AI in General Insurance: What a Small Broker Can Use Today

For a small non-life broker, AI can already draft renewal reminders and client emails, summarise quotes into a comparison table, prepare file notes, and chase insurers and clients on open claims. It cannot decide suitability, give advice or complete disclosure for you. Use it for first drafts and tracking, and keep a licensed person on every decision.

The useful way to think about it is stages in the policy year. Renewal preparation, quote comparison, claims chasing and everyday client correspondence each have a repetitive part that suits a machine and a judgement part that does not. Take a renewal reminder: an assistant can draft one tailored to a print shop's risks for about eight minutes of handler time instead of twenty-five, but the handler still decides the renewal strategy and what goes to which insurer. None of this needs enterprise underwriting models; a business chat plan and your broking system are enough to start.

Follow me on Instagram@sagnikteaches

Where the time goes in a six-person brokerage

Picture a six-person broker handling 900 commercial and personal policies: three account handlers, one claims person, one administrator and a principal. Each handler has about 300 live policies, so roughly 25 renewals fall due every month. If a renewal takes 90 minutes from first reminder to confirmation, that is nearly 38 hours of renewal work per handler per month. Claims add a steady trickle: perhaps 15 open claims at a time, each needing a chase every week or two.

Connect on LinkedInSagnik Bhattacharya

These figures are illustrative, but the shape is typical. A large share of the time is not advice. It is collecting information, writing near-identical emails, building comparison documents and keeping track of who has replied. That is the part AI can take, and the table below sorts the work accordingly.

Subscribe on YouTube@sagnikteaches

Renewal reminders that ask the right questions

Most renewal emails say little more than "your policy is due, please confirm nothing has changed". A better reminder asks specific questions for that kind of risk, which brings back the information a broker needs to avoid a mid-term surprise. An assistant can generate those questions if you give it the policy type and what you already hold. The method is covered in more depth in automating insurance renewals with AI; here is a compact version.

Worked example: a commercial property renewal

Say a client runs a small print shop with a shop-unit lease, stock and two machines, insured on a commercial package. The renewal is 45 days away. The handler pastes the facts that are not personal data into the assistant: the cover sections held (buildings contents, stock, business interruption, employers' liability), the declared sums insured, and last year's notes, with the client's name removed.

You are drafting a renewal reminder for a small commercial insurance broker.
Policy type: commercial package for a print shop (contents, stock, business interruption, employers' liability, public liability).
Held on file: stock sum insured $60,000; machinery $85,000; gross profit declared $240,000; 3 employees.
Write a friendly email to the client, 150 words maximum, that:
1. says the renewal is due in 45 days;
2. asks the client to reply with changes since last year: stock level at its peak, new or replaced machinery, headcount, turnover, any work done off-site, any claims or incidents;
3. says we will not renew terms until we hear back.
Use [first name] and [your name] placeholders. Do not give advice or mention prices.

The output (illustrative) opens with a short greeting, states the 45-day window, and lists six items to confirm. The handler checks that the list matches what the insurer actually asks at renewal, adds the one question specific to the client (for instance whether they began printing for outside events), and sends it. Time: about eight minutes against twenty-five for writing one afresh.

Check before it goes

Verify that no sum insured or figure in the draft is invented, that the reminder asks questions rather than assuming answers, and that nothing reads like advice. Keep the principle that the client declares and the broker records: a reminder that says "we have you down as 3 employees, is that still right?" is fine; one that says "your cover is adequate" is not.

Quote comparison summaries

When three insurers quote a renewal, someone has to read three documents and turn them into one view. AI is good at extracting named fields from a document and arranging them in a table, which is the dull part. It is not good at knowing which exclusion matters for this client. For the policy wording side, see how brokers use AI to compare policy wordings at renewal, and for the quote-to-proposal path see the tutorial on AI insurance quote comparison for brokers.

Worked example: three quotes for a landscaping company

A landscaping company with eight staff receives three quotes for public liability and employers' liability. The handler uploads each quote schedule (with the client name removed) and asks for a table of premium, limit of indemnity, excess, territorial limits, and any exclusion mentioning machinery, subcontractors or work at height. An illustrative result:

The table is a head start, not a finding. The handler opens the three wordings and confirms each cell, paying special attention to "not stated", which often means "look harder" rather than "no restriction". Insurer B is cheapest, but the lower limit and the work-at-height exclusion matter for a tree-surgery-adjacent client, so the person writes the recommendation and the reasons. Record which wording you checked and on what date.

The "unlimited" that came from a header

A common slip is a confident "unlimited" or "included" that came from a header rather than the clause below it. You find it by sampling: open the source page for every cell that drives the recommendation. If the assistant cannot point to the page and clause, treat the cell as unverified.

