A monthly cloud cost review settles five things in 45 to 60 minutes: whether last month's numbers reconcile, why costs moved, who owns them, where budget and forecast stand, and which few actions will be finished before the next meeting. This guide gives you the preparation, an agenda, a variance walk across price, usage, mix, exchange rate and credits, and templates for the decision log and for what finance and management receive.

You don't need a FinOps team to run it. Finance owns the invoice, the cost basis and the currency method; engineering explains what changed technically and makes the changes; management sets priorities and decides which variances it can live with. One person can hold several of these roles. How they map onto the current framework is covered in FinOps Framework™ 2026: what small teams should actually change.

What the review produces

Decide the output up front: a short, versioned monthly report that covers:

  • closed period, data cut-off and version;
  • cost basis, currency and exchange-rate method;
  • variance to last month, budget and forecast;
  • material cost drivers with evidence or an explicit open status;
  • allocated and unallocated cost;
  • commitments, credits and the state of savings work;
  • decisions, accepted risks and actions with owners and dates.

A dashboard is fine for the analysis. It makes a poor record, because filters, data freshness and explanations all change after the meeting.

Preparation: cut-off, exports and owners

Set the data cut-off from each provider's rules

Cost data keeps changing after it first arrives, and each provider documents this differently:

  • AWS: Data Exports refreshes CUR 2.0 at least once a day. After a billing period ends, AWS may update that period's delivery within the first two weeks. Credits and refunds can be added after the bill has been finalised.
  • Azure: For EA and MCA, costs usually reach Cost Management within 8 to 24 hours. Azure typically closes the billing period up to 72 hours after it ends, and usage charges can change until the fifth day after. Until the invoice is issued every cost is an estimate, and Cost Management excludes taxes, support and credits.
  • Google Cloud: The invoice month (invoice.month) can differ from the usage month. Usage at the very end of a month can land on the next month's invoice.

The working rule follows from that: hold the review only after the fifth calendar day of the following month, and version the report, because AWS can still change the previous month afterwards. Agree each of these in writing:

FieldRule to define
Data cut-offWhen the period counts as closed for the review
Late chargesWhether later corrections go into next month or into a new version
Source of truthExport or invoice, and which wins when they differ
Cost typeActual, amortised or effective, chosen for the question
CurrencyReporting currency, rate source, effective date and rounding
CompletenessWhich accounts, subscriptions and projects are in or missing

Pick one cost definition for management trends and stick with it. Actual cost shows how the invoice moves; amortised or effective cost works better when teams answer for their consumption. A zero where cost is simply missing is worse than a blank marked "pending". How late invoices and accruals are handled at month-end is covered in cloud costs at month-end.

The exports and fields you need

ProviderSourceMonth attributionShow separately
AWSCUR 2.0 via Data ExportsPartition BILLING_PERIOD=YYYY-MMline_item_line_item_type: Usage, Credit, Refund, Tax, Fee, SavingsPlanCoveredUsage, SavingsPlanNegation
AzureCost details export, actual and amortisedBillingPeriodStartDateChargeType (Usage, Purchase, Refund), PricingModel (OnDemand, Reservation, Spot, SavingsPlan)
Google CloudBigQuery standard usage cost exportinvoice.monthcost_type (regular, tax, adjustment, rounding error) and the credits field

Which AWS export feeds the baseline is its own decision: AWS CUR 2.0 or FOCUS. For Google Cloud, set up the billing export for cost reporting covers the right export type.

The pre-read pack

Send a pack before the meeting so the time together goes on decisions:

  1. Data cut-off, version, and any missing accounts, subscriptions or projects.
  2. Export total per provider against the invoice or billing total, with tax, support, credits, refunds and Marketplace as separate lines.
  3. The cost bridge from last month to this month.
  4. Anomalies in the period, with class and status.
  5. Allocation coverage and the largest unallocated items.
  6. Budget, forecast and actual on the same cost basis and in the same currency.
  7. Commitments and credits that expire in the coming months.
  8. The action register with each item's current status.

A 45 to 60 minute agenda

BlockMinutesLeadOutputFinOps capability
Data status and reconciliation5FinanceBaseline closed or gap namedData Ingestion
Cost bridge15Engineering with financeDrivers confirmed, probable or unknownReporting & Analytics
Anomalies5–10EngineeringClass and owner per anomalyAnomaly Management
Allocation5FinanceCoverage and an owner for the residualAllocation
Budget and forecast10Finance with budget ownersA decision per varianceBudgeting, Forecasting
Commitments and credits5EngineeringRenew, resize or let expireRate Optimization
Savings pipeline5EngineeringConfirmed status changesUsage Optimization
Decisions and close5ManagementDecision log, sign-off, next dateExecutive Strategy Alignment

The minutes are a suggestion. With no new anomalies or expiring commitments, those blocks shrink to a confirmation.

