Can AIOps Revolutionize Traditional Industries? Insights from a 2018 Shenzhen Talk

1. Current Development and Prospects of the Operations Industry

Operations are ubiquitous, and all future services will run on software‑hardware systems, creating a massive market. The speaker has previously shared talks on machine‑learning‑based intelligent operations, AIOps research issues, and a roadmap for AIOps adoption.

These insights were gathered from half a year of discussions with partners across finance, energy, telecom, manufacturing, and smart city sectors.

Using finance as an example, the current state of AIOps in the industry includes unclear concepts, many implementation doubts, growing project initiatives, building of large‑scale operations data platforms, and attempts to address pain points with AIOps as a response to national AI strategies. 2018 is seen as the "year one" for AIOps in finance and traditional industries.

2. Overview and Value of AIOps

Typical financial failures—such as abnormal bank transfers, login issues, or payment platform outages—cannot be fully eliminated; the focus is on rapid response and swift remediation.

Operations involve massive, ever‑changing interconnections among networks, data centers, and software modules, generating continuous triggers such as hardware faults, software changes, user behavior shifts, and attacks. Human operators cannot process this flood of data quickly enough, so AI is the only viable solution.

AIOps replaces slow, error‑prone human decisions with rapid AI‑driven decisions or proactive fault avoidance. The AIOps ecosystem is described as "eye, brain, heart, hand": the "eye" is the operations big‑data platform, the "brain" processes data and makes AI decisions, the "heart" defines goals (stability, cost‑efficiency, speed), and the "hand" executes actions via automation.

Through this four‑component collaboration, AIOps delivers "stable, cost‑effective, fast" value and dramatically boosts operational productivity. The ultimate form is Autonomous IT Operations—unattended operations.

3. Directions and Application Examples of AIOps in Finance

Successful AIOps deployment requires genuine pain points, an existing operations big‑data platform, basic automation tools, and quantifiable evaluation metrics.

Scenarios are chosen based on short‑term impact, availability of research papers, and product readiness. The speaker lists common AIOps scenarios—fault detection, fault mitigation, fault repair, and fault avoidance—showing their productization levels.

An "out‑of‑the‑box" product for anomaly detection and localization is demonstrated: the system ingests historical KPI curves without parameter tuning, automatically handles missing data, and detects anomalies across CPU, memory, and database metrics.

Two sensitivity‑adjustment methods are provided: a manual slider and an automatic feedback‑driven adjustment that searches for similar anomalous intervals across many curves.

After detection, the system pinpoints the affected machines and metrics, presenting the information on a single screen to aid rapid mitigation.

4. Top Ten Concerns of the Financial Industry About AIOps

1) What exactly is AIOps? How to distinguish genuine AI‑powered solutions? A visual standard is provided.

2) Is it too early to try AIOps? Many institutions have already started projects, so it is not premature.

3) Do internet‑sector cases apply to traditional industries? Good algorithms can be universal; each problem requires specific analysis.

4) Must a top‑level design be completed first? A solid foundation—standardized data, automation, a big‑data platform, compute resources, and mature algorithms—allows incremental progress.

5) Is incomplete data a blocker? Different scenarios have different data needs; start with what you have and iterate.

6) Can internal developers create new AIOps algorithms alone? Breakthroughs require extensive research and collaboration with algorithm experts.

7) Is implementing existing academic algorithms sufficient? Productization introduces new challenges; deep expertise is needed for robust deployment.

8) Can black‑box methods train AIOps models? Effective AIOps requires a modular architecture that matches AI strengths to specific operational problems.

9) Which department should lead AIOps? Cross‑departmental data integration and a dedicated internal AIOps team are essential.

10) Will AIOps replace operations staff? Not in the near term; it will improve work conditions and create new AI‑focused roles.

5. AIOps Challenge

The community organized an AIOps challenge with sponsorship from Sogou, Tencent Games, eBay, Baidu, Alibaba, Microsoft Azure, and Huawei. The preliminary round is finished; the finals are ongoing with top teams achieving over 82% accuracy.

Upcoming final presentations and awards will be held in mid‑May, accompanied by an AIOps academic symposium.

6. Future Outlook of AIOps

AIOps will bring revolutionary changes to operations across all industries, with the ultimate goal of fully unattended operations. The speaker emphasizes that cultivating AIOps talent within traditional sectors through collaboration and experimentation is the key to successful adoption. 2018 is positioned as the "year one" for AIOps in traditional industries.