# Claude Enterprise Smart Reports: Costs, Privacy and Pooled Budgets

Anthropic’s new reporting and budget controls connect Claude usage to actual work, but measuring output is still different from proving return on investment.

By Paras Madan | 2026-10-29


Anthropic released smart reports, surveys and pooled group budgets for Claude Enterprise on October 28. The update gives companies more ways to see what teams produce with Claude, ask employees about the results and allocate spending across a group instead of treating every user as an identical workload.

The useful change is the connection between those three questions. A company can see heavy usage without knowing whether people are finishing valuable work. It can also impose neat individual limits that interrupt the people doing the most useful work.

## What Claude Enterprise smart reports show

In its [announcement](https://claude.com/resources/articles/smart-reports-surveys-pooled-budgets), Anthropic describes reports that connect workstreams, deliverables and spending, then recommend changes such as reusable skills or fixes for recurring friction. A missing connector or repeated tool failure can explain an expensive workflow more usefully than a total token count.

According to the [smart reports documentation](https://support.claude.com/en/articles/16893491-get-started-with-smart-reports), Claude analyzes sampled transcripts from a selected period of up to 28 days. Reported costs use list prices. Organizations get 10 free reports per calendar month, including queued and running reports but excluding failures.

That makes the report a starting point for investigation, rather than a complete accounting record. **A generated deliverable is not automatically a valuable deliverable.** Teams still need to establish whether an output was used, whether it needed substantial correction and whether the underlying task was worth doing.

For example, finding that account briefs are expensive could justify improving a shared workflow. It would not, by itself, establish that those briefs helped win business.

## Availability and employee privacy

Smart reports are Enterprise-only and disabled by default. They are unavailable with customer-managed encryption keys, HIPAA configurations or hybrid deployments running inference on the customer's cloud. With zero-data-retention Claude Code, reports exclude Code but can cover chat and Cowork.

Individual attribution is separately disabled by default. Revealing names requires an explicit action that is logged. Personal conversations are included only in aggregate, without individual sessions or names. Anthropic explicitly says reports should not be used for individual performance evaluation or employment decisions.

Those boundaries matter because a team-level adoption tool can easily be mistaken for an employee scorecard. Before enabling it, an organization should explain what is visible and who can access it, rather than leaving employees to infer the policy from a new dashboard.

## Surveys add context that transcripts cannot

[Enterprise surveys](https://support.claude.com/en/articles/16764057-create-surveys-for-your-organization) let administrators choose questions, audiences and timing. They appear at natural pauses in chat, Cowork or Claude Code, and responses can be exported. Surveys support customer-managed encryption keys but not HIPAA configurations.

They are not anonymous by default in the sense some employees might expect: the documented CSV export includes the respondent's email and group. Responses go to the organization's administrators.

A survey can help distinguish a technically successful session from a useful result. It also introduces self-reporting bias. Combining reported experience with task evidence is more informative than treating either as a definitive productivity number.

## How pooled group budgets change spending

[Pooled budgets](https://support.claude.com/en/articles/17005973-manage-pooled-group-budgets-on-enterprise-plans) create a shared monthly amount alongside each user's individual limit. Requests count against both. Usage stops when the individual limit is reached or all applicable pools are exhausted.

They are available on usage-based Enterprise plans and Standard or Premium seat plans with usage credits enabled. On seat plans, only usage credits draw from the pool; included seat usage remains available after the pool is exhausted.

**Pooling changes allocation, not the price of inference.** A shared budget can accommodate uneven workloads without giving every employee the same spending allowance. It still needs an owner who can decide which work deserves more capacity.

The next useful test is whether these controls lead to better workflows, rather than merely better explanations of a growing bill. Reporting becomes operationally valuable when someone can take a repeated failure, fix the integration or shared instructions, and observe whether the result improves.

I’m Paras. I work at the intersection of AI, engineering and distribution.

I help teams explain complex enterprise AI products through clear technical and executive storytelling.

[Book a 30-minute call](https://cal.com/paras-madan/30min)