How MRH Trowe enabled secure self-service AI agents in financial services
What Changed
MRH Trowe, a German commercial insurance broker, gave about 400 employees secure, self‑service AI agents in the first month of production, using a stack of Strands Agents, Amazon Bedrock AgentCore, and LibreChat. The initial production cost was roughly $14 per seat, with a plan to cut infrastructure costs by ~40% through right‑sizing and scheduled scaling. The solution provides contextual, multi‑step, and institutionally integrated AI while keeping data residency, compliance, and auditability inside a centrally governed environment.
Why It Matters
Enterprise architects can see that a modular, framework‑agnostic stack can deliver secure, cost‑transparent AI at scale in regulated sectors. The use of AgentCore’s consumption model gives clear cost visibility, and the isolation of agent sessions supports compliance and data protection. However, the architecture still requires careful governance of token budgets and model selection to avoid hidden costs.
The Limitation
The cost savings and scalability demonstrated are specific to MRH Trowe’s German insurance context and may not directly translate to other industries or regions without similar regulatory and infrastructure alignment.
What You Can Do
Implement a pilot using Amazon Bedrock AgentCore with an open‑source SDK like Strands Agents, and enforce token‑budget controls via LibreChat or a similar interface.