A simple, repeatable way to sanity-check Azure cost dashboards (Azure native, Power BI, third-party, or custom) before you brief leadership, start a savings plan, or escalate a “why did spend spike” incident.
Most cost dashboards are not “wrong.”
They are unclear.
That sounds like nitpicking until you are in the hot seat trying to explain spend to leadership, or you are a platform owner getting blamed for a spike you did not cause.
So I treat every cost dashboard the same way I treat a monitoring dashboard:
Before I trust the insights, I verify the plumbing.
Here are the three checks I run before I put any weight on the numbers.
Check 1: Scope and allocation are unambiguous
If a dashboard cannot answer “what exactly is included,” it is not a decision tool. It is a vibe.
What I want to see, immediately
Which tenant is in scope (yes, this matters more than people admit).
Which subscriptions are included, ideally with a count and a way to export the list.
Whether the dashboard is covering all subscriptions under the enterprise billing relationship, not just “the ones we remembered to include.”
How shared services are handled (networking, identity, security tooling, logging, hub resources).
Allocation method for shared costs: tagged, split by usage, split by headcount, pushed to a central cost center, or not allocated at all.
Why this breaks in the real world
“All subscriptions” is often shorthand for “all the subscriptions in this one view.”
That is not the same thing as “all subscriptions in the enterprise.”
If you need enterprise-wide coverage, the cleanest way to keep the subscription set from drifting is to anchor your logic in management group structure. It becomes your living inventory of what “counts” as in-scope.
A quick sanity pattern
Use your management group structure to define “in scope” (for example: platform, shared services, business units).
Confirm the dashboard uses that structure, or at least a subscription list derived from it.
Confirm shared services are either:
Clearly shown as shared services, or
Clearly allocated using a stated rule.
If none of that is visible, your dashboard might still be useful, but only for local decisions. Not enterprise calls.
Check 2: The data source and freshness are provable
Cost data is not a live telemetry stream. It arrives, it settles, it backfills, and it sometimes changes after the fact.
If your dashboard refresh story is hand-wavy, you will lose trust the moment the numbers do not match someone’s expectations.
What I verify
Where the data comes from
Azure Cost Management views
Cost export files
A custom pipeline feeding Power BI
A third-party platform using its own ingestion rules
Refresh cadence
Daily? Multiple times per day? Manual?
Data delay expectations
When does “yesterday” become reliable?
Backfill behavior
Do late charges update prior days?
Do you rerun data loads, or do you only append?
Enterprise Agreement vs Microsoft Customer Agreement: why scope can get messy
Both can support enterprise-wide reporting, but their billing hierarchies and “where you point the query” can differ.
What tends to trip teams up:
With an Enterprise Agreement, your billing rollups often align to the enterprise enrollment structure.
With a Microsoft Customer Agreement, your rollups tend to align to billing account structures like billing profiles and invoice sections.
Here is the practical takeaway:
A dashboard can be correct inside its billing scope and still fail “enterprise-wide” coverage.
So the trust check is not “is the chart pretty.”
It is “can you prove the subscription set and billing scope align to the full estate.”
The fastest trust question
Ask the dashboard owner to answer this in one sentence:
“This dashboard refreshes at ___ cadence, uses ___ as the source, and covers ___ subscriptions across ___ billing scope.”
If they cannot answer cleanly, do not argue about the numbers yet. Fix the definition first.
Check 3: Usage is separated from rate, and discounts are handled on purpose
Most “spend spikes” are not mysterious. They are one of these:
You used more.
The rate changed.
A discount treatment changed how the cost is shown.
If a dashboard blends those together, it will create confusion, and it will waste everyone’s time.
What I expect a trustworthy dashboard to show
At minimum, for the top drivers:
Usage quantity trend (consumption)
Effective unit price trend (rate)
A clear statement on how discounts and commitments are treated:
Are commitments amortized?
Are they shown as a separate line?
Are savings netted out or shown separately?
Why this matters
If you do not separate usage from rate, you will make the wrong call:
You might start a right-sizing effort when the real issue was a pricing change.
You might escalate to engineering when the issue was a discount view changing.
You might celebrate “savings” that are just reporting treatment.
A dashboard does not have to do everything, but it must be honest about what it is doing.
Common failure modes (bookmark this)
“All subscriptions” with no subscription list or count
Shared services buried inside business unit totals
Tag-based allocation with no visibility into tag coverage
Refresh timing unknown, or “it updates when it updates”
Costs grouped in ways that hide drivers (too much aggregation)
Commitment savings shown as “reduced spend” with no explanation
Mixing invoice month, usage date, and calendar month without stating it
What I do when a dashboard fails a check
I do not debate the charts. I ask for the minimum proof.
Prove coverage
Export the subscription list used by the dashboard
Map it to the subscription inventory derived from management group structure
Identify what is missing and why
Prove freshness
Show the refresh timestamp
Document expected data delay
Confirm backfills are handled intentionally
Prove drivers
For the top cost items, show usage and effective unit price separately
State how discounts and commitments are displayed
Once those are clear, the dashboard usually becomes “obviously right” or “obviously incomplete.” Either outcome is a win.
Optional: Trust Score (0 to 6)
Score each check from 0 to 2:
Scope and allocation: 0 (unclear), 1 (mostly clear), 2 (provable and exportable)
Data source and freshness: 0 (unknown), 1 (stated but unproven), 2 (stated and provable)
Usage vs rate, discounts treatment: 0 (blended), 1 (partial), 2 (clear and consistent)
0–2: Not safe for decisions
3–4: Use with caution, verify drivers before acting
5–6: Strong enough for planning, forecasting, and leadership updates
Want the Dashboard Trust Checklist + Trust Score worksheet?
Grab it here —> https://tally.so/r/Pd9BG0
It takes ~60 seconds and I’ll send the pack instantly.