Three Core Topics Analyzed: Jointly Discussing Solutions for Industry Transformation

Centered on the intelligent implementation of insurance claims, this sharing session explored how insurance claims can evolve from "process-driven" to "decision-driven" by building vertical data systems and agent products, starting from underlying logic amid the rapid development of AI technology. Below are the compiled core viewpoints and practical achievements.

01 Return to Pain Points: Technological Iteration and Productivity Leap

Taking insurance claims as an example, the industry process has undergone multiple iterations: the early stage relied on the experience-based model of "senior guiding junior", with inconsistent standards and uneven quality; it then entered the era of streamlined and standardized ERP systems, supplemented by assessment systems such as OKRs and KPIs. However, the systems became increasingly bloated after continuous additions, which instead dragged down efficiency.

The emergence of artificial intelligence has broken this dilemma. AI achieves knowledge equalization through the "optimal decision chain", directly outputting the best solutions based on existing knowledge systems, freeing employees from the internal friction of cumbersome processes.

Agents have further achieved collaborative upgrades across points, lines, and planes: industry processes are "lines", industry applications are "planes", and agents, as core callers, determine the upper limit of business efficiency and value.

02 Three Core Elements and Construction of Vertical Data for Insurance Claims

In 2026, the key to AI development lies in the continuous upgrading of three core elements:

Computing Power: Breaking through the bottleneck of Moore's Law with continuously declining costs, providing solid support for large-scale computing and agent decision-making.

Model Logic: Evolving from basic dialogue and reasoning stages to agent collaboration, achieving efficient human-model synergy.

Domain Knowledge System: Large models rely on precise professional knowledge to empower businesses. The current focus is to deeply cultivate scenario-based private domain data, realizing digitalization and in-depth review of business behaviors.

In the field of insurance claims, Jingyou Technology has achieved multiple results in vertical data construction:

uto Insurance Parts Data: Completed full analysis of 360 million vehicles nationwide, covering 5,000 parts and complete assembly relationships.

Personal Injury Claims Data: Refined processing nodes to 56, accurately accumulating processing cycles for regular, large-value disability, and death cases; built injury recognition models that automatically screen unreasonable medications and support dynamic logical error correction.

Pre-Litigation Mediation Element Database: Built a core dispute library in accordance with the "mediation before case filing" requirement, connected to the Supreme People's Court's National Insurance and Property Dispute Mediation Platform, with an annual processing scale exceeding 110 billion yuan. 82% of defendants are insurance companies, effectively acting as a pressure relief valve for social disputes.

03 "Small Entry Point" Breakthrough: 2430 Product Positioning and Five-Layer Architecture

Addressing issues such as bloated enterprise core systems and numerous concerns about full-scale large model implementation, Jingyou achieved breakthroughs through small-entry-point agents: focusing on high-frequency, standardizable scenarios, and prioritizing solving pain points of high case volumes, high risks, and insufficient manpower, such as survey, damage assessment, and minor injury identification.

Accordingly, it launched the "2430" product iterated to version 3.3, aiming for knowledge equalization, enabling one person to complete the workload of three with agent assistance and output optimal solutions.

The architecture adopts a five-layer "hamburger" system:

Perception and Observation Layer: Acting as digital senses, collecting information through hardware;

Computing Execution Layer: Responsible for process execution and data invocation;

Storage and Memory Layer: Undertaking data storage, verification, and recycling

Intelligent Central Layer: Intelligent shunting by Agents, assigning to agents, humans, or experts for processing;

Result Output Layer: Presenting final processing outcomes.

04 One-Stop Claims Platform: From Underlying Engine to Business Layering

Based on the five-layer architecture, Jingyou built a one-stop claims operation platform adopting a three-tier architecture overall:

Bottom Tier: Basic Capability Layer

●  Core Business Engine: Flexibly scheduling Agents, processes, and channel modules;

●  Vertical Domain Agents: Covering scenarios such as case reporting, survey, and vehicle damage assessment;

●  Channel Interaction Layer: Automatically extracting and pre-filling case elements through voice-semantic interaction.

Middle Tier: Business Operation Layer

●  Car Owner End: Realizing rapid reporting and survey via AI, 5G, and AR, meeting customers' demand for efficient case closure

●  Claims End: Covering closed-loop management of the entire process including survey, documentation, personal injury tracking, and pre-litigation mediation.

