Bluesight's Prism Assistant: 20 health systems, recurring drug diversion reports in 15 minutes, not six hours
Bluesight built an agent that answers questions about controlled-substance data for hospital diversion teams and builds reports. A second agent for 340B compliance audits is tested but not yet released.
- Organization
- Bluesight
- Industry
- Healthcare software
- When
- May 2026
- Evidence
- Vendor case study
Reported outcome
By Bluesight's internal measurements across 20 health systems: recurring reports 96 percent faster (about six hours to 15 minutes) and triage time down 90 percent.
- 20Health systems using Prism Assistant
- about six hours to 15 minutes (96 percent faster)Recurring reports
- three hours to roughly 10 minutes (90 percent less)Pre-investigation triage
- 30 minutes to under a minute (97 percent faster)Variance analysis
- 5 minutes to 10 secondsQuery latency after API wrapping
Caveats
Published by the company that sells the agent. Treat numbers as marketing until confirmed elsewhere.
- Published on the AWS blog with Bluesight co-authors. AWS sells the platform, and the numbers are Bluesight's internal measurements.
- The headline claim is 'up to 97% faster'. 'Up to' is a best case, not an average.
- The second agent (GPO prohibition audits) is not released. Its 100 percent and 93 percent scores are on synthetic test data.
- The post's line about the evaluation suite is unclear ("from over 3–15 test cases"). It suggests a small test set.
- The post gives no error rate for the assistant's answers.
What the agent does
Bluesight sells compliance and inventory software to hospitals and pharmacies. Prism Assistant for ControlCheck is a conversational analyst for drug diversion teams, the people who check that controlled substances are not stolen or misused. It queries live ControlCheck data and builds visual reports. It launched in May 2026.
The first version came from a three-day sprint with AWS in September 2025. The team did not give the agent raw database access. They wrapped existing API endpoints in Lambda functions that return structured data, and connected over 10 of them as MCP tools.
Why this works
- The agent reads through the same APIs the product uses, so business logic stays in the application.
- That change cut query latency from 5 minutes to 10 seconds.
- In the second agent, the compliance decision is not the model's opinion. A rule-based scoring pipeline with 13 evidence signals decides. The model gathers data and writes the report.
What to copy
- Wrap existing APIs for the agent. Do not point it at raw tables.
- Keep decisions that a regulator will audit in rules, and let the model gather and explain.
- Report test results on synthetic data as synthetic.
Starter prompt
Written by us, not by Bluesight or AWS. Use read-only tools and keep patient data inside your approved environment.
Question from the diversion analyst: {question}
Read-only tools: {tools}
Plan which tools to call. Call them. Show the numbers each tool returned.
Build the report: a table, the three largest variances, and what data is missing.
Do not say that diversion happened. List items for a person to investigate.
Sources
- Building an agentic AI solution at Bluesight with Amazon Bedrock — AWS, Jul 13, 2026
Added Nov 16, 2026 · Sources last checked Nov 16, 2026.
Get new use cases weekly
One short email. Sources included.
The newsletter is paused. Past issues stay in the archive.
Related use cases
Works Human Intelligence cut its HR browser agent's cost per process by up to 97%
AWS reports the browser operation agent's cost per process fell by up to 97% after prompt caching, shorter sub-agent prompts and a switch from Claude Sonnet 4.5 to Haiku 4.5.
AWS Professional Services built four migration agents; infrastructure code drops from 3-4 weeks to minutes
Infrastructure-as-code development per application fell from 3 to 4 weeks to minutes, from AWS's internal project tracking data, across a portfolio of over 300 applications.
Cohere Health's policy agent cut the time to digitize a clinical policy from 2 h 15 min to 1 h 35 min
Overall time spent on policy digitization fell 30 percent, from 2 hours 15 minutes to 1 hour 35 minutes per policy. Agent deployments went from 3–4 months to 2–6 weeks.