What You Will Learn

• Automate end-to-end agency workflows using Claude Code
• Extract and summarize policies, ACORDs, loss runs, and commission statements
• Connect Claude to email, calendar, AMS, and MCP connectors for live actions
• Build projects, reusable skills, and an agency knowledge base (RAG) with compliance guardrails
• Use Co-Work browser automation for carrier portal entry, quoting, and loss run retrieval
• Implement renewal routines, COI autoresponders, and AI-driven growth systems

Study Guide

Introduction: Why This Course Changes Everything for Your Agency

Let's be honest about something that's been bothering you. You know your agency could be running smoother. You feel it every time you manually type client information into a carrier portal. You feel it when you're digging through emails to find a certificate request. You feel it when you spend an entire afternoon pulling loss runs that should take minutes. The insurance industry hasn't changed much in decades, but the tools we use to run our agencies are finally catching up to what's possible.

This course is about one thing: teaching you how to use Claude Code to automate your entire insurance agency. Not just the writing tasks. Not just the simple stuff. The whole operation. From the moment a lead comes in to the moment a policy renews, there's a way to make that process faster, more accurate, and less dependent on you being in the office.

Here's the truth about AI in insurance right now. Most agents are using it wrong. They're treating it like a fancy typewriter or a search engine. They ask it to write an email and call it a day. But the real power isn't in writing,it's in doing. It's in connecting your AI to your data, your calendar, your email, your agency management system, and your carrier portals. It's in building systems that run themselves while you focus on the relationships and the judgment calls that actually grow your business.

This course walks you through a five-level framework that takes you from basic AI consumption all the way to autonomous growth systems. By the end, you'll have a complete roadmap for transforming your agency into a lean, automated machine. And you'll know exactly where to start, what to prioritize, and how to keep compliance at the center of everything you do.

Level One: The Consumption Phase,Your Foundation

Level One is where every agency starts. This is the phase where you're learning to work with AI in your browser, using it for tasks that make your daily life easier. It's called the consumption phase because you're consuming the AI's capabilities,using it to read, write, summarize, and create. You're not yet connecting it to your systems or letting it take actions. That comes later.

Think of Level One as learning to drive before you get on the highway. You need to understand the basics first. You need to know how the interface works, what the different models can do, and how to get consistent, professional output. Master this level and every level after it becomes dramatically easier.

Pillar One: Getting Comfortable with the Interface

Before you can do anything meaningful, you need to know your way around. The Claude interface has several key areas you'll use constantly. The chat functionality is your primary workspace,it's where you'll type requests and have conversations with the AI. The project workspaces are dedicated areas where you can house context-specific instructions, standard operating procedures, and templates. And the artifacts system is a file management feature that lets you view, download, and iterate on documents the AI creates.

Here's what most people don't realize: the artifacts system is a game-changer. When Claude generates a document, it appears in a preview panel. You can see it, download it, and,here's the magic,when you ask for changes, those changes appear instantly. No more downloading a file, opening it, editing it, and re-uploading it every time you want a tweak. The iteration happens right in the platform.

Now, about models. Claude offers several models, and they're ranked by strength. The high-capability model is designed for complex, difficult tasks that require maximum reasoning power. Think of it as your specialist for the hard stuff. The mid-tier model is your everyday workhorse,it's the default choice for most operations and handles typical tasks exceptionally well. The lightweight model is fast and efficient for simple tasks where advanced reasoning isn't necessary.

You also need to understand the settings. The search and reference chats feature allows the AI to look back at past conversations for additional context when you're working in new chats. This creates continuity across your interactions,you don't have to start from scratch every time. The generate memory from chat history feature means the AI periodically updates its understanding of you and your operational context. It's like having an assistant who remembers your preferences without you having to repeat them.

And here's a useful trick: you can import context from other AI providers. If you've been using ChatGPT and built up a history there, you can export that information and import it into Claude. The AI will "know" you from day one without requiring extensive initial conversations. There's also text-to-speech for dictating prompts and voice mode for more advanced hands-free interaction.

Pillar Two: Generating Professional Outputs

Once you're comfortable with the interface, it's time to start creating. This pillar covers the creation of standard business documents,presentations, word processing documents, diagrams, and data extraction outputs. The key capability here is producing professional, client-facing materials from raw information with minimal manual effort.

