Skip to content

AI use policy for an insurance agency: a one-page template

A one-page AI use policy an independent agency can adopt this week: approved tools, client information, review, and what never gets automated.

Published · 9 min read

Your team is already using AI. Someone has pasted a client email into a chatbot to draft a reply, or asked one to summarize a policy form. That is not a problem by itself. The problem is that nobody has said what is allowed, so each person is making up their own rule.

A short written policy fixes that. It does not need a committee or a consultant. It needs one page, plain words, and an owner. This guide gives you the page.

In short: an agency AI policy answers five questions. Which tools are approved? What information may go into them? What must a licensed person review before it leaves the agency? What may AI never do? Who do you tell when something goes wrong?

This is a starting template, not legal advice. Before you adopt it, share it with your E&O carrier and whoever advises you on privacy and state insurance regulation. Many states have adopted data security rules for licensees, and your carrier contracts may add their own.

Why an agency needs this in writing

Three reasons, in order of how often they come up.

Client information. A dec page, a driver's license number, a loss run and a date of birth all contain personal information that privacy and data security rules protect. Whether a tool keeps what you type, uses it to train a model, or shares it with others depends on the tool and the plan you are on. Staff cannot be expected to know that. The policy should name the tools where client information is allowed and say that it is not allowed anywhere else.

Accuracy. AI tools write confidently, including when they are wrong. A coverage summary that sounds right and misstates a sublimit is worse than no summary. The policy should make review a rule, not a habit.

Authority. Binding, advising and signing are licensed activities, and many states treat quoting as one too. Check your own state's rules. A tool can prepare the work. It cannot hold the license. The policy should say where the line is so nobody has to guess under deadline.

The five decisions to make first

Make these before you edit the template. Each one becomes a line in the policy.

1. Which tools are approved?

List them by name. For each, write down who pays for it, which plan, and whether the provider's terms say your inputs are used for training. If you cannot answer the training question from the provider's own documentation, treat the tool as not approved for client information.

Free personal accounts are the usual gap. A producer's personal chatbot account is not the agency's tool, and the agency has no say over how it handles what goes in.

2. What may go into them?

Sort information into three groups. This is the part of the policy staff will actually use.

GroupExamplesRule
Public or genericA carrier's published appetite guide, a blank ACORD form, a general coverage question, your own marketing copyAny approved tool
Client informationNames, addresses, dates of birth, driver's license numbers, VINs, dec pages, applications, loss runs, claims detailsOnly the tools you have approved for client information
NeverPasswords, MFA codes, full payment card or bank account numbers, anything a carrier or client contract forbids sharingNo AI tool, ever

3. What must a person review?

The simple rule: anything that leaves the agency, and anything that goes into a carrier system, gets read by the person responsible for it before it goes. That covers emails to clients, coverage summaries, proposals, applications, certificates and quotes.

Decide who that person is for each kind of work. For most agencies it is the licensed producer or account manager on the file.

4. What may AI never do?

Write the short list. A sensible default for most agencies:

  • Bind, issue, cancel or change coverage.
  • Tell a client what coverage they need or whether a claim is covered.
  • Sign anything, or accept terms on the agency's behalf.
  • Take or submit a payment.
  • Answer a question on an application that the client has not answered.

The last one matters more than it looks. A missing answer should stay blank and go back to the client. A plausible guess on an application is how a misrepresentation gets into the file.

5. Who owns it?

Name one person. They approve new tools, answer "can I use this for that?", and hear about mistakes. In a small agency this is the owner or the operations lead. If nobody owns the policy, it stops being followed the first busy week.

The one-page policy

Copy this, fill in the brackets, and cut anything that does not fit how you work.

[Agency name] AI use policy

Owner: [Name]. Effective: [Date]. Reviewed: every six months, or when we add a tool.

1. Why we have this. AI tools can take repetitive work off our desks. We use them to prepare work. Licensed people at this agency remain responsible for everything we send, file or tell a client.

2. Approved tools. These are the only AI tools approved for agency work:

  • [Tool], [plan]. Approved for client information: [yes / no].
  • [Tool], [plan]. Approved for client information: [yes / no].

