For independent commercial agencies

Stop losing quotable leads to slow follow up

Learn where AI belongs, then fix the handoffs that are currently costing you

Pull a file

What I build

AI workflows that decide, judge, and act, only where it is safe to.

Not a tool you log into and not a subscription. AI installed into your agency where it fits, built around how your team already works. Start with one workflow for one process. Add more if it helps, or do not. That part is up to you.

A quote request that never sits overnight.
It reads the inbound request, sends a real reply in minutes with the next step, gathers the intake answers, and hands a ready file to the producer. It never quotes. That waits for a person.
A renewal list that builds itself every Monday.
It watches your policy data, surfaces which accounts are coming up and which went quiet, and puts them in front of the right producer early. The producer still decides what to do with each one.
A certificate that drafts itself and waits for a yes.
Routine requests draft from the policy record and sit in a queue for release. Anything asking for wording the policy does not carry stops cold for a person, every time.
The rekeying between systems, gone.
It pulls the fields off an application and stages them in your management system for a person to confirm. Your AMS stays the system of record. Nothing commits without a human yes.

These four are examples, not a menu. Every build is shaped around your team and pointed at the outcome that matters most to you, whether that is speed to lead, quote requests, renewals, or something else entirely. You do not need to know which one you need. That is what the first conversation is for.

Privacy Speed to lead Renewals Service Risk & safety OuterMoon · Systems for independent commercial agencies
Library AI Use Policy Where Your Client Data Goes
Risk and safety · 7 min read

Where Your Client Data Goes

What to ask a vendor before any client information touches their tool, and what a good answer sounds like.

The question that ends most demos

Ask a vendor whether your data is used to train models shared with other customers, and ask for the answer in writing.

Serious vendors answer in one sentence. The ones who talk around it are telling you something.

The four answers you need

  • Where the data physically lives, by country and provider.
  • Who at the vendor can see it, and whether that access is logged.
  • How long they keep it, in days, and how you get it deleted.
  • What happens to it if the company is sold or shuts down.

What good looks like

A SOC 2 Type II report handed over without hesitation. Type II means they were audited over a period rather than on one good day.

Written terms saying the data is yours. A named retention window. Deletion on request, in the contract rather than in an email.

The book export question

Almost every tool wants a book export to be useful. That file is your entire client list, premium, and commission in one place.

Before you send it, ask where it lands, who else can reach it, and whether it gets deleted when you stop paying. A vendor who has thought about this has an answer ready.

Where this leaves your AMS

The safest shape is the one where your management system stays the record. Tools read from it and write back to it. Nothing moves in.

A tool that wants your book to live in its own system is asking for something much bigger than a subscription, and it makes leaving expensive on purpose.

The regulator angle

The NAIC model bulletin on the use of AI has been adopted by roughly 25 states plus the District of Columbia as of early 2026. It is written for insurers rather than agencies, but the third party oversight language matters to you.

The principle it sets is simple. Using someone else's tool does not move the responsibility off you. If the tool misbehaves on your account, it is still your account.

Source: NAIC AI Model Bulletin Implementation Map, March 2026.
Next step

Get the Handoff Map for your agency

Walk me through a normal Tuesday and I send back a one page map of every handoff in your agency, which ones can run on their own, and the one costing you most. Free, and yours to keep either way.

You will not end up on a sales sequence. I do not run one.
Library AI Use Policy When the Output Is Wrong
Risk and safety · 6 min read

When the Output Is Wrong

Made up answers look exactly like correct ones. Here is where they show up in agency work and how to catch them before a client does.

The part that makes this dangerous

A tool that invents a policy limit states it in the same confident tone it uses for a correct one. There is no tremor in the voice.

That is why reading a draft for sense is a different task from checking it against the source, and only the second one catches this.

Where it shows up in an agency

  • A limit or effective date that reads plausibly and does not match the policy.
  • A certificate with wording nobody asked for and the policy does not carry.
  • A summary that drops an exclusion without flagging it.
  • An answer to a coverage question that sounds authoritative and is not grounded in anything.

The three controls that matter

  • Numbers get compared against the source document, not read for plausibility.
  • Anything the system is unsure about routes to a person instead of going out.
  • Coverage questions never get answered by a tool, at any confidence level.

Grounding the answers

The common fix is forcing the tool to look up the answer in your own documents before responding, with a citation you can click.

It sharply reduces invented answers. It does not erase them, which is exactly why a person still signs off.

What to ask a vendor

Ask what happens when the tool does not know. The right answer is that it escalates to your team with the context attached.

A tool that guesses when it is uncertain is worse than one that stops, because a stop is visible and a guess is not.