Let me give you a concrete example. Say you have a substantial homeowners' insurance policy document. It's multiple dense pages of legal language and coverage details. You upload that document to Claude and ask for a summary. The AI processes the policy text and generates a comprehensive summary of coverage details. But here's where it gets interesting: you can then convert that output into a polished, client-ready one-page document.

There's an important implementation tip here. When you're requesting document creation, instruct the AI to work in HTML code first, then convert to PDF. This produces superior visual results compared to direct PDF generation. Why? Because HTML allows for more precise formatting control before finalization. You can specify color palettes and other design elements, ensuring your output documents align with your branding.

The artifact system really shines here. Revisions appear instantly in the platform without requiring file downloads, opening, and re-uploading between each modification cycle. You can iterate five times in the time it would have taken you to do one cycle the old way.

Another example: creating presentations. You can feed Claude your raw notes or a rough outline, and it will structure them into a professional slide deck. You can specify the number of slides, the visual style, and even the tone. For agency owners who regularly present to carriers or at industry events, this alone can save hours of work.

Pillar Three: Document and Data Extraction

This is where things start getting really powerful. A foundational capability of Claude is its ability to ingest carrier documents,declarations pages, certificates, policy forms,and extract structured information. This eliminates the need for manual reading and comparison across multiple documents.

Here's a typical scenario. You upload a declarations page and an ACORD certificate simultaneously. You ask the AI to extract and synthesize all relevant information into a single, readable summary. The AI identifies coverage details, policy limits, named insureds, and other critical data points. What would have taken you fifteen minutes of careful reading and note-taking takes seconds.

But this extraction capability isn't just about saving time on individual tasks. It's the underlying mechanism for every subsequent workflow in this entire framework. Think about it this way: the process of the AI reading through a file, analyzing its content, and giving you an output is literally how every single workflow, every single AI chat, and how every process for a result actually happens.

Let me give you another example. You have a stack of ACORD certificates from different carriers for the same insured. You need to verify that the coverage limits match what was requested. Instead of laying them side by side and squinting at the fine print, you upload them all and ask the AI to compare. It will flag any discrepancies, highlight missing coverages, and give you a clear summary of what's compliant and what isn't.

This capability extends to commission statements, loss runs, and any other document that requires careful reading and analysis. The more you use it, the more you'll realize how much of your day was spent on tasks that can be automated.

Pillar Four: Basic Connectors

Connectors are integrations between Claude and external services. They're significantly simpler to implement than traditional API integrations,typically requiring only OAuth-style authentication through a straightforward sign-in process. You click the connector option, select the platform, and authorize access through your browser. That's it.

You have several connector categories available. Email connectors let the AI read your inbox, draft messages, send emails, manage spam, and organize your email. Calendar connectors allow it to view your schedule, create meetings, block time slots, and provide schedule summaries. Meeting tool connectors pull meeting notes and access meeting details. And media generation connectors create images and videos through connected creative platforms.

Let me show you how this works in practice. You ask the AI to review your day's calendar. It accesses your connected Google Calendar and produces a structured list of all your scheduled events. You ask it to check your email, and it presents your inbox, drafts responses, and executes full email management tasks.

The operational value here is substantial, even at five to ten percent productivity improvement per user. And remember, each connection takes about thirty seconds to set up. There's no excuse not to connect your core tools.

Here's another example. You're preparing for a meeting with a client. You ask the AI to pull up the last three emails you exchanged with them, summarize the key points, and draft a follow-up agenda. It does all of this in seconds because it has access to your email and calendar. You walk into that meeting fully prepared without having spent twenty minutes digging through your inbox.

Pillar Five: Project Workspaces

Projects represent a significant advancement over standard chat sessions. Projects create persistent workspaces that combine custom instructions, context files, and memory systems. This is where you start building your automation infrastructure.

A project has three main components. First, custom instructions,these are detailed system prompts that define how the AI should behave, what it should always do, and what it should never do. Second, context files,these include SOPs, carrier guidelines, style guides, and other reference materials. Third, memory systems,accumulated knowledge about you and your operations.

Let me walk you through a real example. I set up a demonstration project for certificate of insurance (COI) issuance. I uploaded three reference documents: an issuance SOP, a carrier context summary, and a style guide. Then I wrote a comprehensive system prompt outlining workflow specifics, behavioral expectations, verification requirements, and output formats.

