Claude smart reports: Claude reads your transcripts and tallies the costs

On October 28, Anthropic rolled out three new features in Claude Enterprise: Claude smart reports, in-app surveys, and pooled group budgets. All three are generally available on the Enterprise plan, though not in every configuration.

What are Claude smart reports?

They’re reports that Claude writes itself. An admin picks a team, a date range (up to the last 28 days), and the products: chat, Claude Code, Claude Cowork, or chat and Cowork combined. Claude reads a sample of transcripts from that range, groups the work into “workstreams”, and attaches costs at list prices to each one.

Among other things, the report includes (according to the Help Center):

1. The most common use cases, ranked by number of sessions and by spend.
2. Outputs grouped by type (analyses, documentation, content drafts, code), with the average cost per session for each type.
3. Friction points where work gets stuck, such as unconnected connectors or failed tool calls.
4. Sessions that produced nothing useful, shown only as a total count and cost.
5. The most expensive sessions, plus complex, autonomous work.
6. A single list of recommendations, such as which repetitive tasks to turn into shared skills and which rules to add to the repository.

Admins can also ask their own questions about usage and costs, save them as a template, and share any finding as an artifact. A report usually takes a few hours to generate, and the finished one can be downloaded as HTML with interactive charts.

What’s new here from an AI perspective?

The most interesting part is that the model analyzes its own work. Instead of token and session counters, you get a description of what people actually worked on with Claude and what came out of it, plus pointers on what to automate.

Ran Yitzhaki, a developer at Wix, says each report draws on “tens of thousands of sessions” and surfaces the most common problems. John Jannone of DoorDash adds that the reports have become a regular part of his conversations with leaders at the company.

According to TechCrunch, the launch coincides with investor materials in which Anthropic showed an operating profit in the second quarter and 900% net revenue retention, meaning the same customers are spending nine times more than a year earlier. No wonder companies want to know what they’re paying for (especially ahead of Anthropic’s Nasdaq debut).

How do surveys and pooled budgets work?

Surveys are a short card that appears after a Claude response in whatever tool you’re using, and they never interrupt a task in progress. The default questions cover, among other things, your most valuable recent use of Claude and what would break first if the team lost access. According to the Help Center, filling one in takes about 90 seconds, and the answers go to the organization’s admins, not to Anthropic.

Pooled budgets work like this: a group gets one monthly pool, and each member still has their own limit. Every request counts against both, and whichever runs out first stops you. The point is that the most active users don’t get blocked mid-month while their colleagues have budget left over.

The documentation gives a concrete rule of thumb: a pool about 3 times the group’s spend last month, and a personal limit about 3 times the spend of the most active person. In the example, the group spent $1,450 and the most active person $310, so the pool is $4,350 and the limit about $950.

What should you watch out for?

First, privacy. Reports are off by default, and only the organization owner can turn them on. Attributing sessions to specific people is also off by default, and once it’s on, every time a name is revealed it gets logged. Sessions on sensitive topics (health, trade unions, political or religious views, private life) only go into the aggregate analysis. Anthropic stresses that the reports aren’t meant for evaluating employees or making HR decisions (whether every company will stick to that is another matter).

Second, the limits. The free allowance is 10 reports per organization per month. Reports don’t work in organizations with customer-managed encryption keys, a HIPAA configuration, or a hybrid deployment, and with zero data retention for Claude Code they only cover chat and Cowork.

Third, it’s a sample. The report analyzes selected transcripts, not every session, so treat it as a picture of trends rather than an exact accounting.

For people outside large companies, this is more a signal of where the market is heading than something to try out. If you’re just getting started with agents, the Claude Code tutorial is a good place to begin.

FAQ

What are Claude smart reports?

Reports in Claude Enterprise that Claude writes from a sample of a team’s transcripts covering up to the last 28 days. They show what the team worked on, what it cost, where the work got stuck, and what’s worth automating.

How much do smart reports cost?

An organization gets 10 reports a month at no extra charge. Once the limit is reached, you can’t create new reports, and for more you need to contact your Anthropic account manager.

Are smart reports available on the Pro or Team plan?

No. According to Anthropic, all three features are available on the Claude Enterprise plan, and their availability depends on the organization’s configuration.