Cloud cost reaches a spreadsheet or a BI tool such as Power BI in one of two ways: as a download from the console, which freezes the state of one day, or as a query on a running export, which recomputes at every refresh. A download is enough for a single question. A report that comes back every month, combines several clouds or puts cost next to revenue needs the export, because only the export carries every line with its cost measure, its keys and its data cut-off. Whether the figures are right is decided by the data model underneath, not by the tool.
How it works
Every console offers a download. Cost Explorer saves a view as CSV (AWS Cost Explorer and AWS Budgets), Azure cost analysis downloads as PNG, Excel or CSV, and Google's Cost table report downloads as CSV (importing cost data from files). Microsoft describes the download as a summarised snapshot of the table on screen that does not update, names a limit of 15,000 records per CSV in its FAQ, and points to an export for larger scopes. The spreadsheet has limits too: an Excel worksheet holds 1,048,576 rows by 16,384 columns, Google Sheets up to 20 million cells or 100 MB. And the file carries neither the cost measure nor the filters nor the data cut-off unless someone writes them in.
A BI tool reads the export instead, in storage the company owns, by a route each provider documents.
AWS
AWS Data Exports writes CUR 2.0 or FOCUS 1.2 to an S3 bucket (AWS Cost and Usage Report and Data Exports). Amazon Athena queries the files through a Glue schema, and Amazon Quick Sight, the BI part of Amazon Quick, loads the results into its in-memory store, SPICE. The Cloud Intelligence Dashboards sit on this stack: an open-source framework from AWS Enterprise Support under the MIT-0 licence. The CUDOS, Cost Intelligence and KPI dashboards read CUR; the FOCUS Dashboard reads FOCUS data and can add exports from Azure, Google Cloud and OCI. Which export type to pick is weighed in AWS CUR 2.0 or FOCUS 1.2.
Azure
For Power BI Desktop, Microsoft offers the Cost Management connector. It supports Enterprise Agreements and direct Microsoft Customer Agreements, but not pay-as-you-go, indirect MCA or Microsoft Partner Agreements, and it needs Enterprise Administrator (read-only) or Contributor on the billing account or profile. It loads tables such as Usage details and Usage details amortized and fails on requests above one million rows. How much cost data it carries, Microsoft answers two ways: the connector's own page estimates up to USD 5 million of raw cost details, the FinOps toolkit's comparison table up to USD 2 million. The Cost Management app for Power BI exists for EA only and needs a Power BI Pro licence. The FinOps toolkit, open source under the MIT licence, no longer recommends either: both stay available but are not being maintained, and the connector uses an old API version that lacks details such as savings plans. Up to USD 100,000 of monitored spend it points to exports into Data Lake Storage, above that to FinOps hubs on Data Explorer or, where Microsoft Fabric is in use, to Fabric (Azure cost exports).
Google Cloud
Google Cloud delivers line-level data through the export to BigQuery (Google Cloud billing export). Data Studio, as Looker Studio is now called, reads it through its BigQuery connector; how reports are built on it and survive schema changes is covered in dashboards and Google Cloud billing export for cost reporting. The data source decides whether a query runs with the owner's credentials or each viewer's. By default only the owner needs BigQuery roles, and viewers need no access of their own to the export.
The data model underneath
Whatever the tool, every dashboard on top of an export answers the same five questions.
| Question | What the export delivers | What the report has to fix |
|---|---|---|
| What is a row? | one line item per account, service, usage type and period | granularity: hour, day or month |
| Which cost measure? | billed and amortized cost side by side; FOCUS separates Billed Cost and Effective Cost | one measure per visual, named on it |
| How do tags arrive? | CUR 2.0 collapses tags into key-value columns, Google carries labels and tags as key and value structs, FOCUS as a key-value object | the keys in use and their renames |
| Which currency? | each provider's billing currency | one reporting currency with a rate source and rate date |
| Which data state? | several loads a day, earlier months overwritten after finalisation | a full refresh of the months affected and a visible cut-off |
Bringing currencies together is covered in multi-cloud reporting in one currency, and when a month is settled in freshness, restatements and finalisation.
What it means for your cost
The BI route costs money of its own in three places: storage for the export, queries, and licences. How BigQuery and Athena bill storage and queries is set out in data warehouses and analytics; Data Studio bills a report's queries to the project chosen as billing project in the data source. For Amazon Quick Sight AWS lists USD 24 per author and USD 3 per reader a month, SPICE at USD 0.38 per GB a month, and an infrastructure fee of USD 250 per account a month once the account has a Pro user or Q&A features enabled. For Azure, Microsoft's FinOps toolkit gives orders of magnitude at list prices: exports to storage at roughly USD 3 to 5 a month per USD 1 million of monitored spend, FinOps hubs on Data Explorer from USD 120 plus roughly USD 10 per million, and Power BI at USD 20 per user. All of these are list prices in US dollars with no Region stated, read on 2026-10-16.
Cost follows the refresh. Microsoft states that cost and usage data is typically updated in the portal and its APIs within 8 to 24 hours, and so suggests no more than one or two scheduled Power BI refreshes a day. The Cloud Intelligence Dashboards refresh their datasets daily. Loading more often scans the same data again and finds nothing new.
Where it goes wrong
- A refresh that misses corrections. A closed month on the dashboard no longer matches the invoice. The cause: an incremental refresh loads only new partitions, while AWS overwrites the previous month's CUR data after the invoice is finalised, usually mid-month. The Cloud Intelligence Dashboards require a full refresh for that reason, and every restatement needs the months it touches reloaded.
- Cost measures mixed. The dashboard totals differ from Cost Explorer. The cause: one is unblended, the other amortized. Many Cloud Intelligence Dashboards visuals also leave out taxes, refunds and credits.
- Joins on tags that change. A team's cost drops to zero for a month. The cause: a tag key was renamed, or the new key is never read out of the key-value column. In the Power BI connector the tags column arrives as text without its brackets and can be parsed as JSON only after an extra step.
- A dashboard nobody owns. Visuals break after a schema change or miss new services. The cause: the Cloud Intelligence Dashboards do not change when their CloudFormation stack is updated, only through a separate update command. Someone has to own updates, access and views.
Sources
Read on 2026-10-16. AWS Cloud Intelligence Dashboards, CUDOS, CID and KPI, FOCUS Dashboard, Cloud Intelligence Dashboards FAQ, CUR 2.0 data dictionary and Amazon Quick Sight pricing; Microsoft Cost Management connector for Power BI Desktop, Cost Management app for Power BI, choose a Power BI data source (FinOps toolkit), save, share and download views, FinOps toolkit on GitHub and Excel specifications and limits; Google connect Data Studio to BigQuery, standard usage cost export schema and files you can store in Google Drive and Sheets; FinOps Foundation FOCUS Billed Cost, Effective Cost and Tags; Cloud Intelligence Dashboards repository and licence.