When I tested it with a sample COI request, here's what happened. The system performed a five-point verification check on the request. It reported "pass" when all requirements were met. It generated certificate details and a cover email. It created an AMS activity log entry. But here's the impressive part: when I gave it incomplete or non-compliant information, the system correctly refused to issue. It explained that certain details prevented certification, held the issuance, and noted that no coverage was promised.

This demonstrates something crucial: well-configured projects can not only perform designated tasks but also enforce compliance guardrails. The AI isn't just blindly executing,it's checking, verifying, and refusing when something doesn't meet the standards you've set.

One important note about compliance. When you're handling PII (personally identifiable information), you need to use Claude Team or Enterprise plans. These plans offer enhanced security and guarantee that your data won't be used for model training. This is non-negotiable for insurance agencies.

Level Two: The Connection Phase,Bridging to Action

Level Two is where you transition from browser-based consumption to desktop application deployment. This is the phase where you enable direct data connection and initial action-taking capabilities. You're no longer just asking the AI to create things,you're letting it access your systems, analyze your data, and start doing things.

This level introduces several critical capabilities that bridge the gap from passive consumption to active automation. You'll learn about the desktop application, MCP connectors, external data source integration, book of business analytics, and commission reconciliation. Each of these builds on the foundation you established in Level One.

Pillar One: Desktop Application Capabilities

The desktop application differs fundamentally from browser-based access through two additional core features. These are Claude Co-Work and Claude Code.

Claude Co-Work gives you secure access to your computer. It can organize and analyze files, manage screenshots, and perform browser automation,opening browsers and taking actions. It integrates with your projects and workflows. Think of it as the AI working alongside you, handling the tasks that require access to your local environment.

Claude Code operates differently. It functions as an executive assistant or manager capable of taking direct action. For example, it can conduct research,say, identifying high-converting content formats for your marketing. It can use reference materials, like your photos, to create thumbnails. It can execute multi-step creative processes without your involvement. And it can run scheduled routines that check your systems and trigger actions.

Here's a concrete example of a routine. You set up a scheduled routine that runs daily at 6:30 a.m. It checks your agency management system for renewals. If any are found, it sends Slack notifications to your team members about approaching renewal dates. You wake up, and your team already knows what needs attention today. No one had to log in and run a report. It just happened.

Another example: Claude Code can manage your content creation. You ask it to research what types of videos are performing well in the insurance space, create a script based on that research, and generate the video using connected tools. It handles the entire workflow while you focus on other things.

Pillar Two: MCP Connectors

MCP stands for Model Context Protocol. These connectors represent a standardized method for integrating AI with external platforms and services. MCP provides a simplified integration pathway that avoids complex custom coding. This is huge for insurance agencies that don't have dedicated IT departments.

Let me show you how MCP connectors work with a real example. I used a creative platform called Hixfield, connected via MCP, to generate video content for insurance advertisements. Here's the workflow. First, I drafted a sample script for an insurance ad. Then I had the AI determine how long the script takes to speak. Next, I tightened the script to match video length constraints. The AI used the MCP connector to invoke the video generation tool, creating three 15-second clips. Finally, those clips were stitched together into a final advertisement.

The power here is in the number of tools you get access to. MCP connectors provide access to dozens of tools per connection. In the Hixfield example, the platform provided access to 28 interactive tools, 9 read/write tools, and 3 app-only tools,over 40 capabilities accessed through a single connector.

Setting up custom MCP connectors is straightforward. You access the platform's connector management section, name your custom connector, paste the remote MCP server URL provided by the service, and authorize through the browser when prompted. That's it. You now have access to all of that service's tools through Claude.

For insurance agencies, this opens up possibilities like connecting to your AMS, your CRM, or specialized tools for document generation, data analysis, or marketing. The integration possibilities are nearly endless.

Pillar Three: External Data Source Integration

Agencies routinely work with lead spreadsheets, CRM data, and dialing platform information. This pillar addresses connecting the AI to these data sources for live tracking and analysis. This is where you start getting real insights from your data without spending hours in spreadsheets.

Let me walk through a practical demonstration. I had a CSV file called "sample_insurance_leads.csv" containing 40 leads, uploaded to Google Drive. When I asked the AI to locate leads uploaded that day, here's what happened. First, it searched for sheets with "lead" in the title but found no direct matches. Then it recognized the CSV file as the likely candidate based on context. It retrieved the file and provided a direct link. Finally, it analyzed the contents and reported distributions,for example, commercial lines versus personal lines.

