How Claude aims to give companies a clear view of what AI costs and delivers

What do we actually get out of using AI at work, and what does it cost? That is the question Anthropic wants to help companies answer. The company has made three new features generally available in Claude Enterprise, the version of the Claude chatbot aimed at large organizations.

Availability varies by configuration, including for organizations with a HIPAA configuration (US rules for health data) or customer-managed encryption keys. The exact availability is not specified in the announcement, which refers to the Help Center.

A shared budget instead of individual caps

With pooled group budgets, a group gets one shared monthly budget. According to Anthropic, this makes heavy users less likely to hit a limit early in the month while others leave budget unused.

Limits for individuals still exist as a safeguard against unusually high usage. Every request counts toward both the member's budget and the group's budget, and users are stopped at whichever limit is reached first. If you hit a limit, you can ask for more from within Claude, and the request goes to an administrator.

The setup suggests an amount based on the group's recent usage. Administrators with billing permissions receive email alerts at 50, 75, 95 and 100 percent of a budget.

A report on the work and the bill

A smart report shows what a team accomplishes with Claude and what it costs. You choose the team, date range and products. Claude analyzes a sample of the team's transcripts (the saved conversations) and sorts the work into workstreams. Anthropic does not generally disclose how large the sample is.

The report shows usage per workstream, delivered results and the most common friction points, for example a missing connector (a link to another system) or a failed tool call. It also compiles a single list of recommendations, including recurring workflows that could become skills and rules that could be added to the code repository.

Administrators with billing permissions can raise a spending limit directly from the report. Owners can let team leads without administrator permissions run reports for their own groups.

Anthropic writes that the reports are meant to help organizations understand adoption and plan investments. They are not intended for assessing individual performance or for hiring decisions.

Employees' own view

In-app surveys ask the team directly. The default questions cover the most valuable thing someone has done with Claude, how their time use has changed, which use a colleague should copy, and what would grind to a halt if access disappeared. Administrators can customize the questions. Responses can be exported to a spreadsheet.

Surveys can now also be enabled for organizations with customer-managed encryption keys. If an organization has a single retention period for all data, surveys and responses are deleted after that period. Anthropic states that the surveys are treated as customer data.

Praise from customers

On Anthropic's own site, customers such as DoorDash and Docusign say the reports give an overview of AI usage and help with prioritization. These are Anthropic's own selection of quotes, and the features have not been independently tested.