The standard that holds up

Treat every AI output as a draft from a new hire on their first week. Useful, fast, and checked before it leaves the building.

That framing survives an E and O conversation. Trusting the output because it sounded right does not.

Next step

Get the Handoff Map for your agency

Walk me through a normal Tuesday and I send back a one page map of every handoff in your agency, which ones can run on their own, and the one costing you most. Free, and yours to keep either way.

You will not end up on a sales sequence. I do not run one.
Library AI Use Policy Where the Human Check Belongs
Risk and safety · 6 min read

Where the Human Check Belongs

Not everything needs review, and reviewing everything means reviewing nothing. Here is how to place the checkpoint so it holds.

The line that decides it

There is a difference between moving information and judging it. Moving a policy number from one system to another is transport. Deciding whether a changed limit is a problem is judgment.

Transport can run on its own. Judgment stays with a licensed person. Almost every good decision about where to put a checkpoint comes from that line.

What needs a person, every time

  • Anything that answers a coverage question.
  • Anything that binds, quotes, or commits the agency.
  • Any certificate wording that is not already on the policy.
  • Any client facing message where being wrong costs the relationship.

What does not

  • Acknowledging that an inbound request arrived.
  • Pulling documents together into one file.
  • Chasing a loss run on a schedule.
  • Logging what happened to the client record.

Name the role, not the person

A policy that says someone reviews it fails the first time that someone is on holiday. Name the role instead. The account manager checks the certificate against the requirement. The producer approves anything with a number in it.

Then say what they are checking for. Review is not a useful instruction on its own.

Make the checkpoint the easy path

The review step survives when approving takes one tap and the context is already attached. It gets skipped when it means opening three systems.

A draft that arrives with the request, the client record, and the reason it stopped is a checkpoint people use on a busy Thursday.

Write down who owns the output

Every automated workflow needs one named role accountable for what it produces. Work produced by a system and checked by nobody is the exposure.

If a regulator or your E and O carrier ever asks who was responsible, the answer should already be written down.

Next step

Get the Handoff Map for your agency

Walk me through a normal Tuesday and I send back a one page map of every handoff in your agency, which ones can run on their own, and the one costing you most. Free, and yours to keep either way.

You will not end up on a sales sequence. I do not run one.
Free tool

What is the slippage costing you?

Match the sliders to your agency, then answer two honest questions about how things run today. Nothing is uploaded and nothing is stored.

Your volumes
How it runs today
How do renewals get worked?
Who answers a new lead first?
0 hrs
Staff hours reclaimed every week
0
Full time roles you no longer need to fill
$0
Commission protected per year from renewals and leads that stop slipping
Everything above is now shown at half of published industry benchmarks. If even the half number is worth having, the argument is over. Click again for the full figures.

Estimates use published industry benchmarks for certificate handling time, renewal retention lift, and lead response conversion. Your agency is not average, so treat this as the shape of the problem rather than a quote. Run it with your own payroll numbers before you believe it.

Want to see where these hours go?

The maps walk four of these workflows step by step, including where a person still has to decide.

Open the Workflow Maps Read the Library
Hub AI Use Policy Guide Template
Free template

AI Use Policy Template

The nine sections, prefilled. Swap the blanks for your agency and it is done. Written so a producer can read it in five minutes.

One step

Where should I send it?

Put your email in and the full template opens below, ready to copy. You will also get The North Star Guide, one email a week on where AI fits inside a commercial agency.

No spam and no sales sequence. Unsubscribe whenever you want.
Paste it into a doc, fill the blanks, have every person sign it.
Artificial Intelligence Use Policy
______________________ (agency name)  ·  Effective ____________  ·  Review date ____________
1

What this covers

This policy applies to everyone at ______________________ (agency name), including producers, account managers, CSRs, and anyone working under contract. It covers any tool that generates text, reads documents, or drafts communication, whether the agency pays for it or not.

2

Approved tools

Only the tools on the approved list may be used with any client information. The current approved list is kept at ______________________ (where it lives) and is reviewed by ______________________ (role) every ______ months.

If a tool is not on the list, it is not approved. Ask before using it, not after.

3

What never goes in

Do not enter any of the following into a tool that is not on the approved list:

Client names, business names, or addresses
Policy numbers, premiums, or loss history
Social security or federal tax ID numbers
Anything from a loss run, application, or dec page
Bank or payment details

When in doubt, leave it out and ask ______________________ (role).

4

Who checks the output

AI output is a draft. A person reviews it before it leaves the agency.

