AI fits beside a small brokerage's agency management system (AMS) at the edges: drafting client emails and renewal letters, sorting and summarising incoming mail, reading documents and pre-filling submissions, and matching commission statements. The AMS stays the record of truth; AI prepares work for a person, who checks it before it is written back.
That is a different question from "which AMS has the best AI". Plenty of small firms will do better with an inexpensive assistant next to a system they already know than by switching platforms mid-renewal season. The useful first step is to list which layer each daily task lives in, because that answer decides whether you need an AMS add-on, a general assistant or nothing new at all.
The stack in four layers
Most small brokerages run some version of the following. AI is the optional fifth layer that touches the others.
| Layer | What it does | Examples | Where AI can help |
|---|---|---|---|
| Agency management system | Clients, policies, renewals, activities, accounting, commissions | Applied Epic, EZLynx, Vertafore AMS360, HawkSoft | Data entry, email triage, reconciliation, drafting inside the system |
| Quoting and rating | Comparative rating, carrier portals, submissions | Comparative raters inside an AMS, carrier portals | Reading documents to pre-fill; summarising quotes for the client |
| Signatures and documents | Sending forms for signature, storing signed copies | Docusign and similar; modules inside some AMS products | Summarising long agreements; checking a form is complete |
| Communication | Email, phone, text, client portal | Shared inbox, texting add-on | Sorting, summarising threads, drafting replies |
| AI assistants | General drafting and analysis | ChatGPT, Claude, Copilot | Everything that is not in the other layers and does not need live record access |
What changes the decision is where the data lives. If an AI feature lives inside your AMS, it can read and write client records directly and your permissions follow it. If it lives outside, a person carries information between the two, which is safer but slower. Neither is wrong; they suit different tasks.
What the main systems say about AI
Vendor claims move quickly, so read these as a snapshot from product pages in November 2026 and confirm them on a demo.
Vertafore AMS360. Vertafore's page lists five Velocity AI agents inside AMS360, all marked live: an Email Agent that interprets incoming email, summarises it and triggers workflows; a Reconciliation Agent that matches carrier statements to agency transactions; a Policy Intake Agent that turns commercial policy documents into AMS360 data; a Policy Check Agent that compares issued policies with AMS360 data and source documents; and a Coach Agent for how-to questions. Each carries the vendor's own time-saved and accuracy figures, which are claims to test on your own work, not results. Pricing isn't published, so you ask. AMS360 also lists an AI-assisted email marketing tool for client communications and open API support for CRMs and reporting tools.
EZLynx. Its site describes an embedded AI called EVA that aims to reduce manual data entry, automation for quoting, renewals and service requests, and an instant commercial quoting feature it says is now available. Its comparative rater lists connections to more than 330 carriers, which is the company's own claim. Prices aren't published; there are trial and "special pricing" routes.
HawkSoft. Its pages describe built-in accounting with commission-statement import from Excel or CSV, correspondence templates for batch email, texting and e-signature, an Agent Portal for mobile browsers, retention alerts and cross-sell reports, plus partner API connections for rating, marketing automation and phone systems. No AI features appeared on the pages I read, so for HawkSoft users an outside assistant is the likelier route.
Applied Epic. Applied's Epic page introduces Epic Conductor, which it calls a new AI platform inside Epic, and describes AI that fills applications, checks issued policies and reconciles carrier statements within your permissions, with your team approving. The page gives no release or availability date for these, so ask your account manager which are live on your version and get a date in writing for anything that is not.
One lesson from this snapshot: the same word, "agent", covers very different things. A vendor's "agent" might be a scripted workflow with an AI step, or an assistant that edits records. Ask which, and ask what a person has to approve.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA worked example: an eight-person brokerage, month by month
Illustrative figures. Say an eight-person brokerage, five account handlers, one admin, one producer and the principal, mixing small commercial accounts with personal lines. It runs one AMS, an e-signature tool and a shared inbox. Four tasks eat the most time, and the table puts a plausible before and after against each, assuming a person checks every output.