But the capabilities go beyond just reading data. The AI can perform calculation functions, computing total premiums, averages, and distributions across datasets. It can manipulate sheets,converting CSVs to native spreadsheet format, creating multiple tabs, sorting by priority, and duplicating data. And it can integrate with modern CRMs like AgencyZoom or GoHighLevel via MCP connectors or custom integrations, allowing lead retrieval from specific pipeline stages and automated pipeline management.

Here's another example. You have a spreadsheet with all your commercial lines prospects. You ask the AI to sort them by estimated premium, identify the top ten, and draft personalized outreach emails for each one. It does this in minutes. Previously, this task would have taken you hours,compiling the data, analyzing it, and crafting individual messages.

This represents a fundamental shift. Pulling and analyzing data of this nature used to take considerable time and effort. With AI integration, it takes seconds. And the more you use it, the more you'll find ways to leverage this capability.

Pillar Four: Book of Business Analytics

With appropriate AMS access, the AI can perform comprehensive book-of-business analysis with read-only capabilities. This provides agencies with insights that historically required dedicated analytical staff or specialized software. For most agencies, this kind of analysis was simply never done because it was too time-consuming and expensive.

There are three primary analytical functions. First, account scanning,the AI reads every account, capturing lines held, premiums, and last contact dates. It can identify "modal line accounts",for example, auto-only or home-only policies that represent cross-sell opportunities. Second, gap identification,the system flags coverage gaps across the book, highlighting accounts with incomplete protection. Third, concentration risk assessment,evaluating the concentration of risk across carriers, lines, and account types.

The resulting report generation provides producer-ready information packages, management-level numbers reports, and input data for downstream renewal and cross-sell engines.

Here's a common observation in the industry: so many people say that they are numbers-oriented, that they really care about numbers, but nobody even bothers to compile a simple report, look at the numbers, and actually draw some conclusions that could help them out, just because it can be a bit tedious or a bit boring. AI automation removes this barrier. You can now get these insights without the tedium.

For example, you might discover that 40% of your book is auto-only policies with no home coverage. That's a massive cross-sell opportunity. Or you might find that you have too much concentration with one carrier, creating risk if that carrier changes their appetite. These insights are gold for agency growth and stability.

Pillar Five: Commission Reconciliation

Commission reconciliation is a critical but often-overlooked operational task. It's tedious, it's detail-oriented, and it's exactly the kind of task that AI excels at. The AI ingests carrier commission statements,in PDF, CSV, or other formats,and matches them against expected commissions recorded in the AMS.

Let me walk through a workflow demonstration. I provided three files: a sample commission statement in CSV format, the same statement in PDF format, and an agency commission tracker. The AI read all three files simultaneously. It reconciled the tracker rows against the statement data. It flagged discrepancies and mismatches. It generated a completed output file with tracker empty columns filled. And it produced a comprehensive summary of findings.

This capability can identify significant amounts of premium each year being lost through reconciliation errors,revenue that would otherwise go undetected. Think about that. You might be losing money right now because no one has the time to carefully match every commission statement against your records. The AI does this in minutes.

Another example: you receive commission statements from five different carriers each month. Each one has a different format. Some are PDFs, some are CSVs, some are portal exports. You upload them all to the AI and ask it to reconcile against your tracker. It normalizes the different formats, matches the data, and flags any discrepancies. What used to take a full day of careful work takes ten minutes.

This is the kind of task that, once automated, becomes a permanent part of your monthly workflow. You'll never go back to manual reconciliation.

Level Three: The Systematizing Phase,Building Repeatable Processes

Level Three is where you create repeatable systems and assets that the AI can consistently reuse. This is the stage where one-time improvements become permanent operational infrastructure. You're no longer just doing tasks,you're building systems that handle tasks for you.

This level covers skills, browser automation, renewal management routines, CSR microtask automation, quote comparison, and your agency knowledge base. Each of these creates reusable assets that compound in value over time.

Pillar One: Building Your Skills Library

Skills are documented processes that teach the AI to perform specific tasks consistently. Once a skill is established, it can be deployed repeatedly with predictable results. Skills differ from projects in that they are designed for modular, reusable execution.

A skill structure consists of three main components. First, the name,a clear identifier that the AI recognizes as a tool reference. Second, the description,a detailed explanation of what the skill does and when it should be used. This is how the AI decides when to deploy a given skill. Third, the process documentation,step-by-step instructions for executing the task.