Certificates are checked against the policy record by ______________________ (role).
Client emails are approved by ______________________ (role).
Anything with a premium, limit, or coverage statement is approved by a licensed person, without exception.

The person who approves it owns it.

5

Where AI is never used

No tool answers a coverage question for a client.
No tool quotes, binds, or commits the agency to anything.
No tool issues certificate wording the policy does not carry.
No tool speaks to a client without a person having read it first.

These are not preferences. They are the line where our license sits.

6

What gets written down

Every automated client interaction is logged to the client file in ______________________ (your AMS), the same as a phone note.

Each log entry keeps five things: when it happened, who ran it, which tool and version, what it touched, and whether a person reviewed it. Those five are what an E and O carrier asks for after a claim.

Each automated workflow has one named owner: ______________________ (role) is accountable for what it produces.

If a regulator or our E and O carrier asks who was responsible, the file answers.

7

When something goes wrong

If a wrong output reaches a client, tell ______________________ (role) the same day. Do not wait to see whether anyone noticed.

We correct it with the client directly, record what happened, and change the workflow so it cannot repeat.

Nobody is in trouble for reporting a mistake. People are in trouble for hiding one.

8

What we require of any outside tool

If an outside AI tool misbehaves on our account, it is still our account. Responsibility never transfers to the tool.

Before any tool touches client work, ______________________ (role) confirms three things in writing: where our data goes and whether it trains their models, how fast they must tell us about a breach or a bad output, and how we get our data out if we leave.

A tool that cannot answer those three does not touch client work.

9

If someone gets it wrong

First time: a conversation, and we fix whatever made the mistake easy to make.

Repeated or careless: ______________________ (consequence, in line with your handbook). Putting client data into an unapproved tool is treated as seriously as any other data breach.

This applies to everyone, including owners.

The five line version

If nobody reads the full policy, they read this. Put it where people will see it.

  1. Only use tools on the approved list.
  2. Never put client detail into a tool that is not approved.
  3. Every AI draft gets read by a person before it leaves.
  4. Coverage questions, quotes, and binding stay with a licensed person.
  5. If something wrong reaches a client, say so the same day.
Next step

The policy is the easy half

Writing the rule takes an afternoon. Building the workflow so the rule holds on a busy Thursday is the part most agencies never get to. That is the part I build.

You will not end up on a sales sequence. I do not run one.
Hub Who builds this
OuterMoon

AI workflows built around your agency’s biggest operational headaches.

OuterMoon builds AI workflows for independent commercial agencies, shaped around where the week loses the most time. Speed to lead, quote requests, renewals, certificates, whatever costs you most. You can buy one workflow for one process and be done, or keep building as it proves out. There is no one size that fits every agency, so it grows at whatever pace helps you. Every build decides, judges, and acts only where it is safe to, and stops for a person everywhere else.

Where the instinct comes from

I come from a world where if a system failed, it could mean injury or death, mine or someone else's. That kind of work changes how you think. You stop trusting anything that runs on hope. You build the check in before you need it, because the cost of finding out later is a person getting hurt.

That way of seeing never left. I look at a process and I see where it will break, under load, at 5pm on a Friday, when the person who usually catches it is out. I see how the pieces move and where one quiet failure turns into a real one.

I do not know insurance the way you do. OuterMoon’s job is the systems: ones that do not need babysitting and never make a call they should not be making.

What I believe about AI in an agency

01

AI is safe where it moves information. Decisions that could cost money, a client, or a license keep a person making the final call. A tool can sort, draft, route, and remind. A licensed person quotes, binds, and answers coverage. The whole job is knowing exactly where that line sits and building it in.

02

A wrong answer sounds exactly as confident as a right one. So every build has a person at the point where being wrong would cost you.

The first step

You do not need to know what to automate

That is the whole point of the first conversation. You walk me through a normal Tuesday, I map where the work slips, and I show you the one thing worth building first. Free, and yours to keep either way.

Hub Privacy
OuterMoon

Privacy policy

Plain English, because that is how everything here works. This page says what gets collected on this site, where it goes, and how to get it removed.

What gets collected

An email address, if you choose to give one. That happens in four places: unlocking the AI policy template, joining the newsletter, leaving a question with Vega, or filling in the Handoff Map sheet. The sheet also collects your answers to its questions, and a name and agency name if you add them. Nothing is collected without you typing it in and pressing a button.

What it is used for

To send you the thing you asked for, and to send The North Star Guide newsletter if you signed up for it. Sheet answers are used to prepare your Handoff Map. That is the whole list.

Where it is stored

Submissions are processed through n8n, an automation service, and stored in Notion as a contact list and in Beehiiv, the newsletter platform, if you subscribed. A notification also goes to a private OuterMoon Slack channel so requests get seen quickly.

