A hospital’s back-office systems, from scheduling portals to document archives, increasingly run on rented cloud capacity, and the monthly invoice for that capacity is rarely audited with the rigour applied to supply purchases. FinOps supplies the missing method. This checklist is meant for an administrator or IT manager who wants to inspect the state of cost governance in one afternoon. It is non-clinical and informational; it concerns the invoice and the systems behind it, not patient care. Work through the sections in order and note a yes, no or unclear against each line.
Section A: Is the data available at all?
What you can conclude depends on which records you can open, so start with the plumbing.
- Is detailed billing export switched on for every account? At least one major provider fills in only two months of history once you enable the export, so anything older than that is gone unless you switched it on early.
- If your provider uses cost allocation tags, were they activated, not merely applied? The Atlas notes that on AWS a new tag key needs about 24 hours before it is listed and about 24 more before it can be switched on, and then twelve months of history can be backfilled.
- Is there a standard layout for the data? The open FOCUS specification, currently at version 1.4 according to the Atlas, normalises column names across suppliers.
- Can someone distinguish cash cost from amortised cost? The specification keeps BilledCost (cash) apart from EffectiveCost (accrual-style).
Section B: Do the numbers reconcile?
- Does the total of allocated lines match the invoice you actually pay? The Atlas’s worked lab ends with a tidy reconciliation in which the billed figure of 11,950 equals the payable figure; yours should end the same way, or you should know why not.
- Is there a written rule for shared charges such as support plans, taxes and common platforms, which tags cannot assign?
- What share of cost remains genuinely unowned? Record the percentage, even if it is unflattering. The Atlas’s lab left 5.1 percent unowned after a sharing rule, which is a useful yardstick for scale rather than a target.
- Are credits, taxes and adjustments visible as separate charge categories instead of blended into usage?
Section C: Who watches, and how often?
Cadence is where audits most often find a gap. The Atlas proposes a rhythm worth testing your practice against: a daily glance for anomalies, a weekly cost review per team, a monthly session covering showback, closing and the forecast, a quarterly look at commitments and key indicators, and a yearly sweep of budgets and renewals.
- Is there a named person responsible for the daily anomaly check?
- Do you know the latency of your provider’s anomaly alerts? The Atlas reports that one provider’s built-in detector may trail events by as much as 24 hours and wants a ten-day warm-up of history.
- Does each team hold a standing weekly slot, even a short one?
- Are decisions from those meetings recorded with an owner and a date?
Section D: Commitments and discounts
- When a one- or three-year purchase was last made, was it sized to the lowest hourly level of steady usage, not to a daily average? The Atlas cautions that day-level totals flatter what a commitment can absorb, since eligibility is tested every single hour and an hour’s unspent portion simply evaporates.
- Was planned change, such as a system retirement, subtracted before buying?
- Is utilisation of the commitment checked daily during the first weeks? The Atlas gives a simple example in which 9.80 used out of a 10.00 hourly commitment equals 98 percent.
- Is the headline metric effective savings rate, not coverage alone?
- Was any resource-based commitment sized before an optimisation exercise and then left unchanged? The result is a bill for capacity that was trimmed away months ago.
Section E: Rightsizing and waste controls
- Do resizing decisions rely on utilisation data that includes peaks, not averages?
- Is there a written guardrail so that any slowdown in an important service triggers a rollback?
- Are changes made one dimension at a time and observed across a full business cycle?
- If a container platform is used, does your allocation model know that the OpenCost specification charges the greater of requested and used capacity, with the remainder counted as idle?
- Is there a policy-as-code tool or equivalent that marks resources first and acts later?
Section F: Culture and decision rights
- Does the organisation treat the practice as a cost-cutting drive, or as value management? Spending more can be the right answer.
- Has a central team appointed itself the cost police, taking over chores that engineers ought to own? The Atlas files that under things to avoid.
- Has anyone assumed that buying a tool completed the job?
How to run the afternoon
Book two hours and invite three roles: the administrator of the cloud accounts, a finance contact who receives the invoice, and a manager accountable for at least one application. Share the list in advance. Ask the account administrator to bring a screen they can navigate and the finance contact to bring last month’s invoice, because the reconciliation questions in Section B are impossible to answer from memory.
Agree before you start that the purpose is to find gaps, not apportion blame. Audits of this kind tend to go sour when a team feels the list is a tool for fault-finding. Read each line aloud, record the answer and the evidence behind it, and move on without debating remedies. Remedies belong in the closing ten minutes.
Evidence to attach to each answer
A bare yes is weak. For each line, write down where the proof lives. A screenshot of the export settings covers the data questions in Section A. A saved reconciliation sheet covers Section B. Calendar invitations, or their absence, cover Section C. The purchase record plus the hourly usage chart covers Section D. For Section E, the change log of the last three resizing decisions will show whether guardrails existed. Where no evidence exists, mark the line unclear, which is honest and keeps the score from flattering you.
Findings you are likely to meet
Several patterns recur in audits like this. Exports were turned on late, so the history starts mid-year. Tag coverage looks respectable until someone checks which tags were activated for reporting. Forecasts are built from last year’s line and ignore the launch scheduled for next quarter. Commitments were bought against yesterday’s footprint, then a clean-up project shrank the footprint and left the purchase oversized. And almost every group discovers that the weekly review exists on paper but quietly lapsed.
One more Atlas theme: size alone says little about a bill, and the useful figure is what each unit of work cost. An audit that notes the total rose and stops there has not done its job; add a line for the unit measure your organisation uses, for example cost per appointment request handled or per report generated, and record its direction across the last three months.
Scoring your answers
Count the no and unclear answers per section. A section with more than half unresolved is the place to spend your next quarter’s attention. The Atlas grades maturity per capability as crawl, walk or run, instead of giving the whole practice one grade, so do the same: you may be at walk on reporting and crawl on forecasting, and that is normal.
If the Atlas definition is new to you, its entry on FinOps is the reference this checklist leans on throughout. For a perspective closer to home, our article on the practical money questions families face after a cancer diagnosis is a reminder that clear cost information matters to the people the systems ultimately serve, and a roadmap for cloud cost optimisation offers a complementary technical angle.
After the audit
Choose three failed items, assign each to a named owner, and set a date no more than a month away. Repeat the audit after the next quarter closes. Keep the filled-in sheet; the difference between two audits is the clearest evidence of progress you can show a finance committee.
