Every June, the first monsoon showers reach Marine Drive and, right on schedule, the traffic to city guides triples. That is the season our fictional team picked to chase NoOps — the ambition of buying and automating so much of the operational side that no separate operations function survives. Call them the crew behind a Mumbai events-and-guides platform: two developers, a content lead, and a founder who does the accounts. They are invented, but every number in their story is drawn from how these arrangements actually behave, because the point of the story is not the drama. It is the arithmetic the team skipped on the whiteboard in week one and had to learn by August.
The NoOps Pitch That Opened the Quarter
It began, as these things do, with a paragraph in a vendor proposal: fully managed hosting, managed queues, a support tier with a name like Platinum. The founder read “managed” as “over” — no rota, no 3 a.m. messages, no retainer for an agency. The developers were more cautious, having carried pagers before, but even they framed the goal as deletion. When Netflix’s engineers began owning their own deploys and their own alerts around 2012, the word for it was the same, and the engineer John Allspaw observed then that the operations had merely been rearranged rather than retired. Twelve years of debate later, that observation remains the sharpest sentence ever written about the subject, and the team had not read it.
Counting the Hours the Platform Actually Needed
The first honest exercise was an inventory. Everything the platform demanded on a recurring basis went into one sheet: dependency upgrades, certificate renewals, cache tuning before festival weekends, the monthly export of the events database, access reviews when freelancers rolled off, and the unpredictable pages — a queue that wedges, a search index that stops updating, a third-party ticketing API that starts returning errors during a sold-out concert week.
Tallied over three months, the load came to roughly thirty-nine hours a month, about a third of one engineer. The founder circled that number in green: a third of an engineer, easily absorbed. What nobody had worked out yet was the difference between the work and the coverage — the staffing needed to guarantee that whichever hour the fault arrives, somebody is awake, briefed, and authorised to act.
The Line Items Nobody Had Budgeted
Between the inventory and the coverage sum sat a tier of costs that never appears on a procurement page. Retaining judgement still needs a person: when the queue wedges during a sold-out week, somebody must decide whether to shed load, and “the vendor’s support tier” is a phone queue, not a decision. Access review for freelancers rolling off a project cannot be delegated to a platform, because the platform does not know which freelancer owes which draft. And seasonal spikes cut both ways — the same monsoon that triples readership triples the cost of being wrong, since a stalled guide during peak planning hours burns the year’s best traffic. None of these items is ops in the job-title sense. All of them are ops in the sense that decides whether readers come back, which is the sense that pays.
The Arithmetic of Covering a Clock That Never Stops
Coverage is where the NoOps case usually breaks, and it broke here. Keeping one responder reachable around the clock, with a ceiling of twenty-five percent of working hours spent on call — the proportion Google treats as the most one site can ask of a person — works out to a roster of about eight at a single location. Thirty-nine hours of actual toil had quietly been priced against a hundred and ninety-odd person-hours of standing cover every month, a ratio of one to five. Nobody proposed hiring eight people. Instead the burden would have spread across three, which meant each of them absorbing pages at unsupportable rates. Practitioners watch a second number for exactly this reason: pages per shift, with two in a twelve-hour window treated as the sustainable maximum. Their August — two outages during Ganesh Chaturthi week, both resolved by the same exhausted developer — would have breached that line four times over.
What the Team Changed Instead
The September rewrite of the plan kept every managed service and dropped the fantasy. Four moves did the work:
- Every recurring chore from the inventory got an owner and, where possible, a script; the sum of duty hours plus pages became a tracked figure, reviewed monthly.
- Business hours were declared the official response window for everything short of a full outage, which converted most pages into tickets.
- The two components with the worst failure history moved behind queues and caches so that a vendor’s bad minute degraded the experience instead of ending it.
- One weekend drill per quarter rehearsed the restore of the events database, because the team’s real existential risk was data, not uptime.
By December the toil figure sat near twenty hours a month, one person could hold the on-call phone for a whole week without breaching the page budget, and the founder had stopped describing the platform as self-running. It was not zero-ops. It was small, explicit, and survivable — which is what the original ambition, honestly graded, actually awards.
Where the NoOps Idea Still Earned Its Keep
The team’s verdict was neither the vendor’s nor the cynic’s. For a property with spiky, seasonal load — and nothing is spikier than city guides during a monsoon or a festival fortnight — renting the machinery and pressing it into shape beats running your own. The label deserves credit for forcing the question: if operations could be bought, which parts refused to be? Their answer matches what practitioners grade on maturity ladders, where the top rung is not a deleted function but a deliberately minimal one, and the bottom rung is a slide deck. Readers weighing the same trade will find the Ops Atlas entry on NoOps a fair-minded summary of both meanings the term carries and the metrics that separate them.
Plan the content side of a monsoon traffic spike with our guide to putting together a complete Mumbai sightseeing itinerary, and keep the seasonal context handy through our city guide to heritage, food and neighbourhoods. For deciding how much runway your editorial calendar really needs before the rains arrive, see how many days visitors should budget for the city — the same staging logic that smooths a tourist’s week smooths a release calendar.
The Takeaway From One Monsoon
Thirty-nine hours of work hid a hundred and ninety hours of coverage, and a proposal promising the end of operations turned out to promise only the end of a job title. Teams that count the duty hours, cap the pager load, and rehearse the restores finish with a better bargain than the slogan offered: operations sized to the actual load, owned on paper, and quiet for weeks at a stretch. That is the version worth wanting, whatever you choose to call it. The crew from our story kept the word on their roadmap anyway — not as a destination they expected to reach, but as a nudge to re-run the inventory every quarter and ask which chore had become a script since the last read. Run that loop long enough and the function shrinks on its own, which is the only way it has ever genuinely shrunk.
