Supported billing export formats

AWS, Google Cloud, Azure, and OCI can all emit FOCUS. In AWS it is an export type in Data Exports; in Google Cloud it is a BigQuery billing export view; in Azure it is a Cost Management export.

BilledCost

What was invoiced.

EffectiveCost

Amortized, the one to analyze.

PricingQuantity / PricingUnit

The quantity and unit the rate applies to.

ServiceName / SkuId

What was bought.

ChargeCategory

Usage, Purchase, Tax, Credit, Adjustment.

ChargePeriodStart

When. Drives the day-count normalization.

Tags

Allocation metadata, as a map.

Bring this one if you can. It needs no provider-specific interpretation, which means fewer places for a mapping to be wrong, and it works identically across clouds.

Billing and Cost Management → Data Exports → Create export → CUR 2.0. Choose CSV (or gzipped CSV) and include resource IDs if you want resource-level drill-down.

line_item_unblended_cost

What was invoiced, and the cost of every ordinary usage line.

savings_plan_savings_plan_effective_cost

Amortized cost of usage a Savings Plan covered.

reservation_effective_cost

Amortized cost of usage a Reserved Instance covered.

line_item_line_item_type

Decides which of the three applies to each line: Usage, DiscountedUsage, SavingsPlanCoveredUsage, fees, Tax, Credit.

line_item_usage_amount / pricing_unit

Quantity and unit.

product_servicecode

The service.

line_item_usage_type / line_item_operation

Together, the SKU.

resource_tags

Tags, as a nested map. The user_ prefix AWS adds to each key is removed.

The amortized cost is built line by line from the line item type, the way Cost Explorer builds its amortized view: covered usage at its effective cost, Savings Plan negation and upfront-fee lines at zero, and unused commitment kept on its fee line. CUR 2.0 has a fixed column set, which is why it is easier to work with than the legacy format. If your export is Parquet, re-export as CSV. Parquet support is not in this version.

The older Cost and Usage Report, still widely deployed. Columns use slash-separated names and the set changes month to month depending on what you used.

lineItem/UnblendedCost

What was charged.

reservation/EffectiveCost / savingsPlan/SavingsPlanEffectiveCost

Amortized, chosen per line item type exactly as for CUR 2.0.

lineItem/UsageAmount

Quantity.

lineItem/ProductCode

The service.

lineItem/UsageType

The SKU.

resourceTags/user:Key

One flattened column per tag key.

The varying column set is the reason a diff across months can find tag columns in one file and not the other. The mapping panel in the tool shows exactly what was matched in each file.

Billing → Billing export → BigQuery export. Use the detailed export if you want resource-level data. BigQuery cannot export nested or repeated columns (labels, credits) to CSV, so run the query below once per month instead: it flattens one invoice month to exactly the columns the tool reads. Save the result as CSV.

cost

What was charged, after negotiated discounts.

discount_credits

Committed use, sustained use and spending-based discount credits, negative. Netted into cost: they are part of the price.

other_credits

Promotions, free tier and the other credit types, negative. Kept apart, as a reconciling line.

usage_amount_in_pricing_units / usage_pricing_unit

Quantity and unit.

service_description / sku_description

The service and SKU. A SKU whose name starts "Commitment" is a commitment fee, filed with purchases.

project_id

The account dimension.

labels

Resource labels, as JSON.

usage_start_time

The day. Drives the day-count normalization.

currency

The billing account currency. Amounts are shown in it, and two files in different currencies are flagged.

-- Finitizer Cloud Bill Diff: one month of the GCP billing export,
-- flattened to the columns the tool reads. Run once per month,
-- changing the invoice month, and save each result as CSV.
-- On the standard (not detailed) export, change resource.name below to
-- CAST(NULL AS STRING): only the detailed export has resource.name, and
-- deleting the line would shift the GROUP BY positions.
-- discount_credits are committed use, sustained use and spending-based
-- discounts, part of the price; other_credits are promotions, free tier
-- and the rest.
SELECT
  TIMESTAMP_TRUNC(usage_start_time, DAY)                       AS usage_start_time,
  service.id                                                   AS service_id,
  service.description                                          AS service_description,
  sku.id                                                       AS sku_id,
  sku.description                                              AS sku_description,
  project.id                                                   AS project_id,
  location.region                                              AS location_region,
  resource.name                                                AS resource_name,
  TO_JSON_STRING(labels)                                       AS labels,
  cost_type,
  currency,
  usage.pricing_unit                                           AS usage_pricing_unit,
  SUM(usage.amount_in_pricing_units)                           AS usage_amount_in_pricing_units,
  SUM(cost)                                                    AS cost,
  SUM(cost_at_list)                                            AS cost_at_list,
  SUM(IFNULL((SELECT SUM(c.amount) FROM UNNEST(credits) AS c WHERE c.type IN ('COMMITTED_USAGE_DISCOUNT', 'COMMITTED_USAGE_DISCOUNT_DOLLAR_BASE', 'SUSTAINED_USAGE_DISCOUNT', 'DISCOUNT', 'FEE_UTILIZATION_OFFSET')), 0)) AS discount_credits,
  SUM(IFNULL((SELECT SUM(c.amount) FROM UNNEST(credits) AS c WHERE IFNULL(c.type, '') NOT IN ('COMMITTED_USAGE_DISCOUNT', 'COMMITTED_USAGE_DISCOUNT_DOLLAR_BASE', 'SUSTAINED_USAGE_DISCOUNT', 'DISCOUNT', 'FEE_UTILIZATION_OFFSET')), 0)) AS other_credits
FROM `my-project.my_dataset.gcp_billing_export_resource_v1_XXXXXX_XXXXXX_XXXXXX`
WHERE invoice.month = '202608'
  -- The table is partitioned by the day each row was loaded. This bound
  -- reads the month and its late rows, not every month in the table.
  AND _PARTITIONTIME >= TIMESTAMP('2026-08-01')
  AND _PARTITIONTIME < TIMESTAMP('2026-09-15')
GROUP BY 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
ORDER BY 1;

Committed use, sustained use and discount credits are netted into the SKU they discount, so its rate is the one you actually paid and a commitment ending shows as a rate rise. Promotions and free tier are kept apart: with "Include tax, credits, and purchases" ticked they net in too; unticked, they show as a reconciling line. A file from the older query, with one summed credits_amount column, cannot tell the two apart, so all of it is kept apart and the tool says so. The query reads only the partitions that hold the month. A large month may exceed what the console will save as a local CSV; save the result to a table and export that table to Cloud Storage as CSV, which works because every column is now flat.