Changxin Fund Deploys Tencent WorkBuddy: A Team of Under 300 Achieves 80% Monthly Active Usage, Building an Organization-Level AI Model for Public Funds

According to Analysys's report for the third quarter of 2026, the domestic desktop AI-native office agent market has exploded, and AI office work is formally moving from concept to large-scale industrial deployment. Beyond consumer enthusiasm, B-end deployment in financial institutions under strict compliance requirements offers greater industry reference value. Midsize mutual fund manager Changxin Fund deployed Tencent WorkBuddy in June 2026, achieving coverage across all business lines within a team of fewer than 300 people, with internal monthly active usage reaching 80%. It moved beyond the use of a simple Q&A tool and explored a mutual fund AI deployment model in which "30% depends on the tool and 70% on the organization." Security and compliance first: the underlying logic of financial institutions' AI selection. The financial industry has extremely high thresholds for data security, operational auditing, and risk control, and security and compliance are a veto item in AI tool selection. Changxin Fund General Manager Qin Bo said that institution-grade AI tools must core satisfy the hard standards of "data does not leave the domain, and behavior is traceable." This is also the main reason Changxin Fund chose WorkBuddy: the product supports VPC and private deployment, permissions can be finely controlled down to individuals, departments, connectors, and data sources, and audit logs are retained across the entire chain; unlike ordinary conversational bots, WorkBuddy's "experts + skills + connectors" system can connect to enterprise internal systems, solidify complex business processes into reusable and transferable workflows, and truly enable AI to execute tasks rather than merely answer questions; it also supports flexible scheduling of multiple models, allowing switching between cloud and local models according to task sensitivity; and the relatively low onboarding threshold also reduces internal promotion resistance. In actual deployment, Changxin Fund continued its original data classification and grading management system, mandatorily invoked local models in sensitive business scenarios, and conducted permission assessment and control on encapsulated internal system interfaces; data security was incorporated into departmental and individual assessments with a veto system, building a solid security foundation for AI applications from both institutional and technical levels. Changxin Fund Chief Information Officer Shen Wei proposed that organization-level AI deployment must cross four practical thresholds: first, refined management of data and permission boundaries; second, precipitating individual usage techniques into organizational assets reusable across the company; third, controlling the reliability of AI output and avoiding the risk of large-model hallucinations; fourth, driving a shift in work habits across all staff. "The AI tool itself accounts for only 30%, and the remaining 70% lies in data governance, institutional development, knowledge management, and organizational mobilization." Real efficiency gains on the business front line: from information handling to professional decision-making. After the security foundation was established, AI became deeply embedded in core business scenarios such as investment research, marketing, and middle- and back-office operations, and the working models of business personnel underwent tangible changes. Fan Xiaoyuan, Director of the Separately Managed Account Business Department, introduced that separately managed account operation reports involve extensive data retrieval, template organization, and formatting and proofreading, and in the past a report took five to six hours to complete manually. With WorkBuddy solidifying business workflows to complete data retrieval, cross-checking, and document drafting, the reporting cycle is now compressed to one to two hours, and business personnel are only responsible for logic review and final judgment, greatly easing the pressure of high-frequency repetitive document processing. In quantitative research scenarios, the efficiency improvement is even more pronounced. Quantitative researcher Du Tingting encapsulated daily work such as intensive research report reading, factor mining, and factor reproduction into standardized Skill workflows. In the past, reproducing a factor from a sell-side research report often took researchers several days to read the literature, retrieve data, and conduct backtesting and validation; now, by inputting the research report into the system, AI automatically completes the entire process of intensive literature reading, factor extraction, parallel coding, backtesting validation, and secondary verification, and for a round of testing dozens of factors, AI needs only a few hours. Du Tingting acknowledged that AI lowers the threshold for factor reproduction and will accelerate the decay in the effectiveness of public factors, but for teams with deep research accumulation, AI is an important opportunity to amplify research capabilities. Equity researcher Li Jing's focus has also shifted from "information porter" to "information judge." During earnings season, it previously took two to three days to read announcements and organize working papers; now, by uploading financial reports and announcements, AI completes structured decomposition of information, flags abnormal indicators, and outputs research working papers with source evidence in only one hour. The team precipitates individual practical experience into shared Skills, turning the pitfalls individuals have encountered into experience for the whole team, for example by adding a rule to the announcement parsing Skill to "verify the source and scope of information" to avoid the problem of AI citing unverified forecast data, thereby