Myriad Genetics rebuilt its AI platform after costs ballooned. Now, it can process documents in seconds.

Every year, Myriad Genetics processes an estimated 1.5 million diagnostic tests. Courtesy of Myriad Genetics

Myriad Genetics used AI to speed up document processing, but the tool was expensive and too slow.

AWS worked with Myriad to rebuild the tool, called Image Genius, using Amazon Bedrock.

Image Genius now saves some of Myriad's staffers 300 hours each month.

Myriad Genetics processes an estimated 1.5 million diagnostic tests annually, including screening for hereditary cancer and prenatal conditions, tumor profiling, and assessments of how genes may affect the body's response to mental health medications.

It can take months for the Salt Lake City-based molecular diagnostic testing company to process all the paperwork necessary to get paid, a process known as revenue cycle management, which requires health organizations like Myriad to dedicate staff to software, medical coding , and regulatory compliance to manage billing for medical services.

To speed up document processing, Myriad internally developed an AI tool to automate the company's diagnostic, insurance, and reimbursement operations. After launching a version of that tool in 2024 that proved too costly to maintain, Myriad brought in cloud computing giant Amazon Web Services to lower the project's ongoing costs.

With new artificial intelligence capabilities, Myriad's end-to-end document processing has decreased from an average of 10 minutes to 20 seconds per document. This is helping to accelerate repayment cycles, said Martyna Shallenberg, Myriad's senior director of software engineering.

"The sooner we get cash, the better," Shallenberg told Business Insider. She said that typically, it can take up to 18 months for companies like Myriad to receive full reimbursement for a testing order because of the complexities of processing insurance claims .

Shallenberg — along with the Myriad's chief technology officer, his predecessor, and an expert at AWS — spoke with Business Insider about the evolution of the project over nearly three years.

The tech

In the fourth quarter of 2023, Myriad conceptualized a tool that could use AI to classify pathology reports, automate document tagging, and extract Medicare keywords , Kevin Haas, Myriad's former CTO, told Business Insider in a March interview.

A week after speaking with Business Insider, Haas — who worked at Myriad for eight years and spent five years in the CTO role — departed to become chief development and technology officer at cancer diagnostics company Veracyte.

Haas said he wanted to use AI to speed up prior authorization , a process where a doctor secures approval from an insurance provider before ordering one of Myriad's tests for their patient. Once a doctor made this request, an authorization specialist had to sort through the thousands of health plans and insurance payers that handle these claims.

Myriad also needs to act quickly. Once a sample is collected by a doctor, prior authorization is often required within days, or even hours, to secure health insurance coverage for patients, said Shallenberg. If prior authorization for the diagnostic tests isn't obtained, an insurer could deny the claim, potentially delaying patient care or resulting in a loss of revenue for Myriad, she added.

In the second quarter of 2024, Haas and Myriad worked with consulting giant PwC to build the first version of its intelligent document processing tool, called Image Genius, which used machine learning service Amazon Textract and natural language processing service Amazon Comprehend to automate document classification, extract data from pathology reports, process insurance cards, and reduce manual document handling, said Shallenberg.

Myriad rolled out Image Genius for its hereditary cancer business in the third quarter of 2024, to automate medical document classification and streamline revenue cycle workflows, a Myriad spokesperson told Business Insider.

Myriad Genetics is headquartered in Salt Lake City. Courtesy of Myriad Genetics

A cost- and data-conscious pivot

In the fourth quarter of 2024, as Myriad planned to expand Image Genius to additional business units, it ran into some issues around costs, longer-than-expected processing times, and difficulty extracting the right clinical information from dense pathology reports, said Haas.

Haas said Myriad determined that the internally developed AI tool needed constant maintenance and upgrading. "The inputs are as vast as the millions of samples we get every year," Haas said.

Myriad also discovered that the AI system wasn't always astute enough to differentiate a testing order from a doctor's medical notes, said Shallenberg. "It was actually the hardest part of the process, separating those documents," she added.

Shallenberg said that Myriad spent the first quarter of 2025 refining the AI model's prompts to improve the accuracy of its outputs. They also provided more data samples to give the system greater context, but decided results from this current iteration of Image Genius weren't worth the cost, said Shallenberg. AWS Comprehend, which uses machine learning to identify clinical notes, pathology reports, insurance information, and other data from unstructured text and data, was "very expensive," she said.

Myriad pivoted that same quarter to begin working with the AWS Generative AI Innovation Center , a team of scientists and AI experts that helps customers steer AI projects from conception to full rollout.

By the spring of 2025, AWS and Myriad rebuilt Image Genius on Amazon Bedrock , which allows companies to access foundational models from vendors like Anthropic and OpenAI, and Amazon Nova, the tech giant's family of AI models, Dr. Rowland Illing, AWS' global chief medical officer, told Business Insider.

Illing said this architecture allowed Myriad to tinker with various models and pick the best model based on speed, accuracy, and cost. "Model choice is really important," he said.

The outcome

By the third quarter...

Indexed and credited by AIPROPX. Originating outlet: Business Insider. Open at source →