2026/11/1912
The application of large models in finance is evolving from basic general knowledge Q&A and copyediting toward increasingly complex and specialized core business scenarios.
Building on SenseTime's SenseNova large model ecosystem and its experience in financial applications, SenseTime has launched Raccoon Work. SenseTime's Raccoon Work is an AI work assistant developed by SenseTime, providing individuals, teams and enterprises with intelligent workplace solutions that combine analytical and creative capabilities, supported across mobile and desktop platforms as well as private deployment. Developed as Raccoon Work's industry-specific edition for finance, it leverages SenseTime's leading capabilities in multimodal document analysis, complex knowledge reasoning, multi-agent collaboration and rigorous traceability to fully connect credit risk control, insurance claims, compliance management, business analysis and investment research insights five core scenarios, helping financial institutions turn large-model capabilities into robust and efficient productivity.
In finance, SenseTime continues to focus on credit risk control, insurance claims, compliance management, business analysis and investment research insights to deepen AI applications across five core scenarios, building industry expertise and solution capabilities while extending large models from general workplace assistance into more deeply embedded financial business processes. Drawing on its technological expertise and industry experience in financial technology, SenseTime was recently named to the 2026 IDC China Financial Technology 50 list, further demonstrating its sustained investment in and service capabilities for the financial sector. SenseTime Named to the 2026 IDC China Financial Technology 50 List, further demonstrating its sustained investment in and service capabilities for the financial sector.
Scenario 1 | Credit Risk Control
Full-Process Credit Risk Assessment Driven by Multisource Data
Credit operations span pre-loan, during-loan and post-loan stages. Every day, credit review and risk control personnel must handle vast amounts of corporate information, including financial statements that can run to hundreds of pages, bank transaction records containing tens of thousands of lines, complex collateral and pledge agreements, and real-time public opinion and corporate registration information. Relying on manual verification is not only time-consuming and labor-intensive; cross-document validation is also highly challenging, making it easy to overlook potential risks such as related-party transactions, hidden liabilities and embezzlement of funds.
Raccoon Work workflow:
Multisource data aggregation → Cross-validation and risk reasoning → Risk profiles and review recommendations
- Comprehensive Collection and In-Depth Analysis of Corporate Information: Rapidly analyze scanned financial statements, audit reports, transaction details, corporate registration information and litigation-related public opinion, with high-fidelity extraction from complex cross-page tables and notes.
Comprehensive Collection and In-Depth Analysis of Corporate Information: Rapidly analyze scanned financial statements, audit reports, transaction details, corporate registration information and litigation-related public opinion, with high-fidelity extraction from complex cross-page tables and notes.
- Multidimensional Financial Diagnostics and In-Depth Analysis of Fund Flows: Automatically calculate key financial indicators, including profitability, solvency and operating efficiency, while analyzing the timing of fund inflows and outflows, counterparty concentration and abnormal transfers of large amounts to verify the consistency between transaction records and reported revenue.
Multidimensional Financial Diagnostics and In-Depth Analysis of Fund Flows: Automatically calculate key financial indicators, including profitability, solvency and operating efficiency, while analyzing the timing of fund inflows and outflows, counterparty concentration and abnormal transfers of large amounts to verify the consistency between transaction records and reported revenue.
- Dynamic Post-Lending Alerts: Continuously monitors corporate litigation, changes in collateral, negative public sentiment and unusual account fluctuations, automatically generating post-lending monitoring leads and risk alerts.
Dynamic Post-Lending Alerts: Continuously monitors corporate litigation, changes in collateral, negative public sentiment and unusual account fluctuations, automatically generating post-lending monitoring leads and risk alerts.
Business Value:
This helps financial institutions upgrade from reviewing fragmented materials to building comprehensive risk profiles. It not only significantly shortens the due diligence and review cycle for individual companies, but also generates review recommendations that include source documents, calculation processes and investigation logic, ensuring that every risk-control conclusion is supported by verifiable evidence.
Customer Practice | SenseTime and Ping An Bank Explore the Use of Large Models to Enhance Corporate Credit Risk Management
In the area of corporate credit risk management, SenseTime and Ping An Bank have collaborated on the development of large-model applications, focusing on the labor-intensive post-lending management process and exploring an intelligent risk-management model that combines multimodal agents with specialized financial knowledge systems.
The project uses multimodal agents to analyze various types of business materials, including post-lending reports, financial tracking records, on-site visit reports, collateral monitoring records and the stated use of credit. It converts unstructured information into analyzable risk signals and draws on a financial knowledge foundation to integrate policies, regulations, contracts and historical cases, improving the professionalism and traceability of risk assessment. At the same time, three types of agents—for multimodal analysis, scenario-based checks and risk reporting—work together to enable intelligent processing from information analysis and risk identification to report generation.