| Task | Volume a month | Minutes each now | Minutes with AI plus review | Hours saved |
|---|---|---|---|---|
| Triage and log inbound email | 900 | 2.0 | 1.2 | 12.0 |
| Draft renewal and information-request letters | 180 | 9 | 5 | 12.0 |
| Pre-fill submissions from client documents | 60 | 25 | 15 | 10.0 |
| Check commission statements | 14 statements | 75 | 45 | 7.0 |
| Total | 41.0 |
Forty-one hours a month is roughly a quarter of one person's working time. At an internal cost of $35 an hour, that is about $1,435 a month of capacity, less review time already counted in the "with AI" column. Against that, subscriptions: say two business-plan assistant seats around $50 a month, plus any AMS add-on at whatever your vendor quotes. The sums only work if people use the tools, which is why the next sections matter. For a year-one view with set-up and training time, see what AI costs a small insurance brokerage in year one, and for the time side where AI saves time in a small brokerage.
Where AI slots in, job by job
Drafting letters and emails
This is the lowest-risk, highest-volume use and needs no integration at all. A handler pastes the non-identifying details of a renewal, asks for a draft, and edits it. The prompt below shows the shape:
Draft a renewal reminder email to a small landscaping client whose
commercial policy renews in 45 days. Tone: plain and friendly. Include: the
renewal date, three things we need from them (updated vehicle list,
payroll estimate, any new equipment over $10,000), and a deadline 21
days before renewal. Do not mention premium amounts or cover limits.
Leave [first name] and [your name] as placeholders.The output is a usable draft; a person adds the client's name and checks the three requests are right for that account. The value is consistency: every handler's renewal email asks for the same things in the same order.
Triage and summaries
Email is where AMS vendors' agents are aimed. If your system offers an email agent, test it on a week of real mail. Check three things: does it log the email to the right client, does it classify requests correctly (certificate request, claim notice, billing question, change request), and does it flag anything urgent? A misfiled claim notice is the dangerous failure, so insist that the agent cannot close or delete anything. If your system lacks such a feature, a shared inbox plus an assistant that summarises long threads is a cheaper start. Before switching it on for everyone, compare its labels for the test week with what your most experienced handler would have done.
Reading documents and pre-filling submissions
Prior-year declarations, loss runs and application forms contain the data you key into carrier portals. AI document reading can pre-fill fields; the saving comes from reviewing rather than typing. The risk is a plausible wrong value, such as a revenue figure from the wrong year. See AI submission intake for how to build the check.
Commission statements
Matching carrier statements to expected commissions is repetitive, numeric and easy to verify, which makes it a good AI task. Some AMS vendors now bundle it; if yours doesn't, you can still do it with an assistant and a spreadsheet. The routine is laid out in reconciling insurer commission statements with AI.
The integration checklist before you buy or switch on anything
Run through these questions with the vendor, in writing where possible. Each has a reason, and a vague answer tells you something.
- Is it live or announced? Ask for a demo on a sandbox of your own data, not slides. "General availability expected" means it isn't available yet.
- What does it read? Whole client records, only the open email, only uploaded files? Narrower is safer.
- What does it write? Notes, activities, client fields, policy data? Anything that changes policy or accounting data should need a person's approval.
- Can I switch it off per role or per client? A producer may want drafting help that an accountant should not have near ledger data.
- Where does the data go? Ask where it is processed, whether it is used to train models, and how long it is kept. Get the answer in the contract, not on a call.
- Is there an audit trail? You need to see what the AI changed, who approved it and when, for errors-and-omissions purposes.
- What does it cost per user and per use? Some vendors price by seat, some by credits or volume. Ask for the cost at your volumes: 900 emails a month is not the same as 90.
- What happens when I leave? Can you export what the AI generated, and is it deleted from the vendor's side?
- Who supports it? If the answer is "the AMS vendor, but they route it to the AI partner", ask how fast.
Ask a sample of these on your first call. A vendor who answers six of the nine clearly is probably ready; one who answers two with "we will get back to you" is not. A fuller buying list is in the insurance broker AI app buying checklist.
Costs you will actually see
Three kinds of cost, only one of which appears on a pricing page.
| Cost | Typical shape | How to estimate |
|---|---|---|
| Subscriptions | General assistants: ChatGPT Business and Claude Team Standard seats both list at $25 per user a month, or $20 billed annually, with a two-seat minimum. AMS add-ons: usually quote only. | Seats times price; add the vendor's quote for add-ons |
| Set-up and training | Time to test, write rules, train staff | About 10 to 20 hours of the principal's and one handler's time in the first month |
| Review time | The checking that stays human | Count it inside the minutes per task; never assume zero |
| Switching costs | Only if you change AMS | Data migration, carrier re-connections, a month of lower productivity |
The cheapest sensible experiment costs under $100 a month plus staff time: two business-plan seats for the two handlers who write the most email, and a clear rule on what may be pasted. Run it for a month, measure minutes, and decide.
