FINOPS · November 16, 2026

FinOps for business: how to read your cloud bill

How to read a cloud bill on Azure and AWS: tagging, non-production environments, purchase model and measuring without hurting performance.

An invoice with a magnifying glass and a bar chart, over a background of wavy lines

What should you look at first in a cloud bill?

A cloud bill is read by answering four questions: what you pay for (services and resources), who generates it (account, team or environment), how much of it you actually use, and which purchasing model you pay under (usage price or commitment). Until you can answer them, any cut is a guess. Here is how to do it on Azure and AWS with their official tools, and what to check so you don't hurt performance or availability.

The FinOps Foundation describes FinOps as an operational framework and cultural practice that maximises the business value of technology and creates financial accountability through collaboration. Its framework organises the work in three phases: Inform, Optimize and Operate.

Each cloud has its own starting point:

Azure: Cost Management lets you analyse costs, set budget and anomaly alerts, and export the data. To optimise existing resources, Microsoft says to start with the cost recommendations in Azure Advisor.
AWS: Cost Explorer lets you view and analyse costs and usage, with up to 13 months of history plus a forecast. The Cost and Usage Reports give the most complete detail: line items by product or resource and by the tags you define, delivered to an S3 bucket you own. AWS now recommends receiving them as CUR 2.0 through Data Exports.

Two cautions before drawing conclusions. The open month is an estimate: on Azure, costs are estimated until the invoice is issued, and AWS reports keep updating during the month until they are finalised. And the tool does not always match the invoice: Azure Cost Management, for instance, does not include taxes or credits (Cost Management data). Always compare against the closed invoice.

How do you attribute cost to a team, product or environment?

With well-separated accounts or subscriptions, plus tags. The FinOps Foundation calls this allocation: splitting cost among those responsible for each component, directly or as shared cost, relying on accounts, tags and metadata. Without it, the bill is a total nobody can explain.

Three details that often surprise people:

On AWS, a tag does not show up in Cost Explorer or in the reports until you activate it as a cost allocation tag, and it can take up to 24 hours to appear (cost allocation tags).
On Azure, the tag has to be on the resource itself: it is not inherited automatically from the resource group, and it is not applied to historical data, only from the moment it exists. Cost Management offers tag inheritance and cost allocation rules to split shared costs; allocation does not change the invoice.
Some resources do not support tags, or do not include them in usage data. That spend is attributed by structure (account or subscription), not by tag.

[Nubyron's view] The first useful indicator is how much of the spend has no owner. You do not need to tag everything to start: a coherent account structure and a few tags (environment, product or team, owner) already make the bill readable. Shared costs, such as a central network, can be budgeted centrally instead of being split, a practice the FinOps Foundation lists as common when starting out. AWS advises against putting sensitive information in tags.

What do non-production environments cost, and how do you control them?

Development, test and pre-production accumulate unused spend when nobody switches them off. In usage optimisation, the FinOps Foundation names pre-production resources as the biggest target for scheduled start and stop, and asks for it to be planned from launch where feasible. It also counts stranded storage volumes and excessive backups or snapshots as waste.

[Nubyron's view] Four questions to start with:

Can you separate non-production cost from production cost? Microsoft notes that many organisations use subscriptions to keep dev/test apart from production.
Which environments stay on outside the hours when they are used?
Who owns each environment, and until when should it exist?
Which disks and snapshots still exist with no resource attached? Before deleting them, confirm they are not copies your retention or recovery policy requires: a deletion has no way back.

[Nubyron's view] Switching off is only safe if the environment can be brought back identically: if it depends on manual steps, nobody dares to turn it off.

Does your purchasing model match your real load?

There are two ways to pay for the same capacity: at usage price, or with a commitment in exchange for a discount (reservations and savings plans on Azure; Reserved Instances and Savings Plans on AWS). The FinOps Foundation stresses one nuance of rate optimisation: a commitment is not a resource, it works like a coupon. If no resources use it, you pay anyway.