Claims chasing without losing the thread

Claims go quiet. The insurer is waiting on an assessor, the assessor on the client, and the client has forgotten. A broker's value is in keeping everyone moving, and that is largely a tracking and follow-up job. The detail on routine enquiries is in how insurance brokers use AI to handle claims enquiries. Here is the chasing loop.

1. Keep a single status list: claim reference, client first name, insurer, what we are waiting for, who owes it, date last chased.
2. Each Monday, paste the rows where the date last chased is over seven days old into the assistant, with the email thread summary for each.
3. Ask for a short, polite chaser per row, addressed to whoever owes the next step, with the specific document or decision named.
4. Read each draft, correct the facts, send it, and update the list yourself.

Worked example: a water-damage claim

Take a yoga studio with a burst-pipe claim: the insurer wants a repair estimate and photographs, the client has sent the photographs but not the estimate, and nine days have passed. The assistant drafts a chaser to the client: it thanks them for the photographs, explains that the insurer cannot move on the claim until the repairer's written estimate arrives, and asks for it by a stated date. It also drafts a short update to the insurer's handler confirming the photographs are on file and the estimate is pending. The handler changes the date to one the client has agreed and sends both. The point is not elegant writing. It is that no claim waits for nine more days because nobody wrote the email.

What must stay with a person: any question about whether the loss is covered, anything the client is upset or frightened about, and any dispute with the insurer. Do not let an assistant tell a client that a claim will be paid. It does not know, and neither does the broker until the insurer says so.

Everyday client emails and file notes

The unglamorous win is correspondence. A handler gets 40 to 60 emails a day, many of them simple: certificate requests, change of address, "when does my policy end". Give the assistant your agency's standard wording once (in a project, saved instructions or a shared prompt file) and a draft takes seconds. A prompt that works well:

Draft a reply from [your name] at a small insurance broker to the email below.
Rules: plain English, under 120 words, no advice on cover, no promises about claims or premiums.
If the email asks for a change to cover, say we will review it and come back, and list what we need from them.
If it asks about a claim, say we will check with the insurer and reply by the end of the next working day.
Email: [pasted text with the policy number removed]

A sample output to a request such as "can you send my certificate to my landlord" would confirm that you will send it to the address supplied, ask for the landlord's contact and any wording they require, and say the certificate follows once received. The handler checks that the certificate actually exists on file before sending.

File notes follow the same logic. After a call, paste a transcript or your own bullet notes and ask for a dated note with decisions and actions. Then check that the note records what the client actually said, particularly any instruction to change or not to change cover, because an instruction misrecorded is worse than none. Date the note, name who wrote it and who checked it, and file it against the policy rather than in a personal inbox.

Mid-term changes and the renewal call that follows

Two more jobs fit the same pattern. The first is the mid-term change. A tutoring centre writes to say it has taken on a second site and hired four tutors. The assistant can turn that email into a checklist of what the broker needs before approaching the insurer: address and construction of the new site, contents value, tutor headcount and wage roll change, and whether any of the tutors drive on business. It then drafts the message asking the client for those items. It cannot say whether the change affects cover. That judgement, and the request to the insurer, remain with the handler.

The second is preparing for the renewal call itself. Before speaking to a client with a high premium increase, the handler pastes the facts: expiring premium $6,300, renewal terms $7,450, claims count one, change in turnover from $410,000 to $520,000. The assistant calculates the increase, which is $1,150 or about 18 per cent, and lists three honest talking points: part of the rise tracks the higher turnover, one claim in the period, and alternatives quoted. The handler checks the arithmetic, which in this case holds, and adds the market context that only someone who has been on the phone to insurers all month knows. Walking in with the numbers and a list takes ten minutes of preparation, not forty.

How a mistake shows up

Suppose a handler asks an assistant to summarise a long wording and it reports that terrorism cover is included. The clause actually says it is included only for sums insured below a stated threshold, and the client is above it. The summary read cleanly, so nobody questioned it until a second handler reviewed the file. The remedy is process, not suspicion of the tool: any summary that reaches a client or a recommendation must cite the clause it came from, and the person checks the clause rather than the summary. Build that into the template, and require the assistant to quote the sentence it relied on.

How to tell whether it is working

Track three numbers for a quarter: minutes per renewal, share of renewals where the client replied before the 30-day mark, and the number of corrections a reviewer made to AI drafts. If the minutes fall but the corrections rise, the saving is being eaten by rework and the prompts need tightening. If the reply rate improves, the better questions are doing their job. Share the numbers with the team in plain terms, because people adopt tools they have seen working on their own files.

