AI for independent insurance agencies: what to automate first

For an independent insurance agency, AI pays off first on the paperwork around a policy, not the advice. In a 5 to 30 person P&C agency, the first jobs to hand to AI are certificates of insurance, quote intake, renewal prep, routine service requests and document chasing, because that work is mostly reading, sorting, drafting and asking again. Coverage advice, binding and anything that can turn into an E&O claim stay with a licensed person, and you measure the result in service hours per policy, certificate turnaround and retention.

How are independent agents using AI today?

The Big "I" 2026 Agency Universe Study, with 1,376 respondents, found that 46% of independent agencies report using AI, up from 15% in 2024. Reported uses include marketing content (49%), coverage form analysis (43%) and contract reviews (35%). A Big "I" Agents Council for Technology report drawn from the same study found only 21% of agencies have a written AI policy.

Much of that use is individual, not agency-wide. In Liberty Mutual's 2026 Independent Agency Growth Study of nearly 1,200 principals and staff, 41% of agents used AI weekly, but only 14% said their agency had implemented an AI tool. Agents who use AI estimated they save 4 hours a week. That is self-reported, and it measures individuals using AI tools, not an agency's service desk.

Size matters too. Insurance Journal reports that Reagan Consulting's Best Practices Study found 11.5% of firms under $1.25 million in revenue invested in AI in 2025, against 84.2% of brokers over $100 million.

So the gap is not whether your staff can use AI. It is whether the agency's own work runs through it, with a record of what it did.

What should an agency automate first?

Good insurance agency automation starts with jobs that arrive the same way every time, follow a rule, and end in a document a person can check before it goes out. Five fit.

1. Certificates of insurance

A request comes in by email: a holder name, a project, and often a contract clause asking for additional insured status, a waiver of subrogation or primary and non-contributory wording. Someone opens the account, checks the policy and endorsements, fills in the certificate and sends it.

AI can read the request, match it to the account, compare the requested wording with the endorsements on file, and draft the certificate. Requests the policy supports go to a CSR to issue. Requests it does not support go to a licensed person.

That split matters. Under the NCOIL Certificates of Insurance Model Act, a certificate "is not a policy of insurance" and may not alter, amend or extend coverage, and one issued in violation of the act is null and void. A fast certificate that promises coverage the policy does not give is worse than a slow one. For the full request workflow, see the certificate of insurance request process.

2. Quote intake

New business arrives as a web form, an email with attachments or a forwarded referral. AI can pull the facts into one intake record (entity, operations, locations, payroll, vehicles, loss history), list what is missing, such as loss runs or the current declarations page, and ask the prospect for it. The producer starts from a complete file instead of an inbox thread.

3. Renewals

Pull every policy renewing in the next 90 days. For each one, AI can compare the renewal with the expiring policy (premium, limits, deductibles, forms), flag anything that moved past a threshold you set, and draft the client email. A producer decides which accounts to remarket and has the coverage conversation.

4. Service requests

Add a driver, remove a vehicle, change a mailing address, send an ID card, update a mortgagee. AI can sort the inbox by request type, collect what the carrier needs (VIN, license number, effective date) and prepare the change. A licensed person confirms any change that touches coverage before it goes to the carrier.

5. Document chasing

Signed applications, loss runs, inspection responses, no-loss letters, audit payroll. The rule is simple: what is missing, who owes it and how often to ask. That is a schedule, not a judgment call, and it is one of the five jobs your team still does by hand. A person should hear about it only when the asking has not worked.

Job	What AI does	What a licensed person still does	What to measure
Certificates	Reads the request, checks it against endorsements on file, drafts the certificate	Decides on any wording the policy does not support	Hours from request to sent
Quote intake	Builds one intake record, lists gaps, asks the prospect	Recommends coverage, picks markets, presents the quote	Days from first contact to complete file
Renewals	Compares renewal to expiring, flags changes, drafts outreach	Makes the remarket call, reviews coverage with the client	Share reviewed 60 days before renewal
Service requests	Sorts by type, collects carrier data, prepares the change	Confirms any coverage change before it is submitted	Requests per policy, time to close
Document chasing	Tracks what is owed and sends reminders on a schedule	Decides whether to go ahead without it	Open items older than 14 days

We left out a column for hours each job takes today. The only figures we found came from vendor blogs with no survey behind them.

