Unique Research · Insight · August 1, 2023 · Interviews: Moxiang, Youlianyun, Feidi, EntropyJane, DeepQing — characteristics, scene roles, challenges/outlook
Global AI fever (Inflection $1B+; MosaicML $1B+ Databricks buy; Moody’s×Microsoft/OpenAI risk assistant; Mizuho first Japan bank on Azure OpenAI) contrasts China AIGC’s unclear models + compute limits (Lightcone). Unique Research’s ongoing founder/investor conversations map finance-specific thresholds.
Characteristics
PBOC FinTech plan (2019–21): deep IT/AI/big-data/blockchain fusion to optimize markets, raise service, widen coverage, prevent risk—potential to disrupt models/processes/products. Prior AI/ML mainly credit/fraud scoring; LLMs unlock unstructured data. Finance holds multi-industry historical data for domain models—but caution on security/compliance. The Business: global FinTech AI market $9.15B (2022) → $11.59B (2023, CAGR 26.8%) → $31.71B (2027, CAGR 28.6%).
Thresholds — talent
Need compound talent + business know-how. Feidi Qiu Huihui: finance’s value = capital to real economy + asset appreciation for capital holders; densest information industry on professionalism/scene frequency. EntropyJane Fei Binjie: core challenge is talent—research + data science depth plus Spark/Flink/model ops; scarce; quality > headcount (OpenAI-style small super-compound teams). Youlianyun Zhang Yan: architects who know finance + tech; product landing needs scene understanding; role design unlocks compound talent.
Thresholds — data, tech, security & compliance
Accuracy stakes extreme; privacy critical; lawful acquisition/use. Domain private knowledge needed to train vertical LLMs (EntropyJane). Tech barriers + regulatory scrutiny shape deploy (private/offline often required).
Scene roles & outlook
AI across analytics, CS, supply chain, trade advice, content (Yunchuang/Yunlian/Yunshu-class), decision platforms, virtual advisors. Challenges: talent scarcity, data quality/compliance, model explainability, ROI proof under no-trial-error culture. Outlook: mature lanes (risk, marketing, fraud, content) land first; disruptive leaps slower; user-centric digital transformation over pure tech novelty.
6 CN images; attach on ingest.
— END —