A trader who cannot size a position has no business taking one, and the same rule governs the infrastructure fashion of the season: NoOps, the plan to automate and outsource operational work until no dedicated ops team remains. The idea has circulated since 2011, and most teams that burned money on it did so for the oldest reason in the markets — they took the position before pricing it. What follows is the sizing worksheet: six steps that turn a slogan into numbers you can defend, whether the position you are considering is “hire nobody”, “disband the current team”, or the quieter middle path most firms should actually hold.

Step 1: Inventory the Duties NoOps Would Take Over

Open a spreadsheet and list every recurring operational task the business performs: patches, capacity reviews, access requests, certificate renewals, vendor calls, the 2 a.m. restarts nobody admits to. Attach an estimate of hours per month to each line, and a second column for pages — genuine alerts that woke a human. The discipline here mirrors a trade journal: no line may say “miscellaneous”. Sum the hours and the pages, and you hold the number that anchors everything else: residual work, the operations still physically performed by your people after every automation and contract already in place. A firm that cannot produce this figure is not evaluating NoOps; it is guessing with headcount, which is how the expensive guesses happen. Two practical notes on building the inventory. Pull the ticket and incident exports rather than interviewing from memory, because recollection rounds down with remarkable consistency. And add a column for who performs each duty whenever the named person is unavailable — the “nobody” entries in that column are your first findings, not your last.

Step 2: Convert Work into Coverage, Because They Are Not the Same

Here is where most back-of-envelope plans fail, and the error is mathematical, not technical. Consider a service whose genuine workload totals thirty-nine hours in a month — a third of a single full-time person, and every instinct says “tiny”. Now attempt twenty-four-hour coverage: humans sleep, fall ill, take leave and hit statutory limits, so continuous staffing of that same monthly load demands about 195 person-hours of scheduled seats. Field guidance from Google puts the single-site minimum for such a rota near eight people, and even then each engineer’s exposure is capped at two pages per twelve-hour shift to keep the rotation humane. The thirty-nine hours may collapse under automation. The 195 rarely do, because the rota exists for the hours automation cannot predict, not the ones it can. Any proposal that removes the department while keeping the around-the-clock obligation has simply moved the second number onto contractors’ invoices. The ratio itself deserves a cell on the sheet: divide required coverage by raw workload, and the resulting multiple — here five to one — is the price of the clock, not the price of the work.

Step 3: Test the Ceiling Your Dependencies Impose

Two multiplications belong in every honest worksheet. The first is availability. Your service cannot outperform the combination of everything it requires, so multiply the promised figures: take one component offering 99.95 percent, a second at 99.95, a third at 99.99, and the product lands near 99.89 — a ceiling no contract of yours can lift, because the weakest necessary link gets a vote. Copying a vendor’s headline number into your own customer-facing objective is the operations equivalent of assuming yesterday’s fill on today’s order. The second multiplication is headroom. For each managed component, compute one minus peak-over-quota and log it monthly; the ratio tells you how much of the wall remains before the platform refuses your next burst. Trading desks call this knowing your buying power. The cloud equivalent is quieter and the penalties for ignoring it are identical: an unexpected rejection at the worst possible moment.

Step 4: Audit the Contingencies as If the Auditors Were Coming

Two dates belong on the worksheet, and both are checked, not asserted. The first is restore readiness: a recovery counts only when it has been executed — timed, logged, witnessed — within the last ninety days; anything older is a claim, not a capability. The second is runtime currency: how many days of margin separate your runtime from the provider’s deprecation clock, with 180 days as the conventional floor below which upgrade work becomes urgent rather than scheduled. Neither metric appears on any invoice, which is precisely why they appear here. Managed platforms fail differently than owned ones — rarely, suddenly, and usually at the provider’s convenience rather than yours — and the contingencies that blunt those failures are exactly the ones firms stop rehearsing the moment a department is renamed. Assign each date an owner by name, and put both on the same quarterly review as the coverage rota; a contingency nobody schedules is a contingency nobody believes.

Step 5: Price the Whole Position, People Included

Unit economics, then, computed the way the field increasingly computes them: total unit cost per million requests, people hours included. A worked example that circulates among practitioners put a service at 177.6 for a million requests, with roughly sixty-six percent of that figure being human time — coordination, investigation, the unbillable glue between systems. The proportion, more than the absolute number, is the finding. A platform bill can shrink to almost nothing while the payroll quietly expands to absorb the residue, and firms that track only the invoice will book the quarter as a triumph. Price the position by all-in cost or do not price it at all; partial accounting is slippage, and slippage compounds. A serviceable approximation before any formal study: value the Step 1 hours at loaded monthly cost, then add the share of engineering time your last survey attributed to operational interruptions. Decision-makers accept approximations with visible assumptions far more readily than invoices with invisible ones.

Step 6: Price the Exit Before You Are In

Every infrastructure position is a position, with an entry, a holding cost and — for those who plan — an exit. Write down what departure from each critical vendor would cost in engineering weeks and in rupees, refresh the estimate quarterly, and file it beside the contracts. The exhibit every worksheet should cite is Google’s IoT Core, a managed service retired in 2023, which handed its users an exit on Google’s timetable at Google’s prices. Regulation is shifting part of that burden: under the EU Data Act, switching and egress charges become unenforceable for cloud customers from early 2027, which trims the invoice of leaving. The engineering weeks remain untouched by any statute, and they are the real cost. An unpriced exit is not neutral; it is a short option you did not know you had sold, and options sold unknowingly are always exercised at the worst moment.

Reading the NoOps Worksheet Like a Trader

Six numbers now sit where the slogan used to be:

• residual work, in hours and in pages received;
• coverage, in person-hours the rota truly demands;
• a dependency ceiling, computed by multiplying the SLAs;
• two contingency dates — the last rehearsed restore and the runtime-currency margin;
• an all-in unit cost, people time included;
• a priced, dated exit from every critical vendor.

The decision rule writes itself. If residual work is already small, the spiky workload is standard, and the exit is priced, the managed-heavy, ops-light position is probably sound — take it deliberately, with a review date. If the coverage requirement is large, the workload steady and heavy, or the exit unpriced, the honest call is the middle path: automate what the numbers say, keep a small accountable core, and review the position next quarter like any other holding. For the reference behind every figure and term above — origins, cases, counter-arguments — the Ops Atlas publishes a NoOps entry worth studying before you commit capital or headcount.

Two earlier pieces on this site slot naturally alongside the worksheet: our tour of confidential computing platforms compared feature by feature shows how quickly vendor choices stack into dependencies, and our breakdown of spending, subscriptions, startup costs and fraud risks is the same all-in-costing discipline applied to your personal balance sheet. The worksheet changes; the rule — price the position before you take it — never does.