How a Chinese City Bank Integrated DevOps, AI, and Big Data to Transform Operations

Project Background

The bank selected its mobile‑banking system as a pilot to implement a unified development‑testing‑operations (DevOps) workflow that satisfies strict production‑environment security requirements.

Integration of DevOps and ITIL

DevOps practices were combined with ITIL 4 process controls (change, configuration, incident management) to identify key control points across requirement, design, development, testing and operations. Risks are captured, standardized and automated, enabling consistent handling while preserving existing infrastructure and organizational structures.

Artificial‑Intelligence‑Driven Operations

Four AI‑enabled scenarios are implemented:

Single‑KPI anomaly detection – deep‑learning dynamic baselines, reinforcement‑learning‑based self‑healing, and transfer learning for multi‑metric adaptation.

Multi‑KPI alarm aggregation – Monte‑Carlo tree search, random‑forest classification, CUSUM, difference‑in‑differences (DiD) and spectral methods to merge storm‑like alerts.

Fault root‑cause analysis – decision‑tree impact analysis, A/B testing on historical data, and automated fault‑propagation graph construction.

Predictive alerting – machine‑learning models trained on historical logs to forecast capacity, performance bottlenecks and potential failures.

Business Availability Management

Critical business systems and their key paths are identified on top of existing device monitoring. Automated inspections and KPI monitoring reduce outage time, while automated remediation tools enable rapid response to critical events.

Distributed Computing Framework (Apache Spark) and FOCUS Algorithm

The FOCUS algorithm runs on an Apache Spark cluster to process massive multidimensional data efficiently. Spark’s in‑memory computation accelerates iterative analytics, making it suitable for machine‑learning workloads. FOCUS extends Spark MLlib’s decision‑tree module with custom classes ( FocusDecisionTree, FocusClassifier, FocusRule) that perform attribute localization rather than classification.

Platform Architecture

The delivery platform is organized into four layers:

Infrastructure layer – elastic on‑premise and cloud resources that eliminate environment drift.

Tool‑platform layer – end‑to‑end tools for source‑code management, automated build, test and deployment.

Pipeline‑engine layer – orchestrates the tool platform according to the bank’s existing process‑control system.

Process‑control layer – integrates with ITIL‑based change, configuration and incident workflows.

Key Technical Components

Five‑layer DevOps‑ITIL platform – supports code management, automated build, test, deployment and aligns with process‑control systems.

Intelligent operations engine – implements the four AI scenarios using reinforcement learning, random forests, CUSUM, DiD, spectral analysis and Monte‑Carlo tree search.

Standardized inspection & emergency operations – automated state inspection and predefined emergency workflows reduce fault‑resolution time.

Application Effects

After deploying the five‑dimensional management model:

Automated deployment time for mobile‑banking services decreased by roughly tenfold.

Production release success rate continuously improved.

Critical system availability exceeded 99.99% for all major banking services.

Online service ticketing accelerated fault reporting and increased user satisfaction, especially during the 2020 pandemic.

Structured knowledge‑base derived from product catalogs enhanced cross‑team problem‑solving efficiency.

Conclusion

Deep integration of DevOps, ITIL, AI and big‑data analytics enabled the bank to establish a unified development‑testing‑operations pipeline, achieve near‑perfect system availability, and deliver faster, more reliable services to customers.