Three more jobs worth testing, each with a result you can check
Certificate and document requests. A contractor client emails asking for a certificate for a new site. An assistant reads the request, pulls out the certificate holder, the wording required and the deadline, and drafts the reply; the handler issues the certificate from the AMS. The check is simple: the holder name on the certificate matches the request letter exactly. Time saved is small per request, perhaps a minute, but certificates arrive in dozens, and the reduction in re-typed holder names is the real prize.
Claim notices. A client emails a photo of damage and a few lines of description. AI can produce a structured notice for the carrier: date, location, what happened, what was damaged, who else was involved. It must not decide whether the loss is covered or tell the client what will be paid, and it must not suggest the client has a claim at all. A person reads the notice, sends it, and tells the client what happens next. Here the AI is a note-taker, and the value is a complete first notice rather than three follow-up emails asking for the date.
Policy summaries for clients. At renewal, a client wants a plain-English note on what changed. An assistant can compare last year's and this year's schedules if you paste both with identifiers removed, and list the differences in limits, deductibles and exclusions. The danger is a missed exclusion, because a model summarising a long wording can skip a clause. Treat the summary as a checklist for the handler, who reads the wording itself. How brokers use AI to compare policy wordings at renewal has the full method.
What a four-week pilot looks like in numbers
To make the decision concrete, take the eight-person firm above and run a four-week pilot on just two tasks, renewal letters and email triage, with two handlers. Before the pilot, time both handlers for one week: say 45 renewal letters at nine minutes each (6.75 hours) and 210 emails at two minutes each (7 hours), a total of 13.75 hours for the pair. In week three, with templates and an assistant, the same volume takes 45 letters at five minutes (3.75 hours) and 210 emails at 1.3 minutes (4.55 hours), a total of 8.3 hours. The difference, 5.45 hours a week per pair, is about 22 hours across the pilot month, which is worth roughly $770 at $35 an hour against a cost of about $50 in subscriptions.
Now audit quality. Pick 20 outputs at random and mark each as correct, needs a minor edit, or wrong. Suppose 14 are correct, five need an edit and one is wrong. That one wrong output is the finding that matters: what was it, why did the reviewer not catch it, and what rule prevents it next time? If the wrong output went to a client, the pilot has already told you where the review step is weak. Fix that before adding people.
These figures are made up to show the method; your own will differ. What should not differ is the habit of measuring minutes and errors before and after, because that is the evidence you will need when a vendor's account manager says their add-on saves 80 per cent of the time. Compare it with what you measured on your own mail.
What to do with client data
Write a one-page rule and put it on the wall. Allowed: draft text with placeholders, summaries of non-identifying documents, generic templates. Not allowed in a consumer chatbot: policy numbers, personal identifiers, medical or claim details, bank details. Allowed only in tools your contract covers: anything the vendor processes inside the AMS. If you have customers in the EU, data-protection law applies to what you send to any AI vendor, so check the vendor's data-processing terms with whoever advises you on compliance.
Insurance work also carries professional duties. If an AI-drafted email gives wrong advice about cover, the broker remains responsible, so a licensed person reads every client-facing message about coverage before it goes out. Speak to your errors-and-omissions insurer or adviser about their view on AI-assisted work; some will ask how you supervise it.
Four ways stacks go wrong
Two assistants doing the same job. A firm turned on the AMS email agent and kept a separate assistant for summaries; staff used neither consistently. Pick one per task.
The "coming soon" purchase. A firm paid for a higher AMS tier because AI features were announced, and used none of them for six months. Buy what is live.
No owner. Tools without a named person drift: prompts rot, rules aren't updated, new hires don't know the rule. Name one person.
Auto-write to records. An agent that updated contact details from an email signature overwrote a correct mobile number. Keep write access to notes and drafts until a month of spot checks shows no errors.
A sensible order for a small firm
- Week 1: write the data rule; pick two handlers; give them business-plan seats; build three templates (renewal, information request, claim acknowledgement).
