Upgrade & Secure Your Future with DevOps, SRE, DevSecOps, MLOps!
We spend hours scrolling social media and waste money on things we forget, but won’t spend 30 minutes a day earning certifications that can change our lives.
Master in DevOps, SRE, DevSecOps & MLOps by DevOps School!
Learn from Guru Rajesh Kumar and double your salary in just one year.
Every operations team knows the sound of a pager going off at an inconvenient hour. What fewer teams measure is how much of their month the pager has quietly claimed. This checklist is built for AIOps practitioners who suspect that urgent versus important is being decided by whichever notification arrived last. Work through it with your own incident history open, and mark each item as a pass, a partial or a gap.
The core idea is simple. Urgent work shouts: a paging alert, a customer escalation, a failing deploy. Important work waits: retiring a flaky detection rule, cleaning the training data behind an anomaly model, writing down what a runbook should really say. A team that only answers the shouting never reaches the waiting, and the waiting work is exactly what would have made the shouting rarer.
Step one: list the last month of interruptions
Pull the previous thirty days of pages, tickets and chat escalations into one flat list. Do not filter or clean it. For each entry, write two short notes: how quickly someone had to act, and what the item changed once it was closed. An alert that fired, was acknowledged and closed itself leaves no durable change. A tuning session that removed a noisy rule leaves a lasting one.
- Pass: every interruption has both notes written beside it.
- Partial: you can reconstruct the timing but not the outcome.
- Gap: no consolidated list exists, only scattered tool views.
Step two: sort every item into four boxes
Draw a grid with time pressure on one axis and lasting value on the other. Urgent and important goes top right: a production outage with real impact. Urgent but not important is the noisy page for a host that restarts itself. Important but not urgent is the model retraining you keep postponing. Neither is the dashboard tweak somebody asked for in passing.
Most teams discover that the urgent-but-unimportant box is the largest, and that is the finding worth taking to your next planning meeting. It is not a failure of discipline. It is what happens when automation is excellent at generating signals and nobody has been asked to rank them.
- Pass: each of the four boxes has at least a handful of real examples.
- Partial: items cluster in one box because the sorting was done by gut feel.
- Gap: nobody agrees which box an alert belongs in.
Step three: check what the urgent box is costing
Take the urgent-but-unimportant items and ask which of them could be absorbed by a rule, a runbook step or a suppression window. The mechanics of cutting alert volume are laid out in a guide to reducing event noise through smarter technology, so here the audit question is only whether anyone owns the work of turning repeat pages into permanent fixes. If the answer is no, the box will refill forever.
Be concrete. Suppose a disk-usage page fires every Tuesday after a batch job. The urgent response is to clear space. The important response is to change the job or the threshold so the page never fires. Write both actions on the card and notice which one actually got done.
- Pass: repeat pages are tagged and reviewed on a fixed rhythm.
- Partial: repeat pages are noticed but not tracked.
- Gap: each page is treated as new every time.
Step four: protect a block for the important box
Sorting without scheduling changes nothing. Put the important-but-not-urgent items on the calendar before the week fills up, and treat that block like an outage window: it moves only if something in the urgent-and-important box genuinely needs it. Many teams find that a modest, recurring slot beats an occasional heroic cleanup week that always gets cancelled.
A helpful rule is to schedule the important work first and let the urgent work fill the gaps around it, rather than the reverse. Urgent items have a habit of expanding to consume whatever time they are given, while important items never announce themselves.
- Pass: a recurring block exists and has been defended at least twice.
- Partial: a block exists but is regularly given away.
- Gap: important work happens only when nothing else is on fire.
Step five: sanity-check your model-related backlog
AIOps adds its own flavour of important-but-quiet work. Detection models drift, training data goes stale, and correlation rules accumulate exceptions nobody remembers adding. None of this pages anyone. It simply makes the next page a little less trustworthy. Review the age of your last tuning pass and the count of exceptions attached to your top rules, and decide whether either deserves a place in the protected block.
If you want a reference point for how this kind of judgement plays out in a very different setting, consider how a city-events site handles a flood of listings and forum posts. Picture an events website for Jaipur with browsable listings, places to go, a talk board and a journal section: its editors must choose each day what needs a quick response and what can wait for a deeper tidy. Browse the JaipurOrbit site to see how clearly the quick browsing paths are kept apart from the slower editorial ones.
Step six: write the rule down
Finish the audit by turning your findings into a one-paragraph policy: which kinds of alerts may interrupt deep work, which wait for the next review, and who can overrule the policy. It does not need to be long. It needs to be unambiguous enough that a new on-call engineer can apply it at three in the morning without asking permission.
For the decision-making side of this, see our discussion of how AI-assisted insight can support IT choices; the same ideas apply to ranking work, not only to diagnosing incidents.
Scoring your audit
Count your passes, partials and gaps. Mostly passes means your team already sorts deliberately and only needs to keep the habit alive. A spread of partials is the most common result and points to one or two cheap fixes, usually the recurring block and the repeat-page tags. A column of gaps is not a verdict on the team; it is a map of where to start, and starting with step one costs only an afternoon.
Revisit the audit after one full month. The urgent-but-unimportant box should shrink, the important box should show finished items instead of intentions, and the pager should feel less like a boss and more like a tool. That shift, repeated quietly over several cycles, is what separates teams that run their systems from teams that merely react to them.