Let me give you a concrete example of a policy summary skill. The name is "policy-summary." Its purpose is to convert a policy or declarations page PDF into a one-page summary comprehensible to a non-insurance person. It deploys when a user uploads a policy document and requests a summary,the AI automatically invokes this skill.

Here are several essential skills you should consider building:

Policy Summary,converts policy documents to plain-language one-pagers.
COI Intake,processes certificate requests, verifies information, prepares compliant responses.
FNOL Intake,structures first notice of loss data, flags missing information, guides claim initiation.
Endorsement Prep,prepares endorsement or change requests with coverage implications.
Quote Comparison,generates client-ready comparison proposals from multiple carrier quotes.
Skill Creator,a meta-skill enabling the AI to create new skills and improve existing ones.

The Skills Library effectively functions as an entire automation library that the AI can easily access and reliably perform certain processes. You're building your own personal automation infrastructure.

Here's another example. You have a standard process for handling endorsement requests. You document it as a skill with clear steps: identify what was requested, determine what information is missing, draft client-facing communications, prepare documentation for CSR review. Now every time an endorsement request comes in, the AI knows exactly what to do. It doesn't have to be walked through the process each time.

Pillar Two: Browser Automation with Co-Work

Co-Work provides browser automation capabilities that exceed what traditional APIs can achieve. It can click through portals, navigate websites, and take actions exactly as a human user would. This is powerful because many insurance tasks require navigating carrier portals that don't have APIs.

There are two operating modes. The first is "ask before acting",the AI pauses for approval before each action. This is appropriate for compliance-sensitive tasks where you want to maintain control. The second is "act without asking",continuous operation without per-action approval. This is appropriate only for non-sensitive tasks like organizing files or analyzing publicly available information.

The computer use capability means the AI can take control of your computer, navigating through applications and completing tasks that are tedious or time-consuming.

Let me share a demonstration example. I asked the AI to analyze a diagram containing all 21 use cases across five levels. Without any guidance on how to proceed, the AI autonomously opened the Chrome browser, navigated to the correct tab, took screenshots of the diagram, analyzed the content, and produced a structured summary of all pillars.

This demonstrates the AI's ability to independently determine task execution strategy and complete multi-step processes without user micromanagement. It's not just following instructions,it's figuring out how to accomplish the goal.

Another example: you need to pull information from a carrier portal for a client. The AI logs in, navigates to the correct section, extracts the needed information, and compiles it into a summary. You review and approve. What used to take fifteen minutes of clicking through screens takes two minutes.

Pillar Three: Renewal Management Routines

Renewal management represents one of the highest-value automation opportunities in your agency. Routines are automated workflows that execute on schedules or through triggers. They run in the background and make sure nothing falls through the cracks.

Let me give you a detailed specification for a renewal monitoring routine. The schedule is daily, customizable to specific times,for example, before work hours. The trigger is time-based, though it can also be webhooks or other triggers.

Here's the instruction set:

Identify renewals occurring within the next 90 days.
Prioritize by premium amount and days remaining.
Segment into three categories: 90-61 days out, 60-31 days out, and 30-0 days out.
Draft outreach templates tailored to each timeline.
Emphasize urgency for the 30-day group.
Include account name, coverage details, expiration date, and producer assignment in each draft.
Present drafts for producer review and approval before sending.
Mark processed policies to avoid duplicate outreach.
Confirm briefly if no renewals are due.

The schedule flexibility includes one-time execution, hourly, weekday, weekly, custom intervals, and minute-level precision time settings. You can have it run at 7:22 a.m. if that's what works for you.

Routine implementation requires connector access to your AMS,you may need to set up a custom connector if your AMS isn't natively supported. You also need a repository,a folder structure where the routine's code and context reside. And you need clear procedural instructions.

Here's another example. You set up a routine that runs every Friday afternoon. It reviews all policies expiring in the next 60 days, checks which ones have been contacted, and compiles a list of those that haven't been touched. This list goes to your sales team so they can prioritize their outreach. No one has to remember to run this report,it just appears in their inbox.

Pillar Four: CSR Microtask Automation

Customer service representatives handle numerous repetitive tasks that consume significant time. This pillar addresses automating common microtasks that eat up your CSRs' day.