- Week 2: time four tasks without AI; start using the templates and time them again.
- Weeks 3 and 4: ask your AMS vendor for a sandbox demo of any live AI features; run the integration checklist.
- Month 2: decide on one add-on based on minutes saved; set a review date for three months out.
Changing how a small team works is half the job, so ask the two pilot handlers to write down what annoyed them in week one; that list becomes the training material for everyone else. If you would like to map your own stack with someone who has done it, that is what my AI implementation consultation covers; you can ask for it through the contact page.
Questions brokers ask before adding AI to the stack
Do I need to replace my agency management system to use AI?
No. Most small brokerages get more from AI that sits beside the system, handling drafting, email sorting and document reading, than from a migration. Replace the system only for reasons unrelated to AI, such as accounting limits or poor carrier downloads.
Can a general chatbot touch client files safely?
Only on a business plan whose terms you have read, with a written rule on what may be pasted. Business plans from ChatGPT and Claude do not train on business content by default. Strip policy numbers and personal identifiers from anything you paste until you are sure of the terms.
What should I ask my AMS vendor about AI?
Ask which features are live rather than announced, what they cost per user, what data the AI reads, whether outputs write back to client records, and whether you can turn features off by role. Ask for a date and a customer reference for anything described as coming soon.
How do I measure whether an AI add-on pays for itself?
Time two or three tasks for a fortnight before turning it on, then again after. Count minutes per submission, per renewal letter and per statement line. Compare the saved minutes, valued at the staff cost, with the monthly price and the review time still needed.
Further reads
- How to Automate Insurance Renewals With AI and Keep More Clients — A renewal routine that uses the drafting step above.
- How to Track Contract Renewals and Deadlines With AI — Track your own software and contract renewals.
- AI Email Triage for Shared Inboxes: Sales, Support, and Invoices — Set up triage for a shared inbox without an AMS agent.
- AI Software Contracts: Auto-Renewals, Price Rises, Notice Periods — Read vendor contracts before you commit.
- Best AI Customer Support Software for Small Teams in 2026 — Client-service AI that may overlap with your AMS.
- AI Document Extraction vs Traditional OCR: Which Should You Use? — Why AI document reading differs from the scanner you own.
- AI Change Management for Small Teams: A Practical Plan — Get the team using the tools after the pilot.
- Best AI Tools for Insurance Agents: Seven for Quotes, Renewals and Calls — Seven AI tools for independent insurance agencies, with costs, test scripts, a six-person agency sum and the E&O checks before you buy.
- HubSpot for Insurance Agents: Pipeline, Renewals and What Breeze Adds — Step by step: one deal per policy, a new-business pipeline, a renewals pipeline, a 90-day view, and a three-job test of Breeze Assistant.
- Loss Adjuster Software: Claim Files and Reports, and Where AI Fits — What a small adjusting firm needs from claims software, with verified prices and a clear line on where AI drafts and the adjuster signs.
- Zoho CRM for Insurance Agents: Policies, Renewals and Zia — A step-by-step Policies module and renewal rules in Zoho CRM, the edition limits that matter, and what Zia adds for an insurance agency.
- AI Tools for Life Insurance Agents: Follow-Up Within the Rules — A rule-first method for AI follow-up in life insurance: guard-railed prompts, replacement flags, recording consent and a pre-send gate.
- ChatGPT for Insurance Agents: Ten Safe Uses and Three to Avoid — Ten uses an agency owner can approve for ChatGPT, and three to ban outright, with the client-data and policy rules that go with them.
- Copilot for Insurance Agents: Outlook and Word Tasks for the Agency — Email triage, client letters from policy wording and meeting recaps: which Copilot tasks an insurance agency can hand over, and at what cost per seat.
- Does Agency E&O Cover AI Mistakes? What to Ask Your Carrier — A per-tool list of questions for your E&O carrier, the exclusion wording to look for, and the written AI policy that gives you evidence. Not legal advice.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Vertafore AMS360 product page, Applied Epic product page and HawkSoft agency management system pages, read 4 November 2026; EZLynx home page (live site behind a bot check, read from the web.archive.org copy of 10 October 2026); ChatGPT Business and Claude pricing pages, read 4 November 2026. Vendors' own performance claims are quoted as claims.