Show Your Rough AIOps Design Early: A Plain-Language Story

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.


Get Started Now!

A small operations team once spent three weeks building what they were sure would be a neat piece of AIOps: a model that watched service metrics and raised a flag when something drifted from normal. They worked quietly, tuned it until the charts looked beautiful, and then presented it to the on-call rota. Within ten minutes the room had found the problem. The model treated every Monday morning traffic surge as an anomaly, and the people who lived through those mornings had known that for years. Nobody had asked them. This is the plain-language story of why showing an unfinished AIOps design early beats polishing it alone.

What an unfinished AIOps design actually looks like

Let us strip the jargon first. AIOps, in everyday terms, means using statistical models and automation to help people run IT systems: spotting odd behaviour, grouping related alerts, suggesting where a fault may be. A rough draft of such work is not a finished product with a dashboard. It might be a one-page sketch of which signals the model will read, a short list of the situations it should react to, and a few sample outputs pasted into a document.

That is the level at which feedback is cheapest. A page of signals can be crossed out in seconds. A trained model wired into paging, with a month of tuning behind it, resists criticism because everyone has already paid for it. The whole idea of this article is simple: the earlier the draft leaves your desk, the less it costs to change course.

Why polishing alone hides the big flaws

When you work by yourself, you check the things you already know to check. The metrics line up, the code runs, the graph is smooth. What you cannot see are the assumptions that live in other people’s heads. In the story above, the missing fact was a weekly rhythm. In another team it might be a nightly batch job, a planned failover test or a vendor maintenance window that quietly changes what normal means.

Private polishing also builds attachment. After days of careful work you start defending choices instead of questioning them. A colleague who sees a rough sketch has no such attachment and, more importantly, you have none either. Nobody feels bad about a sketch being wrong. Everyone feels awkward about a nearly finished system being wrong.

There is a quieter effect too. Unfinished work invites contribution, while finished work invites approval or rejection. A reviewer handed a polished model usually says something like “looks good” because the effort is visible and the flaws are buried. A reviewer handed a rough draft tends to say “have you thought about the month-end close?” That second sentence is the one that saves a quarter.

Name the kind of feedback you want

Handing over the draft covers one half. The remaining half is saying what you hope the reader will do with it. Without that, reviewers wander into whatever catches their eye: variable names, chart colours, the font in your document. Put a brief line above the draft saying where your eyes need help, and attention goes to the right place. Here are requests that work well for an AIOps sketch:

  • “Which of these signals do you not trust?” People who run the systems know which metrics lie during deploys or restarts.
  • “What would you want to see before you believed an alert like this?” This reveals the evidence an operator needs, which shapes the output format.
  • “What normal situation will this misread?” It invites the reviewer to hunt for false alarms from their own experience.
  • “What is missing entirely?” A broad question, best saved for a reviewer who knows the estate well.

Equally useful is saying what you are not asking about. If the thresholds are placeholders, say so. If the dashboard styling is temporary, say that too. Reviewers relax when they know which parts are deliberately rough and can spend their attention on the parts that are not.

A worked example, step by step

Take the team from our opening story and rewind. Suppose that before touching any model they had written a half-page note. Stage one lists the services in scope and the five or six signals that would feed the model. Stage two lists the behaviours it should flag: sudden latency jumps, error rates rising together across related services, disk growth that outpaces its usual curve. Stage three is a table of three or four historical incidents with a guess at what the model would have said about each.

That note takes an afternoon. They send it to two people: one on-call engineer and one person from a neighbouring team who consumes their services. The message reads, in effect, “this is rough, I have built nothing yet, tell me which signals mislead and which incident I have forgotten.” The replies come back the same day. The engineer points out the Monday surge. The neighbour mentions a data import that doubles load every month end. Two lines are added to the draft and the design improves before a single line of training code exists.

Compare the cost. Three weeks of work thrown into reverse, or one afternoon and two short conversations. The arithmetic is not subtle, yet teams repeat the expensive path because showing something unfinished feels exposed.

