Case Study · AI Agent · Specialty Insurance

Submission triage from 4.2 days to 18 minutes with 12 AI agents

A specialist MGA's underwriting team was drowning in incoming submissions. We redesigned the entire ingestion workflow using a 12-agent AI pipeline — without replacing a single underwriter.

Sector Specialty Insurance · MGA

Engagement Project Delivery

Timeline 14 weeks

Stack Azure · LangChain · GPT-4o

The Challenge

Underwriters buried in inboxes

A leading specialty MGA operating in the Lloyd's Market was receiving upwards of 400 new submissions per week. Each submission required manual extraction, risk categorisation, sanctions checking, capacity routing, and preliminary underwriting notes — all before a senior underwriter could make a meaningful decision.

The Problem

Manual triage was creating a 4-day bottleneck at the start of every deal

400+ weekly submissions processed entirely by hand

Average 4.2 days from receipt to underwriter desk

28% of submissions declined without review due to capacity

Inconsistent triage quality across the team

No audit trail on routing decisions — FCA compliance risk

Senior underwriters spending 40% of time on admin not decisions

Our Approach

12-agent pipeline that reads, scores, and routes — before any human touches it

Multi-format document ingestion across email, PDF, ACORD SLIP

Automated entity extraction, sanctions screening, and capacity check

Risk scoring across 6 underwriting criteria, XAI-traced

Intelligent routing to the right underwriter — with context pack

Full audit log — every decision explained and traceable

Human-in-the-loop override on every flagged decision

The Architecture

12 specialised agents, one seamless pipeline

Each agent is purpose-built for a single task. No monolithic prompts. No hallucination chains. Each step in the pipeline produces a structured, auditable output that feeds the next.

Agent Pipeline — Submission Ingestion to Underwriter Desk

Agent 01

Ingestion & Format Parse

→

Agent 02

Document Classification

→

Agent 03

Entity Extraction

→

Agent 04

Sanctions Screen

→

Agent 05

Capacity Check

→

Agent 06

Risk Profiling

Agent 07

Appetite Scoring

→

Agent 08

Historical Comparison

→

Agent 09

Priority Scoring

→

Agent 10

XAI Summariser

→

Agent 11

Routing Decision

→

Agent 12

Context Pack Builder

"We didn't want AI to replace our underwriters. We wanted it to remove everything that was stopping them from doing their best work. That's exactly what this system delivered."

Head of Underwriting Operations · Specialty MGA (name withheld)

Delivery Timeline

From discovery to production in 14 weeks

Weeks 1–2 · Discovery

Process mapping and data audit

Shadow sessions with the triage team, full mapping of the submission lifecycle, data quality assessment across 6 months of historical submissions, and identification of the 12 discrete tasks that made up the manual workflow.

Weeks 3–5 · Architecture

Agent design and data foundation

Designed the 12-agent pipeline architecture, defined the structured output schema for each agent, built the Azure infrastructure, and established the XAI tracing framework that would follow every decision through the pipeline.

Weeks 6–10 · Build & Test

Agent development and integration

Built and tested each agent against a hold-out validation set of 500 real submissions. Iterated on each agent until precision hit target thresholds. Integrated the pipeline with the existing CRM and underwriting platform.

Weeks 11–14 · Deployment

Parallel run, refinement, and handover

2-week parallel run alongside manual process, calibration of routing rules with senior underwriters, full documentation of the system, and knowledge transfer to the in-house team. Ongoing support SLA in place for 90 days post-launch.

Results

Measurable impact from day one

4.2d → 18min

End-to-end triage time, measured across 6 weeks post-launch

+34%

Increase in submissions reviewed by underwriters within 24 hours

40%

Reduction in time senior underwriters spend on admin tasks

100%

Audit-traceable triage decisions — full FCA-compliant XAI log

Underwriters displaced — capacity freed to focus on complex risk

92%

Routing accuracy vs senior underwriter benchmark (3-month average)

Technology

Built on open, enterprise-grade architecture

☁️

Microsoft Azure

Cloud infrastructure

🔗

LangChain

Agent orchestration

🤖

GPT-4o

Document reasoning

🗄️

Azure SQL + Blob

Data & document store

📊

Power BI

Ops dashboard

🔍

Azure AI Search

Historical RAG lookup

🔐

Azure Key Vault

Secrets & compliance

📋

XAI Engine (custom)

Explainability layer