What is never done with it

Your address is never sold, rented, or shared with anyone outside the tools named above. There is no advertising on this site. Visits are counted with Plausible, a privacy friendly analytics service that uses no cookies and cannot identify you. Nothing you type is sent to it.

Getting removed

Every newsletter email has an unsubscribe link. To have your information deleted entirely, send a message through LinkedIn or reply to any newsletter email and ask. It gets removed from Notion and Beehiiv, usually within a few days.

Questions

Ask through the DM link anywhere on this site. Last updated July 2026.

Hub The Handoff Map
Free, and yours to keep

The Handoff Map

Walk me through a normal Tuesday at your agency. I'll send back a map of every place AI should never touch and where it should never have the final say, the two fixes you can make this week, and one workflow built for your setup with the instructions to turn it on.

1 The map A one page diagram of how work moves through your agency, from a quote request landing to a renewal going out. Every handoff marked as either something a system can carry or something that needs your judgment.
2 Two fixes you can make this week Two changes inside the tools your team already runs. Most agencies have at least one setting or connection sitting switched off, and finding it takes a conversation, not a project.
3 One workflow, built for you I'll build the workflow and send you a ready to deploy file. About 5 to 10 minutes to get it set up depending on the size, with every step explained in plain language and no jargon.

What I need from you

  • Fifteen minutes on a call, or thirty if you want to go through it properly.
  • Roughly how a quote request reaches a producer today. Who touches it and what they open.
  • How you find out a renewal is coming up.
  • What happens when a certificate request comes in after hours.
  • The name of your management system, and anything else your team lives in daily.

If you fill in the sheet below first, the call stays a tight fifteen and the map comes back sharper.

What it costs

  • Nothing.
  • No book upload, no data, no access to any of your systems.
  • No contract and no follow up sequence.
  • Everything is yours whether we ever work together or not.

I keep no copy of anything you tell me beyond the notes I need to draw the map.

Optional, about two minutes

The sheet

Answer what you can. Skip anything you are not sure about. This exists so the call is fifteen minutes instead of thirty, and so the map comes back sharper.

0 of 12 answered
1How many people work at the agency?
2What management system do you run?
3How does a new quote request usually reach a producer?
4Honestly, how long before someone replies to an inbound request?
5How do you find out a renewal is coming up?
6How far out do you usually start working a renewal?
7Roughly how many certificate requests come in a week?
8What happens when a certificate request arrives after hours?
9What else does your team live in every day?pick any that apply
10Have you tried an AI or automation tool before?
11Which of these costs you the most right now?
12Does anyone there run automation software already?
Anything the questions missed about how your week runs? optional
Sending this books nothing. You still pick your own time. Nothing here is shared with anyone.

Fifteen minutes, and you keep all three

Pick a time that suits you. Fill in the sheet first if you want the short version of the call.

Book the call Ask me something first
Am I going to get sold to?No. There is no deck and no follow up sequence. If I think I can help I will say so once, at the end.
Do I need to know anything about AI?No. I ask about your work, not about technology. Knowing how your office runs is the only expertise this needs.
Do you need access to my systems?No. Nothing gets connected, uploaded, or shared. This is a conversation and a drawing.
What if we already tried AI and it did not work?That is the most useful call I can have. Knowing what failed and where tells me more than a clean slate would.
OuterMoon  ·  Resource Library
OuterMoon Resource Library
· Build 1.0
Free access No signup
LibraryAll Resources

Everything I know about AI in an agency

Everything here is about AI and where it belongs in an agency. What it can carry, what it must never touch, and how to tell the difference. Every figure sourced.

Risk & Safety

Where AI belongs nowhere near the work, and how to write that down.
The AI Use Policy for an Independent Agency
The written rule for what AI may touch, who signs off, and what never leaves without a person reading it. Seven sections, and a template you can adopt today.
Guide7 sectionsFree read
View →
Where AI Can Touch Certificate Work
Which certificate requests a system can safely draft on its own, which ones have to stop for a person, and the one limit no AI platform has solved yet.
Carousel9 slidesFree read
View →
What AI Can Safely Do With an Inbound Lead
Eight things a system can do the minute a quote request lands, and the line where it has to hand over. Score your own setup as you read.
Carousel10 slidesFree read
View →
How AI Connects to Your Agency Stack
Where an AI system plugs in, what it reads, what it writes back, and why your management system stays the system of record.
Carousel12 slidesFree read
View →
12 AI Questions To Ask Any Vendor
The twelve questions I would ask about the AI itself. What it does with your data, where it stops for a person, and what it does at 2am when something breaks.
Carousel13 slidesFree read
View →
What AI Does Not Replace
The useful question is what AI should never be handed. Where I stop every build, and the test I use to decide.
Carousel8 slidesFree read
View →

Start with the AI use policy

The nine sections, who signs off, and a one page version you can adopt today.

