Skip to main content

FinOps Analytics

/usage is the unified operational and FinOps analytics workspace. Every view uses the same time range and filters, so a model, provider, key, deployment or attribution selection stays consistent while you investigate.

Open the page​

Select FinOps Analytics in the main sidebar or open /usage. Every signed-in role can access FinOps Analytics. Returned records are constrained by role and by the active organization. /reporting-drillthrough redirects here, to the Attribution view, and keeps its query parameters.

The URL keeps the view (tab), period, bucket and filters, so an investigation can be bookmarked or shared with another user who has access to the same data.

Page layout​

  • Views panel: a side panel titled FinOps Analytics that lists the views in three groups, followed by Saved views.

    • Analyze: Cost overview, Usage & reliability, Model analysis, Client key analysis, Deployments & providers, Usage heatmap.
    • Allocate: Attribution, Subscriptions.
    • Control: Guardrails.

    Collapse views panel hides it and Show views panel brings it back; the browser remembers the choice. Below 1180 px wide, the panel and its saved views are hidden, and a View drop-down in the toolbar switches views instead.

  • Header: a FinOps Analytics / <group> breadcrumb and the view title.

    • An amber count appears when the range has data issues. Select it to list them: unpriced requests link to model pricing, and up to five detected anomalies are listed.
    • Refresh reloads the data.
    • Export downloads the current view as a PDF or PowerPoint report. Preview last report shows the most recent export.

  • Toolbar: date range, the resolved bucket, search and Filters. Active filters appear below it as removable chips. See Time, filters and saved views.

Views​

  • Cost overview: the default view.
    • Cost intelligence summary: recorded spend, requests and tokens, with a per-bucket spend strip.
    • Links to unpriced requests and to company subscription charges, shown as a separate ledger and never added to gateway spend.
    • Unit economics: Cost / 1,000 requests, Cost / 1M tokens and Input served from cache, compared with the previous equal period.
    • Where the money goes: a provider donut. Select a provider to open its requests.
    • What your context contains: uncached input, cached input and output tokens.
    • Shortcuts to Model economics, Client key spend, Cost attribution and Tokenomics. Tokenomics opens with the model and agent filters carried over.
  • Usage & reliability: KPI tiles for requests, average latency, error rate and tokens.
    • Usage explorer: switch between Spend, Requests and Tokens; it compares the total with the previous period, marks anomalies, and opens a bucket's requests on selection.
    • Also includes Latency distribution (P50/P95/P99), Error rate over time, Cost vs tokens and Top endpoints by requests.
    • Usage anomaly detection lists cost, token and error-rate spikes against the range median. Each spike links to its requests.
  • Model analysis: model KPI tiles; Top models overview, where you select a model to open its requests.
    • Also includes Model usage share over time, Cost vs reliability vs volume, Insights & recommendations, Latency distribution by model, Token efficiency and Model reliability.
  • Client key analysis: key KPI tiles, including keys near their budget limit; Key insights & governance alerts.
    • Spend by client key: select a key to filter the page.
    • Also includes Top key usage over time, Top keys overview, Budget consumption trend, Key cost vs reliability and Latency distribution by key.
  • Deployments & providers: routing totals; Provider routing flow (from client keys to models to providers).
    • Also includes a Providers table, Spend by deployment, Latency distribution by provider, Error rate by provider and a Deployments table.
  • Usage heatmap: tiles for Busiest hour, Quietest window, Business hours share and Weekend share. A weekday × hour grid in UTC shows requests, cost or failure rate. Select a populated cell to inspect that time slice in Request Logs.
  • Attribution: choose Attribute by (organization, team, application, user, customer, cost center, environment, agent, deployment, provider or model), then Run report.
    • KPI tiles: Attributed spend, Unallocated spend, Chargeback accuracy, active cost centers, teams and applications.
    • Numbered charts run from spend over time to a chargeback statement preview.
    • Select a group for Underlying requests: a paged list (10, 25, 50 or 100 per page) with the total match count and links to Request Logs.
    • Export CSV downloads provider cost, billed cost, requests and tokens per group. Export produces the statement as PDF or PowerPoint.
    • Platform admins also see scheduled delivery and rollup maintenance; creating schedules requires a write-capable role.
  • Subscriptions: company-paid subscription finances and automatic organization API coverage.
    • Verified invoices are allocated over service periods; current monthly commitments stay separate.
    • The report includes provider usage by connection/product, archived invoice history, account/employee evidence and an executive breakdown.
    • Manage accounts & connections opens AI Subscriptions. Cost CSV and Provider usage CSV download the evidence.
    • Personal/private accounts and device data are excluded.
    • Shared account costs are counted once and stay whole-account amounts when filtered by user or team.
    • See Subscription reports for filters and downloads.
  • Guardrails: the Guardrails monitor shows evaluations, blocked requests, pass rate and added latency for the selected scope and period. It includes Where blocks come from and a Guardrail performance table. Export data downloads the selected range.

