6000 Tests of Fund Companies: Very Few AI Customer Service Agents with High "Intelligence"

In the first month of 2026, China’s public mutual fund total assets surpassed 37.7 trillion yuan, marking a new milestone for the tenth consecutive month. Behind this growth, approximately 800 million retail investors entrust their hard-earned money to fund management companies.

However, most retail investors are not fully familiar with the complex trading rules of funds, and these rules often evolve or change suddenly. The recent extreme case of China Universal Ruilian Silver LOF valuation plummeting (see “Confused Investors: ‘Common Sense’ Meets Sudden Rule Changes”) exemplifies this. When markets are unpredictable and fund net values fluctuate wildly, who can provide them with fast, warm, professional support?

The first point of contact for ordinary investors is certainly not the busy fund managers, nor even human customer service reps, but AI customer service. A global survey released by SimCorp, a fintech company, on January 19, 2026, shows that 70% of buy-side institutions are already using AI in front-office operations. For domestic fund companies, this is not just technological iteration but a crucial path to solving the paradox of managing hundreds of trillions of yuan while providing personalized services.

Is AI customer service truly attentive, professional, and timely? When your fund drops in value, you want to ask what happened; when redemption funds are delayed, you want to check why; when you want to modify a regular investment plan but don’t know how—can the AI customer service that responds first really help?

To explore the real capabilities of fintech at the service frontline, the Southern Weekend New Financial Research Center evaluation team conducted a test from February 3 to February 9, 2026. Using ordinary investors as testers, they performed 5,985 interactions with AI customer service across 15 leading public fund companies during different times of day and non-working hours. The inquiries covered five high-frequency issues: fee queries, transaction operations, exception handling, complaint feedback, and investor education, totaling 19 questions.

The results were surprising: AI customer service capabilities did not grow proportionally with fund size. Even among top fund companies managing trillions, the ability of AI to answer independently varied nearly threefold. Some AI systems failed to recognize basic business logic; most could not clearly explain “why” questions or handle complaints effectively; many either gave vague responses, evaded questions, or refused to answer. During weekday trading hours (9:15 to 15:30), some AI systems kept customers waiting for 22 minutes without transferring to human support.

Four AI Customer Service “IQs” Under Scrutiny

The evaluation selected 15 top public fund companies based on their fund sizes at the end of January 2026 (see chart below).

To comprehensively assess AI responsiveness, testing spanned seven days from February 3 to February 9, covering trading days and non-trading days, morning, noon, and evening periods.

The team designed 19 questions across five scenarios—fee queries, transaction operations, exception handling, complaint feedback, and investor education—common issues faced by retail investors. Each question was tested three times per day (morning, noon, evening) over seven days, totaling 21 tests per institution and 5,985 interactions, with results verified for consistency.

The findings showed that the AI response capabilities among fund companies varied up to three times, with China Asset Management (China AMC) scoring only 31 points.

Only one institution performed excellently. Whether it was “how to modify a regular investment plan,” “whether FOFs (funds of funds) charge double management fees,” or “how to unfreeze shares,” their AI customer service provided precise guidance or broke down complex issues into simple steps for customers to choose and understand.

Nine other institutions achieved a 70%–90% independent answer rate. However, when questions involved detailed explanations or touched on compliance boundaries, the stability of their AI systems declined.

The biggest gaps appeared among the bottom five. For example, Huaxia Fund and Penghua Fund’s AI answer rates hovered around 50%, while Jiashili Fund, ICBC Credit Suisse, and China AMC ranged between 30% and 45%.

Take China AMC as an example: its AI customer service could only independently answer four questions—linking/changing bank cards, quick redemption limits, modifying regular investment plans, and changing dividend options. For more complex inquiries like “what to do if shares are frozen,” “how to transfer outside to inside,” or “how long does a QDII redemption take,” the AI would immediately transfer to human support. Interestingly, China AMC’s user experience for manual transfer was better—when AI couldn’t respond, the system automatically queued the user for human support without needing to click “Transfer to Human.”

Waiting 22 Minutes Without Resolution During Holidays

When AI cannot resolve an issue, transferring to human support is the natural solution. However, in this test, that route was not always smooth—even for institutions with high AI IQ.

In terms of access speed, only six institutions, including China Asset Management, achieved instant connection; E Fund was also quick, taking only 0.08 minutes.

But exceptions exist. For example, Guotai Fund’s AI answered 84.2% of questions well, which is good. Yet, during weekday trading hours, attempts to transfer to human support resulted in a “queue full” message. After waiting 22 minutes, the call was still not connected.

There is no direct correlation between AI answering ability and transfer efficiency. The results showed different combinations:

• Guotai Fund exemplifies “strong AI, slow human transfer”: 84% of questions were answered independently, so most investors didn’t need to transfer. But the process was not smooth—average wait time was 3.27 minutes, and the 22-minute wait indicates the transfer channel is not seamless.

• Jiashili Fund’s AI answered only 42% of questions, meaning over half required human help. However, their transfer to human support was almost instantaneous, and they maintained some manual support even on weekends.

What truly reassures investors is when AI is weak and no support is available on weekends. China AMC is a case in point: with only 31% answer rate, nearly 70% of questions require transfer. The average wait for transfer is 0.93 minutes, which is acceptable, but support is unavailable on Saturdays and Sundays. This poses a challenge for busy working investors.

In fact, weekend manual support is generally lacking across the industry. Of the 15 top fund companies, only three offer live human support on weekends; the remaining 12 rely solely on AI.

Global market linkages mean weekends often see major macro policy releases and overseas market volatility, as well as investor activities like adjusting portfolios and checking funds. For users encountering account issues or fund transfer problems during non-trading days, this results in up to 48 hours without human assistance. When AI faces complex questions, the common response is to transfer to human support, causing issues to pile up until the next trading day.