Next step

Get the Handoff Map for your agency

Walk me through a normal Tuesday and I send back a one page map of every handoff in your agency, which ones can run on their own, and the one costing you most. Free, and yours to keep either way.

You will not end up on a sales sequence. I do not run one.
15 resources · Speed to lead Renewals Service Risk & safety OuterMoon · Systems for independent commercial agencies
OuterMoon  ·  AI Insurance Library
Plain English Glossary
· 41 terms
Free read No signup
Library Plain English Glossary
Reference

What AI does inside an agency, term by term

Every term here earns its place by touching AI. Each one says what it is, what a system does with it, and where a person has to stay in the loop so nothing goes out unchecked.

41 terms
ACORD formService
The standard industry application and certificate forms.
Where AI fits: Standard layout makes these the easiest documents for AI to read, which is why nearly every setup starts here. Accuracy drops on scans and handwriting, so the person reviewing compares the extracted numbers against the form rather than reading for sense.
Agentalso: agentic AI
A tool that takes several steps on its own toward a goal rather than answering one question at a time. Nothing to do with an insurance agent.
Where AI fits: More steps taken unattended means more places an error can travel before anyone sees it. The safe pattern is letting it run the steps and stopping it at the point where something would leave your office.
AI use policyRisk
Your written rules for which tools staff may use, what data never goes in, and who approves output before it leaves.
Where AI fits: This is the document that governs everything else on this page. The 2026 Big "I" ACT report found 55 percent of agencies have none while two thirds plan to use AI more this year.
AMSalso: agency management systemService
The system holding your policies, clients, activity notes, and accounting.
Where AI fits: It stays your system of record and AI sits on top of it, reading from it and writing back. A tool that wants your book to live in its own system instead is asking for something much bigger than a subscription.
Appetite matchingSpeed to lead
Comparing a risk against what each carrier currently wants to write.
Where AI fits: AI returns a shortlist in seconds instead of from memory. The whole value sits in how current the appetite data is, so a producer confirms the top two or three before anything goes out. Stale data produces a confident list of markets that will decline you.
Audit trailRisk
A record of what the system did, in what order, and who approved it.
Where AI fits: A trail shows the sequence. It does not by itself show who owned the decision, so the approval gets recorded at the moment it happens rather than reconstructed later.
Book of businessalso: your bookRenewals
Every policy your agency has on the books, with premium and commission attached.
Where AI fits: Almost every AI tool wants a book export to be useful, which makes it the file to ask about. Where does it live, who else can see it, can you get it deleted, and what happens to it if the vendor is acquired.
Certificate trackingalso: COI trackingService
Monitoring third party certificates against contract requirements and chasing what is missing.
Where AI fits: AI reads the certificate, compares limits and dates against requirements, matches holder names, and flags expirations. As of 2026 no major platform fully automates endorsement interpretation, so complex endorsements route to a person on every serious system. If a vendor claims otherwise, ask them to prove it on your files.
Claims status inquiryService
The where is my claim call from an insured who wants an update rather than a decision.
Where AI fits: AI answers from live claim and policy data at any hour. Status is fact and is safe to answer directly. Anything about coverage, fault, or outcome is not status, and the setup should hand those to a person instead of attempting them.
COIalso: certificate of insuranceService
Proof of coverage issued to a third party, usually because a contract requires it.
Where AI fits: AI drafts it from the policy record on request, including after hours, and holds it in a queue for a person to release. The release step is the control. A tool that issues whatever wording a requester asks for will eventually issue wording your policy does not support.
Confidence score
A number a tool attaches to its own output, meant to show how sure it is.
Where AI fits: Useful for routing, not for trusting. The practical use is sending anything below a threshold to a person automatically, so the low confidence items land on a desk instead of going out.
Copilotalso: assistant
A tool that sits alongside a person and suggests rather than working unattended.
Where AI fits: A good default for a first rollout, because the person stays in the seat and nothing goes out unreviewed.
Data retentionRisk
How long a vendor keeps what you send them, and whether you can have it deleted.
Where AI fits: A vague answer here is the clearest warning sign in a demo. Ask for the retention period in days and get it in writing before any client information goes near the tool.
Deterministic and non deterministic
A deterministic system gives the same answer every time. AI language tools do not, so the same request can come back worded differently.
Where AI fits: It is why testing one output tells you little about the next. Consistency comes from the checks around the tool, not from the tool.
Direct bill reconciliationalso: commission reconService
Checking carrier commission statements against what your agency is owed.
Where AI fits: AI compares line by line and surfaces the variances. Repetitive comparison is what these tools are genuinely good at, and it is the rare case where the output is a dollar figure your accounting person can verify directly.
Document classificationService
Sorting what arrives into what it is. Loss run, ACORD, policy, endorsement, invoice.
Where AI fits: This is the foundation nearly everything else sits on, because nothing can be routed until it is identified. Anything the system cannot classify with confidence goes to a person rather than being filed as a guess.
E and O exposurealso: errors and omissionsRisk
The risk of a claim that your agency made a professional mistake.
Where AI fits: AI does not change who is responsible. Quote and bind authority stays with the licensed person, and AI output is treated as a draft that person verifies, the same as intake notes from a CSR.
Email triageService
Reading what arrives in a shared inbox, working out what each message needs, and routing it.
Where AI fits: AI parses incoming mail, creates the request record, and assigns it. Where it fails is tone, because an urgent message written casually can be sorted low. The queue still needs a person watching it rather than trusting the sort.
Endorsement processingService
Handling a mid term change to a policy, from request through paperwork and record update.
Where AI fits: AI reads the request, drafts the paperwork, and updates the record. Interpreting what an endorsement does has not been solved. Over a thousand additional insured forms exist with state by state differences, which is why the complicated ones go to a person.
