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.
The most useful meeting our imaginary AIOps team ever held had no slides. The platform lead had spent weeks tuning anomaly detection, and the on-call rotation still grumbled. So one Tuesday she closed the laptop and asked the room a single question: what would make this easier? Nobody mentioned model accuracy. What came out instead was a list of small frictions, the kind that never reach a roadmap because they feel too minor to file a ticket for. If you work with AIOps tooling, the eight items below may sound familiar.
Before the details, a word on tone. The question works best when it is genuinely open. If the person asking already has a favourite answer, the room senses it and offers that answer back. The platform lead in our story came in with no preferred outcome, which is why the replies were candid and, frankly, a little funny.
I am telling this as a story because that is how it landed for the people involved. Each fix is tiny. Together they changed how the team felt about the whole AIOps practice, which had been drifting toward the “clever but tiring” category.
1. A summary line that survives the pager screen
The first answer came from the engineer who gets paged most. Alert titles were cut off on her phone, so the one fact she needed, which service was affected, sat in the truncated part. Nobody had decided to hide it; the template simply grew over time. The fix took twenty minutes: put the service name first, the symptom second, and everything else in the body. She said the pager suddenly felt like it was written for a human.
Other teams could copy this without any tooling change at all. Pick the three most recent pages, read them on a small screen, and ask whether a half-awake person could decide what to do from the first line alone. A no usually points at the template, not at the tired person holding the phone.
2. A named place for “why did the model flag this”
An analyst admitted she opened four tabs every time an anomaly fired, just to learn which signals pushed the score up. The team added a short explanation field to each AIOps event, filled automatically from the top contributing signals. It did not make the model smarter. It made the model easier to argue with, which is a different and underrated virtue. The team had already been writing about decision support, and a post on how AI can support everyday IT decision-making had circulated in their channel a week earlier, so the idea of showing reasoning, not just verdicts, landed easily.
3. One shared sentence for “this can wait”
Several people confessed they were unsure when it was acceptable to leave an event until morning. Every engineer had a private rule, and the private rules disagreed. The team wrote one short paragraph describing what “can wait” means for their stack and pinned it where events arrive. Disputes at handover dropped, mostly because the argument had already been had, once, in daylight.
It also helped the newcomers. Instead of guessing at an unwritten norm, they could read a paragraph, disagree with it if they wanted, and propose an edit. The paragraph was revised twice in the first month, which the team treated as a sign that it was alive rather than a sign that it was wrong.
4. Fewer clicks to acknowledge a known issue
A junior engineer pointed out that acknowledging an already-known event required the same five clicks as investigating a new one. Under pressure she skipped acknowledgement entirely, so the dashboard looked noisier than reality. The team added a one-step acknowledge that carried a link to the open incident. If you are already working on trimming unnecessary events, the thinking in a piece about calming IT event noise with sharper tooling pairs naturally with this kind of click-count audit.
5. A rehearsal environment that does not feel like a stage
The newest member said she hesitated to try remediation steps because every trial happened on a shared staging system where other people were testing. Asking what would make it easier produced a throwaway sandbox that resets itself nightly. Nothing was invented here; the team merely made trying things cheap, and trying things is how people learn an AIOps platform.
One side effect surprised them: senior engineers began using the sandbox too. They rehearsed risky changes there before touching production, and the habit spread without anyone announcing it. When a tool makes caution cheap, careful behaviour becomes the default instead of a personal virtue.
6. Plain labels on automated actions
When the platform restarted a service on its own, the log line said only that a policy had fired. Engineers reading it later could not tell which policy, or why. Renaming policies in the language of the outcome, for example “restart stuck worker after repeated health-check failures”, turned the audit trail into something a person could read over coffee. A small naming change, a big drop in confusion.
7. A weekly ten-minute review slot, not a monthly meeting
The team lead assumed monthly reviews of automation behaviour were enough. The engineers said they forgot half the odd cases by then. A short weekly slot, with a shared note listing three surprising events, replaced the big meeting. Notes got shorter, memories got fresher, and the slot was easy to protect because it was easy to defend.
The weekly note also became a quiet archive. When a strange event repeated months later, someone searched the notes, found the earlier entry, and saved an hour of detective work. Nobody had planned the archive; it simply grew out of a habit small enough to keep.
8. A way to say “this suggestion was unhelpful” in one tap
Finally, someone asked for feedback on the platform’s recommendations that cost no more effort than a thumbs-down. Feedback had been possible before, but only through a form nobody opened. Once a single tap existed, the team received more signal in a week than in the previous quarter. The tap also reminded everyone that recommendations are suggestions to be judged, not instructions to be obeyed.
What the exercise actually taught them
Looking back, the eight fixes shared a shape. Each answer came from someone doing a routine job, each fix was cheap, and each removed a small recurring irritation that had been quietly eating attention. None of them required new AIOps capability. Several simply reflected knowledge the team already had but never pooled.
If you try this yourself, a few habits help. Ask the question out loud rather than by survey, because people soften written answers. Ask the quietest person first. Write down every answer, including the ones that sound trivial, and pick the cheapest one to try that same week. Then, and this part matters, tell the team which suggestion you tried and what happened, so the next round of answers is honest.
A side note on where teams look for plain-language reading beyond their own field. When an engineer wanted to understand how a different industry explains complicated topics to ordinary readers, she browsed a plain-language urology information site that also lists hospitals and doctors without ranking them, and which states that it does not offer medical services. She was not after medical content; she was studying how a site organises its sections so that a nervous newcomer can find a starting point. The takeaway for dashboards was simple: a clear front door matters more than cleverness behind it.
That is the whole method. Ask what would make this easier, listen without defending the current setup, and pick one small change. Teams that do this once usually do it again, because the first round quietly proves that the people closest to the work already know where the friction lives.