This “9-to-5” service model contrasts sharply with the 24/7 demand for financial management, leaving investors to cope with anxiety alone.

Responses That Miss the Point or Avoid the Issue

Overall, AI customer service performs relatively well on operational questions. All 15 institutions can answer questions like “how to modify a regular investment plan” or “fund conversion,” as these are rule-based and straightforward.

However, once questions shift from “how to do” to “why,” AI responses decline sharply.

For example, only one institution provided a clear, complete explanation for “Does FOF charge double management fees?” Ten others either answered vaguely or transferred to human support. Take GF Fund: their AI only explained ETF management fee rules, missing the core issue about FOF management fees.

Fund investors are often passive recipients of market fluctuations. They need not only “guidance” but also “explanation” and “companionship.”

This is where AI’s ability to explain becomes a true dividing line. It requires not just storing rules but understanding the underlying logic and translating it into plain language. Most institutions’ AI systems currently lack this capability.

Even more frustrating for investors are situations where complaints cannot be made.

For example, when asked “The fund’s net value has been declining for quarters, holdings unchanged, fund manager inactive—how to give feedback?” only E Fund and eight others provided complaint channels or quick “transfer to human” options; seven, including Tianhong Fund, failed to offer effective guidance. Interaction records show that most AI systems, when faced with words like “inaction” or “complaint,” respond with templated replies such as “please check the fund announcement” or “the fund manager will adjust holdings based on market conditions.” In terms of emotional support and companionship, AI still cannot replace human support effectively.

Poor responses to explanation and complaint questions point to a deeper issue: the lack of investor education (“投教”). The data shows that the average answer rate for investor education questions is only 50%, significantly lower than the 91.7% for account management questions. This means that when investors most want to understand, AI often cannot help. Every proactive question is an opportunity to clarify product rules and build correct understanding. But if the window is closed due to unhelpful responses, the opportunity is lost.

The cost of closing this window may be greater than expected.

Starting January 1, 2026, the “Regulations on the Management of Public Fund Sales Expenses” officially took effect; and from March 1, the “Guidelines for Benchmarking Public Fund Performance” were implemented (see “Are Investors Overcharged? The Illusion of Overly Lenient Benchmarks”). These new rules focus on fee transparency and benchmark constraints. When rules are updated and generate high-frequency investor questions, if AI customer service remains stuck on template responses like “please refer to the webpage,” it not only fails to educate but may deepen investor confusion and dissatisfaction. Moreover, if the knowledge base cannot be quickly updated, outdated information may be disseminated at critical moments, increasing information gaps and complaint risks.

Business Card: Dim or Bright?

The intelligence level of AI customer service does not fully reflect a fund company’s digital transformation. Dong Ximiao, Chief Economist at UnionPay and Deputy Director of Shanghai Financial and Development Laboratory, reminded the Southern Weekend New Financial Research Center that AI customer service is just one specific application of AI and cannot comprehensively represent a financial institution’s overall digital progress.

However, it is the most widely accessible link to retail investors. When major fund companies boast in their annual reports about “increasing technological investment,” “advancing digital transformation,” and “building AI platforms,” their intelligent customer service may still be unable to answer basic questions like “Where is my money?” This stark contrast was the starting point for this evaluation.

Based on this test, we developed a hexagonal capability framework: information disclosure and query services, complaint and issue feedback, investor education, handling of holdings and transaction anomalies, account management, and AI-to-human transfer efficiency. Under this framework, the capabilities of the 15 institutions’ AI customer service were dissected, revealing the root causes of gaps.

Only one fund company’s AI answered 97% of questions during trading hours, demonstrating near-perfect “knowledge coverage and process closure.” Another achieved an average transfer wait time close to zero, indicating that when AI cannot complete a task, human backup can respond swiftly.

But not all institutions have reached this level. For example, China Construction Bank Fund’s AI failed in multiple areas: unable to answer questions about purchase/redemption fees, net value and yield inquiries, off-to-on transfers, net value and holdings discrepancies, ETF vs. index fund fee differences, overseas market net updates, double management fees for FOFs, quick redemption limits, failed regular investment deductions, undistributed dividends, trading suspensions, unshown holdings after purchase, share freezes—all unanswerable.

What does this mean? When investors want to understand transaction costs, AI cannot provide answers; when holdings display anomalies, explanations are unavailable; when regular investment deductions fail, troubleshooting is impossible; when redemption funds are delayed, no help is offered. These are precisely the issues most concerning to investors and most likely to trigger complaints.

If multiple query rules require frequent manual transfer, the human support queue will swell, further degrading user experience and creating a vicious cycle. This points to a systemic gap: insufficient knowledge coverage, incomplete process guidance, and unstable human backup.

It’s worth noting that China Construction Bank Fund is not an isolated case. ICBC Credit Suisse Fund also performed poorly. Both are “big four” bank-affiliated fund companies. Notably, before the new CIO took office on January 16, 2026, the position had been vacant for half a year. How to quickly break down internal data silos and address the shortfall in core business knowledge is the most urgent digital challenge for the new CIO.

High-quality human advisors are always scarce, yet nearly 800 million retail investors should not remain in a “service vacuum” for long. Dong Ximiao pointed out that this is precisely where AI should play a “universal benefit” role—helping to “reduce operational costs, strengthen risk control, and improve user experience,” addressing the “first mile” and “last mile” of financial services.

When an investor checks holdings late at night, a novice is confused about trading rules, or a veteran faces an emergency—these interactions with customer service are the “business card” of a fund company. Whether this card is dull or bright depends on every investor’s experience stacking up over time.

Southern Weekend New Financial Research Center