Claude Enterprise launches smart reports, in-app surveys and pooled group budgets for admins
Anthropic has added three administration features to Claude Enterprise that address a question many companies now ask after the first year of broad AI deployment: what the money is actually buying. Announced on October 28, 2026, smart reports use Claude itself to analyze a sample of a team's sessions and turn them into workstreams, costs, deliverables and friction points; in-app surveys let administrators ask employees directly what Claude changed in their work; and pooled group budgets give a whole team one shared monthly spending amount instead of forcing administrators to split money into rigid per-person limits.
Anthropic's announcement describes all three features as generally available for Claude Enterprise, although availability depends on how an organization is configured. Companies using customer-managed encryption keys or a HIPAA configuration face restrictions, and the help center pages for surveys and pooled budgets still describe those two features as beta at the time of writing. None of the three features carries a separate price in Anthropic's announcement; smart reports come with a monthly allowance, and the budgets control spending that organizations already pay for under their Enterprise agreement.
The release continues a series of spending and visibility controls that Anthropic has been building for enterprise administrators. In July 2026 the company added richer admin analytics, model-level entitlements and spend alerts at 75% and 90% of an organization-level limit, and on October 29 it launched a Usage and Cost Analytics API in beta for Claude Console and Enterprise organizations. The new features move one step further, from counting tokens and dollars to describing what the work was.
··········
HOW SMART REPORTS ANALYZE TEAM USAGE AND COST.
A smart report starts from a scope chosen by the administrator: a team, group, department or cost center, a date range of up to the last 28 days, and the products to include, which can be chat, Claude Code, Claude Cowork or the unified chat and Cowork experience. Claude then reads a sample of the transcripts in that scope, and the setup screen shows an estimated sample size before the report is created. According to Anthropic's documentation, a report usually takes a few hours to generate, and the requester receives an email when it is ready.
The output groups sessions into workstreams, meaning the kinds of work a team used Claude for, and attaches spend at list price to each one. The report also shows the deliverables produced, grouped by output type such as analysis, documentation, content drafts and code, the average cost per session for each output type, the outcome of each task, and the most expensive sessions. A dedicated section lists the most common pain points by category, including a missing connector, output that did not match the request, approval or sign-in gating, tool failures and rework loops. Another section counts sessions that produced nothing usable or were personal or off-topic, but only as aggregate figures, without summaries or drilldowns.
Smart reports also rank complex and autonomous work by task complexity, estimated time saved, autonomous run time and required expertise, and they flag work that is new to the organization as well as cases where employees used Claude to automate parts of their own jobs. Each report ends with a consolidated list of recommendations, which can suggest turning recurring workflows into shared skills, highlight examples of complex work worth copying, or propose rules to add to a code repository. Recommendations can be shared as an artifact, and administrators with billing permission can raise the spend limit of people whose work is being held back directly from the report.
Administrators can go beyond the standard sections with custom questions about model usage and cost. Each custom insight is a question that Claude answers for every analyzed session, with answers formatted as categories, yes or no, or a scale, and the answers are shown with their share of sessions and spend. Questions and insights can be saved as templates for later reports, and supporting files can be attached to give Claude context about the team.
··········
WHO CAN RUN SMART REPORTS AND WHAT THE LIMITS ARE.
Smart reports are off by default. An Owner or Primary Owner has to enable them under Organization settings, Capabilities, in the Analytics section. Once enabled, Primary Owners, Owners, Admins and custom roles with Analytics view access can create and view reports. Anthropic also added delegated access, which lets team leads or department heads run reports for their own groups, departments or cost centers without becoming organization administrators. A delegate can be assigned up to 50 groups, departments or cost centers of each kind, never gets a whole-organization option, sees only reports they ran or that were shared with them, and cannot rename, archive or delete reports.
........
Smart reports setting | Value |
Plan | Claude Enterprise |
Default state | Off, enabled by an Owner |
Reports included | 10 per organization per month |
Date range | Up to the last 28 days |
Products | Chat, Claude Code, Claude Cowork |
Delegate scope | Up to 50 groups, departments or cost centers per type |
Typical generation time | A few hours |
Not available with | CMEK, HIPAA configuration, hybrid deployment |
........
Each organization can run up to 10 reports per month at no charge. Queued and in-progress reports count toward that allowance, failed reports do not, and the counter resets on the first day of each calendar month; organizations that need more must contact their Anthropic account team. Delegates draw from the same monthly allowance as administrators, so a company that hands report access to many team leads will need to coordinate how those ten reports are used.
The feature is not available to Enterprise organizations that use customer-managed encryption keys, a HIPAA configuration or a hybrid deployment that runs inference on their own cloud provider. Organizations that use zero data retention for Claude Code can still run reports, but those reports cover chat and Cowork and exclude Claude Code sessions.
··········
PRIVACY GUARDRAILS AND INDIVIDUAL ATTRIBUTION.
Because smart reports read real employee transcripts, the privacy design matters as much as the analytics. Anthropic states in both the announcement and the help center that the reports are meant for understanding adoption and planning investment, and that they should not be used for evaluating individual performance or making employment decisions. That is a usage policy statement rather than a technical restriction, but the product does include controls that limit how much individual detail a report exposes.
Attribution to individual users is off by default. With it disabled, drilldowns show a generic "User" label and hide session IDs. If an Owner turns on the "Allow attribution to individual users" setting, viewers can reveal names and session IDs through an in-report control, and each reveal is logged. Some guardrails cannot be disabled at all: custom questions cannot target restricted topics, sessions identified as involving those topics are used only in aggregate without a session summary, and personal conversations are excluded from individual sessions and names. The restricted list covers protected characteristics, health and medical information, union activity, harassment, whistleblower and legal matters, political or religious views, personal life including family, relationships and personal finances, and sentiment about a specific named person.