Chart cards​

Most chart cards have:

  • a Chart / Table switch over the same data
  • a download button: an image in chart view, CSV in table view
  • an expand button that opens a larger dialog with the same switch and download

Time, filters and saved views​

The date button opens a popover:

  • Presets: Last 24h, Today, Yesterday, Last 7 days, Last 30 days and Last 90 days.
  • Custom range: From (UTC) and To (UTC) dates, then Apply range.
  • Bucket: Auto, Hourly, Daily or Weekly. Auto uses hourly buckets for short incident windows, daily buckets for normal reviews and weekly buckets for long histories. The toolbar shows the resolved bucket.

Search sits next to the date button.

Filters opens the other filters in three groups, plus an auto-refresh setting:

  • Traffic: models, keys, providers, statuses.
  • Ownership: teams, users, agents, cost centers, tags.
  • Infrastructure: environments, deployments.
  • Auto-refresh: off, or every 15, 30 or 60 seconds.

Select Apply filters to run them. Each active filter becomes a chip; remove a chip to clear that filter, or use Clear all.

Cost-center values come from the effective managed assignment snapshotted on each spend record. Assignment precedence is Application → Team → Organization → User. Governed custom FinOps dimensions are retained in the record's dimension snapshot, but the current Attribute by options contain only the fixed built-in dimensions.

Saved views​

Saved views in the views panel lists My views and Shared with you. The view whose filters are on screen is highlighted. A saved view stores the query, not a copy of the result; opening it reruns the query against currently visible records.

  • + opens Save view with these fields:

    • Name
    • Visibility: Only me, or Share with team for one of your teams
    • Use as my default
    • Auto-refresh: off, 30 seconds, 1 minute, 5 minutes or 15 minutes

    It also summarises the filters being saved.

  • Opening /usage without parameters loads your default view.

  • When one of your own views is applied, Edit view changes its name, visibility, default and refresh. Replace with the filters on screen updates its query. Delete view removes it, including for the team it was shared with.

  • Copy a link to these filters copies a /usage link for the filters on screen.

  • Saving, editing and deleting views requires a write-capable role.

Drill through to request evidence​

  1. Set the time window before interpreting totals, then apply filters.
  2. Choose the view that matches the question:
    • Usage & reliability for when
    • the analysis views for concentration
    • Attribution for ownership
    • Usage heatmap for recurring time patterns
  3. Select a trend bucket, anomaly, heatmap cell, ranked model, provider slice, endpoint or attribution group. FinOps Analytics opens /logs with the time and relevant filters.
  4. In Request Logs, open a row to inspect the request ID, model/provider, key, organization/team/application, customer, tags, token classes, cost, latency and guardrail decision.
  5. Copy the URL or request ID when handing evidence to an application owner, finance reviewer or incident responder.

Long, simple queries use hourly or daily spend rollups. Complex filters automatically use request events. If a rollup is not current, the workspace falls back to raw events rather than displaying incomplete totals.

The rollup generator defaults to hourly, daily, and monthly output. Quarterly rollups are supported but are not generated by default.

Investigate a cost spike​

  1. Select the incident range and use Auto or Hourly buckets.
  2. In Usage & reliability, check whether cost, requests, tokens and errors rose together, and review Usage anomaly detection.
  3. Open Model analysis to find expensive-model concentration.
  4. Open Client key analysis and Deployments & providers to find callers and routing concentration.
  5. Use Usage heatmap to compare the affected weekday and hour with normal traffic.
  6. Select the anomalous bucket or ranked row to inspect individual requests.
  7. Check the data-issues count in the header before treating estimated cost as complete.

Subscription reports​

In Subscriptions, the following apply:

  • the date range
  • provider
  • account, plan and provider search
  • team and user filters

Request-only filters do not apply and are hidden. The bucket control is replaced by Company accounts. Moving into or out of Subscriptions clears request-only filters, provider and search. The Cost overview links to this report without adding the potentially overlapping gateway and provider ledgers.

The executive breakdown has primary and secondary dimensions, explicit shared-department buckets, source-account drill-down and Export breakdown for every matching group. PDF and PowerPoint downloads remove interactive controls, record report scope, and paginate full registered tables with repeated headings. Use Preview last report in the Export menu to inspect report content.

Role and scope behavior​

  • ADMIN and ADMIN_VIEWER can query platform-wide data. Selecting an organization in the header narrows FinOps Analytics to that organization.
  • Other roles are always constrained to SpendLog records attributed to their own user ID. If they select an active organization, both the user and organization constraints apply.
  • Team membership controls which teams a saved view can be shared with; it does not by itself broaden the SpendLog query.
  • Viewer roles can inspect and export permitted results but cannot save views or create Attribution schedules.
  • Filter option lists are derived from records in the current time and access scope. A missing option usually means no visible matching data in that period, not that the entity does not exist.

Why customers use FinOps Analytics​

  • Engineering identifies expensive or failing keys, models, deployments, and recurring traffic windows.
  • FinOps attributes provider and billed cost to accountable business owners and exports repeatable evidence.
  • Application owners see their own consumption without receiving platform-wide access.
  • Security and governance teams correlate guardrail outcomes and anomalies with individual requests.
  • Finance reviewers move from cost-center totals to request evidence, then use Chargeback for accounting-period statements.