Anthropic has shipped a set of admin tools aimed at one question every finance team now asks: what are we getting for our Claude bill? On October 28, 2026, the company made smart reports, in-app surveys and pooled group budgets generally available for Claude Enterprise. Smart reports show what work a team does with Claude and what it costs. Surveys collect the team's own view of the value. Pooled budgets let admins set one shared monthly budget per team.
TechCrunch framed the launch as a response to large customers worried about return on investment. That fits what we have seen in token-cost coverage all month, from Claude Code token-efficiency tactics to the Ramp AI Index spend data.
TL;DR: the three tools #
| Tool | What it does | Why admins care |
|---|---|---|
| Smart reports | Pick a team, date range and products; Claude analyzes a sample of transcripts and groups work into workstreams with spend, deliverables and friction | Shows what the money bought and where it leaks |
| In-app surveys | Short cards after a reply asking about value; admins customize questions and audience | Adds the team's own view next to usage data |
| Pooled group budgets | One shared monthly budget per group, with individual limits as a guardrail | Fewer people stuck mid-month while others leave budget unused |
| Availability | Generally available for Claude Enterprise; varies by configuration (HIPAA, customer-managed keys) | Check the Help Center for eligibility |
How do smart reports work? #
According to Anthropic's announcement, you choose a team, a date range and the products to include. Claude then analyzes a sample of the team's transcripts and groups the work into workstreams. For each one you see the spend, the deliverables produced, and the most common points of friction, such as a missing connector or a failed tool call.
Each report ends with one consolidated list of recommendations. These include recurring workflows to turn into skills and rules to add to a code repository. Admins can share any finding as an artifact with the people who can act on it, ask custom questions about model usage and cost, and save questions as templates for later reports. If a spend limit is holding someone back, admins with billing permission can raise it directly from the report.
Owners can let team leads who are not admins run reports for their own groups. The thinking: the people closest to the work should act on the findings. For background on turning repeated workflows into reusable instructions, see our guide to agent skills.
Anthropic includes a guardrail in the copy: smart reports "aren't for evaluating individual performance or making employment decisions." That is a policy statement, not a technical lock. Whether it holds in practice depends on how each company uses the reports, so employers should set their own rules before rollout.
What do the in-app surveys ask? #
Surveys appear as a short card after a reply, inside the tool the person is already using. The default questions focus on value:
- The most valuable thing someone has done with Claude recently.
- How Claude changed the time that work took.
- Which use of Claude a teammate should copy.
- If the team lost access tomorrow, what would break or slow down first.
Admins can customize or remove questions and choose who receives a survey, when it runs and where it appears. Responses show up as they arrive and export to a spreadsheet. Surveys can now be enabled for organizations using customer-managed encryption keys, and for organizations with a single retention period, surveys and responses are deleted on that schedule. Anthropic says surveys are treated as customer data and accessed only as necessary to operate the service.
The last default question is the useful one for budget fights. "What would break first" gives a finance lead evidence of dependence that a raw token count cannot.
What changes with pooled group budgets? #
Budget cap lid on a coin jar representing a Claude Enterprise pooled group budget with individual guardrails
Until now, per-person limits could strand budget: one heavy Claude Code user hits the cap on the 12th while others never use theirs. A pooled group budget gives the whole group one shared monthly budget. Member limits still work as a guardrail against unusually high individual use. Every request counts against both, and a member stops at whichever runs out first.
When someone hits a limit, they can request more from inside Claude and the request goes to admins for review. The setup suggests an amount based on the group's recent usage, and admins with billing permission get email alerts at 50%, 75%, 95% and 100% of a budget.
Practical setup advice, from our reading of the feature:
- Start with the suggested amount, then compare it to your last two months of invoices.
- Keep individual limits meaningfully above the average user, so they catch runaway agent loops rather than normal work.
- Route limit requests to someone who can read the smart report for that team, so approvals are informed.
- Treat the 95% alert as the moment to look at the report, not the 100% one.
