Question

A SaaS company runs a multi-tenant summarization feature on Amazon Bedrock using a single on-demand foundation model shared across all customers. Finance wants to allocate the monthly Bedrock spend back to each tenant based on their actual usage, but the application makes all InvokeModel calls from one shared backend service. The team needs a way to attribute input and output token consumption to individual tenants without deploying separate models or accounts. Which approach best meets this requirement with the least operational overhead?

Accepted Answer

Enable Bedrock model invocation logging to Amazon CloudWatch and S3, capture the input/output token counts from each invocation record along with a tenant identifier the application writes to the request, and aggregate token usage per tenant.

Answer

Provision dedicated Provisioned Throughput model units for each tenant and use cost allocation tags to split the monthly commitment charges across tenants.

Answer

Rely on AWS Cost Explorer with cost allocation tags applied directly to the InvokeModel API calls to break down spend by tenant automatically.

Answer

Deploy a separate AWS account per tenant with AWS Organizations consolidated billing so each tenant's Bedrock charges appear on its own account bill.