{"resourceId":"aws-bedrock-tokenomics-invocation-data-lake-20261110","versions":[{"version":"external-68342420a81c71ffacec2dc3f16dca75a5abb78c99a08b31a6ab3dc78c9986fb","resource":{"id":"aws-bedrock-tokenomics-invocation-data-lake-20261110","title":"AWS publishes a Bedrock invocation data lake for org-wide token and cost visibility that centralizes full prompts and responses","organization":"Amazon Web Services","sector":"Cross-SLED AI cost governance","geography":"Not region-specific in source","publishedAt":"2026-11-10","publicationDate":"2026-11-10","eventDate":"2026-11-10","sourceName":"AWS Artificial Intelligence Blog","sourceLabel":"Amazon Bedrock tokenomics data lake","sourceUrl":"https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-tokenomics-data-lake/","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["governance-procurement","data-security","operating-model"],"finding":"AWS's November 10, 2026 post provides CloudFormation templates that forward Bedrock invocation logs from linked accounts to a central account, flatten them with Lambda, store Parquet in S3 via Firehose, and query them in Athena, with optional joins to the Cost and Usage Report by IAM ARN and a Quick Sight dashboard. Each record includes model, account, IAM identity, token counts, stop reason and prompt/response text. AWS notes the bucket is KMS-encrypted and that prompts might contain PII, recommending IAM and bucket policy review. It frames this as a tokenomics (AI FinOps) practice.","sledRelevance":"State and university AI programs face pressure to show where AI spend goes. Per-identity token and cost data supports chargeback across agencies or departments. But centralizing every prompt and response creates a new sensitive data store that may fall under records retention, public records and privacy rules.","evidence":"Vendor sample architecture. No customer results or cost figures.","architectureImplications":"Multi-account log forwarding (CloudWatch Logs destination), Lambda transform, Firehose to S3 Parquet, Glue/Athena, optional CUR 2.0 join and Quick Sight dashboard; Bedrock logging must be enabled per account.","governanceImplications":"Enables per-team accountability for AI spend; also requires a records and retention decision for stored prompts and completions.","securityPrivacyImplications":"Concentrates full prompts and responses, which may include PII, PHI or student records, in one bucket; least-privilege access and retention limits are essential.","caveats":"Vendor sample; not a managed service. Prompt-retention implications for public records are our interpretation. GovCloud deployment not discussed.","streamIds":["aws"],"roles":{"sales":"Interpretation: lead with AI spend accountability, then raise the data-retention question early. The customer problem is AI budgets growing without visibility by department. Stakeholders are the CIO, budget office, records officer and privacy officer. Discovery should ask whether agencies need chargeback and whether stored prompts would be public records. Avoid promising savings; the post provides visibility, not reductions. A bounded first step is two accounts with redaction agreed.","engineering":"Interpretation: fit is good for multi-account Bedrock estates. Engineers deploy the central stack first, then linked stacks, enable Bedrock logging per account and join to CUR 2.0. Prerequisites are Organizations structure, CloudWatch delivery and KMS keys. Constraints include storing full prompt text by default; teams may need to drop or redact text fields. Proof of value is a cost-by-team report reconciled to the bill, plus an access review of the bucket.","delivery":"Interpretation: delivery work is a FinOps capability with a records decision. Dependencies are records officer guidance on prompt retention and an access model. Ownership splits between the FinOps or platform team and a data owner. Governance checkpoints include retention schedule approval and quarterly access reviews. Proposed acceptance criteria: monthly AI cost by department within an agreed reconciliation tolerance, and a documented retention period. Risks include a sensitive prompt archive and public-records exposure."},"roleDetails":{"sales":{"customerProblem":"Public-sector AI programs cannot see AI spend by department or model, and centralized logging creates a sensitive data store.","stakeholders":["CIO","Budget office","Records officer","Privacy officer"],"discoveryQuestions":["Do you need AI chargeback across departments?","Would stored prompts be subject to public records or retention rules?","Who may read prompt text?"],"valueHypothesis":"Per-identity token and cost visibility supports accountability and model choices; savings are not shown.","potentialEngagement":"Deploy the data lake for two accounts with a redaction and retention decision.","unsupportedClaims":["Cost savings","Compliance of storing prompts","GovCloud suitability"]},"engineering":{"fit":"Multi-account Bedrock estates.","architectureIntegration":"CloudWatch Logs destination, Lambda, Firehose, S3 Parquet, Athena, optional CUR 2.0 and Quick Sight.","prerequisites":["AWS Organizations structure","Bedrock logging enabled per account","KMS keys"],"constraints":["Stores full prompts and responses by default","Sample, not managed service","GovCloud not discussed"],"security":"KMS encryption, least-privilege IAM, consider redacting text fields and setting S3 lifecycle retention.","proofOfValue":["Reconcile cost-by-team to bill","Access review of bucket"]},"delivery":{"work":"Stand up AI cost visibility with a prompt-retention decision.","dependencies":["Records guidance","Access model","CUR 2.0"],"ownership":"FinOps/platform team; data owner for prompt archive.","skillsAdoption":"Finance and IT staff learn token economics.","governanceCheckpoints":["Retention schedule approval","Quarterly access review"],"acceptanceCriteria":["Proposed: monthly cost by department reconciled within tolerance","Proposed: documented retention period applied"],"risks":["Sensitive prompt archive","Public-records exposure","Cost data misread as savings"]}},"retrievedAt":"2026-11-11T04:00:06Z","enrichedAt":"2026-11-11T04:05:52Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"No accessibility content.","procurementImplications":"Supports chargeback and model-right-sizing decisions; no cost figures given.","operatingModelImplications":"Requires a FinOps or platform owner and a data owner for the prompt archive.","sourceVerification":{"openedUrl":"https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-tokenomics-data-lake/","referenceExcerpt":"Prompts might contain personally identifiable information (PII), depending on how your users interact with Amazon Bedrock.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}