Why is Anthropic doing this now? #
TechCrunch cites a Jellyfish report covering 276,000 customers: Claude Code and other Anthropic-model tools account for nearly a third of proposed code changes submitted for review, and spending on them has grown about 10x since the start of the year, now equal to about 2% of total engineering cost. Jellyfish chief scientist Nicholas Arcolano told the outlet most CFOs did not plan for that growth. That squares with earlier research on AI coding agents adding code faster than finished software, where review remained the bottleneck.
TechCrunch also reports, from investor financials circulating this week, an operating profit in Q2 and 900% net revenue retention, which it attributes largely to Claude Code. We have not verified those figures against a filing. TechCrunch's own read is that the reporting tools are a kind of "FDE-lite," a lighter version of the forward-deployed engineering help labs are investing in. The commercial logic is plain: customers who can prove value renew and expand, and customers who cannot prove it cut. Our read of the Microsoft-Anthropic spend claim and Meta and Microsoft trimming internal Claude use shows how real that risk is.
What smart reports cannot tell you #
These are limits we infer from the announcement, not claims Anthropic made:
- Sampling. Reports analyze a sample of transcripts, so rare but expensive workflows may be missed. A Wix engineer quoted by Anthropic says each report is based on tens of thousands of sessions, but that is a customer's description.
- Attribution. Spend per workstream tells you cost, not business value. A cheap workstream can be worthless; an expensive one can be the best investment.
- Model judgment. Claude is classifying work done with Claude. Treat the grouping as a strong hint and spot-check it.
- Privacy culture. Even with the no-evaluation statement, employees who learn their transcripts feed a report may change how they use the tool. Tell them up front.
- Scope. Only Claude Enterprise is covered. Usage on other vendors' tools or personal accounts does not appear.
How should a team read a smart report? #
A report is only useful if someone turns it into a decision. A simple routine, based on the sections Anthropic describes, keeps it practical.
First, read the workstream list top to bottom by spend. The top three workstreams usually account for most of the bill, and each should have a named owner who can say whether the output justified the cost. Second, look at the friction column. Anthropic gives the examples of a missing connector or a failed tool call. Both are cheap to fix and they waste tokens, because an agent that retries a failing tool burns spend without producing a deliverable. Third, take the recommendation list and pick one recurring workflow to turn into a skill and one rule to add to the code repository. Small, concrete changes beat a long improvement plan.
Finally, compare the report with the survey answers. If the report shows a workstream with high spend and the survey shows nobody names it as valuable, you have found either a hidden cost or a use nobody wants to defend. If the opposite is true, with low spend and strong survey praise, that is a candidate to roll out to other teams.
What the customers say #
Anthropic's page carries quotes from enterprise users. DoorDash's AI enablement lead John Jannone said that even being close to how teams use Claude, smart reports "still surfaced things I didn't know." A Docusign quote says the visibility helps show leadership real outcomes and fix connector bottlenecks. Other quoted titles on the page include Monday's head of applied AI and Deloitte's managing director of AI transformation. These are vendor-selected testimonials, so read them as intended use cases rather than independent evaluations.
What to do this week #
- If you run Claude Enterprise: pick one team with high spend, run a smart report for the last 30 days, and bring the recommendation list to that team's lead.
- If you are a team lead: ask your owner for report access, then convert the top recurring workflow into a skill.
- If you are on finance: pair the report's spend-by-workstream numbers with the survey's "what would break" answers before deciding on renewal budgets.
- If you are evaluating vendors: ask rivals for the equivalent: workstream-level spend, friction analysis and shared team budgets. OurOpenAI discount-poaching coverage shows price pressure is already part of these negotiations.
This launch is also one of several Claude Enterprise releases this week, including artifacts for legal work. Anthropic's pace here matters as it prepares a public listing; see our look at its IPO governance.
Accurate as of October 28, 2026, based on Anthropic's announcement and TechCrunch's report. Pricing and rollout details were not published in either source.