Exposure data collectionSpeed to lead
Gathering the detail a carrier prices from. Property values, payroll, vehicle schedules.
Where AI fits: AI pulls these off documents you already hold rather than asking the client again. An extraction error here follows the account all the way to a policy, so this is the place a person checks numbers against the source before anything is submitted.
FNOLalso: first notice of lossService
The first report of a claim from an insured.
Where AI fits: AI takes the notice at any hour, pulls the policy references, and routes it with a structured summary. Intake and routing only. The handling decision is not the tool's to make and the setup should not let it try.
Generative AIalso: gen AI
AI that produces new material such as a draft email or a summary, rather than only sorting what exists.
Where AI fits: This is the category most agency tools sit in. It writes drafts. It does not decide anything, and a setup that lets a draft leave without approval has skipped the only control that matters.
Hallucinationalso: made up outputRisk
When a tool states something untrue in the same confident tone it uses for everything else.
Where AI fits: A wrong policy limit reads exactly like a right one. The control is comparing every number against the source document before it leaves your office, which is a different task from reading the draft for sense.
Human in the loopalso: human reviewRisk
A person reviews and approves before work goes anywhere. The tool drafts, the person decides.
Where AI fits: In practice this means naming the role rather than a person, and saying what they check. The account manager checks the certificate against the requirement. The producer checks the client email. Anything the system is unsure about goes to that queue automatically instead of going out.
Integrationalso: APIService
A connection that lets two systems pass information without a person retyping it.
Where AI fits: The gap in most agencies is not that tools lack connections, it is that each one connects to the AMS and none connect to each other, so a person still carries information across. Ask which systems a tool writes back to, by name.
Loss runsService
A carrier report of an insured's claim history, needed for most commercial submissions.
Where AI fits: AI sends the request, chases it on a schedule, and reformats what comes back into submission format. Reading what the history means for the risk stays with your producer.
Modelalso: LLM, large language model
The engine underneath a tool like ChatGPT. It predicts likely text based on what it learned from.
Where AI fits: The model is not the product. Two tools can run the same model and be worth completely different amounts depending on what is built around it and where the checks sit.
OCRalso: document extractionService
Reading text off a scanned page or PDF and turning it into usable data.
Where AI fits: This is what reads an ACORD form or a loss run. Accuracy drops on poor scans and handwriting, so anything below a confidence threshold should route to a person rather than being filed.
Orchestration
Coordinating several tools and steps so they run as one connected process, with something managing the order and the handoffs.
Where AI fits: This is usually the difference between speeding up a step and removing it. One clever tool finishes its own task faster and hands off to a person anyway.
Pilotalso: proof of concept
A limited trial on a small slice of work before committing.
Where AI fits: Most pilots look good and stall on the way to daily use, because the demo ran on clean material and real files are messy. Pilot on your worst inputs, not your best.
Policy checkingService
Comparing the issued policy against what was quoted or bound.
Where AI fits: AI lists every difference in seconds, which is the tedious part. Whether a difference is a problem is a coverage judgment and stays with your licensed person, so the output is a list to review rather than a verdict.
Prompt
What you type in. The instruction you give the tool.
Where AI fits: Most disappointing results come from vague instructions rather than bad tools. Specific instructions with an example attached get much closer on the first try.
Quiet accountalso: lapsing accountRenewals
A client with no meaningful contact in a long stretch. Still on the books, no longer engaged.
Where AI fits: AI flags accounts with no logged activity over a set window and drafts a check in. The flagging is the useful part, and it only works if your activity notes are current, so a system pointed at a thin file will find nothing.
Renewal outreachalso: renewal sequenceRenewals
Contact that goes to a client ahead of their renewal date.
Where AI fits: Drafting the outreach is the easy half and most tools stop there. Surfacing which accounts are coming up and which have had no contact is the half that matters, and that is a reporting job against your policy data.
Shadow AIRisk
Staff using AI tools you have not approved and do not know about.
Where AI fits: It is already in most offices, usually as someone pasting client detail into a free tool to draft an email. One approved tool and a written rule solves it. A ban moves it somewhere you cannot see.
SOC 2Risk
An independent audit of how a vendor handles security and data. Type II means checked over a period, not on one day.
Where AI fits: Asking for the report is a normal request. A serious vendor hands it over.
Speed to leadalso: lead response timeSpeed to lead
The time between an inbound inquiry arriving and a real reply going back out.
Where AI fits: AI acknowledges the inbound in seconds, asks the qualifying questions, and puts a booked call or routed lead on a producer's desk. A fast reply is not a quote, so anything with a number in it still comes from a person.
Structured and unstructured data
Structured data sits in fields, like your policy records. Unstructured data is loose, like an email, a PDF, or a phone note.
Where AI fits: Most of the time your team loses is spent turning unstructured material into structured records. That conversion is where these tools pay off first.
Submission intakeSpeed to lead
Gathering what a carrier needs to quote a commercial risk.
Where AI fits: AI reads the ACORD, the loss runs, and the supporting documents and fills the fields so a producer reviews instead of types. Extraction accuracy drops on poor scans, so the review step stays.
Submission triageSpeed to lead
Scoring an incoming submission for completeness and routing it.
Where AI fits: AI sorts and routes. It does not decide whether to write the risk, and a vendor describing this as AI underwriting is overselling it.
Training dataRisk
The material a tool learned from. Separately, some vendors train on what you type in, which is the one worth asking about.
Where AI fits: If a vendor trains on your input, client information can end up shaping a product other people use. Ask in writing whether your data is used for training, and keep client identifying detail out of any tool that has not answered.
No term matches that.
Try a shorter word, or clear the filters.