In practical applications, these capabilities are gradually expanding to other credit processes, including pre-lending due diligence, financial statement analysis and account transaction-flow analysis. Due diligence reports can be generated in as little as 30 minutes, monitoring coverage exceeds 99%, and the time required for account transaction-flow analysis has been reduced from four hours to 15 minutes.
In July 2026, Ping An Bank's corporate post-lending AI application project received honors including The Asian Banker's China Best Corporate Risk Technology Project Award and China Human-Machine Collaboration Achievement Award, demonstrating the value of AI-human collaboration in financial risk management.
Scenario 2 | Insurance Claims
Intelligent Review from Claims-Material Analysis to Liability Determination and Benefit Calculation
In health and life insurance claims, reviewers handle hospital outpatient records in a wide range of formats, handwritten prescriptions, machine-printed inpatient invoices, detailed expense lists and complex policy liability clauses. Traditional OCR has limited accuracy when processing creases, obscured seals and blurred handwriting. In addition, claims rules cover the medical insurance drug catalogue, deductibles, benefit payment ratios and pre-existing-condition determinations, placing a heavy burden on manual review.
Raccoon Work workflow:
Claims Material Collection → Intelligent Analysis and Rule-Based Assessment → Liability Determination, Benefit Calculation and Risk Identification
- Multimodal Recognition of Complex Medical Materials: Classifies medical documents such as medical records, invoices, expense lists and discharge summaries, and performs high-precision structured extraction, with support for recognizing some handwritten content.
Multimodal Recognition of Complex Medical Materials: Classifies medical documents such as medical records, invoices, expense lists and discharge summaries, and performs high-precision structured extraction, with support for recognizing some handwritten content.
- Medical Knowledge Reasoning and Intelligent Policy-Clause Matching: Reasons over the chief complaint, diagnosis, expense items, treatment process, medical history and the scope of policy coverage to automatically determine insurance liability and identify claims risks.
Medical Knowledge Reasoning and Intelligent Policy-Clause Matching: Reasons over the chief complaint, diagnosis, expense items, treatment process, medical history and the scope of policy coverage to automatically determine insurance liability and identify claims risks.
- Itemized Benefit Calculation and Fully Automated Claims Assessment: Based on policy formulas, it automatically excludes self-pay medications, deducts the excess, applies benefit payment ratios and enforces coverage limits, generating a clearly itemized calculation statement.
Itemized Benefit Calculation and Fully Automated Claims Assessment: Based on policy formulas, it automatically excludes self-pay medications, deducts the excess, applies benefit payment ratios and enforces coverage limits, generating a clearly itemized calculation statement.
- Abnormal-Claim Investigation: Automatically identifies risk indicators such as excessive medical treatment, split hospitalizations, inverted expense patterns and logical anomalies in treatment dates, producing a list of abnormal risks for focused review by claims experts.
Abnormal-Claim Investigation: Automatically identifies risk indicators such as excessive medical treatment, split hospitalizations, inverted expense patterns and logical anomalies in treatment dates, producing a list of abnormal risks for focused review by claims experts.
Business Value:
Significantly reduces the burdensome data-entry and calculation workload of frontline claims reviewers, while improving the consistency and accuracy of decisions on complex claims and helping institutions uphold their risk-control standards for payouts.
Customer Case Study | SenseTime and China Pacific Insurance Harness Multimodal Large Models to Enable Intelligent Claims Processing
In the insurance claims sector, SenseTime and China Pacific Insurance have partnered to explore the application of multimodal large models in group insurance claims review. To address challenges in traditional claims processes, including the wide variety of supporting documents, time-consuming information extraction, and heavy manual review workloads, the two parties are using multimodal large models to intelligently classify claims materials and extract key information. Combined with insurance policy terms, the solution analyzes and reasons over information from multiple sources to help determine whether claims meet payment requirements, generate explainable claims assessment recommendations, and provide a basis for manual review.
The solution has now completed live testing in group insurance projects. While maintaining review quality, it effectively reduces the processing time for individual cases, identifies the risk of unreasonable payouts, and improves claims review efficiency and risk management capabilities, providing intelligent support for the scaled operation of group insurance business.
