AI Agents for E&S Insurance: The Next Operating Model for Carriers

Excess and Surplus insurance has always been where the standard insurance playbook breaks down.

The risks are more complex. The submissions are messier. The coverage questions are less straightforward. The documents come in through broker emails, PDFs, spreadsheets, ACORD forms, loss runs, supplemental applications, inspection reports, and policy files. The work depends on speed, judgment, specialization, and operational discipline.

That is exactly why E&S is one of the strongest markets for production-grade AI agents.

The opportunity is no longer about using AI to answer basic questions or summarize a document. For carriers, MGAs, brokers, and TPAs, the real opportunity is to connect customer and broker interactions directly into governed operational execution across underwriting, claims, servicing, and back-office workflows.

In other words: AI should not just talk about the work. It should do the work.

Why E&S is such a strong fit for AI

The E&S market has grown quickly because it serves risks that admitted markets often cannot or will not cover. NAIC describes surplus lines as a specialty market for risks that are not available through the admitted market, and reports that the U.S. surplus lines market reached about $131 billion in direct premiums written in 2024, representing roughly 12% of the U.S. P&C market. (NAIC) WSIA and AM Best similarly reported that surplus lines direct premium written reached a record $129.8 billion in 2024, up 12.3% year over year. (WSIA)

That growth creates operational pressure.

E&S teams are receiving more submissions, more broker inquiries, more documents, more follow-ups, and more edge cases. At the same time, the work is hard to standardize. Every risk can look slightly different. Every submission may arrive with a different document set. Every quote, decline, endorsement, renewal, or claim requires context.

This is the operating gap AI agents can close.

Not by replacing underwriting or claims judgment, but by removing the operational drag around that judgment: collecting missing information, extracting structured data, routing work, answering status questions, preparing files, enforcing process rules, and creating a traceable record of what happened.

From point automation to an AI operating layer

Many insurance organizations already use automation. The problem is that most automation is fragmented.

A chatbot sits in one place. An OCR tool sits somewhere else. A workflow rule runs in another system. A claims intake form feeds a different process. An underwriter still has to open the email, download the attachments, inspect the SOV, read the loss runs, check appetite, ask for missing information, and manually re-enter data into the system of record.

That is not an AI operating model. It is another layer of operational debt.

At Notch, we see the bigger opportunity as connecting broker, policyholder, and internal interactions into compliant operational execution across the insurer’s core workflows. That means AI agents that can support conversational workflows, internal co-pilots, and back-office operations together rather than as disconnected tools. Notch’s platform is already built around this model: conversational agents for broker, partner, and policyholder interactions; internal co-pilots for adjusters, underwriters, and service teams; and back-office agents that ingest documents, extract structured data, classify work, route requests, and prioritize time-sensitive operations.

For E&S, that architecture matters because the process is not linear. A broker question may require checking submission status. A submission may require document extraction. A missing document may require an outbound follow-up. A claims intake may require coverage logic, routing, and system creation. A servicing request may require authentication, policy validation, and a back-office action.

The winning AI strategy is not one agent for one task. It is an operating layer where agents share context, enforce controls, and move work forward.

Customer-facing agents: better broker and policyholder experiences

In E&S, “customer-facing” often means broker-facing.

Brokers want fast answers. Did you receive the submission? Is it in appetite? What documents are missing? Who is reviewing it? When can I expect terms? Has the endorsement been processed? What is the status of the claim?

These questions are high-volume, time-sensitive, and often trapped inside inboxes and portals. Customer-facing AI agents can resolve them without forcing brokers to wait for manual follow-up.

A broker-facing submission agent can receive inbound submissions, identify the insured, classify the line of business, read the attachments, detect missing information, and ask targeted follow-up questions. A status agent can respond to broker inquiries with grounded answers based on the actual file. A servicing agent can help with certificates, endorsements, cancellations, renewals, policy documents, and billing questions. A claims intake agent can collect structured FNOL information across voice, email, web, documents, and APIs.

The important distinction is that these agents are not generic chatbots. They need to operate inside the carrier’s workflow, policies, permissions, and systems. In regulated environments, an AI agent must authenticate users when needed, enforce authorization, validate policies, log decisions, and escalate safely when confidence is low or judgment is required. Notch’s regulated-industry architecture is designed around that principle: the AI agent sits between the customer interface and the client’s systems, with authentication, authorization, policy validation, workflow execution, and audit logging built into the process.

That is what allows AI to move from answering questions to safely executing work.

