Lumıary
Vendor reportedLive since Apr 2026Added Oct 2026

Anterior

Anterior uses Llama 4 on Amazon Bedrock to sort clinical records for prior-authorization review

Anterior, a clinician-led AI company for health insurers, uses open-weight Llama 4 models on Amazon Bedrock to split and label long clinical record packets. AWS reports the platform cut manual clinical review time by 75 percent in prior authorization, with 99.24 percent clinical accuracy.

Move fasterCut costsImprove quality, reduce risk

Key figures

Copied exactly from the sources. Copy any figure with its citation.

75

percent

Reduction in manual clinical review time in prior authorization review, per AWS1

99.24

percent

Clinical accuracy maintained in prior authorization review, per AWS1

as high as 97 percent

Author identification accuracy of the Llama models on clinician-curated synthetic cases (best result), versus 93.5 percent for the proprietary comparison model1

98.4

percent

Description faithfulness of the Llama models on the same synthetic cases1

What it took

CostNot disclosed
TeamNot disclosed
TimelineSix weeks from initial integration to deployment1

The problem

Health insurers review prior-authorization requests, which are doctors' requests to approve a treatment before it happens. Each request comes with a packet of clinical records that can run to hundreds of pages and mixes many document types. Reviewers lose time finding the right pages. Many of Anterior's customers also require that no patient data leaves their own AWS environment.

What they built

A document-sorting step for Anterior's platform. A scanned packet is first turned into text page by page. A Llama model then finds where each document starts and ends, labels its type, and pulls out details such as title, author, creation date and a short description. Llama is Meta's open-weight model family, meaning the model files are published and a company can run them itself.

How they did it

Anterior ran the models on Amazon Bedrock, AWS's managed service for AI models, inside each customer's own AWS account. It compared Llama 4 Maverick 17B and Llama 4 Scout 17B against an unnamed large proprietary model, using the same prompts, data and scoring. The test ran fully inside AWS. AWS says the move from first integration to deployment took six weeks.

Results

On clinician-curated synthetic cases, both Llama models matched the proprietary baseline. They did better on author identification, reaching as high as 97 percent against 93.5 percent, and scored 98.4 percent on description faithfulness. AWS says Llama is now part of Anterior's production workflow for multiple enterprise customers. For prior-authorization review overall, AWS reports a 75 percent cut in manual clinical review time with 99.24 percent clinical accuracy.

Lessons for your project

  1. Test a smaller open-weight model against your current large model on your own documents before assuming you need the biggest one. Anterior found the smaller models matched the baseline.

  2. Check data-residency rules early. Many of Anterior's customers needed all AI processing to stay inside their own AWS environment, which shaped the whole design.

  3. Automate the step before the decision, here finding and labeling the right pages. This gives clinicians a cleaner file without handing the judgment to a model.

  4. Run the comparison on identical prompts, data and scoring so the result is credible to buyers and auditors.

Caveats

  • The only source is an AWS Startups customer story. No independent source confirms the figures.
  • The 75 percent and 99.24 percent figures describe the Anterior platform in prior authorization review as a whole, not the Llama document step alone. The page does not say how they were measured.
  • The accuracy figures for author identification and description faithfulness come from a test on clinician-curated synthetic cases, not live patient records.
  • The page names two models evaluated and does not say which one runs in production.
  • The English page carries no date. The dated copy is AWS's Chinese-language version of the story, April 22, 2026, which is used as the deployed date and published date. Deployment may have started earlier.
  • The page also cites roughly 30 million dollars in annual savings for a regional healthcare organization and a KLAS finding of 155 seconds for cancer care approvals. These are secondhand and are not counted here.
  • The health insurers that use the system are not named.

Sources

  1. Anterior reduces clinical review time by 75% with Amazon Bedrock and Llama

    AWS Startups · Undated · Vendor case study · startups.aws.com

  2. Anterior 使用 Amazon Bedrock 和 Llama 将临床审查时间缩短 75%

    AWS Startups (Chinese-language edition) · 2026-04-22 · Vendor case study · aws.amazon.com

Summarized in our own words from the sources above. Spotted an error? See corrections.

Related cases