrepreneurship (2nd company); Wall Street background. Since ~2000 focus: FI client acquisition & marketing. Post-pandemic banks need digital care/precision marketing — offerings evolved digital employees/content → AI interactive services.
Zhang: DB-layer tech combining LLMs + graph; “deep data era” beyond big data — symbolism × connectionism; high-performance graph for deep value mining.
Cao: GPT for intelligent investment research / product design. Path: research-report DB → 2018–19 smart research → GPT. Mutual funds/brokerages in pipeline; 2024 many institutions finally budget digitalization prioritizing GPT for unstructured info.
Qiu: Content digitization + AIGC for online wealth management / buy-side services; broker digitization; partnering Yingmi Fund on seller→buyer shift.
FI AI adoption
Diao: Pre-AI digitalization already hard (CCB work with BCG/McKinsey; WeCom rollout). AI now feels like tool upgrade — leaders already accept online engagement; cost/efficiency narrative is natural. Adoption order: joint-stock leaders (CMB, Ping An) → other joint-stocks → city banks → SOE banks last (watch then provincial pilots).
Zhang: Banks less profitable than assumed (annual reports); regulated license business. Mid/large banks moving on intelligent cost-out. Data limits → local deploy/train; graph + vector DBs + LLMs as full stack rebuild — AI ≠ LLM only. Focus: internal users (RMs, risk, ops) SaaS.
Cao: Library metaphor — from top-5 search hits to top-2,000 with cite/summarize/multi-agent reports. GPT-4 thought experiments strong; may erase junior analysts while amplifying seniors. Unstructured→structured (charts→Markdown) lifts usable research data from ~20–30% toward ~100% — high willingness to pay, few vendors.
Qiu: Joint-stock banks lead with private LLMs on internal KB; SOE banks / top brokers next. LLM often unlocks work they “should have done” — top-down after regulator push for domestic models. Tier-2: watch + small trials, compliance-first; “create conditions to pilot.” Tier-3: copy later.
Biz vs tech drive; product managers
Diao: ~7B bank “person-times” (5–6 cards/person) — tiny change triggers org resource fights. Bottleneck: business people can’t write tech-ready requirements after offline-only careers.
Zhang: Align demand → solution → then tech selection; keep all departments communicating.
Cao: Missing PMs more than models — GPT-native workflow thinkers, not 3-year traditional analyst tracks. Pessimistic that clients will self-discover in time for startup runway → bias to self-abstract business processes.
Qiu: Still mostly business-driven; 2024 shift to business+tech/mid-platform co-drive (who owns budget?). Leaders added PMs in data/IT and tech-side budgets — vendors must help abstract cross-dept needs. Stage 3 (true tech/AI drive) still ahead while clients sediment existing logic.
Core value / 2023–24 lines
Diao: Cheap always-on online engagement without spam; soft content marketing banks never needed when branches filled themselves — collect interaction data for segment models.
Zhang: Compute-layer infra (graph DB) under GPU shortage; explainable/auditable finance needs beyond pure LLM for risk; LLM stronger for interact/marketing/CS.
Cao: RAG citeable library vs forgetting 80% of 1,000 reports/year — mature for professional “big C / small B.”
Qiu: Clients prefer efficiency over pure cost-cut; long-tail companionship/education in volatile markets; light data-API deploy before heavy mid-platforms. 140M stock / 700M fund / 700M deposit accounts — AI+data required.
One-liners: Diao — “let the spark burn” in 2024; Zhang — big data→deep data, digitize→intelligentize, ecosystem with model/data/app builders; Cao — 2023 “hot” but no killer app yet; 2024 “calm mind,” cost/benefit sober; Qiu — rational optimism, user-centric certainty, embrace ecosystem vs solo.
Inventory type kept ranking despite panel form.
— END —