What challenges did Stripe face before AI implementation?

Compliance teams reviewed thousands of transactions daily while spending up to 80% of time navigating fragmented systems. Stripe needed to scale compliance operations without proportional headcount increases while maintaining regulatory quality and auditability.

What was the impact of the AI solution?

The system reduced median review handling time by 26%. Reviewer helpfulness ratings remained above 96%. The design preserved auditability for regulatory scrutiny and enabled scaling beyond proportional headcount.

Source

Evidence: Low35/100

Stripe: Production-grade AI agents for financial compliance using Amazon Bedrock (26% faster reviews)

Use case typeCompliance automationUpdated Jun 26, 2026

Stripe built a production-grade AI agent system for financial compliance on AWS to help compliance teams review thousands of transactions daily without proportional headcount growth. The system uses Amazon Bedrock with a ReAct-style agent framework, task decomposition into a DAG of sub-tasks, a dedicated agent service, and an LLM proxy with prompt caching and model fallbacks.

Organization: Stripe
Industry: Finance
Location: United States
Published: June 2026

Reported outcomes

−26%

median review handling timeTime & speed

+96%reviewer helpfulness ratings

Strategic outcomes

Risk & compliancePreserved auditability for regulatory scrutinyScale & capacityScaled compliance operations beyond proportional headcount

Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Median review handling time: 26% decrease

AWS Machine Learning BlogJun 26, 2026Blog postExplicit claimLow evidence strength

reduced median review handling time by 26 percent

Normalized claim

Reviewer helpfulness ratings: 96% increase

AWS Machine Learning BlogJun 26, 2026Blog postExplicit claimLow evidence strength

over 96 percent helpfulness ratings

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Stripe
Provider: AWS
Maturity: Unknown
Linked source: AWS Machine Learning Blog

No explicit deployment-stage evidence found.

Customer identity supportedSource describes one deploymentMaturity evidence evaluated

Primary read

Use case focus

Showing 3 of 3

1Compliance automation
2Workflow orchestration
3Risk assessment

Compliance teams reviewed thousands of transactions daily while spending up to 80% of time navigating fragmented systems.
Stripe needed to scale compliance operations without proportional headcount increases while maintaining regulatory quality and auditability.

Stripe decomposed reviews into bite-sized sub-tasks coordinated via a DAG orchestrator.
Stripe implemented a ReAct agent that iteratively reasons and performs tool calls with observation checkpoints anchored on factual tool outputs.
Stripe built a dedicated agent service and an LLM proxy microservice supporting prompt caching, model fallbacks, and long-running stateful agent sessions.
Human oversight remained in control with configurable approval workflows and full audit trails.

Technologies

Amazon Bedrock Prompt caching LLM proxy ReAct agent framework

The system reduced median review handling time by 26%.
Reviewer helpfulness ratings remained above 96%.
The design preserved auditability for regulatory scrutiny and enabled scaling beyond proportional headcount.

Architecture

Stripe built an agentic review architecture with task decomposition into a DAG, a ReAct agent using Amazon Bedrock for reasoning, a dedicated agent service for long-running/stateful sessions, an LLM proxy for prompt caching and model fallbacks, and full audit logging with human approval checkpoints.

Sources & evidence1

Evidence: Low35/100Evidence strength