Internal agents: giving underwriters, adjusters, and service teams leverage

The highest-value AI agent in E&S may not be the one the customer sees.

It may be the internal co-pilot sitting beside the underwriter, adjuster, service rep, or operations team.

E&S files are information-dense. An underwriter may need to understand the insured’s operations, compare the submission against appetite, read supplemental applications, inspect SOVs, evaluate prior losses, review exclusions, and decide what is missing. An adjuster may need to review claim notes, medical reports, photos, policy language, prior communications, litigation indicators, and next steps.

Internal AI agents give teams a way to query these long files in natural language and receive structured, traceable answers grounded in source data. They can summarize the submission, flag missing documents, highlight contradictions, suggest broker follow-up questions, identify referral triggers, and prepare the file for review. Notch’s internal positioning already frames the platform as a co-pilot for operations teams that allows adjusters, underwriters, and service representatives to query long claim files, policy documents, and submission materials, with structured answers grounded in source data.

This is especially powerful in E&S because expertise is scarce. Senior underwriters and claims leaders should not spend their time hunting through PDFs or chasing incomplete submissions. They should spend their time making the decisions that require expertise.

AI agents make that possible by preparing the work before the expert touches it.

Back-office agents: where the economics compound

Back-office workflows are where AI becomes more than a customer experience investment.

Every E&S organization has operational work that is necessary, repetitive, and expensive: ingesting submissions, extracting data, clearing accounts, checking appetite, routing files, creating records, preparing quote packages, reconciling bordereaux, reviewing endorsements, validating compliance artifacts, and prioritizing urgent work.

Back-office AI agents can take inbound emails and attachments, extract the structured data, classify the request, check completeness, route it to the right queue, populate downstream systems, and preserve an audit trail. They can normalize bordereaux, reconcile premium and commissions, identify missing fields, and flag exceptions. They can prepare claims or underwriting files so the human reviewer starts from a structured, decision-ready view instead of a raw pile of documents.

This is where the operating model changes.

A carrier does not need ten different point tools that each automate a fragment. It needs a governed system where agents can ingest, reason, act, and hand off work with continuity. Notch’s FNOL architecture is a good example of this pattern: rather than treating intake as a stateless task, Notch structures FNOL as a long-horizon, policy-aware workflow with persistent claim state, confidence-scored extraction, deterministic fallbacks, escalation paths, and full decision traceability.

The same principle applies across E&S operations. When the first interaction creates a clean, structured, auditable record, every downstream workflow becomes easier to automate.

The carrier opportunity: speed with control

The promise of AI in E&S is not simply “faster.”

Speed without control creates risk. In E&S, a fast but wrong answer can damage broker trust, create compliance exposure, or send a complex risk down the wrong path. Buyers should look for AI systems that improve speed while strengthening governance.

That means agents should be able to:

Work across voice, email, chat, documents, portals, and APIs
Extract structured data from unstructured submissions and claims files
Enforce business rules, authority rules, and escalation thresholds
Integrate with core systems, CRMs, policy systems, claims platforms, and identity providers
Maintain a complete audit trail of actions, decisions, handoffs, and confidence levels
Support human override, review, and exception handling
Learn from operational patterns without becoming an uncontrolled black box

This is the difference between AI as a feature and AI as infrastructure.

For carriers and MGAs, the opportunity is to increase submission throughput, reduce manual rework, improve broker responsiveness, accelerate claims intake, raise downstream automation ceilings, and give expert teams more leverage. For policyholders and brokers, the opportunity is a more responsive experience. For leadership, the opportunity is better visibility into where work is getting stuck and which workflows are creating operational drag.

What you should look for

If you are evaluating AI for E&S operations, the key question is not whether the system can summarize a document or answer a broker question in a demo.

The better question is: can it operate safely inside your real business?

Can it handle messy submissions? Can it ask for missing documents? Can it read the SOV, loss runs, ACORD forms, and supplemental applications? Can it distinguish between a simple servicing request and a sensitive action that requires authentication? Can it integrate with your systems of record? Can it explain what it did? Can your team test it, govern it, monitor it, and improve it over time?

Most importantly, can it compound?

A point solution may automate one task. An AI operating system should make every additional workflow more valuable because agents share context, controls, and data across the insurance lifecycle.

That is the real opportunity for E&S carriers.

The future of E&S operations will not be defined by who adds the most AI features. It will be defined by who can turn complex, fragmented, document-heavy work into governed execution at scale.

That is where AI agents move from experimentation to production.

And that is where the market is going.