According to incomplete statistics from Zero One Think Tank, 23 financial AI agent tenders were publicly disclosed in October, a month-on-month decrease of more than one-third. There were 15 winning bid and transaction announcements, up by one from the previous month, of which 12 disclosed amounts, totaling 16.528 million yuan—a new low since the second half of the year.

A closer look at the procurement targets reveals that data, evaluation, platform subscriptions, and module additions—content that in the past was mostly bundled into a single platform project—each had its own independent number and price in October. SDIC Securities, for instance, purchased its client-facing "Master Stock Diagnosis" in four separate orders: the foundation, the knowledge base, the agent itself, and the data.

The winning bid list featured no perennial winners either. Among the 12 named winning bidders, after deduplication by project, only Today's Investment Data Technology and Global Data Solutions each won two contracts.

 01  Nine Types of Entities Placed Orders, Platform-Type Share Rose

From the perspective of tendering entities, there were nine types in October. Banks (including the rural credit system) accounted for 11, or 47.8%; securities firms, 4; insurers, 2; and the remaining 6 were distributed among guarantee companies, finance companies, funds, fintech subsidiaries, financial leasing companies, and city commercial bank cooperative alliances.

Compared with September, the number of categories did not change; what changed was the long-tail group—wealth management subsidiaries, trusts, AMCs, and investment companies did not place orders, while finance companies, guarantees, funds, and city commercial bank cooperative alliances were added.

Grouped by whether the procurement target is delivered as a platform, foundation, or development framework, there were 8 platform-type tenders, accounting for 34.8%; new platform-type projects still accounted for less than 20%.

Among them, Industrial Bank aims to build a financial AI agent platform with advanced cognition and operational processing capabilities, enabling low-cost development and management of agent applications, support for mainstream large models, flexible plugin integration and invocation, knowledge base construction, autonomous process planning, chain-of-thought and visual orchestration, and other functions, to accelerate the development and construction of the bank's internal agent applications.

Jiangsu Rural Commercial United Bank plans to build a mobile AI capability platform, including a multimodal interaction engine, an intent understanding middle platform, an AI streaming gateway, an agent scheduling hub, a large-model security fence, and an agent high-code platform; as well as a mobile banking AI assistant and a wealth AI assistant.

Shanghai Rural Commercial Bank issued a POC solicitation announcement for an agent development platform project, addressing enterprises' lack of capabilities in agent application building, evaluation, release, and permission management, and aiming to achieve full lifecycle management of AI agents and the accumulation of agent application assets.

Bank of Shanghai's intelligent foundation capability construction is divided into 4 lots, including procurement of agent implementation modules; Taiping FinTech subscribes to HiAgent platform software services annually for independent deployment by various professional companies within the group; Taiping Petrochemical Financial Leasing is upgrading and expanding its existing AI agent platform, adding four agents in the first phase.

Technical terms are also being phrased differently. Industrial Bank requires the platform to connect to the bank's business and data platforms, and Changshu Rural Commercial Bank requires desktop agents to connect to internal systems via A2A, MCP, CLI, and other methods. What clients want is not a platform that can demo, but one that can connect to existing systems and be managed.

 02  Data-Related Needs Listed Separately, Evaluations Have Mandatory Basis

Data used by agents was often written into the technical requirements of platform projects in the past. In October, 3 tenders directly targeted data or knowledge supply.

SDIC UBS Fund procured "agent external information data services" so that internal office agents can access market quotes, funds, corporate registration, and external research reports. The announcement stipulated no fewer than 80 users, an annual cost of no more than 100,000 yuan, and no more than three information service providers winning bids.

SDIC Securities' announcement was more direct: individual stock diagnosis is a core segmented scenario of AI client-facing intelligent Q&A, and to build the "Master Stock Diagnosis" agent, data services must be procured from third parties. Anhui Provincial Credit Financing Guarantee Group launched "key technology research on the construction of high-quality financing guarantee datasets for agent applications."

This is a typical "continuous data supply" approach: the ontology is developed once, and data is subscribed by service period and renewed annually. None of the three amounts is large, but this expense will follow the agent continuously.

In terms of agent security evaluation, this month saw an announcement that wrote evaluation basis into a mandatory list: Zheshang Securities engaged a third-party institution to conduct compliance and security evaluation of deployed agents, and the tender document attached a mandatory basis list.

This list includes GB/T 45654-2025 "Cybersecurity Technology Basic Security Requirements for Generative Artificial Intelligence Services," the "Artificial Intelligence Security Governance Framework 2.0," the "Cybersecurity Law," the "Data Security Law," the "Personal Information Protection Law," the "Interim Measures for the Management of Generative Artificial Intelligence Services," and the "Measures for the Management of Network and Information Security in the Securities and Futures Industry," among others.

In the same month, CITIC Securities subscribed to agent security detection tools, China Minsheng Bank procured AI email security agent software, and Bank of Shanghai separately listed a large-model security platform module in its intelligent foundation capability construction.

In new platform projects, similar clauses are written in more detail. Changshu Rural Commercial Bank requires permission management, operation traces, and audit traceability; Zhongyuan Bank requires controlled execution in a security sandbox; and Jiangsu Rural Commercial United Bank's construction content includes a large-model security fence and assistance in completing algorithm filing.

On one hand, third-party acceptance is being sought for agents already online; on the other hand, control clauses are being written for new platforms. How far procurement has progressed can be seen from the clauses.

 03  Winning Bids: Today's Investment and Global Data Solutions Each Win Two

The 15 winning bid announcements correspond to 12 named winning bidders. After deduplication by project, only Today's Investment Data Technology (two contracts from SDIC Securities) and Global Data Solutions (Huatai Securities, Hengfeng Bank) each won two, while the others won one project each.

Platform subscriptions, module additions, industry applications, data services, and security evaluations are inherently not something the same type of supplier can dominate across the board.

SDIC Securities completed bidding for four needs. A 1.65 million yuan desktop AI intelligent workbench license was awarded to Volcano Engine; a 2.05 million yuan agent platform upgrade and AI knowledge base went to Tencent Cloud; the 180,200 yuan "Master Stock Diagnosis" agent and the 201,400 yuan stock diagnosis data service both went to Today's Investment.

The foundation, knowledge base, client-facing application, and continuous data supply were split into four orders, each of which can be compared separately in price and can also have its supplier changed separately. For the buyer, this is easier to control than packaging a platform all at once, and also facilitates flexible arrangement.

Among the winning bid data, the largest single amount was not the agent itself, but the data foundation that the agent needs to connect to.

Zhejiang Rural Commercial United Bank's data analysis and modeling platform software permanent license was won by China Electronics Goldinfo for 3.5 million yuan, including domestic adaptation, dynamic desensitization, data operations management, and connection to the bank's agents.

Other relatively large contracts: Global Data Solutions won 2.735 million yuan to undertake Huatai Securities' institutional client data operations platform; Hangzhou Yuancheng won 2.39 million yuan to develop CITIC Financial Asset's non-performing asset agent; Tencent Cloud won 2.05 million yuan to undertake SDIC Securities' agent platform upgrade and AI knowledge base; Volcano Engine won 1.65 million yuan for its AI intelligent workbench license service; China Insurance Investment won 984,400 yuan for computing power leasing and agent development services; and Jiaokong Capital won 789,000 yuan for an investment decision model agent.

Data note: The monthly series uses a retrospectively revised statistical baseline; the September tender count is 36 (the September observation draft used a 33-item basis).

-End-