Claude Enterprise gets team budgets and “off-topic” flags

Anthropic on October 28 made smart reports generally available on Claude Enterprise: Claude reads a sample of a team’s transcripts, sorts the work into workstreams and puts spend, deliverables and friction points against each one, Anthropic said in its announcement. Mashable reported that the same reports list sessions with no usable output, and sessions classed as “personal or off-topic,” as “inefficiencies” for whatever team a report covers. Anthropic says the reports aren’t for evaluating individual performance or making employment decisions.

In-app surveys and pooled group budgets became generally available the same day. All three depend on configuration, including for organizations with a HIPAA configuration or customer-managed encryption keys, and eligibility and setup steps are in the Help Center. Organizations get up to 10 smart reports a month for free and pay extra fees after that, according to Mashable.

What a report covers

A report starts from three choices: a team, a date range and the products to include. Mashable put the window at up to the last 28 days and the products at chat, Claude Code, Claude Cowork and the “unified chat and Cowork experience.” Claude Duet isn’t available on Enterprise. Friction points come as concrete cases, such as a missing connector or a failed tool call. The report ends with one list of recommendations, for example recurring workflows to turn into skills and rules to add to a code repository. Admins can ask their own questions about model usage and cost and save them as templates for later reports. If a spend limit is holding someone back, an admin with billing permission can raise it from the report.

Owners can let team leads who aren’t admins run reports for their own groups, and any finding can be shared as an artifact with the people who can act on it.

Down to the session

Mashable says the charts are interactive: clicking a workstream opens its sessions, which can be ranked and filtered by cost, product, date and type of output. Reports also list the obstacles people ran into and how often, next to the inefficiencies. The outlet linked the feature to worker unease. It pointed to a recent survey in which one in four U.S. managers said they use AI for layoff decisions “often or all the time,” and to an allegation from a group of former Meta employees in July that Meta relied on improperly tested AI tools to evaluate workers’ productivity and track their token use.

The grouping step looks a lot like Clio, a system Anthropic describes as privacy-preserving analysis of real-world Claude use, which it renamed Anthropic Insights on August 24. In Clio, Claude pulls facets such as topic and language from each conversation, clusters similar conversations by theme and writes summaries that leave out private details. A topic has to reach a minimum number of unique users or conversations before it shows up. The Clio paper describes analyzing aggregated patterns across millions of conversations without human reviewers reading the raw text. A smart report covers a single team, and its sessions are one click away.

One budget, two limits

A pooled group budget gives a group one shared monthly budget. The point is that heavy users are less likely to hit an individual limit early in the month while colleagues leave budget unspent. Member limits still act as a guardrail against unusually high individual use: every request counts against both, and a member stops at whichever runs out first. A bigger pool on its own doesn’t lift anyone past their member limit.

Someone who hits a limit can ask for more from inside Claude, and the request goes to admins for review. During setup, Claude suggests a budget amount based on the group’s recent usage, and admins with billing permission get email alerts at 50%, 75%, 95% and 100% of the budget. The caps and spend alerts Claude Enterprise added in July tracked spend by group and user. The shared pool is what’s new.

Surveys after a reply

In-app surveys appear to members as a short card after a reply, in the tool they’re already working in. Admins decide who gets a survey, when it runs and where it appears, can edit or remove questions, and can export responses to a spreadsheet. Surveys can now be turned on for organizations that use customer-managed encryption keys. Where an organization sets one retention period for all data, survey responses are deleted on that schedule. Anthropic treats surveys as customer data and says it accesses them “only as necessary to operate the service.”

The default questions are about value: the most valuable thing someone has done with Claude recently, how Claude changed the time that work took, which use a teammate should copy and, if the team lost access tomorrow, what would break or slow down first.