Platform

Is the AI spend actually returning anything?

Nearly every organisation now has a Copilot, Cursor or Claude line on the budget and no defensible answer about what it produced. Pacia pulls real usage from each provider’s own API — seats, active users, acceptance, cost — and sets it against the cycle time, size and rework of the work those tools were involved in.

AI adoption — last 30 days
Seats assigned
44 across 3 providers
Active in the last 7 days
38 of 44
Never activated
6 seats
Cost, month to date
Tracked per provider
Cycle time, AI-exposed work
Compared, labelled as correlation

What the AI views cover

Usage on one side, delivery evidence on the other

Pacia compares AI-exposed work with the rest over the same period and the same measures. The observed difference is useful evidence, provided it remains labelled as correlation rather than causation.

How AI spend is set against delivery evidence

Provider analyticsSeats, active users, acceptance and cost where available
Same period, same delivery measures
AI-exposed workCycle time · size · rework
Other workCycle time · size · rework
Observed differenceSample shownCorrelation, not causation

The comparison can support an investment decision without pretending to prove that a tool caused the difference. Provider coverage and reporting lag remain visible beside it.

How the provider connections work

  • GitHub Copilot

    Read from GitHub’s own organisation usage API — seats, active users and acceptance — rather than inferred from commit patterns.

  • Cursor

    Team usage and adoption pulled from Cursor’s team API, sitting in the same trends as the other providers.

  • Claude — two different key types

    A Claude Console organisation uses an Admin API key (usage and estimated cost). A Claude Enterprise organisation uses an Analytics API key with read:analytics (seats, daily active users, sessions, accepted and rejected tool actions, lines changed, commits, pull requests). The two are not interchangeable; Pacia detects which you have entered and validates it against the matching endpoint.

  • Reporting lag is respected

    Console data can lag around an hour. Enterprise data lags longer, so Pacia syncs through three days before the current date rather than showing a half-populated today as a real drop.

  • What is genuinely not covered

    Anthropic’s organisation APIs do not include Claude usage hosted through Amazon Bedrock, Google Vertex AI or Microsoft Foundry. If that is where your usage runs, it will not appear, and we would rather say so up front.

  • Keys are encrypted and never returned

    Every provider key is encrypted at rest and never included in any API response. Enter it only in the provider password field — never over email or chat.

Common questions

Can Pacia prove AI made us faster?

No, and neither can anything else. What it can do is show whether AI-exposed work has different cycle time, size and rework characteristics from the rest, over a real sample, with the sample size visible. That is a correlation and Pacia labels it as one. It is enough to make an investment argument honestly; it is not proof.

Do you need access to prompts or generated code?

No. Pacia reads the aggregate organisation analytics each provider exposes. It does not see prompts, completions or generated code.

Which AI providers are supported today?

GitHub Copilot, Cursor and Claude Code, including both Claude Console and Claude Enterprise organisations. Additional providers are added as their organisation analytics APIs become available.

Find out what your AI tooling is actually producing