MRH Trowe: 400 Employees on AI Agents at €11 per Seat
In short
MRH Trowe equipped 400 employees with self-configured AI agents at around €11 per seat in the first month, productive within four weeks. The case proves that AI agents are fast, affordable and secure even in regulated industries.
The problem: Shadow AI at an insurance broker
MRH Trowe is an owner-managed insurance broker with around 400 employees, focused on Germany, Switzerland and Austria. The company was an early cloud pioneer among German brokers – and that openness led to a classic challenge: individual teams experimented with AI on their own, without central control. This so-called shadow AI is often the biggest barrier to productive AI use.
The case comes at the right time: according to Bitkom 2026, 41 percent of German companies already use AI, but a third struggle with unexpected costs. An IW Köln study puts the productivity advantage for AI users at 8 to 12 percent. MRH Trowe delivers reliable figures precisely into this tension.
What they built
MRH Trowe combined open-source components with AWS infrastructure: Strands Agents as the SDK, Amazon Bedrock AgentCore as the runtime with session isolation, and LibreChat as the UI with token budgets. This architecture lets employees configure their own agents without IT having to approve every step.
Governance from day one
Thanks to confidential computing, data stays in Frankfurt (EU), identity management runs via Microsoft Entra ID, and every action is auditable. This makes self-service possible without sacrificing compliance.
The numbers that matter
Around 400 employees gained access. The cost per seat in the first month was about €11 – including infrastructure and tokens. The first agent was productive within four weeks. These are exactly the facts often missing from the AI debate.
- 400 employees with access to self-configured AI agents
- Approximately €11 per seat per month in the first month
- Productive in four weeks
- Data processing in Frankfurt (EU), confidential computing
Why €11 is not the whole story
The number is remarkable, but it is not a licence for naive budget planning. The analysis by The Clarity Today points out that the $14 comprises infrastructure and token costs – without a detailed breakdown. Model costs scale with usage, and announced infrastructure savings of 40 percent have not yet materialised.
Compliance as an enabler, not a blocker
The case shows that regulation and speed are not mutually exclusive. MRH Trowe introduced central governance that ended shadow AI while enabling self-service. For decision-makers, this means: if you want to introduce AI agents, you must think about compliance from the start – then it becomes an accelerator, not a brake.
The first productive agent: meetings that document themselves
The first productive agent turns Microsoft Teams meetings into structured minutes – with date, participants, agenda and action items. A perfect example of a tightly scoped use case that delivers quick value without major integration projects.
Every question will first be answered by AI before humans step in.
What you can take away
The MRH Trowe case offers three concrete lessons for your business. First, AI agents are affordable. Second, they become productive in weeks, not quarters. Third, compliance is not a reason to wait, but a design criterion. To explore how to make AI a leadership priority, listen to this Swiss podcast on AI in SMEs.
- Start with a tightly scoped use case, such as meeting minutes.
- Use self-service with clear governance rules instead of central approval for every request.
- Calculate total costs realistically, including token consumption and infrastructure.
- Make AI a leadership issue: the biggest hurdle is rarely technology, but management.
Your next step
Pick a repetitive process that currently eats manual time – such as minutes, data matching or standard requests. Define the desired output and check which governance you need. That is enough for a valid test.
Frequently asked questions
- What distinguishes AI agents from traditional chatbots?
- AI agents autonomously execute multi-step tasks and access internal systems. Chatbots mostly answer single questions. At MRH Trowe, employees configure their own agents for recurring workflows.
- How much does an AI agent cost per employee?
- In the MRH Trowe case, costs were around €11 per seat per month in the first month, including infrastructure and tokens. No detailed breakdown was published.
- How quickly can AI agents become productive?
- MRH Trowe reports that the first agent was productive within four weeks. A tightly scoped use case and clear governance are decisive factors.
- What technology does MRH Trowe use?
- The company combines Strands Agents as an open-source SDK, Amazon Bedrock AgentCore as the runtime with session isolation, and LibreChat as the interface with token budgets.
- How does MRH Trowe ensure compliance and data protection?
- Through confidential computing, data processing in Frankfurt (EU), identity management via Microsoft Entra ID, and complete audit trails. Central governance also ends uncontrolled shadow AI.
- Is the €11 per seat figure transferable to other companies?
- The figure is a real reference point, but not a guaranteed value. Model costs scale with usage, and announced infrastructure savings of 40 percent have not yet been realised.
Sources
Would you like to explore this topic for your company?
Check Availability