AIOps: Background, Scenarios, Capability Building, and Practical Implementation by JD Digital Operations Team

Since Gartner introduced the AIOps concept in 2016, platformization and intelligence have become major trends in operations, which can be divided into five stages: manual/scripted operations, standardized tool operations, platform automation, DevOps, and AIOps.

Automation improves efficiency but cannot adapt to new problems; AI provides solutions to these pain points, giving rise to AIOps, which uses big data and AI to learn from massive monitoring data and complex IT environments, automatically detecting anomalies, locating faults, and predicting risks.

The development of AIOps is driven by the accumulation of standard operation data from mature platforms such as CMDB, monitoring systems, and workflow centers, as well as the increasing complexity and scale of monitoring data that challenge traditional automation.

Key focus areas of AIOps include:

Empowering DevOps : Using AI to handle problems that automation cannot solve.

Real‑time analysis and processing : AI algorithms provide rapid diagnosis and operational suggestions, dramatically reducing mean time to detect (MTTD) and mean time to repair (MTTR).

Reducing alarm noise : Data correlation improves filtering of false alerts.

Fault cause analysis and prediction : Massive data analysis identifies root causes and predicts future incidents.

The JD Digital Operations team assigns distinct roles—operations engineers, development engineers, and algorithm engineers—each essential for delivering AIOps solutions.

Typical AIOps scenarios are organized around three pillars:

Quality assurance : anomaly detection, fault diagnosis, fault prediction, and self‑healing.

Cost management : metric monitoring, anomaly detection, resource optimization, capacity planning, and performance tuning.

Efficiency improvement : intelligent prediction, change management, Q&A, and decision support.

The capability‑building roadmap starts with single‑scenario pilots, progresses to multiple linked AI modules, then to a fully orchestrated, core‑AI‑driven platform that balances cost, quality, and efficiency across the entire operation lifecycle.

Internally, the team provides four product platforms: metric identification, alarm identification, log interpretation, and fault investigation, along with scenario‑specific model files and containerized deployment solutions.

In 2020, the team accelerated AIOps adoption by designing a "mortise‑and‑tenon" algorithm that integrates metric values and log texts, enhancing scenario coverage, automatic orchestration accuracy, and extensibility.

The AIOps solution consists of three progressive modules:

Fault detection : rapid identification of anomalies in time‑series monitoring data.

Fault localization : precise pinpointing of root causes in complex systems using a dynamic knowledge graph and reinforcement‑learning algorithms.

Fault repair : recommendation of intelligent remediation actions based on expert knowledge, with risk indicators and, for some cases, automated self‑healing.

Metric anomaly detection leverages waveform analysis and spatio‑temporal features to capture complex patterns without manual thresholds, while log‑analysis extracts real‑time anomalies and validates root‑cause hypotheses.

The knowledge graph continuously ingests configuration, log, alarm, and change data, providing a dynamic, extensible foundation for fault analysis; reinforcement learning searches the graph globally to ensure accurate root‑cause identification.

When the search completes, the algorithm ranks and scores candidate causes, generates a fault‑analysis report, and stores it for post‑mortem review.

During intelligent fault repair, expert experience guides the generation of actionable recommendations and risk metrics; the knowledge graph validates these suggestions across the full call‑chain, enabling partial self‑healing in certain scenarios.

The JD Digital Operations team combines AI, big data, and knowledge‑graph technologies to deliver end‑to‑end AIOps solutions that improve detection speed, reduce false alarms, accelerate root‑cause analysis, and enable automated remediation, thereby enhancing service quality and business availability.