So the question is not which commitment to buy, but which part of your load is stable:

Load that runs constantly is the candidate for a commitment.
Variable load needs more care: the FinOps Foundation links spend-based commitments (such as Savings Plans) to variable usage, and Spot only to loads that tolerate interruption, because the provider can reclaim that capacity.
Optimise usage before you commit, without waiting for everything to be optimised: the FinOps Foundation calls that the ideal, but warns against delaying rate decisions for it. If you buy against a load you later reduce, you pay for what you no longer use. It also recommends spreading purchases over the year for flexibility.

Tools help but do not decide: Cost Explorer offers Reserved Instance recommendations, and Azure Advisor's are based on your usage patterns. [Nubyron's view] Check them against what is about to change in the coming months: migrations, projects ending, new environments.

How do you measure without hurting performance or availability?

[Nubyron's view] By measuring before you change, and changing with a written way back. A resource with low utilisation is not always surplus: it can be the replica in another zone, the reserve for a peak, or the node that takes over when another fails.

The FinOps Foundation warns that resizing a resource usually means recreating or reconfiguring it, with possible service disruption, and that it must be coordinated with engineering. It also asks for FinOps data to be accessible, timely and accurate.

How we do it at Nubyron:

Read-only first: a role you create, limited to billing, usage and configuration. We do not ask for keys or passwords.
A representative observation window: month-end closes, campaigns, backups and batch jobs. A daily average hides the peaks.
Separating idle from reserved: check whether the resource is part of high availability, recovery or peak capacity.
One change at a time, in an agreed window, with performance and availability measured before and after, and the way back written down before we start.
Production changes only with your written authorisation. If a saving puts the service at risk, we tell you before proposing it.

What does a cloud bill review deliver?

In Nubyron's review for AWS or Azure you receive:

An executive and technical report.
The potential saving identified, prioritised and with its risk. It is potential: it is not a saving until it is applied and measured.
A 30-day action plan and a call to go through the results.

We work with a read-only role that you create: if you cannot create it, we cannot do the review. We do not modify any resource; if you want the changes applied, they are quoted separately, with your written authorisation and the way back defined. The price is fixed and in writing, after a scoping call, and if there is no saving to find, the report is still delivered and charged.

If you want a broader look (cost, security and architecture), the read-only technical review includes an optional cost module.

Sources (accessed on 16/11/2026)

FinOps Foundation: Framework
FinOps Foundation: Phases
FinOps Foundation: Principles
FinOps Foundation: Allocation
FinOps Foundation: Usage Optimization
FinOps Foundation: Rate Optimization
Microsoft Learn: Overview of Cost Management
Microsoft Learn: Understand Cost Management data
AWS: Analyzing your costs and usage with AWS Cost Explorer
AWS: What are AWS Cost and Usage Reports?
AWS: What is AWS Data Exports?
AWS: Organizing and tracking costs using AWS cost allocation tags

Each provider's documentation changes: always check the current version. Anything marked [Nubyron's view] is our own judgement, not the documentation's.

Frequently asked questions

What is FinOps?

It is the operational framework and cultural practice that the FinOps Foundation describes to maximise the business value of technology and create financial accountability through collaboration. It starts by understanding what each thing costs and who generates it.

Where can I see the detail of my bill on Azure and AWS?

On Azure, in Cost Management, using Cost Analysis and data exports. On AWS, in Cost Explorer and, for the most complete detail, in the Cost and Usage Reports, which AWS recommends receiving as CUR 2.0 through Data Exports.

Do I need to tag everything to get started?

In our view, no. A coherent account or subscription structure and a few tags (environment, product or team, owner) are enough to start. On Azure a tag does not apply to historical data, so add them early.

Does reducing cloud costs affect performance?

It does not have to, but it can: resizing a resource usually means recreating or reconfiguring it, and that can interrupt service. That is why you measure before and after, change in an agreed window and write down the way back.

Is a saving guaranteed with a review?

No. A review identifies potential saving, prioritised and with its risk. It is not a saving until it is applied and measured, and applying the changes is not part of the review: it is quoted separately.

Do you want a second pair of eyes on your cloud bill?

We review your AWS or Azure bill in read-only mode and deliver a report with the potential saving, prioritised and with its risk. We change nothing in your environment.

Find out how much you could optimise

See the full cloud review ↗