Reports & Exports

How do I export AI spend in FOCUS format?

Export AI spend aligned with the FinOps Open Cost and Usage Specification — the same data model used by AWS Cost Explorer, Azure Cost Management, and Google Cloud Billing. Load it into Apptio, CloudZero, or a data warehouse alongside your cloud billing data.

In the app: Reports → FOCUS Export · Growth plan and aboveOpen in app →No account? Start free

Cognocient AI exports your AI spend in a CSV aligned with the FinOps Open Cost and Usage Specification (FOCUS) — the same open schema the major cloud providers offer for their own billing exports. Load it into Apptio, CloudZero, or a data warehouse alongside your cloud billing data.

Cognocient AI's FOCUS export is undergoing conformance validation against the FinOps Foundation's own open-source validator. We describe this export as "FOCUS-aligned" rather than certifying a specific conformance level (e.g. "FOCUS 1.2") until that validation is complete and independently confirmed.

Logging cost data and reporting it to finance are two different problems. Tools that expose a raw per-call cost value (for example, a response header showing the cost of a single API call) solve the first problem, but someone still has to build and maintain the layer that turns that raw data into cost per feature, cost per customer, and trends over time that a CFO can actually read. Cognocient AI's FOCUS export is built to close that second gap directly, without requiring a custom reporting pipeline.

Board-ready PDF reports

Finance and board audiences don't want raw API logs. Cognocient AI generates one-click PDF reports that translate your AI spend data into CFO-friendly language — cost by department, waste recovery opportunity, cost-per-outcome, and 3-month trend.

SectionContent
Executive summaryTotal AI spend, MoM change, and waste recovery opportunity highlighted in 3 numbers.
Spend by departmentHorizontal bar chart showing each department's share, with trend arrows vs. prior period.
Investment vs. WasteDonut chart showing classified spend. What percentage is generating measurable ROI.
Top 5 cost reductionsAI-generated recommendation summary — each with estimated monthly saving and effort.
Cost per outcomeCost per resolved ticket, generated contract, or any outcome you track via the outcomes API.
Model efficiencyWhich models are over-provisioned for their actual task complexity.

Generate a board report from the dashboard: Reports → Board Report → Select date range → Download PDF

Schedule monthly board reports to be automatically emailed to stakeholders. Go to Intelligence → Reports → Scheduled Delivery and configure the recipient list. Recipients receive the PDF attachment with no Cognocient AI login required. Available on Growth plan and above.

What is FOCUS?

FOCUS (FinOps Open Cost and Usage Specification) is the open billing standard from the FinOps Foundation. It normalises cost and usage data across providers into one set of columns.

The major cloud providers offer FOCUS-format billing exports, and FinOps platforms such as Apptio Cloudability and CloudZero accept FOCUS data. A FOCUS-aligned AI export can therefore sit next to your cloud billing data instead of needing a separate pipeline. You can also load the CSV into a data warehouse such as BigQuery, Snowflake or Redshift.

Downloading a FOCUS export

Download from the dashboard: the Reports page or the Chargeback page both have a FOCUS Export card — pick a date range and click Download FOCUS CSV.

Or pull via API:

curl "https://api.cognocient.com/api/chargeback/exports/focus?from=2026-05-01&to=2026-05-31&version=1.2" \
  -H "Authorization: Bearer sk-cog-YOUR-PROXY-KEY" \
  -o cognocient-focus-2026-05.csv

The export is CSV only — there is no JSON response format for this endpoint today. from and to are inclusive calendar days in UTC. version picks the FOCUS column set: 1.2 (the default, what most FinOps tools read today) or 1.4.

How rows are built

There is one row per UTC day, provider, model, tag set and meter. A meter is one thing the provider prices separately:

Meter (SkuMeter)What it countsPricingUnit
Input TokensPrompt tokens not served from the provider's prompt cache1000000 Tokens
Cache Read Input TokensPrompt tokens read from the provider's prompt cache1000000 Tokens
Cache Write Input Tokens (5 min)Prompt tokens written to the prompt cache with the default 5-minute lifetime (Anthropic, Bedrock, Vertex), billed at 1.25× input1000000 Tokens
Cache Write Input Tokens (1 hour)Prompt tokens written to Anthropic's 1-hour cache (direct, Bedrock, Vertex), billed at 2× input1000000 Tokens
Output TokensCompletion tokens, including reasoning tokens1000000 Tokens
Images / Audio Seconds / Characters / SearchesImage generation, transcription, text-to-speech, rerankthe same unit

A single API call can produce several meters, for example input, cache-read and output tokens. Each meter has its own list price, so splitting them is what lets ListUnitPrice × PricingQuantity = ListCost hold on every row. This follows the AI billing examples in the FOCUS 1.5 working draft.

Left out of the export: Cognocient AI cache hits and MCP gateway calls (no provider charge), and failed calls that consumed nothing.