The variance walk: price, usage, mix, FX and credits

"Compute went up" tells nobody what to do. Split the change from the last closed month to this one into effects that add up. Per SKU, with quantity q, effective unit price p in pricing currency and exchange rate k, index 0 for last month and 1 for this month:

Usage effect   = (q1 - q0) × p0 × k0
Price effect   = q1 × (p1 - p0) × k0
FX effect      = q1 × p1 × (k1 - k0)
Sum of all     = q1 × p1 × k1 - q0 × p0 × k0

Some movements need their own lines:

  • Mix: shifts between SKUs, regions or instance types with different unit prices. Run the walk at SKU level; the difference from the same walk at service level is the mix effect.
  • New or removed: SKUs with no prior-month value form their own block, not a usage effect.
  • Credits, refunds and one-off purchases: separate lines, split by line type.
  • Scope or allocation change: new accounts, moved subscriptions, changed rules.
  • Unexplained: the residual, shown explicitly.

For AWS, one query against CUR 2.0 in Athena gives you the inputs. Grouping by line type keeps Savings Plans lines, credits, refunds and tax apart:

SELECT
  billing_period,
  line_item_product_code,
  line_item_usage_type,
  line_item_line_item_type,
  SUM(line_item_usage_amount)  AS usage_amount,
  SUM(line_item_unblended_cost) AS unblended_cost
FROM cur2
WHERE billing_period IN ('{previous_month}', '{month}')
GROUP BY 1, 2, 3, 4
ORDER BY unblended_cost DESC;

Only call a cause confirmed when cost and usage data, a documented deployment or a contract change backs it up. Something happening in the same week is not proof.

Driver statusEvidenceOwnerTreatment
ConfirmedQuery, deployment or contract changeTeam able to actExplain in the report, add an action if needed
ProbableSome evidence, not completeInvestigation ownerMark open and set a date
UnknownNo defensible causeTechnical ownerDon't guess; narrow the scope and keep looking

Exchange-rate effects appear even when the invoice is already in euros. Paying and reporting AWS and Azure costs in euros shows how to keep them apart from consumption.

Anomaly triage

The native detectors run with delays and different scopes:

  • AWS Cost Anomaly Detection runs roughly three times a day on net unblended cost once billing data is processed. Because it reads Cost Explorer data, detection can take up to 24 hours after the usage. It lists possible root causes ranked by cost impact across service, account, Region and usage type. It does not monitor third-party Marketplace products, except third-party models in Amazon Bedrock; AWS Budgets covers those.
  • Azure evaluates anomalies daily per subscription in Cost analysis smart views, against a forecast built from the last 60 days, 36 hours after the end of the day (UTC). Anomaly alert rules exist only at subscription scope, with a limit of five per subscription.
  • Google Cloud shows anomalies on the billing account's Anomalies dashboard, with thresholds set automatically and a root cause analysis panel; billing admins get an email for every anomaly by default.

In the review, detection matters less than classification. Take each anomaly through this sequence:

  1. Data artefact? Late usage, rerating in an open month, a credit in the wrong period, or a newly connected account.
  2. One-off or run rate? A one-off purchase changes the month; a new run rate changes the forecast.
  3. Expected? A planned launch or migration belongs in the cost bridge, not the anomaly list.
  4. Name an owner and record a decision: accept, fix, or investigate by a date.
YesNoOne-offRun rateYesNoAnomaly from the detectorData artefact?Correct the period or the dataOne-off or run rate?Changes the monthChanges the forecastExpected?Into the cost bridgeName an owner and decide by a date
Anomaly triage: a data artefact is corrected, a one-off stays in the month, a run rate changes the forecast, and what was planned belongs in the cost bridge

Allocation coverage and unallocated cost

Don't wait for complete tagging. Start with stable hierarchies: AWS accounts, Azure subscriptions and resource groups, Google Cloud projects, and known service mappings. Tags refine allocation; they do not replace an ownership rule. In Azure, resource tags are not inherited from the resource group unless tag inheritance is turned on, and they never apply retroactively to older cost data.

Show three numbers side by side:

  • directly allocated: by a stable source or rule;
  • rule-allocated: shared cost distributed by a documented method;
  • unallocated: visible, with an owner for resolving it.

Allocation coverage is allocated cost as a share of total cost for the same period and cost basis. Track it as a trend. When a rule changes, record the date and the effect, or a reallocation will look like a real change in cost. Treat a Kubernetes cluster as one owned block at first; understanding Kubernetes cluster cost explains how to read it without per-pod guesses.

Budget and forecast check

Budget, forecast and actual are three different values:

  • Budget: the approved plan or limit.
  • Forecast: the current projection from known usage and explicit assumptions.
  • Actual: cost under the report's defined monthly method.

Compare them only on the same cost basis and in the same currency, or the method itself creates a variance. Every material variance needs a decision: revise the forecast, prioritise technical work, change scope, or accept it. "Keep an eye on it" only counts with a threshold, an owner and a date to look again. Building the plan itself is covered in cloud budget planning for 2027.

