Rede Mater Dei deploys 12 AI agents on Amazon Bedrock AgentCore and reports 517% ROI

Brazilian hospital network Rede Mater Dei has described how it uses Amazon Bedrock AgentCore to monitor and manage AI agents in its revenue cycle. The company has already launched the first agents from a set of 12 and says it earned 517% ROI in four months, cutting authorization times and speeding up the start of surgeries.

Why hospitals are automating

The main problem Rede Mater Dei is trying to solve is the rise in claim denials from insurers and health plans in Brazil. According to figures cited in the article, the average denial rate at the country's private hospitals rose from 11.89% to 15.89% in 2024, and revenue lost across the market could have reached 10 billion Brazilian reais. For a large hospital network this is more than an accounting metric: every delay in checking contracts, authorizations and invoices hits service times, team workload and cash flow.

The network's own processes had the bottlenecks typical of the industry: a lot of manual work, scattered documents, high turnover in operations teams and constant rechecking at vulnerable stages. In this setting, AI agents are interesting not only as a way to save time but as a tool for standardizing decisions. Once an agent takes part in a critical stage of the revenue cycle, simply launching a model is no longer enough: the company needs to see what exactly the agent did, why it did it and how stably the whole chain works.

How the system is built

Together with A3Data and AWS, Rede Mater Dei built a program of 12 AI agents meant to cover the hospital's entire revenue cycle, from contract rules to authorizations and billing. The base platform is Amazon Bedrock AgentCore, which provides a serverless environment for running agents, tool connections, memory and observability tools. The article calls this set a digital workforce: the agents are supposed to perceive data, make decisions and act as autonomously as possible, but within a managed and verifiable architecture.

• Contracts Agent collects and structures complex contract rules that were previously scattered across different documents.
• Parameterization Agent automatically transfers these rules into the hospital's ERP system and reduces the risk of manual errors.
• Authorization Agent handles requests, checks and communication with insurance companies.
• AgentCore Observability and AgentCore Evaluations add a monitoring layer that can measure correctness, accuracy, helpfulness, safety and context relevance.

The architecture is split into three layers. DEL prepares data and loads it into a structured data lake, AEL handles orchestration and execution of the agents, and TCL covers trust, compliance and traceability. The team puts particular emphasis on evaluating agent performance: Rede Mater Dei uses AgentCore Evaluations as a continuous-improvement loop, where the quality of the multi-agent system can be measured with clear metrics rather than the team's subjective impressions.

What the results are so far

The most striking part of the case is the numbers. According to the project, the first phase, focused on Observability and Evaluations, returned 517% on investment in the first four months. At the same time, authorization time fell by 66% and time to the start of surgeries by 33%. For a hospital network this matters: it is not only about internal efficiency but about more predictable handling of procedures, which directly affect scheduling, revenue and the patient experience.

"Our goal is to increase accuracy, predictability and speed at the critical stages of the revenue cycle," says

Rede Mater Dei vice president Renata Salvador Grande.

The second effect is manageability. Thanks to unified telemetry, the team gets a full audit trail of key decisions: which rules were applied, which actions the agent took and where failures or anomalies occurred. This is especially important in sensitive areas such as contracts, authorizations and billing, where any error can lead to payment denial or regulatory risk. Management also gets real-time KPIs: the volume of automated checks, an estimate of financial impact, processing speed and denial risk by insurer.

What this means

The Rede Mater Dei case shows the market moving from experiments with individual AI tools to operating whole agent systems under tight control. For business, this is a signal: value now lies not only in the agent itself but in the layer of observability, evaluation and audit, without which scaling AI in critical processes would be too risky.