Scenario 3 | Business Operations Analysis
Natural Language-Powered Financial Operations Analysis
Financial institutions have numerous branches and complex product lines, while business operations analysis involves multidimensional indicators such as deposit and loan scale, net interest margin, asset quality, and customer profiles. When business managers and executives want to understand operational changes, they often need to submit requests to technology teams to run queries and write SQL, with turnaround times measured in days or even weeks. Even after receiving the reports, analyzing the causes of anomalous indicators still requires repeated manual slicing and dicing.
Raccoon Work workflow:
Natural-language data retrieval → Core indicator monitoring and drill-down → Anomaly attribution and rapid report generation
- Financial semantic dialogue for instant data retrieval (Text-to-Data): By connecting to core business databases or indicator platforms, the solution accurately understands complex business queries such as “How did the non-performing loan ratio for consumer loans at a certain branch change month over month last month?” and “What are the trends in deposit acquisition costs and interest spreads across different maturities?” It automatically parses the user's intent and retrieves the relevant data.
Financial semantic dialogue for instant data retrieval (Text-to-Data): By connecting to core business databases or indicator platforms, the solution accurately understands complex business queries such as “How did the non-performing loan ratio for consumer loans at a certain branch change month over month last month?” and “What are the trends in deposit acquisition costs and interest spreads across different maturities?” It automatically parses the user's intent and retrieves the relevant data.
- Multidimensional Slicing and In-Depth Attribution of Anomalous Fluctuations: When an indicator changes unexpectedly, such as declining revenue in a business line or lower-than-expected non-interest income, Raccoon Work automatically drills down across multiple dimensions, including region, channel, customer segment, and maturity. It measures each factor's contribution and identifies the primary sources of underperformance.
Multidimensional Slicing and In-Depth Attribution of Anomalous Fluctuations: When an indicator changes unexpectedly, such as declining revenue in a business line or lower-than-expected non-interest income, Raccoon Work automatically drills down across multiple dimensions, including region, channel, customer segment, and maturity. It measures each factor's contribution and identifies the primary sources of underperformance.
- One-Click Generation of Visual Analysis Reports: Combining chart generation with professional financial language, the solution automatically produces weekly and monthly business operations analysis summaries to support assessments of operational trends and recommendations for next-step strategies.
One-Click Generation of Visual Analysis Reports: Combining chart generation with professional financial language, the solution automatically produces weekly and monthly business operations analysis summaries to support assessments of operational trends and recommendations for next-step strategies.
Business Value:
It significantly frees business personnel and data engineers from the routine burden of data requests, reducing the time required to obtain data and prepare an initial report from several days to a few minutes. This enables data to be retrieved and analyzed on demand, making management decision-making more agile.
Customer Case Study | SenseTime and CCB Life Advance Intelligent Data Analysis in Insurance Operations Management
In insurance operations analysis, SenseTime and CCB Life have implemented an intelligent data analysis engine that combines large-model capabilities with insurance business data to improve the efficiency of data retrieval and analysis for business personnel.
The project connects to insurance business data assets, including operational indicators, data dashboards, and wide data warehouse tables. It enables business personnel to query business data in natural language, obtain analytical insights, and generate professional analysis reports with assistance from the system.
The capabilities are now being used in business scenarios including channel premium analysis, audit analysis, and financial analysis. They help business personnel perform data queries and indicator analysis, as well as prepare audit and financial reports, shifting data analysis from reliance on technical specialists toward self-service access by business personnel.
In practical applications, traditional data extraction and analysis processes that once took three to five days have been shortened to minutes, significantly lowering the barrier to data use for business personnel and improving the efficiency of operations analysis and business decision-making.
At the same time, through project implementation, the two parties developed a standardized model for launching use cases, providing a practical foundation for rapidly replicating these AI capabilities across more business scenarios. Building on these innovative practices, CCB Life's "Generative AI Data Decision-Making and Knowledge Worker Assistant Project" was selected for the 2026 IDC China Financial Industry AI Innovation Cases: Top 50, reflecting the results of the two parties' exploration of innovative financial AI applications and the conversion of business value.
Use Case 4 | Investment Research Insights
Intelligent Investment Research Analysis Through Multi-Source Information Fusion
Buy-side and sell-side investment researchers are surrounded every day by a vast volume of announcements, financial reports, broker research reports and industry news. In rapidly changing markets, the ability to quickly map industry-chain transmission, distill core investment logic and compare competitors' financial performance is critical to investment research efficiency.
Raccoon Work workflow:
Industry and company information aggregation → Industry-chain and investment-logic analysis → Risk assessment and multi-company comparison → Traceable report generation
- Comprehensive scanning of market announcements and professional research reports: Automatically monitors announcements from listed companies across the market and core research reports, accurately extracting key events such as earnings forecasts, major restructurings, senior management changes and share reductions.