Commitments and credits calendar

EventWhat the provider gives youWhat goes into the review
AWS Savings Plans expiryEmail alerts 1, 7, 30 or 60 days before expiry and on the expiry date, to up to 10 recipients; queued purchases for planned buysEnd date, utilisation, renewal decision with approver
Azure reservation expiryAutomatic renewal is on by default at purchase; renewal price shown 30 days before expiry; notification 30 days ahead for terms of a year or more; pay-as-you-go rates after expiryConfirm or switch off every active renewal deliberately
CreditsAWS shows credits as their own lines and can add them after finalisation; Azure Cost Management shows no credits; Google lists them in the credits fieldRemaining balance, expiry date, and the bill after expiry
One-off purchasesAWS: Fee, SavingsPlanUpfrontFee; Azure: ChargeType PurchaseSeparate in the walk, billed and amortised side by side

High coverage is not worth much if it leaves you paying for commitment you don't use. Commitment purchases therefore go into the same action register as technical changes, with term, scope, break-even assumption and approver.

Savings pipeline: estimated, accepted, implemented, verified

A long list of savings ideas looks productive and rarely is. Keep only items whose next step can realistically be done before the next review, and move each through four states:

StatusWhen an item reaches itWhat you record
EstimatedA hypothesis with scope and calculation methodExpected effect as an estimate, never as a saving
AcceptedAn owner with change rights commitsOwner, date, and the approver for commitments
ImplementedThe change is liveDate and evidence such as ticket, commit or purchase
VerifiedBefore/after comparison on a fixed scope and windowMeasured effect, or "inconclusive"

Only verified items count as realised savings in the management report. "Inconclusive" is a valid result and more honest than a forced number.

hypothesis with methodowner commitschange live, with evidencebefore and after measuredeffect not shownEstimatedAcceptedImplementedVerifiedInconclusive
The four states of a savings item; only a verified item counts as a realised saving

Template: decision log

FieldContent
ID and dateRunning number, review month, attendees
SubjectVariance, anomaly, budget line or action
DecisionRevise forecast, start action, change scope, accept, or investigate by a date
Rationale and evidenceQuery, deployment, contract, or explicitly "open"
OwnerPerson or team with change rights
DueNext verifiable milestone
ImpactOn budget, forecast or risk, with method
Status at next reviewDone, deferred with reason, or dropped

The log is the part of the review someone will still need a year from now. Keep it with the versioned report, not in a chat thread.

What finance and management receive

RecipientContentFormat
FinanceReconciled totals per provider and invoice, open differences, cost basis, currency and rate method, inputs for accrualsVersioned report with reconciliation sheet
ManagementTotal cost against budget and forecast, the largest drivers, decisions needed, accepted risksOne page
Engineering teamsCost of their own scope, anomalies, open actionsSaved view or report

Questions about capitalising development work don't belong in the review; take them, with its data, to your adviser: capitalise cloud costs?.

Keeping it to one hour

  • No chart is shown for the first time in the meeting; everything is in the pre-read.
  • Finance sets a materiality threshold. Movements below it go in an appendix.
  • Every block ends with a decision, or an owner with a date. Longer investigations get their own session.
  • The action register is discussed only where status changed, not line by line.
  • Whoever has to decide is in the room; otherwise the item is deferred.

Common mistakes

  • Starting before the cut-off and debating estimates for an open month.
  • Switching between actual and amortised cost across months, so a commitment purchase looks like a cost spike.
  • Losing credits in the total, because Azure Cost Management doesn't show them and nobody adds them from the invoice.
  • Reading an allocation change as a change in a team's spend.
  • Letting an Azure reservation renew automatically after the workload has shrunk.
  • Reporting estimated savings to management as realised.

Cost reports in Costfluent

In Costfluent you save the review view as a cost report, combine reports into dashboards, and schedule a saved report by email daily, weekly or monthly; the email links to the live report. Budgets compare spend with the budget and show where the current period is heading; a cost report can also show a spend forecast; threshold notifications are Planned; cost insights show trends, comparisons and unusual changes, and allocation rules leave unallocated cost visible. Savings opportunities move through estimated, accepted and implemented states; verification against actual cost, for the Custom plan, is Planned. Costs are presented in euros by default, or in GBP or another supported currency, and every converted figure states its ECB exchange-rate method. On the Custom plan, invoice reconciliation compares issued provider invoices with the cost ingested for the same month.

Sources and method

Checked on 2026-10-10. AWS Data Exports delivery and refresh, CUR 2.0 line item columns, understanding your bill, Cost Anomaly Detection and Savings Plans alerts; Microsoft understand Cost Management data, cost details fields, identify anomalies and automatically renew reservations; Google Cloud standard usage cost export and cost anomalies; FinOps Foundation™ Framework domains and capabilities. The agenda, variance-walk formulas and templates are Costfluent recommendations for a lean, repeatable monthly process, not accounting or tax advice.