Hospital pharmacy teams spend a staggering amount of time chasing paperwork that has nothing to do with patient care. AWS and Bluesight just showed how much of that burden AI can take off their plates, using 340B compliance as the proving ground.
What Prism Actually Solves
A single 340B covered entity can burn more than 4,000 staff hours a year just checking whether drug purchases through a Group Purchasing Organization qualify for an exception. That means cross-referencing FDA shortage notices, pharmacy association records, inventory levels, and back-order reports from other hospitals, all by hand.
Federal rules make this stricter than it sounds. Disproportionate Share, Children’s, and Free-Standing Cancer hospitals can’t buy outpatient drugs through GPO contracts unless supply conditions genuinely block the non-GPO route, and teams have to document every exception they claim.
Bluesight’s answer is Prism, an AI layer that pulls data from across its product suite so compliance staff aren’t stitching together spreadsheets from three different systems.
How the AI Layer Works Under the Hood
Prism Assistant, the ControlCheck-focused piece, is already live across 20 health systems. It gives diversion teams a conversational interface for querying medication transaction data instead of building reports and dashboards from scratch.
The bigger piece, a multi-product GPO compliance agent, is due later in 2026. It coordinates specialist agents that each pull one type of evidence: purchase records from CostCheck, shortage data from ShortageCheck, and eligibility checks from 340BCheck. A coordinator agent then assembles everything into an audit-ready report.
Built on Amazon Bedrock AgentCore and running Anthropic’s Claude models, the system was reportedly connected end-to-end within a single day, with every planned feature working by day two. In testing against synthetic data, Bluesight reported a 100% invoice discovery rate and 93% evidence-justification accuracy, above its own 85% target. Those numbers came from synthetic data, not live hospital operations, so they’re a starting point rather than proof of production performance.
Why This Matters Beyond One Vendor
The compliance determination itself isn’t left to an AI’s judgment call. Bluesight built a rule-based scoring pipeline with 13 evidence signals underneath the AI layer, so the language model handles data gathering and report writing while a deterministic system makes the actual call. That distinction matters for hospitals that need to explain their methodology to regulators, not just show a confident answer.
It’s also a signal for where healthcare AI is heading generally: less flashy chatbot demos, more quiet infrastructure work aimed at the paperwork nobody wanted to do anyway.
Conclusion: A Test Case Worth Watching
Prism isn’t finished, and its headline accuracy numbers still need validation against real hospital data. But the direction is clear: AI-assisted 340B compliance could meaningfully cut the thousands of hours hospitals currently spend on manual reviews, freeing pharmacy teams for work that actually needs a human. Health systems evaluating compliance tools should keep an eye on how Prism performs once it moves from synthetic testing to live deployments later this year.
