How China Bank Executed a Full-Scale IT Architecture Transformation: Practical Framework and Lessons

1. Overview of Implementation

When a large‑scale architecture transformation moves from blueprint to reality, the core challenge is breaking a multi‑year, hundreds‑of‑systems strategic goal into definable, deliverable, measurable projects and application migration tasks. The article focuses on the execution layer, decomposing the process from program‑level coordination to single‑application migration, offering a proven operational framework and key decision points for peers.

2. Implementation Framework

The transformation is summarized as “one program, two key activities, three enablement capabilities”. A program‑level office led by the head‑office fintech department coordinates “stabilizing the foundation, rebuilding the engine, empowering the business”. The work is split into three parallel project categories:

Infrastructure and platform projects (cloud platform, technology‑mid‑platform, security upgrades).

Core capability decoupling and migration projects (customer centre, product centre, trading engine, etc.).

Business‑enablement and pilot projects that build innovative applications on the new platform.

The three enablement capabilities provide enterprise‑grade platform services, shared technical components, and unified monitoring/operations to ensure consistent engineering standards.

2. Project Coordination: Classification and Precise “Drip” Management

Because of massive scale, complex dependencies, diverse technology stacks and domestic‑overseas differences, the bank adopts a classification‑by‑policy and precise‑drip management strategy. First, all applications are inventoried and tagged according to business criticality, technical coupling, data sensitivity and continuity requirements, forming an “application asset panorama” that drives priority, resource allocation and renovation strategy.

A closed‑loop technical governance system is applied before, during and after implementation:

Pre‑implementation: “package‑style” technology selection standards, “one line, one vendor” rule, and a black‑white‑gray open‑source admission list to curb fragmentation; a “38+106” standard‑specification system and online implementation craftsmanship enforce hard constraints.

During implementation: cross‑department technical reviews and architecture decision mechanisms enable rapid decisions.

Post‑implementation: batch “landing‑target” checks audit non‑functional compliance, interface release accuracy and timeliness, and track remediation.

Different system types receive tailored migration routes:

Core transaction systems (deposits, loans, payments): long cycles, high safety, “logical lift‑and‑shift, functional freeze”, prioritize tech‑stack replacement and distributed base migration, verify with traffic mirroring and parallel run.

Highly coupled complex systems: focus on domain modeling and dependency governance, “decouple first, migrate step‑by‑step”, split monoliths into micro‑services.

General business and management systems: “cloud‑first, encourage refactor”, agile iteration, serve as “gold test” for new capabilities.

Edge and non‑core systems: “eliminate or refactor”, replace low‑value legacy, evaluate mature services for specialized needs.

Inter‑project dependencies are managed through a “dependency map” and a version‑train release mechanism; for example, platform projects must deliver stable versions before application‑renovation projects, and the customer‑centre project must synchronize interface and data migration plans with dependent credit‑system projects. Regular cross‑project coordination meetings resolve resource conflicts and interface disputes.

3. Application Migration: Seven‑Step Practical Process

The bank distilled a seven‑step migration method: Assessment → Design → Decoupling → Refactor → Verification → Switch → Operations.

Deep assessment and metrics: analyse architectural decay (circular dependencies, fan‑in/out), test coverage, non‑functional baselines, and quantify expected business value (e.g., 20 % reduction in transaction latency).

Target architecture and design: define target form (micro‑services, modular monolith), produce detailed design documents, and pass mandatory technical review.

Dependency decoupling and interface governance: use analysis tools to identify internal/external dependencies, isolate external ones with “anti‑corruption layers”, refactor internal modules for high cohesion, register all interfaces on a unified API management platform.

Incremental refactor and CI: avoid “big‑bang” rewrite; adopt “killer‑mode” or parallel mode to gradually implement new services, enforce DevOps pipelines with automated quality gates (code scan, security, license checks).

Multi‑layer verification: establish a four‑layer test suite (unit → automated API → non‑functional stress/chaos → production‑traffic simulation). Simulation compares replicated production traffic with the new version to lower core‑system switch risk.

Controlled switch and rollback: detailed data migration, traffic cut‑over and rollback plans; use blue‑green, canary releases; monitor with clear business and technical success criteria.

Post‑go‑live operation and feedback: a few weeks of “intensive operation” monitors latency, error rate, traffic; issues are fed back to design for continuous improvement.

4. Platform and Tooling

To support large‑scale transformation, the bank built an enterprise‑grade technology platform that penetrates infrastructure and base software, unifies the tech stack, and abstracts underlying complexity. Key tools include an architecture‑asset management system (“one‑book‑of‑products”), a unified DevOps platform with quality and security gates, an API lifecycle management platform, data‑migration comparison tools, and an intelligent operations platform. These enable end‑to‑end standardisation and automation from governance to delivery and operations.

5. Metrics and Closed‑Loop Improvement

A three‑level measurement system drives continuous improvement: project‑level metrics (delivery schedule, budget compliance, requirement delivery rate); application‑level metrics (TPS, response time, SLA, CPU/memory usage, defect density); organization‑level metrics (deployment frequency, change failure rate, MTTR, developer productivity). Regular reviews of these data guide organisational management upgrades.

6. Insights and Outlook

The case shows that digital transformation success is roughly 30 % technology and 70 % governance and engineering. A decisive strategic commitment, systematic top‑down design, rigorous engineering implementation, fine‑grained technical governance and ecosystem collaboration are essential. Looking ahead, the bank plans to deepen AI integration, building autonomous, intelligent, agile digital infrastructure to support the real economy and high‑quality industry development.