FOCUS columns in the export

FOCUS ColumnCognocient AI sourceExample value
BillingAccountId / BillingAccountNameYour Cognocient AI customer ID and company name (your sign-in email if no company name is set)cbee6a71-..., Acme Inc.
BillingCurrencyAlways USDUSD
BillingPeriodStart / BillingPeriodEndThe calendar month the charge falls in, UTC, end-exclusive2026-05-01T00:00:00Z, 2026-06-01T00:00:00Z
ChargePeriodStart / ChargePeriodEndThe UTC day the usage happened, end-exclusive2026-05-14T00:00:00Z, 2026-05-15T00:00:00Z
ChargeCategory / ChargeFrequencyAlways Usage / Usage-BasedUsage
ChargeClassNull (no corrections are emitted)(empty)
ChargeDescriptionMeter and modelInput Tokens - gpt-4o-mini
InvoiceIdNull — usage rows aren't tied to an invoice(empty)
InvoiceIssuerNameThe company that bills you for the usage. Cognocient AI passes provider cost through with no markup; its own subscription is not in this exportOpenAI, AWS
ProviderName (1.2) / ServiceProviderName, HostProviderName (1.4)Same company as InvoiceIssuerNameAWS
PublisherName (1.2 only)The company that made the model, even when a cloud platform hosts itAnthropic
ServiceNameThe provider's API product, not the modelOpenAI API, Claude API, Amazon Bedrock
ServiceCategory / ServiceSubcategoryAlways AI and Machine Learning / Generative AI
SkuIdProvider, model and meteropenai/gpt-4o-mini/input
SkuPriceIdSkuId plus the unit price appliedopenai/gpt-4o-mini/input@0.15
SkuMeterThe meter (table above)Cache Read Input Tokens
SkuPriceDetailsJSON: x_ModelDeveloper, x_ModelId, x_TokenDirection (Input/Output), x_TokenCacheAction (Uncached/Read/Write). The x_ prefix is required until FOCUS defines these properties{"x_ModelDeveloper":"OpenAI","x_ModelId":"gpt-4o-mini",...}
PricingCategoryAlways StandardStandard
PricingUnit / PricingQuantityPriced block and quantity in it1000000 Tokens, 0.0184
ListUnitPriceList price per PricingUnit applied to the calls0.15
ListCostListUnitPrice × PricingQuantity0.00276
BilledCost / EffectiveCost / ContractedCostThe cost recorded for the calls. Can be below ListCost, e.g. Batch API calls at 50%0.00276
ConsumedQuantity / ConsumedUnitRaw amount consumed18400.0, Tokens
TagsOne JSON object from your attribution headers (keys feature, department, user, project, gl_account, customer; null when the row has none){"department":"customer-success","feature":"ticket-resolver"}
x_PriceSourcerecorded: prices stored on the calls. estimated: calls from before 2026-10-28, priced with today's rates rescaled to the cost recorded at the time. derived: non-token calls, priced as cost ÷ quantityrecorded

Custom columns use the x_ prefix that FOCUS requires and come after every FOCUS column.

Column changes on 2026-10-28. The export now has one row per meter (input, cache-read, cache-write and output tokens, or a non-token unit) with the FOCUS SKU and price columns: SkuId, SkuPriceId, SkuMeter, SkuPriceDetails, ListUnitPrice, PricingCategory and x_PriceSource. PricingUnit is 1000000 Tokens. ServiceName is the provider's API product (the model is in SkuPriceDetails), and provider names are company names (OpenAI, AWS). Removed: x-cog-* and Tags/x-cost-* (replaced by the JSON Tags column), x_InputTokenCount, x_OutputTokenCount, x_TotalTokenCount and x_ModelId (use ConsumedQuantity by SkuMeter, and SkuPriceDetails), ChargeType (use ChargeCategory), UsageQuantity/UsageUnit (now ConsumedQuantity/ConsumedUnit), and ResourceId, ResourceType, SubAccountId and SubAccountName (departments are in Tags). Periods are full UTC timestamps with one row per day. Pipelines built on the old layout need updating.

Changes on 2026-10-30. Anthropic's ServiceName is now Claude API (was Anthropic API), the name Anthropic's own Console bills it under. A new meter, Cache Write Input Tokens (1 hour) (SkuId suffix cache-write-1h), carries 1-hour cache writes at 2× input, and the 5-minute meter is renamed Cache Write Input Tokens (5 min) (SkuId suffix cache-write-5m, was Cache Write Input Tokens / cache-write). The names match what the converter produces from Anthropic Console exports. BillingAccountName is never empty: it falls back to your sign-in email. To check an export yourself, use the free FOCUS checker or the open-source ai-bill-to-focus CLI.

Reconciliation workflow

Join Cognocient AI FOCUS exports against your cloud billing exports using BillingAccountId for cross-source reconciliation in a single FinOps dashboard.

  1. Download the Cognocient AI FOCUS export for the billing period.
  2. Download your AWS CUR / Azure Cost export / GCP BigQuery billing export.
  3. Load both into your FinOps platform or data warehouse (BigQuery, Snowflake, Redshift).
  4. Join on BillingAccountId + BillingPeriodStart to get a unified view of all cloud + AI spend.
  5. Filter on the feature or department key in Tags to generate per-feature or per-department chargebacks.

If you use Apptio Cloudability or CloudZero, load the CSV as a custom or FOCUS data source. Column names follow FOCUS conventions, but we haven't certified the import with any specific platform, so check the column mapping on your first load.

On the Business plan, FinOps connectors send this export to Vantage, CloudZero or an S3 bucket every day, so no one has to download it.

Frequently asked questions

On this page