The problem being solved

One of the first problems teams meet when adopting a language model platform at organisational scale is that the bill arrives as a single line. The monthly total is visible, but which team, application or user it came from is not. The granular cost attribution AWS introduced for Amazon Bedrock closes exactly that gap.

How the feature works is direct: every inference request is automatically traced back to the IAM principal that made the call. The new line_item_iam_principal column thereby gives per-user and per-application visibility. With optional cost allocation tags, spend can additionally be aggregated by team, project or tenant using AWS Cost Explorer.

The scope of the approach is not limited to applications you build yourself. Usage can be tracked at the granularity you want for any Bedrock-powered service or application, including third-party tools such as Claude Code or Codex.

Setup: CUR 2.0

Before analysing costs you need to set up a Cost and Usage Report, that is a CUR 2.0 data export, and connect it to Amazon Athena. The prerequisites the post lists are:

  • an AWS account with billing console access,
  • IAM permissions for Cost and Usage Reports, S3 and Athena,
  • an S3 bucket for CUR data storage,
  • basic familiarity with SQL and the AWS Management Console,
  • and optionally Claude Code or Kiro-CLI for automated setup.

The critical detail for Bedrock cost attribution to work is that IAM principal data must be enabled in the CUR 2.0 export. Otherwise the line_item_iam_principal column and the associated IAM principal tags are not populated.

Athena or CUDOS?

Once the data is ready, two routes for analysis are offered, and they are complements rather than alternatives.

Athena allows the CUR data to be queried directly with SQL. That route is advantageous for flexibility in aggregations, integration with different business intelligence tools and internal chargeback processes. Breaking Bedrock costs down by IAM principal and usage type, for example, is possible this way.

CUDOS dashboards, by contrast, come with pre-built visuals tailored to an organisation's specific structure. The post notes that the dashboards' new capabilities include granular Bedrock cost and usage data. In short, Athena offers flexibility and CUDOS offers a fast start.

Why it matters

Attribution of this kind looks like a technical detail, but it touches one of the practical obstacles to enterprise AI adoption. A team's ability to expand its use of models often depends on being able to show where the cost is going. As the total bill grows, so does the burden of defending the spend.

Tracking at the identity level is also a governance tool. When it is visible how many tokens each application consumes, unexpected increases are noticed early and cost can be assigned to a specific team or project. That visibility has its own cost, though: the CUR export, S3 storage and query layer form an additional structure to build and maintain. At small scale the return on that overhead may be limited; the threshold where attribution becomes worthwhile is the scale at which the bill genuinely has to be split between teams.