What challenges did Amazon Global Engineering Services face before AI implementation?

Operational readiness testing for new fulfillment centers required about 2,000 hours of manual effort per facility. Teams had to verify more than 200,000 components across 10,500 workstations. Manual review made it difficult to focus on missing components and scale testing efficiently.

What was the impact of the AI solution?

The prototype achieved 92% precision on representative test modules. Latency was 2 to 5 seconds per image. The solution reduced total testing time by 60% versus manual operational readiness testing.

Source

ProductionEvidence: Medium50/100

Amazon Global Engineering Services: Automated operational readiness testing with Amazon Bedrock Nova Pro

Use case typeComputer vision inspectionUpdated Feb 10, 2026

Amazon Global Engineering Services (GES) built an Intelligent Operational Readiness (IORA) solution to automate testing for new fulfillment centers. The system uses Amazon Bedrock with Amazon Nova Pro for real-time image-based object detection and Anthropic Claude Sonnet 4.0 via Bedrock to generate standardized UIN descriptions and detection rules. Amazon API Gateway, AWS Lambda, Amazon S3, and Amazon DynamoDB orchestrate and store the workflow, allowing testers to verify installation status, detect defects, and review results with a production UI.

Organization: Amazon Global Engineering Services
Industry: Logistics
Location: United States
Published: February 2026

Reported outcomes

Latency per image: 2–5 seconds

Time & speed

Planned next steps

Field teams could focus only on missing components, with the article noting that 40% coverage translates to 40% time reduction.

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

Normalized claim

Precision: 92%

AWS Machine Learning BlogFeb 10, 2026Blog postExplicit claimMedium evidence strength

We achieved 92% precision on a representative set of test modules

Normalized claim

Latency per image: 2-5 seconds

AWS Machine Learning BlogFeb 10, 2026Blog postExplicit claimMedium evidence strength

with 2–5 seconds latency per image

Normalized claim

Total testing time: 60%

AWS Machine Learning BlogFeb 10, 2026Blog postExplicit claimMedium evidence strength

IORA reduces the total testing time by 60%

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

Customer: Amazon Global Engineering Services
Provider: AWS
Maturity: Production
Linked source: AWS Machine Learning Blog

Amazon Global Engineering Services (GES) built an Intelligent Operational Readiness (IORA) solution to automate testing for new fulfillment centers

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 2 of 2

1Computer vision inspection
2Operations optimization

Amazon GES developed an AI-powered Intelligent Operational Readiness (IORA) pipeline.
Amazon Nova Pro performs real-time image detection with bounding boxes and installation-status verification.
Anthropic Claude Sonnet 4.0 generates detailed component descriptions from reference images and helps create standardized detection parameters and false-positive rules.
Amazon API Gateway routes requests to AWS Lambda functions, which process images and call Amazon Bedrock.
Module images, reference images, results, and structured verification data are stored in Amazon S3 and Amazon DynamoDB with encryption enabled.

Technologies

Amazon Bedrock Amazon Nova Pro Anthropic Claude Sonnet 4.0 Amazon API Gateway AWS Lambda Amazon S3 Amazon DynamoDB AWS Key Management Service AWS Identity and Access Management

The prototype achieved 92% precision on representative test modules.
Latency was 2 to 5 seconds per image.
The solution reduced total testing time by 60% versus manual operational readiness testing.

Architecture

A serverless orchestration flow uses Amazon API Gateway to invoke AWS Lambda functions. Lambda calls Amazon Bedrock for two paths: batch generation of standardized UIN descriptions and detection rules using Anthropic Claude Sonnet 4.0, and real-time image analysis using Amazon Nova Pro to detect UINs, bounding boxes, installation status, defects, and confidence scores. Images and structured results are stored in Amazon S3 and Amazon DynamoDB, protected with AWS KMS and IAM-based access control.

Sources & evidence1

Evidence: Medium50/100Evidence strength