Tianji Zhichong Platform Powers New AI Agent Collaboration for Brokerage Services

AI technologies, especially large‑model based solutions, are reshaping decision‑making in the securities industry, and a "human + AI" collaboration model has become the norm. Guangfa Securities identified three key bottlenecks in traditional sales: scattered information, slow response, and uneven professional judgment, prompting the need for an end‑to‑end intelligent decision‑support system.

Unified Intelligent Service Foundation and the "Two Accuracies" Goal

Guangfa Securities introduced Volcano Engine’s Intelligent Marketing Agent not as a separate bot but as a unified service layer that empowers both C‑end customers and B‑end employees through a "central control brain". The project’s core objectives are "routing accuracy" (correctly directing a user query to the appropriate agent or workflow) and "answer accuracy" (providing responses based on reliable knowledge and data).

For routing, the system builds an intent‑recognition and routing engine that quickly classifies user questions—often expressed in colloquial or ambiguous language—into specific financial intents such as product inquiry, rule clarification, or investment analysis. Simple intents are dispatched instantly, complex ones trigger deep‑reasoning, and unclear queries invoke clarification mechanisms. The engine can be continuously updated with new intents and distribution strategies.

For answer accuracy, the Agent leverages Volcano Engine’s accumulated terminology library that maps financial colloquialisms to standard expressions, and a Q&A knowledge base that stores standard answers for high‑frequency concepts and processes. This foundation supports both generation and retrieval of reliable responses.

Dual‑Scenario Deployment: Customers and Employees

In the customer‑facing scenario, the platform aims to provide more than simple Q&A it delivers professional services covering product selection, risk‑return comparison, and thematic opportunity assessment. By using the MCP protocol to connect internal financial data services, the system bridges data, middleware, agent, and presentation layers. Natural‑language inputs are mapped to concrete products, conditions are automatically parsed, data APIs are invoked, and structured analysis results with visual reports are generated, dramatically lowering the barrier for ordinary investors to obtain expert analysis.

In the employee‑facing scenario, the same capabilities support daily tasks of relationship managers and business staff. Queries about market quotes, account opening, margin financing, product recommendations, etc., enter a unified entry point and are automatically routed to the appropriate business capability, eliminating the need for staff to switch between multiple systems. The platform introduces a "1+N" intelligent‑agent collaboration matrix: a central control engine oversees nine major business scenarios and coordinates more than 40 vertical agents, enabling quick access to fund data, fees, ratings, as well as complex operations such as portfolio analysis, client‑account diagnostics, and compliance checks. Evaluations show intent‑recognition accuracy exceeding 90% and routing response times measured in seconds.

Strategic Insight and Future Direction

The practice demonstrates a broader principle: when an enterprise already possesses multiple vertical AI capabilities, the next step is not to add more entry points but to organize those capabilities into a business‑oriented collaborative system through unified intent planning, knowledge support, and data‑connection mechanisms.

Guangfa Securities and Volcano Engine plan to deepen cooperation in intelligent marketing and related scenarios, advancing financial services from a "human finds tool" model to a "service understands and proactively responds to the human" paradigm.