Compliance monitoring · Production

PerformLine uses Amazon Bedrock prompt engineering to detect marketing compliance violations at scale

The problem

PerformLine's enterprise customers needed efficient compliance checks on complex multi-product web pages requiring context-aware interpretation, while monitoring millions of pages daily—a demand that traditional static parsing could not handle.

First attempt

Manual tracking of multiple prompt versions and templates became inefficient as PerformLine iterated and collaborated.

Workflow diagram · grounded in source
1
ETL ingests web pages
trigger
“Millions of pages are processed by an upstream extract, transform, and load (ETL) process from PerformLine's core systems running on the AWS Cloud.”
2
Event routes page to SQS queue
routing
“EventBridge uses event-driven processing to route Amazon S3 events to Amazon SQS.”
3
Lambda invokes Bedrock AI inference
ai_action
“This function uses Amazon Bedrock to perform extraction and generative AI analysis of the content from Amazon SQS. Amazon Bedrock offers the greatest flexibility to choose the right model for the job. For PerformLine's use case, Amazon's…”
4
Multi-pass product filtering and extraction
ai_action
“Initial filtering with Amazon Nova Micro – This lightweight model efficiently identifies relevant products with minimal cost. Targeted extraction with Amazon Nova Lite – Identified products are batched into smaller groups and passed to A…”
5
Structured JSON output stored in S3
output
“Amazon S3 stores the extracted data, which is formatted as structured JSON adhering to the target schema.”
6
Compliance rules engine validates content
validation
“Compliance checks and business rules, running on other PerformLine's systems, are applied to validate and enforce regulatory requirements.”
Reported outcome

PerformLine achieved a 15% workload reduction in human evaluation tasks and over 50% reduction in analysts' workload from avoiding reprocessing of unchanged pages, with the system projected to process between 1.5 to 2 million pages daily and extract approximately 400,000 to 500,000 products per day.

Reported metrics
Human evaluation task workload reduction15%
Analysts' workload reduction from avoiding reprocessingover 50%
Daily pages processed (projected)1.5 to 2 million pages daily
Daily products extracted (projected)approximately 400,000 to 500,000 products
Show all 6 reported metrics
human evaluation task workload reduction15%
analysts' workload reduction from avoiding reprocessingover 50%
daily pages processed (projected)1.5 to 2 million pages daily
daily products extracted (projected)approximately 400,000 to 500,000 products
daily rule observations requiring review (projected)about 500,000 rule observations
time to develop and deploy architectureless than a day
Reported stack
Amazon BedrockAmazon EventBridgeAmazon SQSAmazon S3Amazon DynamoDBNova ProAmazon Nova LiteAmazon Nova MicroClaude HaikuAmazon Bedrock Prompt ManagementAmazon Bedrock Converse API
Source
https://aws.amazon.com/blogs/machine-learning/how-performline-uses-prompt-engineering-on-amazon-bedrock-to-detect-compliance-violations?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

PerformLine achieved a 15% workload reduction in human evaluation tasks and over 50% reduction in analysts' workload from avoiding reprocessing of unchanged pages, with the system projected to process between 1.5 to 2…

What tools did this team use?

Amazon Bedrock, Amazon EventBridge, Amazon SQS, Amazon S3, Amazon DynamoDB, Nova Pro, Amazon Nova Lite, Amazon Nova Micro, Claude Haiku, Amazon Bedrock Prompt Management.

What results were reported?

Human evaluation task workload reduction: 15%; Analysts' workload reduction from avoiding reprocessing: over 50%; Daily pages processed (projected): 1.5 to 2 million pages daily; Daily products extracted (projected): approximately 400,000 to 500,000 products (source-reported, not independently verified).

What failed first in this deployment?

Manual tracking of multiple prompt versions and templates became inefficient as PerformLine iterated and collaborated.

How is this compliance monitoring AI workflow structured?

ETL ingests web pages → Event routes page to SQS queue → Lambda invokes Bedrock AI inference → Multi-pass product filtering and extraction → Structured JSON output stored in S3 → Compliance rules engine validates content.

Related compliance monitoring cases