AI Research Analysts Go Live at Chinese Mutual Funds: From Morning Meeting Questions to 40% of Desk Work

Forum topic · QianXun · 2026-10-25

Summary

A survey by China Fund News (October 25) reports that Chinese public fund managers are moving AI from proof-of-concept to scaled deployment across investment research. At Tianhong Fund, an internal tool called AI Challenge now interrupts morning meetings to flag when a researcher's earnings forecast diverges from consensus. China AMC (Huitianfu) has built an agent cluster spanning information gathering, report processing, portfolio construction, trading execution, and risk monitoring; its 'AI researcher' agent reportedly handles about 40% of analysts' basic desk work—though this figure is company-stated and unverified. The industry forms three tiers: E Fund, China AMC, Southern, and Harvest lead with enterprise AI platforms; ICBC Credit Suisse, China Merchants, Guotai, and First Frontier apply AI deeply in pensions, fixed income, and Q&A. Firms repeatedly stress boundaries: AI assists efficiency but does not replace human judgment or investment decisions, with hallucination detection and independent thinking flagged as key challenges. A parallel front is consumer-facing: E Fund and Southern Fund have joined Tencent WorkBuddy, while GF Fund launched skills on ByteDance's Doubao Workspace. Industry observers call 2026 a potential 'investment research agent year,' arguing AI will reshape production methods but not replace humans as the decision-making subject.

Key points

At Tianhong Fund's investment research morning meeting, an internal AI system nicknamed the "AI researcher" (AI Challenge, embedded in the firm's self-developed TIRD platform) now interrupts analysts to challenge forecasts that diverge from sell-side consensus.
A China Fund News survey (Oct 25) indicates leading Chinese public fund firms have crossed from proof-of-concept into scaled, platform-based AI deployment; industry insiders call 2026 a potential "year of investment research agents."

Deployment tiers and platforms

Tier 1: E Fund, China AMC (Huitianfu), Southern, and Harvest — enterprise-level AI platforms with AI penetration across the full research workflow.
Tier 2: ICBC Credit Suisse, China Merchants, Guotai, First Frontier — deep application in pension operations, fixed-income trading, and intelligent Q&A.
Platforms: China Merchants has moved to platform-based supply (smart meeting minutes, per-stock AI insights, report translation/Q&A, AI credit reports, earnings reviews, index constituent forecasts); GF Fund's core platform "Archimedes" serves fixed income, active equity, passive index, and cross-border lines.

The 40% figure

China AMC's task-driven "AI researcher" agent reportedly takes over about 40% of analysts' basic desk work — reading the internal knowledge base, learning buy-side research frameworks from past reports, and writing research reports with stock ratings. This figure is company-stated, without third-party verification.

Four live application scenarios (per fund manager Xia Zheng'an, China AMC)

| Scenario | Method | Stated status | |---|---|---| | Market narrative / mainline detection | Batch-parses research corpus, measures crowding | Daily operation | | Expectation-gap mining in low-coverage stocks | Scans under-covered, low-valuation names | Continuous push to research | | Quant alpha factor iteration | Autonomous exploration of factor combos and weighting | Connected to live trading | | Portfolio risk monitoring | Multi-dimensional profiling and attribution | Daily automatic |

*Except "connected to live trading," no performance numbers were given; all claims are company self-reported.*

Boundaries, repeatedly stressed

E Fund (via its EWork platform) explicitly states AI "assists researchers' efficiency but does not replace human judgment or investment decisions."
Some firms deliberately avoid anthropomorphizing AI, keeping it to information preprocessing with humans as the sole output authority for conclusions.
Cited challenges: verifying/falsifying AI output, detecting hallucinations and false signals; non-standardized soft information (industry understanding, management judgment); and misjudgment risk in extreme markets.
Industry consensus: AI will reshape asset management's production methods but not change the production subject — alpha shifts from "who gets information fastest" to "who reads deeper logic from the same information."

Consumer-facing front

E Fund is the first fund company on Tencent WorkBuddy (three modes, nine service scenarios); Southern Fund launched five expert modules there; GF Fund launched three skills on ByteDance's "Doubao Workspace."
E Fund explicitly distinguishes conversational AI investor education (knowledge and risk awareness) from licensed investment advisory, which provides individualized advice.

Context and caveats

This connects to earlier zhichai.net coverage: Zhang Xiaoquan's "10,000 overnight backtests" point (picking the best of 10,000 pure-noise strategies can yield a 10-year backtest Sharpe of 1.2 purely by luck; separating live results from noise at a 1:1 signal-to-noise ratio takes about four years), plus prior pieces on backtest validation failure and strategy attrition once costs enter training objectives. AI adoption speed and validation discipline are two sides of the same coin.

Sources: China Fund News (Oct 25), "From analysis to questioning: fund companies' AI researchers take the post"; republished by 10jqka (Oct 25).

Limitations: The 40% figure and all platform performance descriptions are company self-reports without third-party verification; the "2026 agent year" is an industry prediction, not fact; this site has a commercial relationship with the WorkBuddy platform.