Specialty insurance has long resisted automation because risks are unique, data arrives as messy attachments, and expert judgment drives decisions. CFC, a global MGA focused on cyber, SME, and specialty risks, put an agentic system into production that turns inbound emails into quotes in under 15 minutes, compressing a process that used to take days. The initiative reframes AI as a scale strategy, not a cost-cutting exercise, aiming to triple revenue without adding headcount.
The team redesigned the workflow from the inbox outward, stood up specialized agents on a modern lakehouse foundation, and moved from prototype to production in a single quarter. Delivery leaned on an API-first estate, a culture already comfortable with compute making commercial decisions, and the kind of agent tooling and analytics that companies winning at AI build on a data layer first to harden the system with governance, observability, and MLOps.
One underwriting system, two users, shared controls
CFC avoided a parallel AI stack. Agents were treated as another class of user inside the existing underwriting platform, subject to the same rulebook, control gates, audit, and observability as human underwriters. That choice minimized organizational friction and let agents inherit enterprise-grade resilience, security, and disaster recovery.
Every agentic action runs through the same battle-tested services that humans use. The result is consistent rules, consistent controls, and a single operational model that blends machine speed with human oversight.
Turning expert judgment into micro-decisions
The work of an underwriter was decomposed into roughly a hundred micro-decisions, each handled by a specialized agent that produces evidence-backed recommendations. One agent ingests the underwriting guide and translates it into machine-executable instructions. Another continuously monitors live quotes to keep agent behavior aligned with how underwriters actually operate in the market.
Orchestration stitches these agents into a left-to-right workflow, all sitting on a modern data platform with standard governance and LLMOps. Integrations execute through legacy APIs, which kept the system aligned with existing processes while enabling speed.
Shipping fast by prioritizing feasibility
Delivery followed a product mindset that weighted feasibility first in a regulated environment. The team proved the hardest technical uncertainties in stealth, then co-designed the product with underwriters, shipping and iterating quickly. Agentic engineering served as a force multiplier for a small team, enabling same-day changes and rapid exploration of undocumented legacy services.
Speed was enabled by a modernized data foundation, an API-first legacy estate, and a business already trading digitally via API. The MVP went live within months, then expanded to handle new business and renewals across geographies.
Measurable impact, clear guardrails
Impact is tangible and compounding:
- At least 10x reduction in cost to quote, even before model-finops optimization
- Average cycle time of about 14 minutes from inbound email to quote
- Underwriter throughput tracking toward two to five times more quotes
- Better consistency and auditability in classification and risk decisions, which pleases heads of line and capacity providers
- Faster responses that improve broker conversion, since CFC often replies first
Humans remain accountable for the final decision and client push. Agents handle ingestion, extraction, configuration, subjectivities, and tax calculation, while underwriters apply judgment, negotiate price, and maintain broker relationships. Organizational design is still evolving as roles adapt to an agent-augmented workflow.
Why agentic scale beats bolt-on automation in underwriting
Treating agents as first-class users of the existing platform, decomposing judgment into tightly scoped decisions, and focusing relentlessly on feasibility produced real outcomes quickly. The core lesson is to fall in love with the business result, not the technology, and to build on a data foundation and culture ready for machine-speed operations.