Let's start with FNOL intake. When a claim notification arrives, the AI processes the information through the FNOL skill. Even when complete information isn't available, the AI extracts all provided details, flags missing information,for example, which policy applies, auto or home,and generates a next-steps checklist. That checklist might include confirming the policy, verifying coverage, setting up the claim, and requesting photos. It drafts responses that require CSR finalization.

Endorsement preparation works similarly. The AI processes endorsement requests by identifying what was requested, determining what information is missing, drafting client-facing communications, and preparing documentation for CSR review.

Now, here's the key insight about microtask automation. It's important to understand the distinction between absolute and proportional time savings. Cutting a task down from 5 minutes to 2 minutes has only a 3-minute impact per task. But we're measuring in proportions,going from 5 minutes to 2 minutes means you're spending 2.5 times less time. At scale, handling tens of these cases daily creates exponential efficiency gains that directly improve call handling and overall CSR capacity.

Let me give you a concrete example. Your CSR team handles 30 endorsement requests per day. Each one takes about 5 minutes of processing time. That's 150 minutes,2.5 hours,per day. With AI assistance, each request takes 2 minutes. That's 60 minutes,1 hour. You've just reclaimed 1.5 hours per day for your team. Multiply that by 5 days a week, 4 weeks a month, and you're looking at 30 hours of reclaimed time per month. That's almost a full work week.

Another example: FNOL intake. When a claim comes in, the CSR used to spend 10 minutes gathering information, checking the policy, and drafting the initial communication. Now the AI does the initial processing in 2 minutes. The CSR reviews, adds any missing context, and sends. The client gets a faster response, and the CSR handles more claims per day.

Pillar Five: Quote Comparison Engine

Comparing multiple carrier quotes is traditionally a time-intensive, manual process. The quote comparison skill automates this entirely. This is one of those tasks that agents dread because it's so detail-oriented and requires careful reading of multiple documents.

Here's the operational sequence. First, input,multiple carrier quotes for the same insured, uploaded as PDFs. Second, processing,the AI reads each quote, maps plan structures, and normalizes coverage comparisons. Third, output,a polished, client-ready PDF proposal that includes coverage requirement review, side-by-side comparison at a glance, recommended alternatives with rationale, plain-English coverage notes, carriers reviewed but not recommended, next steps for the client, and disclaimers.

In real-world testing, I uploaded five quotes from different carriers. The system produced a six-page, client-ready PDF ranking the top three carriers, dropping two from consideration, and providing full comparative analysis. The entire process,from uploading quotes to final polished deliverable,took minutes.

The leverage equation here is transformative. Five raw quote PDFs plus a five-word instruction yields a complete, professional client proposal. In just a few minutes we went from very raw information from the quotes to these random PDFs and a five-word sentence to instruct the AI to complete the proposal. If that's not leverage and the best use of your time, then I honestly would not know what it is.

Here's another example. A client is deciding between three commercial auto quotes. You upload all three to the AI and ask for a comparison. It produces a clean, branded PDF that explains the differences in plain English, highlights the best value option, and notes any coverage gaps in the cheaper quotes. You send this to your client, and they make an informed decision. The client perceives you as thorough and professional, and you didn't spend an hour building the comparison.

Pillar Six: Building Your Agency Brain

The agency brain consolidates all institutional knowledge in accessible formats. This is your knowledge base,everything your agency knows about carriers, compliance, procedures, and clients, all accessible to the AI.

There are two primary approaches. The first is a project-based knowledge base. This is a simpler setup using the project structure to house carrier information, compliance requirements, and procedural documents. It's suitable for smaller operations with limited data volumes.

The second is a RAG setup,Retrieval Augmented Generation. This is a more powerful approach that matches queries to relevant documents with absolutely immense accuracy. Files such as carrier appetites, compliance disclosures, and agency context are stored and retrieved dynamically.

RAG has several advantages. There are no context window limitations,you can store hundreds of files. The retrieval is precise, based on query matching rather than trying to fit everything into a single context. And it integrates with skills for combined knowledge and execution.

The optimal configuration is to have skills and RAG work in tandem within the code workspace. The AI references the knowledge base for domain expertise while executing skills for process execution. This resolves the trade-off between having sufficient context versus having access to all skills.

Here's an example. A CSR is handling a complex endorsement request. The AI accesses the agency brain to understand how this particular carrier handles endorsements, what the compliance requirements are, and what the standard procedure is. It then executes the endorsement prep skill using that knowledge. The result is a response that's both procedurally correct and carrier-specific.