Do not use a personal account for agency work. To request a new tool, ask [Owner].

3. What may go in.

  • Public or generic material: any approved tool.
  • Client information (names, addresses, dates of birth, license numbers, VINs, policy documents, loss runs, claims details): only the tools marked "yes" above.
  • Never, in any tool: passwords, MFA codes, full payment card or bank account numbers, or anything a carrier or client has told us not to share.

4. Review before it leaves. Before anything prepared with AI goes to a client, a carrier or the file, the person responsible for that account reads it against the source documents and corrects it. If you did not check it, do not send it.

5. What AI does not do here. AI tools do not bind, issue, cancel or change coverage. They do not advise a client on coverage or claims. They do not sign, accept terms or take payment. They do not answer application questions the client has not answered. Missing answers stay blank until the client provides them.

6. Carrier systems. We sign in to carrier systems only with the agency's own credentials, under the carrier's terms. Any tool that works inside a carrier system must be on the approved list.

7. If something goes wrong. If AI-prepared work went out with an error, or client information went into a tool that is not approved for it, tell [Owner] the same day. We fix it and adjust the process. Reporting a mistake promptly is always the right call.

8. Records. We keep the source documents and the final reviewed version in the client file, the same as work done by hand.

Signed: ____________________ Date: __________

How to roll it out in a week

A policy nobody has read is not a policy. The rollout is short.

  1. Monday. Make the five decisions above. List your approved tools and check each provider's data terms.
  2. Tuesday. Fill in the template. Send it to your E&O carrier or broker and ask one question: "Is there anything here that conflicts with our policy conditions?"
  3. Wednesday. Walk the team through it in fifteen minutes. Use two real examples from your own desk: one that is fine, one that is not.
  4. Thursday. Ask each person to name one task they will try with an approved tool, and who reviews it.
  5. Friday. Collect signatures. Put the policy where the team already looks, such as the shared drive or the procedures manual.

Then put a date on the calendar six months out to read it again.

Questions to ask any AI vendor before it goes on the list

These are the questions the policy depends on. A vendor who cannot answer them plainly is not ready for client information.

  • Is what we enter used to train your models, or anyone else's? Where is that written?
  • Where is our data stored, and for how long? Can we delete it?
  • Who at your company can see our clients' information, and under what circumstances?
  • If the tool signs in to carrier sites, where are our credentials kept, and does it work under our own logins?
  • What does the tool do when it is unsure or information is missing?
  • What happens to our data if we cancel?

For a longer list, see 25 questions to ask an AI vendor.

Common questions

Do we need a policy if we only use AI for emails?

Yes, and it can be very short. Emails to clients often contain policy details and personal information, and they leave the agency. The two rules that matter are which tool is approved for client information, and that a person reads every email before it is sent.

Should we ban AI tools until we have this sorted?

A ban is hard to enforce and tends to push use onto personal accounts, where you have the least control. Naming one approved tool and one clear rule about client information usually works better than a ban.

Does using AI affect our E&O coverage?

It depends on your policy and carrier, which is why the rollout includes asking them. In general, E&O carriers care about the same things with AI as without it: documented procedures, review by a licensed person, and a complete file. A written policy and a review step are evidence of both.

How does this apply to a tool like Relay?

The same way it applies to any other tool. Relay goes on the approved list like anything else, and the rules still hold: it works carrier sites under your team's own logins, leaves missing answers blank, and returns work for a person to review before anything is sent or bound. You can read the details on our security and control page and put the vendor questions above to us on a call.

What to do next

Fill in the template today, even roughly. A rough policy your team has read beats a perfect one in a draft folder. Then pick one task to try under it, and use the review routine for AI-prepared documents the first few times.

From Relay, by email

Guides your agency can use, with or without Relay.

Templates and checklists for independent insurance teams. Read these now, and leave your work email to hear when we publish a new one.

We’ll email you when we publish a new guide. Unsubscribe anytime. Privacy

See it quote your carriers.

Thirty minutes. Bring the carriers you quote most and one submission that took too long.

Built for the independent agency.

Piper, the Relay fox, standing with a red cape