How can my insurance agency use AI for renewals and client email without risking an E&O claim?

By Ranjit Bhinge, founder of Blur Studio · Updated · 8 min read

Short answer: Use AI for the work around the advice, not the advice itself. It can draft renewal reminders, sort the shared inbox, chase missing quote documents and write call notes into your agency management system (the AMS, the software that holds your clients and policies). A licensed person should read anything that explains coverage, limits, exclusions or a price before it goes out. That one rule keeps most of the errors and omissions (E&O) risk out of the picture. E&O is the professional liability insurance that covers claims that you made a mistake or left something out.

Ranjit Bhinge

You run a small insurance agency, renewals and the inbox are eating your week, and you want AI to take some of it without creating a claim.

I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.

The line AI should not cross

Insurance is a regulated business, and the person who explains coverage is usually a licensed producer. A language model (the kind of AI behind ChatGPT and Claude) writes fluent text that is sometimes wrong. In most businesses a wrong sentence is a small embarrassment. In an agency, a wrong sentence about whether a loss is covered is the exact thing an E&O claim is built from.

So I split agency work into two piles.

  • Admin that carries no advice: reminders, scheduling, sorting, filing, chasing documents, summaries for your own use, drafts a person reviews. AI is a good fit.
  • Anything that says what a policy does or does not cover: limits, exclusions, whether a claim is covered, whether a client should buy a coverage. A licensed person writes or approves it. AI may draft a first pass for that person, but nothing goes to a client untouched.

The National Association of Insurance Commissioners (NAIC, the group of state insurance regulators) points the same way. Its model bulletin on AI, adopted on December 4, 2023 and addressed to insurance companies, says decisions made with AI must meet the same legal standards as any other decision, including unfair trade practice laws, "regardless of the tools and methods" used. The bulletin is a model that states choose to adopt, and it is written for carriers, not agencies. Even so, your carriers may pass expectations down to you, so ask your carrier contacts and your state department what applies to your agency.

A renewal workflow with 90, 60 and 30 day touches

Renewals are the best first use because the work is repetitive, dated and low on judgment. Here is a workflow that fits a small agency. Each step says who does what.

WhenWhat happensAI doesPerson does
90 days outRenewal list pulled from the AMSSorts by line of business, flags policies with open claims or recent changes, drafts a short "we are starting your renewal" emailReads the list, removes anyone who needs a call instead, approves the batch
60 days outInformation update requestDrafts a request for changes (new vehicles, new locations, new payroll), tracks replies, flags clients who have not answeredPhones the flagged clients, updates the AMS
30 days outOptions and decisionSummarizes the quotes received into a plain comparison table for the producerChecks every figure against the quote documents, writes or approves the coverage explanation, speaks to the client
7 days outLast chanceDrafts a short reminder and a list of unsigned itemsHandles anyone at risk of a lapse

The important detail is the third row. The AI makes the comparison easier to read. A producer checks the numbers and owns the explanation.

The shared inbox

Most agency inboxes hold a mix of certificate requests, policy change requests, billing questions, claim reports and quote documents. AI can read each message and label it, which is called triage, and draft a reply for the common ones. I wrote a broader guide on AI email assistants if you want the general picture.

Set these rules in plain language before you switch anything on.

  • Claim reports are never auto-answered. They go to a person fast, with the AI only marking them as urgent.
  • Certificate and policy change requests get a draft that a service person checks against the policy before sending.
  • Billing questions can get an AI draft that quotes the amount and due date from your system, checked by the person who sends it.
  • Anything mentioning "covered", "denied", "exclusion" or a dispute goes to a licensed person with no AI draft at all.

Quote chasing and document handling

New business stalls on missing items: a prior policy page, driver licence details, a loss run (the carrier's claims history report), a signed application. AI helps in two ways. It reads incoming documents and pulls out the fields you need, a job covered in my guide to AI document processing. And it drafts the polite "we still need these three things" follow-up on a schedule.

The extraction step must be checked. A misread policy number or a swapped date on a loss run can quietly flow into an application. Have the person who submits the application look at the source document, not only the AI's extracted fields.

Sorting new inquiries by how likely they are to buy is a related job. I cover it in AI lead qualification. For an agency, keep it to ranking the callback order. It should not decide who gets quoted.

What your AMS may and may not allow

Your agency management system is where the truth lives, and many agency tools were not built with AI in mind. Before you plan anything, find out three things from your AMS vendor in writing.

  1. Does it have an API (a way for other software to read and write its data), and is it included in your plan or an extra cost?
  2. Can it export renewal lists and client contact data on a schedule, for example to a spreadsheet?
  3. What does the vendor say about sending client data to outside AI tools?

If the answer to the first question is no, you can still start. A scheduled export to a spreadsheet gives an AI tool something to work from, and a person pastes the results back. That is slower, but it keeps you out of risky workarounds. If you are choosing or replacing the system itself, my guide to an insurance agency CRM covers what to look for.

Client data and vendor terms

Insurance files hold names, addresses, driver data, health details on some lines and financial information. Two things to settle before you paste any of it into an AI tool.

