Tagging isn't allocation: why tag compliance stalls and what to do instead

In brief

Tags are one input to cloud cost allocation, not the whole of it. Why 100% tag compliance stalls, what can't be tagged, and a layered approach that works.

Steve Elliott

Steve Elliott · · 7 min read


On this page
  1. Why isn't tag compliance the same as cost allocation?
  2. What can't be tagged?
  3. Why do tag cleanup drives stall?
  4. What does a layered approach look like?
  5. An illustrative example: 81% direct, 100% allocated
  6. Where does FOCUS help, and where doesn't it?
  7. What should you measure instead of tag compliance?
  8. Sources

Tag compliance stalls because some cloud spend can't be tagged, tags reach back into history only partly, and every provider sets different limits on what a tag can hold. In our view the fix is to stop treating tags as the allocation method and start treating them as one of three: account and subscription structure first, tags second, written rules for everything left over.

The FinOps Foundation draws the same line. Its Allocation capability calls for three strategies, allocation, tagging and shared cost, and it says plainly that "some costs cannot be tagged in environments."

Why isn't tag compliance the same as cost allocation?

Allocation answers who pays for each dollar. Tag compliance answers how many resources carry the right labels.

The FinOps Foundation's Allocation capability defines cloud cost allocation as apportioning costs to those responsible for each part of them, using accounts, tags, labels and other metadata. Its Crawl stage allocates by accounts, projects or subscriptions mapped to cost centers from a known list, a method that doesn't depend on tags. Its metadata compliance KPI counts spend tied to accurate metadata or to a recognizable hierarchy. So account structure counts toward compliance, not just tags.

AWS says the same about its own estate. Its tagging whitepaper describes an account-based model that needs the least effort, gives high accuracy for showback and chargeback, and doesn't rely on tagging individual resources.

So a resource with no tags, in an account owned by one team, is already allocated. A perfectly tagged resource in a shared account may still need a rule to split it. In our view, reporting tag compliance as if it were allocation coverage sends teams to fix the wrong resources.

What can't be tagged?

Each provider documents spend that never carries a resource tag, or carries it only some of the time. As of November 2026:

AWSAzureGoogle Cloud
Spend with no resource tagSome commitment fees, including Reserved Instance and Savings Plans subscription fees and unused commitment, can't be tagged before they appear in billingSome resources don't support tags or don't include them in usage dataSupport costs belong to the billing account, not a project
Inheritance from the hierarchyAccount tags from AWS Organizations can be activated for cost allocation (announced December 19, 2025) and cover credits, refunds and some service chargesResources don't inherit resource group or subscription tags; tag inheritance applies them to usage records onlyProject labels are exported separately from resource labels; you can't select project labels when filtering a billing breakdown by label
Rules for shared costCost Categories split charge rules (proportional, fixed, even), shown only on the cost categories details page, not in CUR or Cost ExplorerCost allocation rules move costs between subscriptions, resource groups and tags; purchases, including reservations and savings plans, aren't supportedNo split-charge feature in the billing docs cited; project ancestry and labels in the BigQuery export let you apply your own rules

Every provider now offers a way to apply hierarchy to usage that resource tags miss: AWS account tags, Azure tag inheritance, and, in Google's billing export only, the project label and ancestry columns. Commitment fees are only partly reachable. On AWS, used commitment can be traced to tagged resources after the fact, but unused commitment can't. Azure's allocation rules don't support purchases at all. The unused part needs a rule you write.

Why do tag cleanup drives stall?

Tags don't reach back far. Google's billing export says the cost data for a label "only shows usage from the date that the label was applied." Microsoft says resource tags appear in usage data only while applied to the resource and aren't applied to historical data. Azure's tag inheritance updates the current month. AWS lets the management account backfill cost allocation tag activation for up to 12 months, but only where the tag was actually on the resource at the time. A tagging sprint fixes next month's showback, not last quarter's.

Activation adds delay. On AWS cost allocation tags must be activated before they appear in billing reports. A user-defined tag key can take up to 24 hours to appear for activation and up to 24 more to activate.

Limits differ by provider. AWS allows 50 user-defined tags per resource, keys up to 128 characters and values up to 256. Azure allows 50 tag name-value pairs per resource, with a few resource types limited to 15, and bans characters such as the slash and percent sign in tag names. Google Cloud allows 64 labels per resource, with keys and values of up to 63 characters and only lowercase letters, numbers, underscores and dashes. A key such as CostCenter works on AWS and Azure and fails on Google Cloud. A multicloud tag dictionary has to fit the strictest of the three.

Enforcement is narrower than it looks. AWS Organizations tag policies can block tags with the wrong values, but AWS states: "This capability does not enforce missing tag keys." To block creation without specified tags, AWS points to service control policies, and it offers a required tag keys option aimed at infrastructure-as-code. It also warns that, for services that create and tag resources automatically, enforcement may block dynamic scaling and provisioning.

Nobody owns the leftovers. In our view, the last stretch of untagged spend tends to sit in shared accounts, legacy workloads and short-lived resources, where no single team feels the cost. Chasing it by resource count produces a long list of low-value fixes.

What does a layered approach look like?

We suggest building allocation in three layers, in this order.

