How JD Insurance Rebuilt Its Core System for Scalability and 99.99% Accuracy

Introduction

JD is widely known for its e‑commerce marketplace, but behind it lies a dedicated team of JD Insurance agents who provide protection services rather than physical products. The Insurance Agent Development Department supplies technical and business support to agents, aiming to optimize services, create opportunities, and drive industry growth.

Agents act as a trust bridge between customers and the company; establishing clear regulations and compensation systems—referred to as the "Basic Law"—is essential for maintaining this bridge. The Basic Law defines management, promotion, rewards, and profit‑sharing rules, ensuring agents can operate within a transparent framework and receive substantial commissions.

To improve service efficiency for agents and internal staff, a smart, automated, and personalized system architecture is required. The existing architecture no longer meets business needs, causing frequent system issues, heavy workload, and reduced customer trust, prompting an urgent need for system reconstruction.

Basic Law Overview

The Insurance Basic Law is a set of daily management regulations for insurance marketers (agents and staff), covering activities, promotions, rewards, and benefit distribution. In JD's insurance business, it guides agents' career development, performance assessment, and compensation, directly influencing business stability and market competitiveness.

2.1 Core Content

Daily Management: Standards for signing, changes, termination, and training to ensure team stability and professionalism.

Compensation & Benefits: Defines calculation methods for direct commissions, indirect commissions, non‑cash rewards, and benefits such as pensions.

Performance Assessment: Clear performance requirements and incentives for each career stage, evaluated across metrics like business achievement rate, continuation rate, and team indicators.

2.2 System Business Flow

Challenges

3.1 System Complexity

Data Processing Complexity: Large volumes of customer, policy, and performance data with diverse structures create significant technical challenges.

Customization Requirements: Three versions of the Basic Law involve 36 commission items and 17 grades, demanding deep insurance knowledge and strong system customization capabilities.

Policy & Regulation Changes: Frequent regulatory updates require a flexible, extensible system to keep the Basic Law compliant.

3.2 Business Complexity

Product Diversity: Multiple insurance products (life, property, auto) each have unique terms and rate structures, raising the bar for rule definition and execution.

Sales Channel Diversity: Agents sell through offline and online channels, each with distinct sales models and customer needs, requiring adaptable rule sets.

Risk Management: Insurance carries credit, market, and operational risks; the Basic Law must promote growth while strengthening risk controls.

System Transformation

4.1 Full‑Scenario Architecture

4.2 Refactor Comparison

Comparison of pre‑ and post‑refactor:

Readability: Before – tightly coupled, duplicated, chaotic code; After – low coupling, eliminated duplication, clear logic, reduced complexity.

Flexibility: Before – low modularity, hard to maintain/extend; After – modular design, easy maintenance and extension.

Reliability: Before – 0%; After – 99.99%.

Security: Before – missing error‑handling; After – automatic review and traceability.

Performance: Before – heavy real‑time data dependence, long response time (6 hours); After – pre‑loaded indicators, Drools rule engine, short response time (30 minutes).

Key Breakthrough Points

5.1 Modular Decoupling & Clear Responsibility

Component‑Based Design: System is split into autonomous components (rules, rewards, grade management, performance evaluation, team collaboration), improving maintainability, extensibility, and reuse.

Configurability: A visual configuration interface allows behavior changes without code modifications, simplifying rule customization and reducing operational costs.

Free Combination: Users can assemble component versions like building blocks to quickly create personalized Basic Law solutions, shortening rollout time.

5.2 Introduction of a Rule Engine

Drools is used to handle diverse commission and assessment rules. Business rules are written in DRL, separating them from code for easier maintenance.

Rule‑Code Decoupling: Rules stored as scripts, enabling independent updates.

Common Indicator Extraction & Pre‑Loading: Frequently used metrics are pre‑loaded to reduce real‑time data pressure.

Rule Configuration: Visual UI lets users select indicators, conditions, and operators to compose rules without writing DRL manually; the system auto‑generates the script.

Intuitive Rule Viewing: Configured rules are displayed in a readable format, lowering learning curve and improving verification efficiency.

5.3 Multi‑Dimensional Execution & Data Version Isolation

The Basic Law can be executed per version, team, commission, or assessment component, with data version isolation supporting repeatable runs and cross‑version comparisons, enabling accurate commission forecasting and performance tracking for agents.

5.4 Enhanced R&D Verification Mechanism

5.4.1 Data Verification

Before settlement, the system calculates a reasonable commission range based on preset fee ratios and automatically aggregates direct, indirect, and other commissions. The aggregated result is compared against the range to ensure accuracy, preventing human error and safeguarding both agent earnings and company finances.

5.4.2 Multi‑Dimensional Business Data Analysis

Integration with a BI reporting system provides visual, drag‑and‑drop dashboards for multi‑dimensional data analysis, improving data interpretability, anomaly detection, and cross‑department collaboration.

System Stability Construction

6.1 Alert Mechanism

The system integrates UMP and MDC monitoring, configuring alerts for JVM, CPU, disk, connectivity, RSS memory, load, TCP connections, network I/O, and retransmissions to ensure continuous, stable, and efficient operation of Basic Law tasks.

6.2 Task Execution Trace

Link Monitoring (PFinder): PFinder tracks SpringMVC, JSF, MySQL, JMQ and other middleware, building service topology and enabling rapid performance bottleneck analysis.

Task Chain Monitoring: Critical tasks (team tasks, commission settlement, performance assessment) are logged per node; timeouts trigger retries and, after three failures, email and phone alerts for immediate response.

Summary and Outlook

Since the refactor, system accuracy has risen to 99.99%, supporting over 40,000 agents and changing their perception of the platform. Future plans focus on continuous performance, stability, and security improvements, embracing new technologies to drive smarter, more automated business processes and accelerate growth.