Prior authorization · Production

Myriad Genetics achieves 77% cost reduction and 80% faster document classification with AWS GenAI IDP Accelerator

The problem

Myriad Genetics processed thousands of healthcare documents daily across Women's Health, Oncology, and Mental Health divisions, but their existing Amazon Textract and Amazon Comprehend pipeline cost $15,000 per business unit monthly at 3 cents per page and took 8.5 minutes per document to classify. Information extraction remained entirely manual, requiring up to 10 full-time employees contributing 78 hours daily in the Women's Health unit alone.

First attempt

Despite 94% classification accuracy, the existing Amazon Textract and Amazon Comprehend solution suffered from high operational cost and latency, and misclassified documents due to structural similarities and overlapping content across document types. Information extraction could not be automated because it required contextual understanding to differentiate clinical distinctions and locate information across varying document formats.

Workflow diagram · grounded in source
1
Order event triggers ingestion
trigger
“Document Ingestion: Incoming order events trigger document retrieval from source document management systems, with cache optimization for previously processed documents.”
2
OCR text and layout extraction
ai_action
“Text Extraction: Amazon Textract extracted text, layout information, tables and forms from the normalized documents.”
3
LLM document classification
ai_action
“Classification: The configured LLM analyzed the extracted content based on the customized document classification prompt provided in the config file and classifies documents into appropriate categories.”
4
LLM key information extraction
ai_action
“Key Information Extraction: The configured LLM extracted medical information using extraction prompt provided in the config file.”
5
Structured output to authorization system
output
“Structured Output: The pipeline formatted the results in a structured manner and delivered to Myriad's Authorization System via RESTful operations.”
Reported outcome

The new solution increased classification accuracy from 94% to 98%, reduced classification costs by 77% (from 3.1 to 0.7 cents per page), and cut classification time by 80% (from 8.5 to 1.5 minutes per document).
Automated KIE achieved 90% accuracy matching the manual baseline. Myriad will realize up to $132K in annual savings, and the solution saves 300 hours monthly across Women's Health prior authorizations, reducing each prior authorization submission time by 2 minutes.

Reported metrics
Document classification accuracy94% to 98%
Classification cost reduction77%
Cost per page (old solution)3 cents per page
Monthly cost per business unit (old solution)$15,000 monthly
Show all 17 reported metrics
document classification accuracy94% to 98%
classification cost reduction77%
cost per page (old solution)3 cents per page
monthly cost per business unit (old solution)$15,000 monthly
classification time reduction80%
classification time per document (old solution)8.5 minutes per document
classification time per document (new solution)1.5 minutes per document
KIE extraction accuracy90%
annual savings in classification costsup to $132K
projected monthly savingsmore than $10,000 per month
monthly hours saved (Women's Health)300 hours monthly
prior authorization submission time reduction2 minutes per prior authorization
time saved per prior authorization specialist monthly50 hours per prior authorization specialist
manual KIE employees required (baseline)up to 10 full-time employees contributing 78 hours daily
accuracy gain from negative prompting4%
KIE processing time per documentapproximately 1.3 minutes each
KIE processing cost per page9 cents per page
Reported stack
Amazon BedrockAmazon Nova ProAmazon Nova PremierAmazon TextractAmazon ComprehendGenAI IDP AcceleratorAnthropic Claude Sonnet 3.7Amazon Nova Lite
Source
https://aws.amazon.com/blogs/machine-learning/how-myriad-genetics-achieved-fast-accurate-and-cost-efficient-document-processing-using-the-aws-open-source-generative-ai-intelligent-document-processing-accelerator?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The new solution increased classification accuracy from 94% to 98%, reduced classification costs by 77% (from 3.1 to 0.7 cents per page), and cut classification time by 80% (from 8.5 to 1.5 minutes per document).

What tools did this team use?

Amazon Bedrock, Amazon Nova Pro, Amazon Nova Premier, Amazon Textract, Amazon Comprehend, GenAI IDP Accelerator, Anthropic Claude Sonnet 3.7, Amazon Nova Lite.

What results were reported?

Document classification accuracy: 94% to 98%; Classification cost reduction: 77%; Cost per page (old solution): 3 cents per page; Monthly cost per business unit (old solution): $15,000 monthly (source-reported, not independently verified).

What failed first in this deployment?

Despite 94% classification accuracy, the existing Amazon Textract and Amazon Comprehend solution suffered from high operational cost and latency, and misclassified documents due to structural similarities and overlapp…

How is this prior authorization AI workflow structured?

Order event triggers ingestion → OCR text and layout extraction → LLM document classification → LLM key information extraction → Structured output to authorization system.

Related prior authorization cases