MRH Trowe Lets Employees Build Their Own Agent Factory
Employees at insurance broker MRH Trowe wanted artificial intelligence to help with their daily work. Teams started trying AI tools on their own. That created a risk for the company. Sensitive client and insurance data could end up in software nobody at the firm controlled.
MRH Trowe’s answer was to build one secure system where employees create and use their own AI agents. About 400 employees got access in the first month, Amazon Web Services said in a case study on its customer. The system cost about $14 per employee per month in cloud and AI model fees. MRH Trowe is a commercial insurance broker operating mainly in Germany, Switzerland and Austria.
The first agent writes meeting notes. An employee asks it, in German, for a recent meeting. The agent finds the meeting on the calendar, pulls the transcript and drafts minutes with participants, agenda, topics and action items.
Employees don’t need deep technical skills to build an agent. It takes a few lines of code, the case study said. The firm wants repetitive work automated by the people who used to do it. “If you and your colleague are doing things twice, consider creating a LibreChat agent!” said Leonid Karlinski, a board member at MRH Trowe. LibreChat is the chat tool employees use to reach the agents.
Two safeguards protect client data. Each agent runs under the identity of the employee who is signed in, so it can reach only that person’s calendar and transcripts. All processing stays in AWS’ Frankfurt region, which keeps the data in Germany.
Finance Has AI in the Back Office
Financial firms already lead other sectors on AI. They have reached high adoption in 27 of 75 business tasks, PYMNTS Intelligence found in a March survey of 60 senior technology executives at large U.S. companies, 20 from each sector. Media and advertising reached 16 tasks and healthcare reached 10.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
That adoption clusters in controlled internal work. Revenue recognition leads at 65%. Credit risk assessment and sales forecasting follow at 60% each. Work closer to the customer lags. Churn prediction sits at 30% and identity verification at 20%.
MRH Trowe’s newest agent targets sales growth. Called “talk to your data,” it supports cross-sell and upsell reviews that combine customer records with public information.
The Payoff Still Needs Numbers
The case study reports users and cost. It gives no figures for time saved, revenue gained or accuracy. MRH Trowe tracks adoption on a dashboard showing unique users and cost per user.
A larger firm offers a comparison. Most of the AI builders at BNY now come from outside its engineering teams, the bank said. A contract review agent built there cut legal review time from four hours to one, OpenAI said in a December case study on the bank.
MRH Trowe is leaning on people as much as software. It aims to have 10 to 15 agents created and maintained by its own subject-matter experts by the end of 2026.
More money is coming across the sector. Some 85% of financial services firms expect to raise AI budgets over the next 12 months, PYMNTS reported in August. Fragmented or poor-quality data was the biggest barrier to wider use for 30% of them.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
