Skip to content
Technology transformation expertise firm · ARDIANE Group Innovate, secure, succeed

Glossary

AIOps: definition, possible uses and data conditions

Domain

Cloud and platforms

In brief

AIOps (artificial intelligence for IT operations) refers to applying statistical and machine learning methods to IT operations data (metrics, logs, traces, alerts, tickets) to group alerts, detect anomalies and assist diagnosis. A term of common usage, AIOps draws on monitoring and observability data without replacing either.

Page contents

Definition of AIOps

AIOps, for artificial intelligence for IT operations, refers to applying statistical and machine learning methods to the data produced by IT operations: metrics, logs, traces, alerts, incident tickets, change log. The stated aim is to help teams sort these signals and understand an incident.

The term is one of common usage. Its scope varies from one vendor to another.

Too many signals to read one by one

In a distributed system, a failure can show up in several places at once. One component slows down, those that depend on it degrade, and each triggers its own alerts. The on-call engineer then receives a series of notifications for a single incident, with no indication of which one points to the origin.

AIOps targets this sorting: reducing the number of alerts to examine, flagging deviations that a fixed threshold does not see, linking an incident to what preceded it.

Three possible uses, each with its own data conditions

Use What it can provide What it requires
Alert correlation Grouping into one incident the alerts that appear to share a cause Consistent timestamps, common labels (service, environment, version), a description of the dependencies between services
Anomaly detection Flagging that a measurement is deviating from its usual behaviour, without a manually set threshold A history that covers activity cycles (day, week, period close), and allowance for changes in regime after a production release
Diagnostic assistance Linking the incident to a recent change or to similar past incidents, and suggesting leads A change log linked to services, past incidents documented with their cause

These uses remain possibilities, subject to these conditions. A grouping relies on coincidences and declared dependencies: it suggests an origin, it does not establish it. Anomaly detection can flag a harmless deviation or miss a slow degradation. A model trained on a history of poorly tuned alerts can reproduce their flaws.

Automatic execution of a corrective action is sometimes filed under the same word. It requires a situation recognised with sufficient certainty and an action whose effect is bounded.

AIOps, observability, SRE, MLOps: not to be confused

Monitoring and observability produce the signals, and our article observability vs monitoring distinguishes between them. AIOps consumes them. A standard such as OpenTelemetry comes in upstream: it gives signals common names and attributes, on which their correlation depends.

SRE sets what matters: service level objectives defined from the user’s point of view. AIOps processing does not say which degradation deserves to wake up an on-call engineer.

MLOps goes in the opposite direction. It applies operational practices to machine learning models, whereas AIOps applies machine learning to operations. An AIOps model is, moreover, still a model: it is monitored and re-evaluated like any other.

Example: a database slows down, alerts multiply

Take an online sales platform whose orders database slows down. Alerts are triggered on the response time of the programming interface, on the length of a message queue, on errors in the payment service and on the database itself.

Correlation processing groups them into a single incident, based on their timestamps and the declared dependencies between services. It places the database as the probable origin and flags a schema change deployed shortly before. The on-call engineer checks this hypothesis before acting.

Without a description of the dependencies, the grouping would have relied only on proximity in time. An alert unrelated to the incident, occurring at the same moment, could have been attached to it.

Consistent signals, alerts linked to an action and service level objectives are the subject of our page SRE and observability.

FAQ

Frequently asked questions

Another question? Ask us

What does AIOps stand for?

AIOps stands for artificial intelligence for IT operations. The word refers to the use of statistical and machine learning methods on the metrics, logs, traces, alerts and tickets of an information system. It is a term of common usage, whose scope varies from one vendor to another.

Does AIOps replace monitoring or observability?

No. Monitoring and observability produce the signals: checks, metrics, logs, traces. AIOps takes them as input to group alerts, spot deviations or suggest lines of diagnosis. Incomplete or poorly labelled signals therefore limit what AIOps processing can draw from them.

What data does an AIOps use case need?

Each AIOps use case has its own conditions. Alert correlation requires consistent timestamps, common labels and a description of the dependencies between services. Anomaly detection requires a history that covers activity cycles. Diagnostic assistance requires a change log linked to services and past incidents documented with their cause.

A project to bring to completion?

A first conversation is enough to understand your challenge, identify the right people and tell you frankly what we can bring.