Wain AI/Tech Blog

AI news and trends worldwide, updated nearly every day

Claude Enterprise Makes Shared Group Budgets and Usage Analysis Reports Generally Available - Individual Limits Stay, and Whichever Runs Out First Stops the User

Claude Enterprise Makes Shared Group Budgets and Usage Analysis Reports Generally Available - Individual Limits Stay, and Whichever Runs Out First Stops the User

On October 28, 2026, Anthropic made pooled group budgets, smart reports, and in-app surveys generally available in Claude Enterprise. A group can hold one monthly budget, and each request counts against both that budget and the member's own limit. Smart reports summarize a team's work and spend from a sample of transcripts, but they are off by default and unavailable to organizations using CMEK or a HIPAA configuration.

On October 28, 2026, Anthropic made three admin features generally available in Claude Enterprise1: pooled group budgets, which let a group share one monthly budget; smart reports, in which Claude analyzes what a team does with it and what that costs; and in-app surveys, which put short questionnaires in front of users.

All three are tools for the people who manage Claude once usage spreads across an organization and the questions become who is using how much, and where the money goes. This article focuses on pooled group budgets, which change how budgets are held, and smart reports, which read into what people actually do, and sets out how each works and where the operational limits are.

One budget for the whole group

In Claude Enterprise, a group spend limit gives every member of the group the same monthly cap, applied to each person separately3. Anthropic explains that pooled group budgets are meant to reduce an imbalance: heavy users run into their personal cap well before the month is over while budget assigned to others goes unused1.

A pooled group budget gives the group one common pot of money on top of the personal caps. It works as follows3:

  • Both limits apply: every request is charged to the member’s own monthly cap and to the group’s shared budget, and the member is stopped by whichever is exhausted first. The personal cap stays in place as a check on any one person using far more than the rest
  • An empty pool stops usage: members with no other shared budget available cannot continue until an admin raises the amount or the budget resets with the new month
  • More is granted on request: a user who reaches a limit can ask for more from within Claude, and the request is routed to admins to review. The group’s name and the budget figure stay hidden from users
  • Alerts: people with the Billing permission are emailed as spending crosses half, three quarters, 95%, and all of the budget

Eligible plans are Enterprise plans billed by usage, plus Standard and Premium seat plans that have usage credits turned on. On seat plans, the shared budget covers only the usage-credit portion; usage included with a seat does not count toward it, and members can keep using their seat’s included usage after the shared budget runs out.

The help article cautions that setting the member cap to the budget divided by headcount turns the arrangement back into per-person limits and defeats the purpose of sharing. As a first setting, it suggests roughly three times the group’s spend last month for the shared budget, and roughly three times the heaviest user’s spend for the member cap.

There are operational constraints too. Because the Billing permission covers the entire organization, anyone able to edit one group’s shared budget can edit all of them. Shared budgets are also configurable only from the Organization settings screen; the Admin API does not support them yet. Organizations that automate budget settings will have to use the settings screen for now.

Managing AI usage through budgets is not unique to Claude. The same day, October 28, GitHub announced that Enterprise Cloud licenses will include Copilot, with AI usage charged against the company’s budget from the first request. The further pricing moves from paying per seat to managing actual usage against a budget, the more deciding how to divide that budget becomes part of an admin’s job.

Smart reports read transcripts

Smart reports show what a team accomplished with Claude and how much it cost1. An admin chooses which team, which period, and which products to include; Claude then reads a sampled subset of that team’s transcripts, sorts the work into workstreams, and shows the spend for each workstream, the deliverables produced, and recurring friction such as a connector that was never set up or a tool call that failed. Costs are calculated at list price2.

Each report ends with a list of recommendations, such as repeated workflows worth turning into skills and rules worth adding to a code repository1. It also flags people whose spend limit is holding up their work, and an admin who holds billing rights can lift that limit directly from the report (according to the help article, when viewing the report in its attributed mode). Owners can also hand report creation for a group to that group’s lead without making the lead an admin.

Because Claude reads the content of users’ conversations, privacy is a major factor in deciding whether to use the feature. The help article describes the following design2:

  • Smart reports are off by default, and a Primary Owner or Owner has to turn them on
  • The setting that lets reports tie sessions to named individuals is also off by default. With it off, names appear as “User” in the report and session IDs are hidden. Even with it on, reports still show “User” by default, and every time someone viewing the report unmasks names, it is recorded in a log
  • For restricted topics, including protected characteristics, health and medical matters, and union activity, admins cannot create custom questions, and matching sessions feed only the aggregate analysis without summaries. These two guardrails are always on and cannot be disabled

Anthropic states in both the announcement and the help article that smart reports exist to help understand adoption and plan investment in Claude, and are not a tool for judging individual performance or for employment decisions1.

Some organizations cannot use them. Smart reports are off-limits to organizations that hold their own encryption keys (CMEK), that have a HIPAA setup, or that host inference in their own cloud account under a hybrid arrangement2. Where Claude Code runs with zero data retention, reports are limited to chat and Claude Cowork, with Claude Code excluded. On the zero-data-retention front, Anthropic also announced Enterprise Frontier Safeguards in September, an approach meant to combine zero data retention with misuse detection.

The number of reports is capped as well. Each organization gets up to 10 reports a month at no charge, and a report counts against that number as soon as it is waiting in line or being built (failed ones do not). The date range can reach back as far as the last 28 days, and a report usually takes a few hours to produce.

In-app surveys collect users’ views

In-app surveys show up as a small card beneath a Claude reply, in the same tool the user is already working in1. Out of the box, the questions cover the most useful thing a person has recently done with Claude, how much Claude shortened or lengthened that work, which use a coworker ought to adopt, and what would stop or slow down first if the team lost Claude the next day. Admins can rewrite or drop questions, pick the recipients, the timing, and the placement, and download the answers as a spreadsheet.

Organizations that rely on their own encryption keys can now enable these surveys as well. Where an organization applies a single retention window to everything it stores, survey questions and answers are purged on that schedule too.

What admins should check

Claude Enterprise has been gaining admin features steadily, including inference hooks, which send prompts through the company’s own servers for inspection before they reach the model, added in beta in August. The three features covered here address the cost and usage side of that work.

Admins considering them should check the following first:

  • Whether their plan qualifies for pooled group budgets (a usage-based Enterprise plan, or a seat plan with usage credits turned on)
  • Who should hold the Billing permission that allows editing shared budgets (the permission applies across the whole organization)
  • Whether to turn on smart reports, and if so, whether to let reports name individual users, decided against labor agreements and internal policies
  • Whether their own encryption keys, a HIPAA setup, or a hybrid deployment rules out smart reports for them altogether

Sources

  1. Smart reports, surveys, and pooled group budgets for Claude Enterprise - Anthropic official announcement (October 28, 2026)
  2. Get started with smart reports - Claude Help Center
  3. Manage pooled group budgets on Enterprise plans - Claude Help Center

We publish the latest AI news nearly every day.

Subscribe via RSS Get new posts the moment they go live.

Search other keywords →