Specialty insurance broker puts AI agents on its core broker platform to [i]halve operations cost[/i]

With 500+ employees and regulated entities in seven countries, the broker builds tailored programs for clients with complex risks. Behind every placement sat a 121-person outsourced operations team working 62 procedures across 20 systems, and each new account meant more people keying the same data into more screens. The outsourcing bill rose in step with the book, so growth kept eating into margin.

Leadership set out to cut outsourced operations cost in half and build an operating model that scales with written premium, with agents that do the processing, people who own the judgment calls, and an audit trail on every transaction that the broker's Risk and Compliance team can stand behind.

AI transformation can still work with the legacy systems a business runs on today. The broker's core platform has no API, so Praxent built agents that work its screens the way a trained processor does. That let the broker start replacing outsourced manual work right away, on the platform its teams already know.

In under two weeks the team reviewed more than 50 SOPs, ran a process-discovery workshop with the broker's operations leaders, and put a working agent on the broker's live Opening Memo process. Praxent built the agent runtime around Anthropic's Claude Compouter Use, which reads the broking platform's screens and operates the keyboard and mouse the way a trained processor does. The core system stays untouched while the agents work.

Pilot runs measured model cost near $3 for a 75-step agent run, a fraction of the labor each run replaces. The same runtime carries every workflow that follows, so each new area of the business builds on the last.

The team mapped every procedure into four areas: policy processing, accounting, claims and onboarding, then set aside the 12 that are reporting-only or depend on negotiation, open-ended judgment or a deliberate human fraud check. The remaining 46 are built for agents, with people approving anything that moves money or reaches a client.

The working agent convinced the broker's leadership to commit to agentic operations across policy processing, accounting, claims and onboarding, with policy processing first. The program is built to halve outsourced operations cost, automate up to 83% of the outsourced workload and remove more than 100,000 hours of manual effort a year, while the broker's own people move from keying data to deciding exceptions.

Every agent run produces a verifiable record with screen evidence, so the broker's compliance team can see exactly what each agent did and why.

Praxent builds Claude agents that work the systems your operations team works today, from policy servicing and bureau submissions to cash allocation and claims. Your agents can start work before your modernization finishes. Tell us which workflow keeps adding headcount, and we'll show you what an agent can take on.