Write the policy while it is fresh

The guide walks the nine sections, who signs off, and a one page version you can adopt today.

Read the AI policy guide Get the policy template
Next step

Get the Handoff Map for your agency

Walk me through a normal Tuesday and I send back a one page map of every handoff in your agency, which ones can run on their own, and the one costing you most. Free, and yours to keep either way.

You will not end up on a sales sequence. I do not run one.
41 terms· Back to the library OuterMoon · Systems for independent commercial agencies
OuterMoon  ·  AI Insurance Library
AI Insurance Library
· Guide 01
Free No signup
Library AI Use Policy
Risk and Safety

The AI Use Policy for an Independent Insurance Agency

What to write, who signs off, and how to keep it from sitting in a drawer.

Why most agencies do not have one

Your team is already using these tools. The question is whether they are doing it against a rule you wrote or a rule they guessed at.

55%
of independent agencies have no written AI use policy. Another 23 percent have one in development. Only 13 percent have adopted a formal policy.
2026 Big "I" Agents Council for Technology Tech Trends Report

In the same survey, two thirds of agencies said they plan to increase their AI use in the next twelve months. Thirty eight percent said very likely, thirty percent said somewhat likely.

So the use is going up and the written rules are not.

That same report asked agencies what worries them. Data privacy and compliance risk came first at twenty four percent. Inaccurate outputs came second at twenty two percent.

Kasey Connors, executive director of ACT, put it this way in the release: agents are not worried about the price of AI, they are worried about the cost of getting it wrong.

Both of those concerns are handled by a policy, not by a product. No tool decides for you what a CSR is allowed to paste into a chat window. No tool decides who checks an AI drafted email before it reaches a client. Those are your rules, and if they are not written down, every person on your staff is making them up on their own.

A written policy does three things
  1. Tells your team what is allowed, so they stop guessing.
  2. Tells them what is never allowed, so the worst outcome is off the table.
  3. Names who is accountable for checking output before it leaves the agency.

That is the whole job. It does not need to be long.