What stays with a licensed person?

The NAIC Producer Licensing Model Act draws the line. A person may not "sell, solicit or negotiate" insurance without a license, and to negotiate means to confer with or advise a buyer about the "substantive benefits, terms or conditions" of a contract. An agency employee can work without a license when the work is executive, administrative, managerial or clerical, is "only indirectly related" to the sale, and earns no commission.

Read that as a map. Reading, sorting, drafting and chasing are clerical. Three things are not:

Coverage advice. Which limits, which endorsements, whether a client needs an umbrella or cyber cover.
Binding. Committing a carrier to a risk under the agency's authority.
Anything that creates E&O exposure. A Big "I" webinar run with Swiss Re Corporate Solutions lists the allegations agents most often face: failure to procure coverage, inadequate recommendations, processing errors and coverage lapses. AI that drafts renewals, certificates and service changes sits close to all four, so a person approves before anything goes out.

The same rule holds in what AI can and cannot do in a data room: AI is reliable when it is pointed at the right page, and a person still decides.

What do regulators expect?

The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, adopted December 4, 2023, is addressed to carriers, not agencies. It says decisions that affect consumers and are made or supported by AI must comply with insurance law, including unfair trade practice law, and it expects each insurer to keep a written AI program. As of November 3, 2026, 29 jurisdictions had adopted it. In those states it applies to the carriers you place business with.

The NAIC Insurance Data Security Model Law (#668) does reach agencies. It applies to every licensee, requires a written information security program, and requires due diligence on any third-party service provider with access to nonpublic information. An AI vendor reading your client files is one. In the model text, a licensee with fewer than ten employees, counting independent contractors, is exempt from Section 4, which holds both of those requirements. As of November 3, 2026, 28 jurisdictions had adopted the model. Your state's version may differ.

Only 22% of agents in the Liberty Mutual study said they trust AI with business data and client information. Ask any vendor where your clients' data goes before it goes anywhere. This post is not legal advice. Check your state's rules, or ask counsel.

How do you measure whether it worked?

Take a baseline for four weeks before you switch anything on. Then track the same numbers every month.

Service hours per policy. Service team hours in the month divided by policies in force. If AI is doing its job, this falls, or holds flat while the book grows.
Certificate turnaround. Time from request received to certificate sent. Track the median and the slowest 1 in 10, because the slow ones are the ones a contractor calls about.
Retention. Policies renewed divided by policies up for renewal, and the same for revenue. For reference, the Ohio Insurance Agents Association's summary of the 2025 Big "I" and Reagan Consulting Best Practices Study puts the average retained revenue rate of Best Practices agencies at 96.5%. Those are the channel's top performers, so treat it as a high bar, not a typical agency's number.
Exceptions caught. How often the reviewer changes or stops what AI drafted. If it never happens, nobody is reading. If it happens often, the setup is wrong.

Then add the numbers most agencies have never seen: service hours per account against the commission it pays, the share of renewals reviewed 60 days out, and how long a new-business file sits incomplete. Those show where the time goes and which accounts pay for it.

What dailybuilt does for an agency

dailybuilt is the thinking layer for your company, and for an agency our team does the build. We map the admin work in your agency, set up AI to do the reading, sorting, drafting and chasing in the jobs above, and a person on your team approves before anything goes out. Each month you get a report of the work handled and the numbers from the section above.

It runs on the dailybuilt platform: custom collections for records such as certificate requests and renewals (15 field types), automations with 31 event triggers plus schedules, waits and branches, an inbox for website forms and email, documents and e-sign, every change written to an audit log, full data export, and remote MCP access so Claude or ChatGPT can work in it under your own permissions. dailybuilt is not an agency management system. Your policies stay in yours.

Pricing is a monthly retainer, scoped on an intro call. See how it works for insurance agencies.