Anthropic's admin console and analytics APIs give organisations daily aggregates: active developers, sessions, spend estimates and per-user accept rates. They exclude individual Pro and Max subscription seats entirely, and none of them can see inside a session. For team-level questions about who ships what, you need the session record itself.
What the official tooling provides
Anthropic ships a capable first layer. The admin console's usage and value tabs show org-wide active developers, session counts and estimated productivity figures. The Claude Code analytics API adds daily per-user aggregates: estimated cost, sessions, lines of code, commit counts and tool accept rates. Cost tools such as Datadog and CloudZero already consume these feeds. For finance-level questions, this layer is genuinely useful.
The two structural gaps
Subscription seats are invisible
Every official admin surface covers organisation accounts. Individual Pro and Max subscriptions, which is how a large share of real engineers actually run Claude Code, appear in none of them. Anthropic says so plainly in its own documentation. If half your team codes on personal Max plans, your admin dashboard reports half a team.
Aggregates cannot answer team questions
Daily totals tell you the organisation used the tool. They cannot tell you who is carrying the critical work, which spend produced shipped features, whether anyone is stuck in a rework loop, or who validates model output before committing it. Those answers live inside sessions: the prompts, the tool calls, the files touched, the sequence of what happened. Officially, that level is simply not exposed.
Measuring from the session record instead
The session data exists on every developer machine, and it is far richer than any aggregate. Reading it directly, with consent and on company projects only, changes the questions you can ask:
- Coverage: subscription seats and API accounts land on one ledger, because measurement no longer depends on billing APIs.
- Attribution: cost can follow the work, down to the feature that shipped, instead of stopping at a monthly total.
- Behaviour: steering, validation and rework become visible patterns rather than guesses.
The trade-off is responsibility. Session data is sensitive, so the collection scope has to be narrow, verifiable and symmetric: company repositories only, employees able to see and delete their own record, every managerial read logged. Done that way, session-level measurement answers the questions the official aggregates were never designed for.
Frequently asked questions
Does Anthropic offer any session-level analytics?
No. The console and analytics APIs expose daily aggregates and per-user counts. Session content and structure stay on the machines where the work happened, which is where session-level tools read them, with consent.
Can I see Pro or Max subscription usage in the admin console?
No. Anthropic states that individual subscription accounts are unavailable to the admin APIs. Only organisation seats appear.
Is reading local session data allowed?
The data belongs to your organisation when the work happens on company projects under a policy your team has agreed to. The key is transparent, scoped collection: company repositories only, with employees able to inspect and delete their own data.