Somewhere in your workplace there is a group that answers when the shared printer dies, when the email stops syncing and when a payroll screen freezes an hour before deadline. Nobody outside that group thinks about it much, yet everything else rests on it. The group’s work has a name, ITOps, and for anyone curious about how machine learning is starting to touch it, a plain primer is a good place to begin. This guide tells the story of one imaginary morning, then gathers the key words into a compact lookup list for later.
One imaginary morning in the operations room
It is 8:15. Priya, a made-up junior analyst on a made-up operations team, opens a queue of overnight tickets. Eleven are password resets. Two are people asking for software. One says “the reporting tool shows yesterday’s numbers”, and that one makes her sit up. The first twelve are routine; the last is a break in something that normally works.
Priya’s morning shows the pattern that defines the field. The Ops Atlas, a reference set maintained by xOps School, describes ITOps through its entry on the discipline as the function responsible for keeping IT services up. It handles requests and incidents, hunts down causes of repeat failures, controls change and keeps a map of what exists. If you strip away the jargon, that is a help desk, a repair crew, a gatekeeper and a record keeper sharing one roof.
The stale-numbers ticket turns out to be a scheduled job that stopped after a maintenance change two days earlier. Priya fixes it, then does the part beginners often skip: she writes down what happened and flags it as a candidate for deeper investigation, because a job that stopped once may stop again. Fixing the symptom is incident work; chasing the cause is problem work. Keeping the two apart is the first big idea in this primer.
The word list, in the order the terms turn up
Request
A routine ask with a known answer, such as a new account or a software install. Requests are expected, so teams prepare recipes for them.
Incident
Something that worked and now does not. The goal is to restore service quickly, not to find out who is to blame.
Problem
The underlying cause behind one or several incidents. Problem work asks why the stale report happened, rather than merely restarting it.
Change
Any planned alteration to a live service. The Atlas sorts these into three families: standard changes that are pre-approved and low risk, normal changes that are assessed first, and emergency changes that go ahead immediately because waiting would cause more harm.
Service level terms
An SLA is the promise made to the people who use a service. An OLA is a promise one internal team makes to another, and an underpinning contract, abbreviated UC, is the promise an outside supplier makes. A customer deadline is really a stack of these smaller promises added together.
CMDB
A configuration database: the team’s map of devices, applications and how they connect. Newcomers assume it must list everything. The Atlas takes the opposite view, saying it should hold only what the team’s decisions actually need.
ITIL, ITSM and ITOM
Three terms that cause endless confusion. As the Atlas draws the lines, ITOps is the function made of people and roles, ITSM is the set of practices they follow, and ITOM is the technical tooling that watches systems. ITIL is a well-known body of guidance for those practices. Vendors do not all agree on how the labels overlap, so do not be surprised when two product pages describe them differently.
What machine learning can take on in the ops room
Because this site follows artificial intelligence, the obvious question is whether software can simply take over Priya’s job. Parts of it can help. The Atlas leaves the task of grouping and correlating raw alerts to a neighbouring discipline, AIOps, and keeps only the event layer inside ITOps. Its small event lab makes the point concrete: thirteen alert rows collapsed into two incidents at 8:20, and after a grouping window closed at 8:30 three late alerts opened three more, so five incidents were opened and one resolved by itself.
That is a gentle example of why grouping rules matter. If your window is too short, one outage looks like several; if it is too long, separate problems merge into one. A learning system can tune such windows, but someone still has to decide what counts as one event worth waking a person for.
The Atlas also notes that the next generation of the ITIL guidance, announced by its publisher in January 2026, is described as “AI-native by design” and is being released in phases. The earlier edition, ITIL 4, remains available, and the Atlas calls the whole framework optional. So adopting machine learning in operations is a choice, not a requirement.
A cautionary tale about predicting risky changes
One case in the Atlas is worth a slow read for any beginner. A research team at a bank published a preprint, not yet peer reviewed, describing a model that scores the risk of a proposed change. In the figures the Atlas reports, it caught 63 percent of the changes that went on to cause a major incident, against roughly 20 percent for the bank’s existing questionnaire. Out of 203 changes, 35 caused major incidents, and the model’s precision was 0.55.
Those numbers sound impressive, and the Atlas adds the sobering part. Apply the same approach to 10,000 changes where only 1 percent fail, and only about 6 percent of the flags would be real. Rare events swamp even a good detector with false alarms. For a newcomer, this is a lesson in base rates, and it is as useful as any algorithm detail. A tool can sharpen attention, but a person still has to weigh each warning.
If this reminds you of other decision-support ideas, you might enjoy our piece on how AI supports everyday business decisions, which looks at the same tension between helpful signals and sound judgment.
The anti-patterns beginners can spot early
Spotting trouble takes no seniority. These warning signs from the Atlas are visible from a seat in the queue.
- Heavy procedures forced on a team that runs a single small service.
- Every change handled with the same ceremony, whether trivial or risky.
- New rules invented after each failure, until nobody can follow them.
- Response-time clocks that nobody can explain, such as whether weekends count.
- Monitoring treated as a nice extra, so a silent failure hides an outage.
- Private spreadsheets that duplicate the official records.
The monitoring item deserves a second look. If the alerting layer itself breaks without a sound, nothing tells you that the alarms are off. Treat it as a live service, with owners and checks, and not as a side project.
Two myths worth dropping now
The first myth says that ITIL demands a committee approval for every change. According to the Atlas, standard changes are pre-authorised, so the committee sees only the assessed ones. The second says that ITSM means the ticketing product you bought. The Atlas puts it the other way round: the tool records the practice, and the practice is what people do.
There is a third trap with a similar name. Two unrelated things are abbreviated DORA: a long-running research programme about software delivery, and a European regulation about digital resilience in finance. They are different things, and mixing them up leads to some very confused meetings.
Where to go next
Once the vocabulary feels comfortable, try tracing one real incident from your own world through the stages in this glossary: how it was reported, how it was ranked, who fixed it, and whether anyone asked why it happened. The exercise takes an hour and teaches more than a chapter of definitions.
For a gentler cousin of this topic, the site’s explainer on how image-classification systems learn to recognise pictures shows the same patience with jargon. Operations and machine learning share a habit: both reward people who ask what a signal really means before acting on it.
Back in the imaginary operations room, Priya closes the stale-numbers ticket by 9:30 and opens a note titled “why did the job stop?”. That small note, more than any tool, is what separates a team that merely reacts from one that gradually gets quieter. Keep that image in mind and the glossary above will start to feel like a map rather than a list.