Another example: a producer is preparing for a meeting with a commercial prospect. They ask the AI what carriers the agency typically uses for this type of business. The AI accesses the agency brain, pulls the relevant carrier appetite information, and provides a summary. The producer goes into the meeting knowing exactly which carriers to quote and why.

Level Four: The Delegation Phase,Letting Go

Level Four represents the delegation of significant operational responsibilities to the AI system. At this stage, the AI handles complex, multi-step workflows with human oversight. This is where you start to see the real time savings,the kind that transforms your agency's capacity.

This level covers COI autoresponders, co-work quoting, and loss run retrieval. These are the heavy-duty automation workflows that handle some of the most time-consuming tasks in your agency.

Pillar One: COI Autoresponder

The certificate of insurance workflow is one of the most frequent and time-consuming operational tasks in insurance agencies. Every day, requests come in from clients, contractors, landlords, and other parties who need proof of coverage. Each one requires verification, generation, and delivery. The COI autoresponder automates the full process.

Here's the high-level workflow. A certificate request hits your designated inbox,for example, support@youragency.com. The trigger is a webhook that detects the email. The AI detects the email and analyzes its contents,fully automated. It identifies whether the request is standard or exception-based. Via your AMS API or MCP, it verifies policy status,active, limits, and so on. It generates an ACORD 25 certificate. A notification is sent to a team member via Slack or another channel for review,this is your human-in-the-loop checkpoint. After approval, the certificate is sent to the requester and logged in the AMS.

Here are the key statistics and operational insights. Approximately 85% of COI cases can be handled with minimal human touch. However, human review immediately before sending to clients remains the recommended practice. Compliance-sensitive edge cases,coverage limit mismatches, special wording, non-policy issues,route to CSRs for manual handling. Agencies report reclaiming 50+ hours per month with this implementation.

The design principle here is crucial: most automation should operate with human oversight. Fully autonomous operation is generally inappropriate for compliance-facing tasks. The AI does the heavy lifting, but a human reviews before anything goes out the door.

Here's another example of how this works in practice. A contractor emails requesting a COI for a construction project. The AI receives the email, verifies that the policy is active and has the required limits, generates the ACORD 25, and sends a Slack notification to your CSR. The CSR reviews, clicks approve, and the certificate goes out. The entire process takes 3-5 minutes instead of the 20-30 minutes it used to take.

And for the 15% of cases that are exceptions,maybe the requested limits are higher than the policy, or there's special wording required,the AI flags these and routes them to a human for manual handling. The AI never tries to handle something it shouldn't.

Pillar Two: Co-Work Quoting

Co-Work quoting automates data entry into carrier portals,a task that comparative raters cannot handle because many carriers require information to be entered directly into their proprietary portals. This is one of the most tedious tasks in insurance, and it's ripe for automation.

The capabilities here are impressive. The AI autofills carrier forms with client information. It operates with agency brain context,carrier appetites, typical carriers used, portal differences. It functions like comparative raters but with superior contextual awareness.

Let me share a demonstration with a demo carrier portal. The AI activated browser control functionality, which visually surrounded the tab in a distinct color to indicate active control. The task given was: "Fill in whatever info you can, then report back on what you didn't get. If you got everything, stop and let me review."

The AI autonomously read the form structure, scrolled through the page, filled in all client details,name, address, website, coverage types,selected appropriate coverage options, and reported completion status. Critically, the AI also flagged a form deficiency,the coverage checkboxes lacked a standalone liquor liability option. This level of analytical review demonstrates that the AI isn't just entering data; it's evaluating the process itself.

Human review remains essential before final submission. The AI prepares, compiles, and flags; the professional reviews, validates, and sends.

Here's another example. You have a new commercial prospect with a complex risk profile. You feed the AI all the client information. It navigates to the carrier portal, fills in the application, selects the appropriate coverages, and flags any questions it couldn't answer. You review the submission, add any missing details, and hit submit. What used to take 30 minutes of data entry takes 5 minutes of review.

Pillar Three: Loss Run Retrieval

Loss run retrieval automates what is often a tedious, monotonous data-gathering task involving multiple carrier portals and claims history compilation. Every agent knows the pain of logging into multiple carrier portals, navigating to the loss run section, and downloading reports for each client.