Reports can be shared with individual users through a sharing link, which only the named recipients and administrators who can already view reports are able to open. Administrators can archive and then delete any report.
··········
IN-APP SURVEYS FOR QUALITATIVE FEEDBACK.
The second feature complements the transcript analysis with direct answers from employees. Administrators create surveys under Analytics, Surveys, set a start and end date, choose an audience of all members or specific groups, and decide whether the survey appears in Cowork, in chat, or in both. Members see a short card at a natural break in their work, after a reply, and Anthropic's documentation says a survey never interrupts a running task. Each user sees a given survey once, dismissing the card counts as declining it, and completing it takes about 90 seconds.
Four default questions ask about the most valuable recent task done with Claude and how the outcome improved, whether Claude changed the time the work took, which practice a teammate should copy, and what would break or slow down first without Claude. Administrators can edit or remove questions before launch, and a reach panel estimates audience size, active users and expected responses. Results appear as they arrive and can be exported as a CSV file with one row per response, including the submission time, email, group, surface and one column per question.
Surveys are not anonymous in the sense that the export includes each respondent's email and group, and the card tells users that answers go to their organization's administrators, not to Anthropic. Anthropic says it treats survey content as customer data and accesses it only as necessary to operate the service. The announcement states that surveys can now be enabled for organizations using customer-managed encryption keys, while the help center article still says CMEK customers cannot use them during the beta, so organizations in that configuration should check their admin console before planning a rollout. HIPAA customers are told not to put regulated data such as protected health information into survey responses.
··········
HOW POOLED GROUP BUDGETS CHANGE SPEND LIMITS.
Pooled group budgets address a practical problem with per-person caps. When an administrator divides a team's money into equal individual limits, the heaviest users run out early in the month while colleagues leave their allocation unused. A pooled budget adds one shared monthly amount for the whole group on top of the existing per-member limit, so the team's money follows actual usage.
........
Spend control | Before pooled budgets | With pooled budgets |
Group limit | Same per-member monthly limit for every member | Shared monthly pool plus a per-member guard rail |
When a user stops | Their individual limit runs out | Their own limit or every usable pool runs out |
Admin alerts | 75% and 90% of the org-level limit | 50%, 75%, 95% and 100% of each pool |
Setup | Amount entered manually | Suggested amount from recent usage |
Admin API | Limit monitoring and increase requests | Pools managed in settings only |
........
Every request counts against both the member's individual limit and the group pool, and a member stops at whichever runs out first. When a pool is exhausted, usage pauses for all members of that group until an administrator raises the budget or the new month begins. Users do not see the pool amount or the group name; they see a message with a "Request more" button, and the request goes to administrators for review. Primary Owners, Owners and custom roles with the Billing permission set to "Can manage" can configure pools, and because that permission applies across the organization, anyone who can edit one group's pool can edit all of them.
Anthropic's help center gives concrete guidance for the first setup. It suggests a pooled budget of about three times the group's spend in the previous month and a member limit of about three times the top spender's spend. In its example, a group that spent $1,450 with a top spender at $310 would get a $2,900 pool and a member limit of about $950. The documentation also suggests raising the pool by 50% if it is 75% used before the 20th of the month, and it warns against dividing the pool by the number of members, which would simply recreate individual limits.
Users who belong to several groups draw from the largest pool first by default, though administrators can define a custom order with the most specific team first. A pool can be frozen by setting it to $0, paused while keeping its amount, or removed entirely. One limitation is relevant for larger IT teams: pooled budgets can only be managed in Organization settings and are not available through the Admin API, so organizations that automate spend management with scripts will need to handle them by hand for now.
··········
WHAT CHANGES FOR CLAUDE ENTERPRISE ADMINISTRATORS.
For an organization already on Claude Enterprise, the first concrete step is enabling smart reports, since they are off by default, and deciding whether individual attribution should stay off, which is the default and the safer choice for most works councils and privacy teams. Companies that want department heads to see their own data without broader admin rights can use delegated access, but they should plan the shared allowance of ten reports per month, which can be used up quickly across many teams.
The pooled budgets change mostly affects teams with uneven usage, such as engineering groups where a few people run long Claude Code or Cowork sessions while others use chat occasionally. Moving those groups from equal per-member caps to a shared pool with a higher individual guard rail should reduce the number of blocked users without raising the total budget, and the alerts at 50%, 75%, 95% and 100% give billing administrators earlier warnings than the previous organization-level thresholds. Regulated organizations should check eligibility first: smart reports are unavailable with CMEK, HIPAA configurations and hybrid deployments, and the help center still labels surveys and pooled budgets as beta.
··········
WHAT IS ESTABLISHED AND WHAT REMAINS OPEN.
The documented facts are specific: smart reports are available on Claude Enterprise with an allowance of 10 reports per organization per month, a 28-day maximum window, delegated access for up to 50 scopes per type, and fixed privacy guardrails; surveys run in chat and Cowork with four default questions and CSV export; pooled budgets add a shared monthly amount on top of per-member limits, with alerts at four thresholds and no Admin API support. The customer quotes in Anthropic's announcement, including Wix describing reports based on tens of thousands of sessions, are vendor-selected testimonials rather than independent measurements.
Several points remain open. Smart reports rely on Claude's own classification of sampled transcripts, and Anthropic has not published accuracy figures for how well workstreams, outcomes or estimated time saved match reality, so the cost per deliverable and the "inefficiencies" counts should be read as estimates. The announcement and the help center also disagree on the status of surveys and pooled budgets, generally available in one and beta in the other, and on CMEK support for surveys. Finally, the guidance that reports must not be used for performance evaluation depends on how each customer behaves, which makes internal policy, and not only product settings, part of a responsible rollout.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