  • Use a business plan with training turned off. Anthropic, the company behind Claude, states that for its commercial products (such as its Team and Enterprise plans and its API) it will not use chats to train its models unless the customer joins its Development Partner Program or explicitly opts in. Its consumer plans (Free, Pro, Max) are covered by separate rules. Check the same statement for any tool you choose, and put the answer in your file.
  • Check your own duties. The Gramm-Leach-Bliley Act (GLBA) is the federal law on how financial institutions handle customer information. For people engaged in providing insurance, the law gives enforcement to the state insurance authority of the state where they are domiciled, so your state insurance department is the place to start. State insurance data security rules may add to this. I am not a lawyer. Ask your compliance contact or attorney which rules apply to an agency your size.

The NIST AI Risk Management Framework is a voluntary guide from the US standards agency (the National Institute of Standards and Technology), built around four functions called Govern, Map, Measure and Manage. You do not need to adopt it, but its four words make a decent checklist: who owns the AI use (govern), what it touches (map), how you test it (measure), how you fix problems (manage).

Test it on 20 policies before you trust it

Do not switch a new process on for the whole book. Run it beside your current process for one renewal cycle on a small group.

  1. Pick 20 policies renewing in the next 90 days, a mix of personal and commercial lines.
  2. Do the renewal work the way you do today. Keep that as the real process.
  3. Have the AI produce its version in parallel: the reminder draft, the update request, the quote comparison.
  4. Compare. For each item, mark it as identical, better, worse or wrong. Write down every "wrong".
  5. Count the errors and read each one. A wrong date is one kind of problem. A wrong statement about coverage is another, and it should stop the rollout until you know why.
  6. Only then let AI drafts go out for a small group of clients, with a person reviewing each one.

Worked example with made-up figures: suppose the AI drafts 20 renewal emails and 17 are ready to send, 2 need a small edit and 1 states a wrong effective date. You keep the review step, fix the date source, and run the next 20. Suppose two cycles later the review finds almost nothing. Even then you keep the review on any coverage language.

What to measure

  • Renewals processed on time, meaning before the 30 day mark.
  • Lapses caused by a missed step.
  • Minutes of staff time per renewal, before and after.
  • Draft errors found in review, by type.
  • Client complaints about communication.

If staff time drops but lapses rise, the process is worse, not better.

Who should not do this yet

  • An agency with no written renewal process. AI will automate the mess. Write the steps first.
  • An agency whose data in the AMS is out of date. Drafts built on wrong contact or policy data are wrong at scale.
  • An owner who cannot find time to review the first month of drafts.

For the wider picture of where small businesses get value from AI, see my guide to AI workflow automation for small business and to AI implementation, which covers how a rollout is staged.

How I'd approach this

I would start with renewals, not the inbox, because renewals have dates and a clear success measure: nobody lapses by mistake. I would write the no-advice rule on one page, give every staff member a copy, and make the review step part of the job description rather than a favour.

I would keep the first tools small: a business-plan AI assistant for drafting, plus a scheduled export from the AMS. Only when that works would I connect systems directly, because direct connections are where a quiet mistake can write bad data into your records. If your AMS offers no connection, plan for that from day one. A custom connection can be built, but it should be priced as a project, with testing time included.

And I would tell every client that a licensed person reviews coverage questions. Saying so costs nothing, and it makes the line real for your staff too.

Questions people ask

Can AI answer client coverage questions for my agency?

I would not let it answer them directly. A wrong answer about what a policy covers is the kind of mistake that leads to an E&O claim. Let AI draft a first pass for a licensed person, or answer only non-coverage questions such as office hours, how to request a certificate or where to send a claim. See my guide to AI email assistants for how to set those boundaries.

What should AI never do in an insurance agency?

It should not decide who gets quoted or declined, explain exclusions or limits to a client without a licensed review, bind coverage, or change records in your AMS without a person checking. Keep it to drafting, sorting, reminding and summarizing.

Do I need to tell my E&O carrier I use AI?

Policy wording and carrier expectations differ, so ask your E&O carrier or your broker directly and keep the answer in writing. Policies differ, and this is general information, not legal advice.

Does my agency management system work with AI tools?

It depends on the system. Ask the vendor whether it has an API, whether that API costs extra and whether scheduled exports are possible. If not, a spreadsheet export that a person feeds to an AI tool still works for renewal drafting. If you are weighing a system change, read my insurance agency CRM guide first.

Is it safe to paste client details into ChatGPT or Claude?

Use a business plan, switch off any training on your data and check the vendor's terms before you paste anything sensitive. Anthropic states that on its commercial products (such as Claude for Work and the API) it will not train on your chats unless you join its Development Partner Program or explicitly opt in. Consumer plans follow separate rules. Check the same promise for any other tool. Also check what your state insurance department and the GLBA require of an agency, and ask a lawyer if you are unsure.

How much staff time can AI save on renewals?

It depends on how much of your renewal work is repetitive. Measure your minutes per renewal before you start, run the 20 policy parallel test, and compare. Do not rely on a vendor's claimed savings. They were measured in another office with different habits.

More in this topic: Problem solving: AI strategy and adoption (24 guides) →

Sources

Ready for the next step?

A 30 minute call. If you do not need me, I will tell you.

Ranjit Bhinge You talk to me, Ranjit. I reply within a few hours.

Not sure what you need? Tell me what is going on. I will point you the right way, even if it is not me.

Blur Studio · 30 minute call

Book a 30 minute call

You talk to me, Ranjit. If you do not need me, I will tell you.

  • We talk about your business and what is going on.
  • You leave with a clear next step, even if it is not me.
  • I reply to every message within a few hours.
Loading the calendar…