Myriad Genetics Cut Medical Document Classification Costs 77% With AWS GenAI
Myriad Genetics and AWS reported in November 2025 that a generative-AI document pipeline lifted classification accuracy from 94% to 98% while cutting per-page classification costs 77%. The system reduced classification time from 8.5 minutes to 1.5 minutes per document and was planned for a phased rollout across Myriad's Women's Health, Oncology and Mental Health units.
Myriad Genetics and the AWS Generative AI Innovation Center reported in November 2025 that they had rebuilt part of the genetic-testing company's medical document workflow around Amazon Bedrock and AWS's open-source Generative AI Intelligent Document Processing Accelerator.
The project targets documents used in prior-authorization work, including test request forms, lab results, clinical notes and insurance records. Myriad's earlier pipeline combined Amazon Textract with Amazon Comprehend for classification and still relied on people for key-information extraction.
What changed
The new system uses the accelerator's customizable second deployment pattern. Amazon Textract extracts text, tables and forms, while Amazon Nova Pro classifies documents and Amazon Nova Premier handles more complex information extraction. Myriad's subject-matter experts refined the document definitions and prompts used by the models.
AWS said the classification evaluation covered 1,200 healthcare documents across three document classes. In those tests, classification accuracy rose from 94% to 98%, cost fell from 3.1 cents to 0.7 cents per page, and processing time declined from 8.5 minutes to 1.5 minutes per document.
For key-information extraction, AWS reported a smaller test across 32 documents. That configuration combined Textract's layout, tables and forms features with document images, few-shot examples and Nova Premier. It reached 90% accuracy, matching the stated human-evaluator baseline, and processed a document in about 1.3 minutes. The smaller sample means that result should be read as a reported project benchmark, not a general healthcare-AI performance guarantee.
Operational impact and rollout
AWS said faster classification cut two minutes from each prior-authorization submission in Myriad's Women's Health unit. Across 9,000 monthly submissions, the company projected 300 hours saved each month and up to $132,000 in annual classification-cost savings.
The November case study described a phased rollout beginning with Women's Health, followed by Oncology and Mental Health. It did not provide later production-wide validation for every division, so the reported savings and expansion plan should not be treated as completed deployment across the company.
For data teams, the practical lesson is narrower than simply replacing document rules with a large language model. The project paired model selection with document-class definitions, negative prompts, visual examples, human domain review and explicit cost, latency and accuracy tests. In regulated document workflows, those evaluation and rollout controls are as important as the model itself.
Key Points
- 1AWS reported classification accuracy increasing from 94% to 98% on a 1,200-document evaluation.
- 2Classification cost fell 77%, from 3.1 cents to 0.7 cents per page, while time fell from 8.5 to 1.5 minutes per document.
- 3A separate 32-document extraction test reached 90% accuracy and about 1.3 minutes per document.
- 4The November 2025 case study described a phased rollout and projected up to $132,000 in annual classification savings.
Scoring Rationale
A reproducible enterprise document-AI case study with concrete accuracy, latency and cost measurements, but based mainly on participant-reported benchmarks and a phased rollout rather than independent production-wide validation.
Sources
Primary source and supporting public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
