MRH Trowe enables secure AI agents for 400 AWS employees

MRH Trowe, one of Germany’s leading commercial and industrial insurance brokers, has placed self-service AI agents in the hands of some 400 employees — not in an isolated laboratory environment, but connected to internal systems and sensitive customer data. The implementation, made on AWS solutions, was designed to operate within a governed, auditable and transparent-cost perimeter, according to the report published on AWS machine learning blog.

What does self-service mean for AI agents?

Self-service, here, is not a chatbot of frequent questions. It is the ability of the employee himself — a broker, a sinister analyst, someone from backoffice — to ask a question in natural language and receive an answer that came from the internal systems of the company, not from a generic model disconnected from the reality of the broker. In practice, the agent assumes the consultation work that previously required to open various systems or trigger another area. The number that measures the project is precisely this: about 400 people, which indicates internal adoption on a scale, not a pilot with half a dozen enthusiasts.

Why governance and audit precede AI speed

In the financial sector, a wrong responder is not just an operational annoyance — it is a matter of compliance. Therefore MRH Trowe describes implementation as a governed and auditable environment: interactions are recorded, what the AI can access is defined by the company itself and consumption has a transparent cost. This tripod attacks the main risk of any generational IA deployment in office: the parallel use of public tools, in which an employee pastes customer data into an open chatbot without anyone knowing.

How to control access to sensitive data from customers with AI

Two decisions weigh more than the choice of the model. The first is where the given traffic: keep queries and answers within the company's controlled environment, with AWS services, reduces the exposure of customer information outside the corporate perimeter. The second is the granularity of permissions — each agent should see only what the function of that user would allow to see in a manual consultation. Audit and transparent cost complete the design, making it possible to rebuild who asked what, when and at what cost.

What Adopting AI Agents Self-Service Changes to Brazil

Brokers, insurers and banks operating in Brazil deal with LGPD and with the supervision of agencies such as SUSEP and Central Bank, which places traceability at the center of any AI project. The German case is concerned less with the chosen model and more with the format: Internal AI, with identity, registration and access limit, instead of assistants released by the organization. It is the same movement that appears in other deployments in the financial sector about AWS, as in the codeless trading assistant created by Jefferies — the difference is that in MRH Trowe, the goal is the daily work of hundreds of employees, not a specialized team.

What's the MRH AI agent announcement? Trowe's not responding

The report does not contain actual adoption numbers, saved time, error rate or details which specific models and services are behind the agents. It is also unclear whether the broker intends to take the approach to the final customer care. And it is worth always warning: a governed environment reduces the risk of leaking and misuse, but does not eliminate the chance for the agent to answer something incorrect — human review remains part of the design, not an optional step.

Frequently Asked Questions about AI agents

How many employees use MRH Trowe's AI agents?

Some 400 employees of the German Realtor, according to the account published by AWS.

What does "transparent cost" mean in this deployment?

It means that the consumption of AI is traceable within the company, allowing us to see how much each use or area generates expense, rather than treating the cost as a closed block.

Can Brazilian companies replicate this model?

Yes, from a technical point of view. The biggest obstacle is usually organizational: defining permissions, keeping track of interactions and ensuring that the customer data does not leave the controlled perimeter in line with LGPD.