Comprehensive scanning of market announcements and professional research reports: Automatically monitors announcements from listed companies across the market and core research reports, accurately extracting key events such as earnings forecasts, major restructurings, senior management changes and share reductions.
- Industry-chain mapping and competitive-landscape analysis: Maps upstream and downstream supply-chain dependencies, industry supply and demand trends and competitive barriers to quickly form a comprehensive view of the industry chain.
Industry-chain mapping and competitive-landscape analysis: Maps upstream and downstream supply-chain dependencies, industry supply and demand trends and competitive barriers to quickly form a comprehensive view of the industry chain.
- Peer comparison and investment-logic extraction: Supports side-by-side comparisons of financial metrics, valuation levels and business structures across multiple benchmark companies, distills core investment assumptions and catalysts, and objectively identifies potential risks.
Peer comparison and investment-logic extraction: Supports side-by-side comparisons of financial metrics, valuation levels and business structures across multiple benchmark companies, distills core investment assumptions and catalysts, and objectively identifies potential risks.
- Traceable professional research report generation: Assists in producing industry updates, initial drafts of in-depth company research reports and research meeting notes. Every data point and cited view is linked to the page number of the original announcement or the source research report, enabling one-click verification.
Traceable professional research report generation: Assists in producing industry updates, initial drafts of in-depth company research reports and research meeting notes. Every data point and cited view is linked to the page number of the original announcement or the source research report, enabling one-click verification.
Business value:
Helps investment researchers move beyond repetitive material collection and basic information processing so they can focus more on in-depth analysis and investment decisions, making research more insightful and conclusions more robust.
Use Case 5 | Compliance Management
Intelligent Compliance Review of Policies and Contract Terms
Financial regulatory requirements are becoming increasingly stringent and are updated frequently. Compliance and legal professionals must, on the one hand, keep up with new external regulations from the People's Bank of China, the National Financial Regulatory Administration and other authorities, assess how these requirements should be incorporated internally and revise internal policies. On the other hand, during routine contract reviews, traditional search methods often fail to identify deeper risks such as unequal rights and obligations, conflicting clauses and compliance gaps.
Raccoon Work workflow:
Regulatory requirements and internal policies aggregation → Regulatory-clause analysis and cross-reference → Contract risk review and revision recommendations
- Authoritative policy Q&A and precise clause retrieval: Connects to regulatory databases and internal policies and standards to provide semantic-level legal-clause retrieval, with answers precisely pinpointed to specific document numbers, chapters and clause paragraphs.
Authoritative policy Q&A and precise clause retrieval: Connects to regulatory databases and internal policies and standards to provide semantic-level legal-clause retrieval, with answers precisely pinpointed to specific document numbers, chapters and clause paragraphs.
- Incorporation of external regulations and differentiated policy comparison: When new regulations are issued, Raccoon Work automatically maps their provisions against the institution's current policies clause by clause and analyzes the differences, clearly identifying which policies require additions, repeal or revision, and automatically generating draft revision recommendations.
Incorporation of external regulations and differentiated policy comparison: When new regulations are issued, Raccoon Work automatically maps their provisions against the institution's current policies clause by clause and analyzes the differences, clearly identifying which policies require additions, repeal or revision, and automatically generating draft revision recommendations.
- Intelligent compliance review of business contracts: For various types of financial contracts, deeply scans for potential risks such as unlawful exemptions, implicit guarantees, authority overreach and performance deficiencies, generating a compliance review checklist that includes risk levels, the basis for revisions and recommended alternative clauses.
Intelligent compliance review of business contracts: For various types of financial contracts, deeply scans for potential risks such as unlawful exemptions, implicit guarantees, authority overreach and performance deficiencies, generating a compliance review checklist that includes risk levels, the basis for revisions and recommended alternative clauses.
Business value:
Builds a closed-loop compliance framework spanning regulatory input, internal policy upgrades and business contract execution, reducing delays in policy interpretation and eliminating blind spots in manual reviews.
Combining accumulated expertise in technology and industry practice to support reliable implementation of AI in finance
In demanding financial scenarios, trustworthiness is the foundation of usability. Raccoon Work Financial Edition is backed by three core engineering capabilities:
- Deep Multimodal and Financial Semantic Understanding: Building on SenseTime's expertise in computer vision and large language models, the solution maintains industry-leading parsing accuracy for layout analysis, seals and skewed tables, handwritten medical receipts, and specialized long-form financial documents.
Contact Information: business@sensetime.com