Here's the automated workflow. The AI accesses the agent portal. It executes login using provided credentials. It navigates to the loss run section. It identifies clients from the dashboard with pending loss information. It accesses each client profile. It downloads loss run documents. It compiles claims history for remarketing purposes.

There's an operational caveat here. Some portals and platforms have terms of service restrictions on scraping activities. Agencies must verify compliance with both AI platform terms and third-party portal terms before deploying automated retrieval. You need to make sure you're not violating any agreements.

The core principle remains consistent: by automating the 80-90% of the work that is repetitive, the professional retains responsibility for the 5-10% or maybe 20% of work requiring judgment and final review.

Here's an example. You're remarketing a commercial account with three carriers. You need loss runs from all three. The AI logs into each carrier portal, pulls the loss runs for the past five years, and compiles them into a single document. You review the compilation, add your notes, and send it to the market. What used to take an hour of logging in and out of portals takes 10 minutes.

Another example: you have a book of 50 commercial accounts that need annual loss run reviews. The AI works through them systematically, pulling each one and flagging any that show concerning trends. You review the flagged accounts and prioritize your conversations accordingly.

Level 4.5: Autonomous Growth Systems,Scaling Your Agency

Level 4.5 addresses revenue generation rather than operations. It's designated between Level Four and Level Five because these systems are operationally simpler but strategically more leveraged. This is where you move from saving time to making money.

There are two principal growth systems: the referral partner pipeline and the Facebook ad machine. Both leverage AI to generate leads and grow your book of business.

Growth System One: Referral Partner Pipeline

This is a five-step process for building referral partnerships. Referral partners,loan officers, real estate agents, other professionals who interact with potential insurance clients,are one of the most valuable lead sources for any agency. But building these relationships at scale is time-consuming. AI changes that.

Here are the five steps. First, identify the best referral partners,loan officers, real estate agents, and similar professionals. The AI identifies target profiles based on your criteria. Second, craft messaging for partner outreach. The AI generates personalized messaging templates. Third, source partners,for example, via Google Maps. This involves web scraping and data gathering automation. Fourth, find and verify email addresses. This is bulk email verification. Fifth, send outreach messages and convert to referral relationships. Automated sends can go up to 1,500 emails per day, and the AI books meetings with interested partners.

Skills relevant to this system include find email, scraping, Google Maps processing, and lead pipeline processing.

The reported impact is significant. One agency generated additional premiums of $90,000 per month through this system. Let me put that in perspective. That's over a million dollars in annual premium,from a system that runs largely on its own once set up.

Here's how it works in practice. The AI identifies loan officers in your target geographic area who are active on social media and have a strong online presence. It compiles a list with their contact information. It drafts personalized outreach emails that reference their specific business and explain how a referral partnership would benefit them. It sends these emails and manages follow-ups. When a partner expresses interest, the AI books a meeting on your calendar. You show up, build the relationship, and close the partnership.

Growth System Two: Facebook Ad Machine

The second growth system pairs video and image generation with a testing and scaling engine. This is your AI-powered lead generation machine.

Here's the structure. First, creative generation,AI avatar videos, informative ads, and image creatives produced in minutes. Platforms like Hixfield, connected via MCP, handle this. Second, testing protocol,the Meta ad system tests different creatives to see which ones perform. Third, optimization,poor performers are removed, and winning creatives are scaled.

Here's the industry context that makes this so powerful. Lead vendors use essentially this same strategy,running Meta ads, mixing in aged and shared leads, and selling overpriced lists to agents. Agencies adopting this system get the same method but own their infrastructure, rather than paying premium prices for externally sourced leads. You're cutting out the middleman and keeping the margins.

The strategic sequencing here is important. Level 4.5 should typically be the last stage, and only after operational efficiency has been achieved. If you have a broken system where you can't delegate or systematize,where you haven't prioritized systems in your CRM, tasks like data analysis or comparison renewal tasks, or CSR microtasks,it's a much better and more proven strategy to first handle your operations rather than immediately jump to autonomous growth. You can't scale a broken system.

Here's an example of the Facebook ad machine in action. The AI generates 10 different ad creatives,some with AI avatars speaking to pain points, some with text overlays, some with images. These are loaded into Meta's ad system. The system tests them against your target audience. After a week, three creatives are clearly outperforming the others. The AI scales those three and pauses the underperformers. You're generating leads at a fraction of the cost of buying them from a vendor.

Key Insights and Best Practices

Frequently Asked Questions