Keeping advice, disclosure and data with a person

The boundary is simple to state and easy to cross in a hurry. AI may organise, draft and track. A person must advise, record what the client disclosed, explain what the policy does and does not cover, and be accountable for what is sent. Three working rules help:

• Nothing leaves without a named reader. Every email, comparison and note has a person who has read it against the source.
• The client's declaration is theirs. Never have an assistant fill in a proposal answer from guesswork or last year's file without the client confirming it.
• Minimum personal data. Use first names or reference numbers, drop dates of birth and bank details, and use a business plan whose data terms you have read. Free personal accounts may treat data differently from business plans.

Rules for firms in your position differ by where you operate and the products you sell, and I am not giving compliance advice here. Ask your compliance adviser, and your insurers' agency terms, whether there are limits on using AI tools with client data or in client communications.

A four-week start for a small brokerage

You do not need a project. Start with one handler and one job each week, and measure it.

1. Week 1, renewal reminders. Pick the 25 renewals due in the next 60 days. Draft the reminders with the assistant and time how long each takes. Record how many clients replied within a week compared with last year's rate if you have it.
2. Week 2, quote tables. Use the assistant on every multi-insurer renewal. Verify every cell that drives the recommendation and count the cells you had to correct.
3. Week 3, claims list. Build the status list and run the Monday chase. Count claims that moved after a chase against the previous month.
4. Week 4, emails and notes. Roll the standard-reply prompt to the whole team and agree a short list of email types that always need a handler's judgement before sending.

At the end, total the minutes saved, the corrections made and anything that went wrong. A realistic result for the six-person brokerage above is a saving of several hours per handler per month, not a transformation. The hours matter because they go to the conversations clients value: the mid-term call, the risk review, the awkward claim. If you want an idea of what the software side costs, what AI costs a small brokerage in year one sets it out.

Choosing where AI lives in your setup

You have three options, and most small brokers end up with a mix. The first is a general assistant (ChatGPT, Claude or Copilot) on a business plan, used with copy-and-paste prompts. It costs the least and teaches the team fastest, but needs care with data. The second is AI inside your broking system or CRM, where the draft sits next to the client record. That is convenient, though features and plan limits vary, so ask your vendor what is live today and which data it sees. The third is a specialist broker app for submission intake or comparison; the buying checklist for an insurance broker AI app lists the questions to ask.

A sensible order is to prove the jobs with a general assistant, then see what your broking system already offers, and only then consider a specialist tool for the one job that eats the most time. If you would rather map this with someone, that is what my AI implementation consultation covers. Either way, test with a small, low-risk set of renewals first, and keep the rule that a person reads every word before it leaves the building.

Broker questions about AI in the office

Can AI recommend which policy a client should buy?

It can summarise and tabulate what each quote covers, but the recommendation, the suitability reasoning and the disclosure to the client should come from a licensed person. Treat any AI comparison as a working note you verify against the policy wording before it informs advice.

Is it acceptable to paste client details into a chat assistant?

Only on a plan whose data terms you have read and your agency agreements allow. Business plans such as ChatGPT Business and Claude Team do not train on your content by default, but check the terms yourself. Where possible, remove names, dates of birth and policy numbers and paste only the facts the task needs.

Which job gives the quickest return for a small brokerage?

Renewal preparation usually does. It is repetitive, every client has one each year, the inputs are in your system already, and a person reviews each message before it goes. Quote summaries come second. Claims chasing helps most when you have several open claims per week.

Do clients need to be told that AI is involved?

That depends on your terms of business and the rules that apply to you, so ask your compliance adviser. Many brokers add a short line to their terms saying AI tools may help draft communications and that a person reviews them before they are sent.

Further reads

• Where AI Saves Time in a Small Insurance Brokerage — Where else a small brokerage saves time.
• AI for Insurance Brokerage: Reconcile Insurer Commission Statements — Reconcile insurer commission statements with AI.
• Contractor Insurance Renewal: Gather the Figures Before Your Broker Calls — The figures a client should gather before renewal.
• AI Submission Intake: Stop Re-Keying Data Into Insurer Portals — Stop re-keying proposals into insurer portals.
• What Mortgage Brokers Can Automate With AI, and What Stays Advice — A neighbouring broking sector and its advice boundary.
• AI Governance Checklist for Small Financial Advice Firms — A governance checklist for small advice firms.
• ChatGPT for Insurance Brokers: Client Emails, Summaries and Limits — ChatGPT for broker emails and summaries, and its limits.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business and Claude pricing pages for training defaults (checked 4 November 2026); general broking practice. No regulator guidance is quoted, so confirm your own obligations with a compliance adviser.