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.
Anyone who has spent a week configuring an AIOps platform knows the feeling: a screen full of toggles, each with a perfectly reasonable argument for either setting, and a growing sense that you are burning your best hours on choices nobody will remember next month. That is where sensible defaults earn their keep. A default is simply a setting someone decided in advance, so that the common case needs no debate and any change needs a reason. What follows are the queries that keep surfacing once a team tries the idea on its AIOps stack, answered informally.
Why do defaults matter so much in an AIOps rollout?
Because decision fatigue is real, and operations tooling is full of small decisions. Which services get anomaly detection first? How much history should a model learn from? Who is paged when a score crosses a line? Each one is minor, but dozens of minor choices made on a tired afternoon produce a platform nobody quite understands.
A default flips the burden of proof. Instead of asking “what should this be?” every time, the team asks “is there a reason to differ from the standard?” Usually the answer turns out to be no, and work carries on.
What is a good default for how much history a model learns from?
Pick one training window for every ordinary service and write it down. A window long enough to include a full business cycle, including the quiet weekend and the busy end of the month, is a sensible starting point for most workloads. Services with unusual rhythms, such as a seasonal batch job, can override it, but the override should carry a short note explaining why.
The point is not that the chosen window is perfect. The point is that nobody has to rediscover the question for service number forty.
Should automated actions be on or off by default?
Off, with a clear path to switch them on. A platform that can restart a pod or scale a group by itself is wonderful once trusted, and unsettling before that. A safe out-of-the-box posture is: detect, explain, suggest, and wait for a human to approve. Teams then promote individual actions to automatic one at a time, each with a recorded reason and an owner.
This default also makes onboarding easier to explain. A new engineer can be told in one sentence what the system will and will not do without asking.
What does an opinionated starter configuration look like?
Think of it as a short list of pre-made choices that cover the boring majority:
- One naming pattern for services, so correlation rules can rely on it.
- One severity ladder, with the same wording in every team’s runbooks.
- One standard set of tags (owner, tier, environment) applied at creation time.
- One default retention period for raw events, with a documented exception route.
- One default notification channel per tier, rather than per person.
Nothing on that list is fancy. What counts is sameness across teams, which makes dashboards, reports and handovers behave predictably.
How do defaults relate to the “smart” part of AIOps?
Machine learning features sit on top of the basics. If tags are inconsistent and ownership is unclear, even a strong anomaly model produces findings no one can route. Good defaults are the plumbing that lets the smarter features land somewhere useful. For a broader view of how the discipline hangs together, the site’s article on AI-backed decision support for IT is a reasonable companion read.
How do defaults fit when a team is new to AIOps?
They make the first month calmer. Teams just starting out can lean on an easy walkthrough on adopting AIOps within a tech team for the people side, and use defaults for the settings side. Between the two, the newcomers spend their energy learning what the system is telling them, not wrestling with a hundred switches that nobody has an opinion about yet. When someone does hold a strong view, the override process gives that view a clear, low-drama home.
Who gets to change a default?
Anyone, as long as the change carries a reason and is visible. A lightweight convention works well: the override lives in the service’s own configuration file, with a one-line comment such as “longer window because of monthly billing run”. Reviewers can then see exceptions at a glance, and the standard stays the standard.
It helps to review the list of overrides now and then. If many services override the same setting for the same reason, that is a signal the default itself should move.
Where can I see how other people talk through everyday choices like this?
Outside the engineering world, ordinary online communities are full of people trading “what do you do by default?” answers. One example is a members-run community forum for Kerala with sections for discussions, topics by district, jobs and education; it makes clear it does not speak for any government body. Reading how people there settle small everyday questions is a decent reminder that a good house rule travels well across very different subjects.
What are the common mistakes with defaults?
Three come up repeatedly.
Choosing defaults nobody can find. A standard buried in a wiki page from last year might as well not exist. Keep the list short, and keep it where the configuration is edited.
Treating a default as a law. It marks where people begin, and nobody is forbidden from thinking past it. If a team can never deviate, they will quietly work around the platform instead.
Never revisiting them. Traffic grows, services change shape, and a window or threshold that suited last year’s workload may no longer fit. A quarterly glance is enough.
How do I pick the first defaults without a long workshop?
Start from what your team already does most of the time. Look at the last ten services onboarded, note which settings were identical in nearly all of them, and promote those to defaults. You are not inventing policy; you are writing down habits. Then pick the two or three settings that caused confusion in the past, decide them once, and move on.
If you are unsure about a specific value, choose the more cautious option and note when you plan to reconsider. A cautious default with a review date is far better than an open question that blocks the rollout.
Does this slow down experimentation?
Quite the opposite. When the common settings are settled, experiments can focus on the one variable being tested. A team trying a new correlation rule does not also have to decide retention, tagging and routing; those come pre-chosen. The experiment’s results are cleaner, and rolling back means returning to a known standard rather than a half-remembered configuration.
A short routine to adopt this week
- List the five settings you most often re-decide when onboarding a service.
- For each, write the value you would pick on an ordinary day.
- Record it somewhere the configuration lives, with the owning role and a review date.
- Tell the team that overriding is fine, provided the reason is written beside it.
- After a month, count the overrides and see which ones repeat.
That is the whole practice. It is small on purpose, because a default that takes a meeting to adopt will not survive its first busy week.
Final thought
AIOps promises to take work off people’s plates, and the easiest first step is to take decisions off them. Settle the repeat questions once, make the safe choice the automatic one, write down every exception, and keep your attention for incidents and design problems that really are new. Your future self, returning to a service after six months, will be grateful for a configuration that says what it does and why.