achieving the precipitation and iteration of team knowledge. At the same time, it can conveniently reconcile historical forecast data with actually disclosed values, improving tracking quality and response speed. Every day, the system also automatically parses market-wide research reports and periodic reports, extracts core viewpoints, generates an "AI morning briefing," and pushes it to fund managers, compressing what was originally half a day of information gathering into fragmented reading, so researchers have more time to devote to deep thinking and industry exchange. Top-down + bottom-up: building an institutional AI operating mechanism. AI deployment in financial institutions cannot rely on administrative coercion. Changxin Fund adopted a combined model of "top-down strategic push + bottom-up business innovation" and established a complete operating safeguard system. Senior management treated AI digital transformation as a top-priority project, frequently conducted tiered AI concept training, identified seed users in each business line, and had seed users drive colleagues around them, expanding application scope from point to area. The three major investment research lines set up business-technology co-creation groups, where business units proposed scenario requirements and the IT team completed the implementation and construction of systems, interfaces, and Skills, bridging the gap between business and technology. Supporting this, weekly, biweekly, and monthly supervision and notification mechanisms were established to track the progress of AI scenario deployment and regularly report Token consumption, system usage rates, and the number of precipitated Skills, avoiding AI projects that are "built but not used."At the same time, distinguish between "mandated actions" and "optional actions": the supervision list contains the core scenarios the company requires to be implemented; meanwhile, computing power, tools, and data resources are opened up to encourage frontline employees to spontaneously explore and innovate. Many practical Skills come from employees' original creations—for example, an equity researcher spontaneously built a Skill for monitoring listed company announcements, which was subsequently promoted for use across the entire department. To adapt to the AI era, Changxin Fund specifically established an AI Application Research Department: on one hand, it optimizes AI usage methods, identifies inefficient Token consumption, and guides business personnel in using tools well; on the other hand, it explores the implementation of AI in frontier scenarios such as pre-event risk control and drawdown management. The IT department is also undergoing capability reshaping, with an internal proposal that "IT personnel are divided into two categories: those who can use AI and those who cannot," forcing the technical team to iterate its capabilities and adapt to the working model of the AI era. Human-machine collaboration is the core at the current stage and may reshape the competitive logic of public funds. For the public fund industry, what AI brings is not simple tool replacement but a reconstruction of the underlying competitive logic. Shen Wei judges that future competition in public funds will shift from comparing the scale of investment research personnel to comparing organizational-level AI accumulation capability; the competitive dimension will also upgrade from single-point efficiency to the response speed of the entire business chain. Capabilities brought by public data and general tools will rapidly converge, and the true moat comes from institutions' private high-quality data and accumulated business experience. "Dirty data in, dirty data out"; private-domain data assets determine the quality of AI output. At the same time, AI will both amplify the advantages of leading institutions and bring opportunities for small and medium-sized institutions to overtake on the curve. But no matter how great the efficiency improvement, the principle of human-machine collaboration must never be shaken: AI undertakes first-draft output, clue mining, and process execution, while all key conclusions and external outputs are still judged by humans. The original compliance approval process is not exempted because of AI intervention, and AI is strictly prohibited from directly replacing decision-making. When discussing experience sharing with peer institutions, Shen Wei summarized four key points: safety and compliance are always the bottom line; start from small business pain points and use small-scale practical results to drive promotion; 70% organization and 30% tools—focus not only on purchasing tools but also more on knowledge accumulation and organizational mobilization; and preserve innovation space for frontline employees to unleash creativity on the business side. Qin Bo believes: first, the company should set direction, goals, and boundaries at the strategic level; second, the company and employees should meet each other halfway—the company builds the platform and frontline employees actively participate, with both top-down guidance and attention to cultivating and amplifying bottom-up seed users; third, continue learning, work things out together, build trust, and uphold the bottom line of safety and compliance. At present, the industry has not yet formed a mature quantitative evaluation system for AI input and output. Changxin Fund at this stage encourages exploratory use, continuously observes costs and business output in the background, and does not blindly restrict usage frequency. Changxin Fund's practice shows that true AI productivity lies not in flashy conversational effects, but in AI deeply connecting with enterprises' private data, business processes, and expert experience, accumulating into capability assets that the organization can reuse, letting machines execute and letting people do higher-level thinking and decision-making.