What goes in the policy

There are nine sections. Each one answers a question your staff is already asking silently.

1 Which tools are approved

List them by name. A tool that is not on the list is not approved.

This sounds strict and it is the most useful line in the document. Without it, your staff is using whatever free tool they found, and you have no idea where your client data is going.

Include the version or account type. A paid business account and a free consumer account of the same product often handle your data differently. Name which one you mean. Review the list quarterly, because tools change their terms.

2 What data never goes into an AI tool

This is the section that prevents the worst day.

Never entered into any AI tool

Client names paired with policy details. Dates of birth. Social Security numbers. Tax identification numbers. Driver license numbers. Bank or payment information. Loss run detail tied to a named insured. Anything from a client file that would identify the client.

Say what to do instead. Most work can be done with the identifying details stripped out. An AI tool can draft a renewal email without knowing the client's name. It can summarize a policy form without the insured's address attached.

If a tool is approved specifically for client data because you have reviewed how that vendor stores and handles it, say so by name in this section. Otherwise the default is no.

3 What AI is allowed to do

Be specific. Vague permission produces vague behavior.

Approved uses in most agencies look like this: drafting a first version of an email, summarizing a long document into plain language, cleaning up notes after a call, drafting a checklist, answering a general question about a coverage concept, and reformatting information the agency already has.

Notice what all of those have in common. The AI produces a draft that a person then reads. Nothing on that list reaches a client or a carrier without a human in between.

4 What AI is never allowed to do

Just as specific.

Off the table

AI does not bind coverage. AI does not issue a certificate without a person reviewing it. AI does not send anything to a client, a carrier, or a claimant without a named person approving it. AI does not give a client a coverage answer directly. AI does not make a decision your license makes you responsible for.

The line is simple to state. The tool drafts, the licensed person decides.

5 Who checks the output, and what they check for

This is the section most policies skip, and it is the one that matters when something goes wrong.

Name the role, not the person. The account manager checks the certificate. The producer checks the client email. The person whose name is on the work checks the work.

Then say what they are checking, because people read for tone instead of accuracy. The checker confirms every number against the source document. Limits, dates, premiums, policy numbers, named insureds.

AI states a wrong number with the same confidence as a right one, and it does not flag its own mistakes. Reading for fluency will not catch it. Only comparing to the source will.

6 What gets documented

If the work touched a client file, note that AI was used to produce the draft and who reviewed it. One line in the file is enough.

This costs almost nothing at the time and it is the difference between being able to show your process later and not being able to.

7 Who owns the policy

Name a person, not a committee. That person keeps the approved tool list current, answers questions when something is not covered, and reviews the policy twice a year.

A policy with no owner stops being true within a few months and nobody notices.

What to do after you write it

A policy nobody has read is the same as no policy.

Walk your team through it once, in a meeting, out loud. Fifteen minutes. Take questions, because the questions show you what you left out. Have every person sign it and keep the signed copies. Add it to onboarding so new hires get it in week one.

Put a date on it and review it every six months. The tools change fast enough that a year old policy is describing an agency you no longer run.

The one page version

If the full document is more than you want to start with, write these five lines and expand later.

Five lines that count as a real policy
  1. These are the AI tools we use. Nothing else.
  2. Client identifying information never goes into any of them.
  3. AI writes drafts. People approve anything that leaves the agency.
  4. Every number in an AI draft gets checked against the source before it goes out.
  5. [Name] owns this policy and reviews it every six months.

That is a real policy. It is shorter than most and it covers the failures that happen.

Source

2026 Big "I" Agents Council for Technology Tech Trends Report, released February 19, 2026, based on a national email survey of independent agency members of varying sizes. Reported by IA Magazine and Agency Checklists.

See how AI handles the work, on a sample book

Run it yourself on a sample book. See how AI handles a renewal, a quote request, and a 7pm certificate.

See where AI fits Back to the library
Next step

Get the Handoff Map for your agency

Walk me through a normal Tuesday and I send back a one page map of every handoff in your agency, which ones can run on their own, and the one costing you most. Free, and yours to keep either way.

You will not end up on a sales sequence. I do not run one.
Guide 01 of 06· AI Insurance Library OuterMoon · Systems for independent commercial agencies
Terms
Nothing matches that in this category.
OuterMoon

The North Star Guide to AI in Insurance

One email a week on where AI fits inside a commercial agency, what to keep a person on, and what the vendors leave out of the demo. Written for owners, not engineers.

Free. Unsubscribe anytime. Nothing else lands in your inbox.
Where is work still leaking time in your agency?
Vega brought you here
Not built yet. Coming in the next pass.
Private · this session only