Anthropic launches reports showing what AI spend delivers

Anyone who has approved AI tools at a company knows the scene: the invoice arrives, it climbs every month, and nobody can say exactly what the team is getting in return. On October 28, 2026, Anthropic announced three features for Claude Enterprise that target exactly that gap: smart reports, in-app surveys and pooled group budgets. According to Anthropic's post, as cited by FourWeekMBA and CXOToday, all three are generally available.

It's worth being clear about the scope: this is a product feature, not a new model. And it's on the Enterprise plan. Availability for other plans has not been published.

What Anthropic released

Smart reports: a report on what the team did and what it cost

You pick a team, a time period and the products to include. The report analyzes a sample of the team's conversations and groups the work into streams (the "workstreams" in the post). For each stream, it shows the spend, what was produced and the most common friction points.

At the end comes a single list of recommendations, including workflows that repeat and deserve to become a standardized routine. You can also ask custom questions about usage and cost. People with billing permission can raise a spending limit straight from the report. And the account owner can allow team leads, who are not administrators, to run reports for their own groups.

Usage details come from a secondary source, the site aiparabellum.com, which describes what Anthropic says: a period of up to 28 days, 10 reports per month per organization at no cost, and generation that takes a few hours with an email notification. The report can flag people blocked by a personal spending limit, and personal conversations are included only in the totals. I treat these numbers as unconfirmed in the original post.

In-app surveys

There are four default questions. One of them asks what would break or slow down first if the team lost access tomorrow. The administrator can customize or remove questions and export the answers to a spreadsheet. The post states that the surveys are treated as customer data.

Pooled budget per group

Instead of individual limits only, the administrator sets a single monthly budget per group. Individual limits remain in place as a safety lock. According to the post, each request counts against both, and the person is stopped at whichever runs out first.

Why this matters for a mid-sized company

TechCrunch describes the launch as an attempt to demonstrate the return on Anthropic's tools, a central concern for large customers. The problem, however, is not exclusive to large companies. India Today cites a KPMG report according to which only 12% of companies consistently compare the value they get from AI with what they spend. The same outlet and aitoolsoasis.com cite a Jellyfish report: spending on agentic coding tools has grown 10 times since the beginning of the year and amounts to about 2% of total engineering costs.

In other words: the AI bill is new, grows fast and almost nobody audits it. In a mid-sized company, where every dollar of the budget has an owner, this becomes a tense discussion in the results meeting.

A day-to-day example

Imagine sales, customer support and finance using the same subscription. Today, the invoice arrives consolidated and the owner's question is simple: who is using it, and for what? With a report per team, you can see which streams consume the most and which ones produce deliverables. With a per-group budget, support gets its own monthly ceiling and finance stops being surprised at month-end close. The recommendations on repeated workflows point to what is worth standardizing, and that tends to reduce rework.

This is a use I would put into practice. The research file contains no measured savings case, so I'm not promising any percentage.

What is still unproven

• The post does not say how many customers use the tools or present measured savings, according to FourWeekMBA, which also says it did not contact Anthropic.
• The post cites users at six companies, including Deloitte and DoorDash. These are testimonials, not audited results.
• The reports analyze a sample of the conversations, not all of them.
• Smart reports are off by default, and the organization owner must enable them (aiparabellum.com).
• Organizations with a HIPAA configuration or customer-managed encryption keys may have different availability (FourWeekMBA). aiparabellum.com lists customer-managed keys, HIPAA configurations and hybrid deployments as unsupported.

Data and contract

Anthropic says that, by default, it will not use this data to train models (TechCrunch and other outlets). The Latent Daily suggests checking the contract terms on retention and access. That is the outlet's recommendation, but it makes sense before turning on any report that reads the team's conversations.

Pricing

No separate price has been published. FourWeekMBA says the post gives no price for the three tools beyond availability on Enterprise. aitoolsoasis.com, a secondary source, says usage counts toward the token spend already in place. In other words, the cost structure appears to be the same usage bill the company already pays, but this is not confirmed in the original post.

How to get started without setting up a project

In my reading, the path that fits a mid-sized company has three short steps:

1. Pick just one team, the one that spends the most or the one that complains the most. Run a report and read the recommendations.
2. Define an owner for the review, someone from finance or operations, not only from IT.
3. Apply the per-group budget after you understand consumption, and use the four-question survey to find out what the team would miss.

The feature does not replace a well-designed process. If the workflow is bad, the report will show a bad workflow with a cost attached. Even so, seeing the cost per work stream already solves the earlier problem: paying without knowing for what.