One more habit is worth borrowing from this story: keep the rejected ideas. When a reviewer talks you out of a signal or a rule, jot the reason beside it in the draft instead of deleting the line. Months later, when somebody proposes the same signal again, the note explains why it was dropped, and the conversation starts from evidence rather than from memory.

Handling the awkward feeling

The discomfort is real, and it helps to understand where it comes from. Most of us learned that work is judged when it is complete. Handing over a half-built idea feels like handing over a half-written exam. The reframing that helps is to treat the draft as a question rather than an answer. “Here is how I am thinking of doing this, where does it break?” is a question, and people enjoy answering questions about systems they know.

Some practical ways to make it easier:

  1. Label the draft clearly as a draft, in the title and the first line.
  2. Pick one or two reviewers rather than a crowd. Twelve opinions on a sketch produce noise; two focused ones produce direction.
  3. Set a short window, say a day or two, so the draft does not linger.
  4. Thank people specifically, naming what their comment changed. They will be more willing next time.
  5. Agree with yourself beforehand how you will react when a comment cuts against the plan. Often the answer is to run a tiny test rather than argue.

When to stop asking and start building

Early feedback is not endless consultation. There is a point where the big questions have been answered and the remaining uncertainty can only be resolved by building. A reasonable rule: once reviewers stop surfacing new surprises and start repeating each other, you have enough. Take the sketch forward, build the smallest working version, and return to the same reviewers with something running. The second round is easier because they already understand the context and have seen their earlier points reflected.

For a view on the choices behind AIOps tooling, this piece on how the site supports IT decision-making with machine help gives one view of the topic. On the noise side of the same discipline, an article about taming event floods using smarter tooling is a useful companion, and the predictive angle appears in a write-up on forecasting maintenance needs in IT systems.

The same habit outside IT

Nothing about this habit belongs to IT alone. People who plan community events know it well. Before committing to a hall, a date and a programme, the sensible organiser circulates a loose outline to a couple of regulars and asks what they would change. If you are curious how a regional events site presents that world, KeralaOrbit, an events website for Kerala built around browsable listings, venue pages and a community forum, shows how that world is laid out for visitors.

Take-away for your next AIOps draft

Before the next model, rule set or correlation recipe grows beyond a page, stop and send the page out. Say it is rough. Say which question you want answered. Wait a day. Then fold the answers in and decide what to build. The team in our opening story eventually did exactly this, and their second model shipped with a calendar of known surges that the on-call rota had helped write. It was less impressive on a chart than the first. It was far more useful at three in the morning, which is where AIOps earns its keep.

Related Posts

A Teardown Checklist for a Finished Drift-Monitoring Project

Open a small, finished model-monitoring repository and audit how it was built, from baseline and thresholds to tests and alert wording, then borrow its structure.

Read More

Eight Runbook Phrases Worth Rewriting in Plain Words

Eight bits of insider shorthand common in incident runbooks, and the everyday wording that lets a tired on-call reader act on each one without asking anyone.

Read More

AIOps With a Tiny Toolbox: Three Triage Builds Compared

A storytelling comparison of three incident triage builds made under one strict limit: a single dependency, a small codebase and one shared set of past incidents.

Read More

Spotting Recurring Incident Patterns: An AIOps Explainer

One alert is noise, three of the same kind is a habit. A plain-language look at how AIOps tooling and a simple tally reveal the recurring causes behind repeat incidents.

Read More

Why AIOps Teams Are Slowing Rollouts Down on Purpose

A friendly look at a quiet trend in AIOps: teams letting each automated change prove itself slowly, because the fast version keeps ending in a rollback.

Read More

How to Keep a Done List for AIOps Changes, Step by Step

A six-step method for recording AIOps changes that have shipped and been verified, so your team stops repeating work and can judge progress from evidence.

Read More
0 0 votes
Article Rating
Subscribe
Notify of
guest
0 Comments
Oldest
Newest Most Voted
0
Would love your thoughts, please comment.x
()
x