  1. Hierarchy first. Give each product or team its own accounts, subscriptions or projects wherever you can, and map each one to an owner and cost center. Apply account-level metadata so it reaches usage records: AWS account tags, Azure tag inheritance, Google project labels. We think this layer can allocate a large share of spend without a single resource tag. On AWS, account tags also reach credits, refunds and some service charges that resource tags can't.
  2. Tags second, where the hierarchy is shared. Inside shared accounts and subscriptions, tags carry the application, environment and owner. Keep the required set short, write it down as a tagging strategy, enforce it at provisioning through policy or infrastructure-as-code, and accept that tags won't fix the past.
  3. Rules for the rest. Write down how each untaggable or shared cost line is split: commitment fees, support, tax, credits, central networking, shared clusters. The FinOps Foundation lists fixed, proportional, proxy-metric and even-split methods. It notes that most organizations use a mix: some shared costs are allocated to cost center budgets and others are funded centrally. Pick a method per line, publish it and hold it steady.

This mirrors the FinOps Foundation's maturity path: allocation by account at Crawl, tagging in key areas and shared-cost splits at Walk, and ownership metadata enforced before provisioning at Run.

An illustrative example: 81% direct, 100% allocated

The numbers below are made up to show the mechanics. Take a $1.2 million monthly cloud bill.

LayerSpendHow it's allocatedCumulative direct share
Dedicated accounts and subscriptions$780,000Hierarchy: account owner and cost center65%
Shared accounts, tagged resources$195,000Resource tags81%
Shared accounts, untagged resources$105,000Rule: proportional to each team's direct spend in those accounts81%
Commitment fees, support, tax, credits$120,000Rule: used commitment to the teams whose usage it covered; unused commitment to the commitment owner; support proportional to spend81%
Total$1,200,00081% direct, 100% allocated

Now suppose the same estate reports tag compliance of 58%, measured by resource count, because many resources in dedicated accounts carry no tags. That figure says almost nothing about allocation. Tagging every resource in the dedicated accounts would not move a dollar to a new owner, though it may still help engineers see cost by application. The useful target is the $105,000 of untagged spend in shared accounts, and the test is whether tagging it beats a published rule. Unused commitment has no workload behind it, so we route it to the commitment owner, as in our guide to mapping FOCUS data to the TBM taxonomy.

In our view, finance should see two numbers here: share of spend directly attributed, which the FinOps Foundation's Allocation Accuracy Index measures, and share of spend allocated by any method, including rules.

Where does FOCUS help, and where doesn't it?

The FOCUS specification gives tags a common shape across providers. In version 1.4 the Tags column is a JSON object of finalized tags, both user-defined and provider-defined. Where a key is set at several levels, such as account and resource, the data generator reduces it to one value. FOCUS 1.3 added an Allocated Tags column for charges produced by a provider's split cost allocation.

That helps the hierarchy layer: account-level context arrives already merged into each row. As we read the spec, it standardizes the column, not your key names. If one cloud uses CostCenter and another cost_center, you still reconcile them.

What should you measure instead of tag compliance?

We suggest four measures, reported by spend rather than resource count:

  • Directly attributed share. Spend allocated by hierarchy or tags, divided by total spend.
  • Allocated share. Directly attributed spend plus spend split by a published rule.
  • Unallocated spend in shared accounts. In dollars, with a named owner for the backlog.
  • Rule coverage. Every untaggable line type, such as commitment fees, support, tax and credits, with a written split method.

An allocation is only as useful as the ownership behind it. That is the case for an operating model in which accounts, products, teams and funding lines are mapped and kept current, so every allocation rule points to someone who can act.

Sources

Frequently asked questions

What tag compliance rate should we target?

In our view, target spend, not resources. Aim for a high share of spend directly attributed through account structure and tags, and 100% allocated once written rules cover the rest. Resource-count compliance in shared accounts is a useful guardrail for new deployments. Chasing 100% across the estate rarely pays, because some charges, such as certain commitment fees and support, can't carry resource tags at all.

Can cloud cost allocation tags be applied retroactively?

Only partly. AWS lets the management account backfill tag activation for up to 12 months, but only where the tag was on the resource at the time. Microsoft says resource tags aren't applied to historical usage data, and its tag inheritance updates the current month. Google says label cost data starts from the date the label was applied.

Which cloud costs can't be tagged?

AWS says some commitment fees, including Reserved Instance and Savings Plans subscription fees and unused commitment, can't be tagged in advance. Microsoft says some Azure resources don't support tags or don't include them in usage data. Google Cloud support costs belong to the billing account, not a project. Account-level metadata or written allocation rules are needed to assign these costs.

Should we use accounts or tags for cloud cost allocation?

Both, in that order. Accounts, subscriptions and projects give each team a clean boundary that allocates spend without resource tags, and AWS calls this model the least effort with high accuracy. Tags add application and environment detail inside shared accounts. Written rules cover what neither can reach, such as commitment fees, support and shared networking.

Free playbooks

In-depth guides for product and operations leaders.

Browse the playbooks
Steve Elliott

Steve Elliott

Published November 22, 2026

More on FinOps & Cloud

FOCUS, commitments, shared cost.