Top Tier: Customer Touch Layer

Supporting multiple channels such as WeChat, Mini Programs, and proprietary systems, covering scenarios including accidents, inquiries, claims, and tracking.

Addressing the pain point where claims personnel of small and medium-sized insurance companies need to coordinate with multiple parties such as 4S shops, repair shops, and internal management simultaneously, Jingyou Technology is transforming into an open service platform this year. It will build a tradable industry ecosystem platform in the future, returning to the essence of "doing business well".

05 Scenario Implementation: From Intelligent Reporting to Frictionless Claims

Currently, the platform has implemented multiple scenarios including AI intelligent case reporting, intelligent dispatch and shunting, self-service damage assessment, minor personal injury identification, intelligent mediation, AI quality inspection and re-inspection, intelligent SOP, and omni-channel services. Core applications are as follows:

AI Intelligent Case Reporting

Matches appropriate voice styles based on customers' age, region, and identity, with costs reduced by over half compared to manual + 5G video, supporting 7×24-hour service. AI automatically identifies demands, pre-fills work orders, intelligently dispatches tasks, recommends scripts, and follows up on missing document reminders, solving the problem of no service during non-working hours.

Auto Insurance AI Intelligent Survey

Car owners complete self-service surveys with one-click photo taking. Visual models identify three certificates, automatically extract elements, and OCR pre-fills and archives, covering the entire process of reporting, material collection, and damage estimation. The system has built-in rule quality inspection, verifying document relevance and loss rationality, effectively preventing unreasonable payouts.

AI Damage Assessment for Minor Personal Injuries

Covers 146 common minor injuries, completing vector calculations based on massive images, automatically identifying injury locations, providing mediation plans and exclusive scripts combined with regional factors, and supporting online agreement signing. In pilots with 16 insurance companies, it significantly improved efficiency and reduced losses.

Personal Injury Flash Recording Product

Supports batch uploading of claims images, automatically classifying and extracting information, identifying duplicate invoices, completing medical insurance matching and deductions, and generating professional reports with one click. Supports local box deployment to meet data localization requirements, providing two cooperation modes: rental instead of construction and platform invocation.

06 Technical Foundation: Agent+Skill Architecture and Security & Compliance Design

Jingyou Technology and Volcano Engine have established in-depth technical cooperation to jointly build an intelligent technical foundation for insurance claims. The system abandons the model of solely relying on large model Agents and adopts a layered "Agent+Skill (Intelligent Agent + Skill)" architecture.

The application layer builds vertical scenario agents for insurance claims, and the underlying layer relies on core products of Volcano Engine such as High Agent and Agent Sphere to achieve efficient scheduling and intelligent transformation of agent capabilities. The platform disassesses damage assessment experts' skills into independent Skill units. For example, image processing can be subdivided into modular capabilities such as classification, recognition, and quality verification, enabling flexible combination of capabilities.

Meanwhile, addressing the pain points of bloated and difficult-to-transform core systems of insurance companies, Jingyou built a unified business engine between traditional systems and agents, responsible for process orchestration and intelligent task allocation, forming a closed-loop operation with efficient agent-human collaboration. In frictionless claims scenarios, the automation rate of pre-processes such as reporting, survey, and document collection has reached 90%–95%. Combined with AR photo guidance and intelligent voice interaction, it can effectively comfort accident customers and standardize operation processes.

In-Depth Application of Vector Databases

Addressing the scattered and non-standardized vehicle maintenance data, the platform builds a dedicated vector database based on original maintenance manuals, generating structured troubleshooting paths according to fault codes and phenomena, and providing targeted inspection suggestions combined with regional environments such as high temperature and humidity, highly restoring the professional reasoning logic of senior maintenance technicians.

Security & Compliance Design

The insurance industry is strictly regulated. The system adopts a dual-layer security protection system: Bidirectional Desensitization: Desensitizes private information such as ID cards and medical records throughout the input-output link, strictly complying with the Personal Information Protection Law; Reject Black Boxes: Adds business fallback mechanisms for large model hallucination risks, eliminating problems such as overpayment and misjudgment of injuries through rigid rules.

The entire system, based on the joint technical foundation of Jingyou and Volcano Engine, supports capability reuse across multiple insurance types including auto and non-auto insurance. One engine covers full-scenario claims operations, adhering to deep cultivation in vertical fields to maximize professional value.