FinOps forecasting: budget variance and planning
Updated: 4 November 2026Get a quote for my projectFinOps forecasting explains how cloud spending is expected to move and why actuals differ. Keep cloud cost forecasting separate from the approved budget, then connect both to workload demand and business decisions.

On this page
- Finops forecasting: Budget, forecast, estimate: who owns which number
- Clean the history before projecting it
- Build the variance bridge
- Drivers, scenarios and thresholds
- Deliverables to write into the scope
- Questions for the first workshop
- Reading a proposal for this work
- Official sources
- Questions finance teams ask about forecasts
Finops forecasting: Budget, forecast, estimate: who owns which number
In the FinOps Framework, Planning & Estimating, Forecasting and Budgeting are three separate capabilities under the Quantify Business Value domain. Budgeting is approved funding for planned technology activity. Forecasting is a model of anticipated cost built from historical spend and planned changes. Finance usually owns the first, while engineering and product supply the drivers behind the second. A provider that delivers one spreadsheet called «forecast» and lets it double as the budget has merged two decisions that different people take.
- Budget: the approved envelope, fixed at a decision date and changed only through a holdback or a formal revision.
- Forecast: revised on a cadence. Microsoft's FinOps guidance suggests reviewing forecasts monthly or quarterly, which fits most finance closes.
- Estimate: a number for something not yet running, such as a launch or a migration wave. Once approved, it becomes a line in the forecast.
Clean the history before projecting it
AWS and Microsoft both document how their native forecasts behave, and those limits define what a provider has to add on top.
- AWS Cost Explorer produces a forecast with an 80% prediction interval and returns none when it lacks enough data, which is common for accounts with less than one full billing cycle. A member account added to an organization stays out of the forecast until its spending pattern has been analyzed.
- Discounts are included by default in Cost Explorer forecasts. Refunds and other non-recurring discounts need the net unblended view.
- Microsoft's forecasting guidance for Cost Management starts by choosing between billed and amortized cost, then filters out purchases (Charge type = Purchase) and forecasts them separately. New and deleted subscriptions, resource groups and reconfigured resources are set aside and forecast on their own.
- AWS Budgets can alert on actual or forecasted spend, tracks blended, unblended, net unblended, amortized and net amortized costs, refreshes up to three times a day and supports custom-period budgets aligned to a fiscal year or a project.
Ask the provider which of these treatments it applies and in what order. The order changes the result.
Build the variance bridge
A bridge reconciles budget to actual in named steps: volume, unit rate, product mix, scope (a new service or account), timing (commitment payments, credits, late invoices) and corrections (re-allocations). Every step needs a data source. What has no source goes into a residual line that stays labelled as residual, never renamed «optimization».
An illustrative quarter with invented round numbers shows why the views matter; it describes no customer result. A product is budgeted at 400 (thousand euros) and closes at 452. The bridge reads +22 from higher traffic, +18 from a service launched a month earlier than planned, +20 from an upfront commitment payment and -8 from late credits. Finance reading billed cost sees an overrun of 52, or 13%. In the amortized view the 20 is spread over the commitment term, so only the share falling in this quarter remains. The budget owner needs both views and a written rule on which one the budget is expressed in.

Drivers, scenarios and thresholds
A driver-based model ties each cost line to a business quantity such as requests, active tenants, terabytes stored or batch runs. At the Run maturity stage the Framework describes forecast models as a combination of rolling, trend-based and driver-based methods. The minimum useful set is a base case plus one growth and one contraction scenario, each tied to an observable trigger like a signed launch date or a migration wave. Recompute commitment coverage in every scenario: growth may need capacity beyond the commitment, and contraction leaves paid-for capacity unused.
For thresholds, Microsoft's guidance defines two KPIs. The first compares forecast to actual to test the forecasting method, with a target under 12% variance when there are no anomalies. The second tests whether costs were on target, with a 12% to 20% band and under 12% describing an optimized team or workload. Treat these as starting points, and record the reason when you choose others. The same guidance adds two operational measures: the number of unexpected anomalies in the period and the time taken to react to forecast alerts.
Want a forecast bridge built on your own billing data? Get support on my project.
Deliverables to write into the scope
- A model file with visible formulas and the queries that feed it (Cost Explorer or Cost Management exports, BigQuery billing export), so your own analyst can refresh it.
- An assumption register listing each driver, its source, owner, status (signed, planned or optional) and last review date.
- Three views kept side by side: approved budget, previous forecast and latest forecast, with a revision log explaining each change.
- A month-end bridge template with the residual shown.
- Budget alerts per scope on both actual and forecasted spend, with named recipients.
- A handover session with finance and one engineering lead in which the model is refreshed live from new billing data.
Questions for the first workshop
- Finance: which view do you budget (billed, amortized), when is a period final, and what rule applies to currency conversion?
- Product: which launches are contractually committed, and which are still options?
- Engineering: will a migration run two environments in parallel, and for how long?
- Procurement: which obligations cannot shrink when usage falls (commitments, marketplace contracts, support plans)?
- Everyone: who approves a revised assumption, and where is that approval kept?
Reading a proposal for this work
- It asks for your planning calendar and close dates before proposing a method.
- It shows a bridge template or an anonymized example, and says what it cannot explain when volume data is missing.
- It gives the internal effort expected from your finance and engineering staff, alongside the external fee.
- It quotes no forecast accuracy figure before seeing your data. An accuracy claim made in advance has nothing behind it.
- It separates the forecast work from any purchase or configuration change, which remain your decisions.
Official sources
Official documentation used for this page.
- learn.microsoft.com/en-us/cloud-computing/finops/framework/quantify/forecasting
- www.finops.org/framework/capabilities/forecasting/
- www.finops.org/framework/capabilities/budgeting/
- docs.aws.amazon.com/cost-management/latest/userguide/ce-forecast.html
- docs.aws.amazon.com/cost-management/latest/userguide/budgets-managing-costs.html
- docs.aws.amazon.com/wellarchitected/latest/framework/cost01-bp01.html
Questions finance teams ask about forecasts
Should we use only the previous invoice?
It is a fair baseline for stable workloads. It misses launches, migrations, price changes and commitment payments, so use it as the starting line and add drivers on top.
Is every budget variance an optimization failure?
No. Traffic growth, scope changes, the timing of commitment payments and data corrections each explain part of a variance. Label those before judging efficiency.
How often should the forecast be revised?
Microsoft's guidance suggests monthly or quarterly. Align the cadence with your finance close and revise earlier when a launch date or a migration wave moves.
Related workload decisions
Continue the review with bigquery pricing. Each guide addresses a separate decision within the same cloud cost boundary.
Before explaining a variance, agree the approved envelope and the data used to track it. The cloud budgeting export checklist (French) connects those inputs to budget owners and alert response.
Continue your review
- FinOps consulting: cloud cost review and implementation scope
- Request a scoped FinOps quote
- How your request is handled: from enquiry to proposal
- Cloud cost anomaly detection: design the response process
- Rightsizing and idle resources: plan a controlled cost review
- FinOps showback vs chargeback: choose a financial model
Get support on my project
A specialist in the subject reviews your request and sends you a quote.
Get support on my project