How Agricultural Bank’s Credit Middle Platform Accelerated Delivery with DevOps

1. Background

The credit middle‑platform project is a key digital‑transformation initiative of Agricultural Bank, aiming to replace the traditional siloed credit‑product development model with standardized, reusable credit services deployed on a private PaaS cloud using micro‑service architecture.

Since the agile transformation began in 2017, the team has achieved partial automation in code submission, checking, and building, but still lacks systematic continuous‑delivery tools and processes, leading to three main pain points:

Insufficient end‑to‑end automation: manual intervention is required to stitch together automated steps, causing reliability risks.

Weak quality‑control mechanisms: low unit‑test coverage, manual functional testing, and delayed code‑review and security checks.

Missing digital measurement feedback: no engineering‑level metrics for delivery efficiency or quality, hindering continuous improvement.

2. DevOps Practices

1. "Simplify" – Lightweight Branching Model for Fast Delivery

Adopting a feature‑branch‑based trunk‑release model, developers create short‑lived feature branches, merge quickly to the master, and trigger automated build, test, deployment, and release, enabling multiple deployments from a single build and shifting delivery cadence from monthly to weekly.

2. "Flowing" – Visual, Orchestrated Continuous Deployment Pipeline

The team built the first visual, orchestrated automated deployment pipeline in Agricultural Bank’s pilot projects, supporting parallel task execution, concurrency control, manual intervention, and module‑level reuse, covering development, SIT, and UAT stages.

Development Deployment Pipeline

After generating a primary artifact, it is automatically deployed to development and regression test environments, runs full regression suites, and promotes to an Alpha version upon passing quality gates.

SIT Deployment Pipeline

Testers select an Alpha version for manual and automated testing; successful tests promote the artifact to a Beta version.

UAT Deployment Pipeline

At the iteration’s acceptance meeting, the Beta version is deployed to the user‑acceptance environment, runs critical test cases, and upon meeting entry criteria becomes an RC version, later finalized as PRBL before production.

3. "Linked" – End‑to‑End Change Traceability

Requirements, code, test reports, and artifacts are linked via a build‑based tagging system. Feature branches tie requirements to code; successful builds generate a unique tag that matches the artifact name, and subsequent pipeline stages keep tags and artifact names synchronized, ensuring bidirectional traceability.

After production, change‑order numbers and baseline information are written back to the requirement items, enabling lookup of test reports, code, and artifacts for any released feature.

4. "Fortified" – Triple Quality Gates

Quality checks are embedded in the pipeline to intercept non‑compliant items early.

Continuous Integration Gate : code style check, automated unit tests (coverage > 80% and no regression), and security scanning (no new critical or high‑severity defects).

Development Self‑Test Gate : full interface regression suite must pass 100% before the artifact advances.

System Test Gate : SIT requires 100% automated interface test pass; UAT requires 100% critical‑case pass before promotion to release.

5. "Half‑Effort, Double‑Result" – Full Automation of Interface Testing

Because the platform exposes credit services via APIs, three categories of automated interface tests are maintained: existing interfaces, new interfaces, and core interfaces, each automatically generated per iteration to cover all scenarios and allow selective execution.

6. "Targeted" – Metric‑Driven Delivery Improvement

A unified measurement platform tracks seven R&D stages (demand, coding, build, test, defect, environment, deployment). One key metric, “same‑day merge efficiency,” requires feature‑branch merges to master be completed on the same day, driving higher integration frequency and faster issue resolution.

3. DevOps Outcomes

1. Agile Management Becomes the Norm

DevOps transformation enforced Scrum‑based agile practices, breaking silos between business, development, and testing, and establishing weekly iteration cycles with daily stand‑ups and acceptance meetings, creating an end‑to‑end delivery loop from demand to operation.

2. Delivery Capability Significantly Enhanced

Introducing an in‑house automated API testing tool reduced regression testing time from weeks to minutes, while the full‑pipeline and weekly iterations cut delivery cycles from monthly to weekly and enabled rapid issue fixes.

3. Delivery Quality Continuously Improves

Continuous integration quality gates raised unit‑test line coverage from 0% to 90%, reduced technical debt, and lowered defect counts.

Technical debt items are treated as iteration tasks, steadily decreasing over time.

Defect numbers dropped sharply thanks to strict CI quality controls.

The DevOps continuous‑delivery maturity model is now fully implemented, with the team completing over ten iterations, transitioning from waterfall to agile, and achieving domestic‑leading levels of delivery speed, quality, and efficiency.