Skip to main content
The Usage screen gives Org Admins a complete picture of AI spend across every project, team member, provider, and model in the organization, read from the same ledger that enforces your budgets. Instead of reconciling line items from multiple provider invoices, you get one view — drillable by the dimensions that matter for professional services billing, in the currency your organization reports in. Use it to spot high-spend engagements before they reach a cap, identify which models your team reaches for most, or pull the export your finance team needs.

​
Opening the Usage screen

Select Usage in the main sidebar. The report opens on the last 30 days, grouped by project.
The Usage screen is for Org Admins. Project Admins and members see their own project’s spend on that project’s Budget tab — see Budgets.

​
The currency of the report

Every cost on this screen is in your organization’s reporting currency, and says so: the headline reads Total cost (GBP) and the table column reads Cost (GBP). An Org Admin sets the reporting currency in Org settings → Budgets & alerts → Reporting currency — see Currencies and Exchange Rates. Costs recorded in another currency are converted month by month: each month’s usage at Parlé’s exchange rate for that month, then added together. Changing the reporting currency changes how the report is shown, never what was recorded.
A project can override the reporting currency for its own Budget tab. That override does not affect this screen — the organization’s Usage report is always in the organization’s reporting currency.

​
Dimensions You Can Slice By

Use Group by to break spend down along one axis, and the Projects, People, Models and Providers filters to narrow to a specific subset.

Project

See which client engagements are consuming the most tokens and cost. Personal conversations appear as their own Personal row.

User

See which team members are the heaviest AI users across the org. Useful for understanding adoption patterns and for reviewing individual spend against personal budgets.

Provider

Compare spend across Anthropic, OpenAI, Google Gemini, Azure AI Foundry and Parlé Managed models.

Model

Break spend down by specific model. Identify whether cost-efficient models could substitute for premium ones on lower-stakes tasks.

​
Auto groupings

If your organization uses Auto, two more Group by options show what Auto did with your spend. Every figure is actual billed spend; nothing on this screen is an estimated saving. Auto tier splits spend by the tier Auto chose: How the model was chosen splits the same spend by the reason: Auto’s rule, Deep, Stronger retry, Project starting tier, Moved up after stronger retries, Person’s pick and Background. Combine either grouping with the Models filter, or group by Model, to see which model actually served each tier.
These groupings are on every plan. An organization that has never used Auto sees its spend under Chosen by a person. An organization that stops using Auto can still see how its past spend was split.

​
Prompt cache

Providers charge much less for text they have recently seen (the prompt cache), and a little more to store it for reuse. The Prompt cache block beside the totals shows what that did to your spend for the range and filters you’ve selected:
  • Served from cache: the share of input tokens that came from the cache.
  • Cache reads saved: what the cached input would have cost at the normal input price, minus what it did cost.
  • Cache writes cost extra: what storing text in the cache cost above the normal input price.
  • Net: the two together, shown as Saved an amount, or Cost an amount more when caching cost more than it saved.
These figures are exact, not estimated. Each message recorded the prices it was charged, so the saving is worked out from those prices, in your reporting currency, on the billed amount. The same four figures appear as columns on every grouping. Some usage can’t be measured, for example usage recorded before the cost breakdown existed, usage on provisioned capacity with no per-token price, or usage with no exchange rate. It is counted separately and says so, never guessed.
A high share served from cache usually means people work in long, continuous conversations. Short, one-off conversations write to the cache without reusing it, which is when Net can show a cost.

​
Date range, bucket and metric

  • Range — Last 7 days, Last 30 days, Last 6 months, or Custom with a start and end date.
  • Bucket — chart the period by Day, Week or Month.
  • Metric — chart Cost or Tokens.
Above the chart, four figures summarize the selection: Total cost, Total tokens, Tokens in / out and Ledger entries.

​
Costs that cannot be shown

A cost cell never shows zero for usage that really happened. Instead it reads:
  • Not priced — the usage ran on provisioned Azure capacity, which is billed as a flat commitment rather than per token. An Unpriced column counts those tokens.
  • Cost unavailable — the usage was charged per token but Parlé could not state its cost, typically because an exchange rate was missing. That is a Parlé issue, not a problem with your account, and a Cost unavailable column counts those tokens.
Both columns appear only when there is something to report, with a note under the totals saying how many tokens the cost figures exclude. See Currencies and Exchange Rates.

​
Exporting Data

1

Set up the view you want

Choose the range, the Group by dimension and any filters. The export contains exactly what the Breakdown table shows.
2

Download

Select Download CSV above the table.
Each row in the export is one group (a project, user, provider, model, or one of the Auto groupings), followed by a TOTAL row:
The cost column is headed Cost, not Cost USD — the Currency column says what it is in. Read columns by header name, not by position. Several of the columns above — Project Code, Cached Tokens In, Reasoning Tokens, Unpriced Tokens, Unconverted Tokens and the four prompt cache columns — appear only when there is something to report, so the same report can have a different number of columns from one period to the next. The columns always come in the order shown in the table, so their relative order never changes, but a column’s absolute position can: Currency sits one place further along in a month where a provider first reports cached tokens.
Export monthly, grouped by project, and import by project code into your practice management system to automate client AI cost recovery without manual reconciliation. If you automate the import, match columns on their header names (Project Code, Cost, Currency) rather than on column numbers, and skip the final TOTAL row so the period is not counted twice. See Expense Codes.

​
Foundry reconciliation statement

If you bring your own Azure AI Foundry deployments, the Foundry reconciliation statement sits below the usage report. Pick a Month to see what Parlé metered on each deployment, meter by meter, laid out to sit beside your Azure invoice; Export CSV downloads it. This statement deliberately stays in the units your Azure invoice is billed in:
  • Each line is in the currency its deployment is quoted in, not your reporting currency, and nothing on it is converted. The statement says so — for example “Figures are in GBP — the currency your deployments are quoted in, so they are the amounts to hold against your Azure invoice…”
  • Each deployment’s heading names its currency. If a deployment’s quote currency changed during the month, the heading reads “Quoted in more than one currency — see each line”, and every meter line still states its own.
  • Where any line is not in US dollars, the period total and the per-project cost column are withheld, with a note explaining why. Adding amounts in different currencies would not be meaningful.
  • In the CSV, the columns are headed Rate per 1M Tokens and Cost, and a Currency column at the end gives each line’s currency.

​
One Canonical Ledger

The Usage screen and Parlé’s budget enforcement read from the same underlying token ledger, using the same month-by-month conversion. The spend a project’s Budget tab shows against a cap is the same number that triggers the 80% warning and the pre-flight block. There is no lag, no reconciliation gap, and no difference between “reported spend” and “enforced spend.” When a project shows £420.00 of a £500.00 monthly cap used, enforcement blocks at £500.00 — not at some independently calculated figure.
Token counts reflect the tokens the provider reported for each response. Parlé does not mark up, round, or adjust token counts.
In a billable-hours environment, the number on your usage export needs to match the number on any client invoice or write-off memo. Because Parlé uses one canonical ledger for both reporting and enforcement, you can export usage data and use it directly in billing workflows — with confidence that the numbers reconcile with the limits your team experienced during the engagement.