What Does AI Cost a Small Insurance Brokerage in Year One?

Most small brokerages spend about $3,000 to $18,000 on AI in year one, depending on size: roughly $3,000 for a lean four-person office, $9,000 for eight people, and $18,000 for twelve with AI call handling. Subscriptions run $120 to $950 a month. A third to a half of the total is staff time, not software.

That staff-time share is why brokerage AI budgets go wrong. Software prices are published and easy to add up, so they get budgeted. The hours spent setting up renewal templates, getting compliance sign-off on AI-drafted client letters and cleaning client records in the broker management system don't appear on any invoice, and they arrive in the first three months, before anything has saved a minute.

Follow me on Instagram@sagnikteaches

The line items, one-off and monthly

USD list prices checked in September 2026, annual billing unless stated. Hourly costs are illustrative; use your own loaded rates.

Connect on LinkedInSagnik Bhattacharya

The staff-time lines need a loaded hourly rate, and it's worth working out rather than guessing. An illustrative account handler on $60,000 a year, plus about 25% for employer costs such as pension and insurance, costs $75,000. Divided by roughly 1,680 paid working hours after holidays, that is about $45 an hour, the rate used for the larger scenarios below. Use the figure for whoever will actually do the setup work; if it falls to a director, the hours cost more.

Subscribe on YouTube@codingliquids

On the assistant choice: if the brokerage already runs on Microsoft 365, Copilot can work across Outlook, Teams and your files, which suits renewal and claims correspondence. Microsoft 365 Copilot pricing, Business versus Enterprise explains which licence applies.

Three brokerages, costed for a year

Year two assumes the same tools plus about 15 hours of upkeep, and it shows why year one looks expensive: setup time is a one-off. None of these include a broker management system add-on or outside help; add those from your quotes. If you are weighing a paid pilot with outside help, what an AI pilot project costs gives a sense of the range to expect.

The eight-person year, quarter by quarter

Cash doesn't go out evenly. For the middle brokerage, illustratively:

Almost 70% of the staff time lands in the first quarter. Plan that quarter around the renewal calendar: a brokerage with heavy renewals in one month should not schedule the clean-up for that month.

Where the setup hours actually go

The biggest block is usually renewal correspondence, because it is high-volume and compliance-sensitive. A before-and-after for the renewal chase, illustrative:

• Before: an account handler exports next month's renewals from the broker management system, writes each reminder by hand from an old template, and chases non-responders when they remember. About 12 minutes per policy, with gaps.
• After: a weekly automation pulls renewals due in 45 days, the assistant drafts a reminder from the approved template and the policy details, the handler reviews and sends, and a second automated nudge goes at 21 days. About 4 minutes per policy.

Getting there takes the hours. The prompt, and what it first produced:

Draft a renewal reminder for the client below, using our approved
renewal template (in this Project). Include: policy type, insurer,
renewal date, and the list of information we need from them. Keep
to 150 words. Do not mention premiums, cover changes or insurer
terms unless they appear in the details below.

Client details: [policy type], [insurer], renews [date],
info needed: [list]

Illustrative first draft: "Your commercial combined policy with [insurer] renews on 1 November. To make sure your cover continues without interruption, please confirm your turnover for the last year and any changes to your premises. We expect your premium to remain competitive..."

The last sentence was the problem. The assistant added a reassurance about the premium that no one had given it, which is exactly the kind of line a compliance reviewer rejects. The team tightened the instruction ("never predict or describe price"), added a banned-phrases list to the Project, and had compliance sign off the template, not every letter. That cycle of draft, reject and tighten is where most of the 20 to 60 setup hours go. Automating insurance renewals with AI covers the full renewal workflow.

The record clean-up is the second-biggest block, and an assistant can do the first pass on an export with personal details kept to the minimum: client reference, name, email domain, policy type, renewal date and status. A prompt that works:

Below is an export of our client records. Flag, in a table:
(1) likely duplicates of the same client, with the reason;
(2) active policies with no renewal date or a date in the past;
(3) contact emails that are generic (info@, office@, admin@).
Don't merge or change anything. If two records look similar but
might be different people, mark them CHECK rather than DUPLICATE.

[pasted export]

Illustrative output: "DUPLICATE: refs 10442 and 11873, same surname, same email domain, same address line, both home policies. PAST RENEWAL: ref 10915, status Active, renewal date 14 March last year. GENERIC EMAIL: 37 records use info@ or office@ addresses. CHECK: refs 10230 and 10231, same name and address, one motor and one home policy."

The CHECK pair shows why the instruction matters. They turned out to be a father and son with the same name living at the same address, two separate clients with separate policies. An earlier run without the CHECK rule had listed them as a duplicate, and a merge would have sent one of them the other's renewal. The 37 generic addresses were the real finding: renewal reminders to an office inbox are the ones nobody reads, so each became a phone call to get a named contact.

What the eight-person year buys back

A cost only makes sense next to what it returns. Illustratively, for the eight-person brokerage once the pilots are live:

That is roughly 40 hours a month from quarter two and 47 from quarter three, or around 400 hours across the year at the illustrative $45, which is well over the $9,200 cost on paper. On paper is the catch. Saved minutes scattered across a day don't pay anything unless they turn into more renewals handled per person, fewer lapsed policies, or new business that would otherwise have been turned away. Decide at the start which of those you're aiming for, and measure it: renewals handled per account handler and retention rate are the two numbers that show whether the year paid.

The claims updates in that table are a good example of where the minutes come from. Before, a typical update read "Hi, no news from the insurer yet, will chase", which prompted a call back from the client most times. The approved template the assistant now fills from the claim notes:

"Update on your claim [reference]: the insurer's loss adjuster visited on [date] and their report is due by [date]. Nothing is needed from you at the moment. We'll update you again by [date], or sooner if we hear anything."

It takes about the same time to send, but fewer clients ring to ask what "will chase" means, which is where the ten minutes each come from. The same rule as the renewal letters applies: the assistant never states or hints at a likely settlement.

A quarter-four check for the eight-person brokerage might read, illustratively: renewals handled per account handler up from 38 to 45 a month, retention flat, two lapsed policies traced to the generic-inbox problem rather than to AI. That is a year that paid, because the extra renewals per person came from the saved time. If the handler figure had stayed at 38, the right call would have been to cut the call-answering plan and keep only the seats that feed renewals.

Costs brokerages tend to miss

• Consent and disclosure scripts. Call notes and AI call answering need a line telling clients the call is recorded or that they're speaking to an automated assistant. If you sell to customers in the EU, telling people they are talking to a chatbot has been a legal duty since 2 August 2026.
• Record keeping. AI call summaries and drafted letters become part of the client file. Decide where they're stored and for how long, and budget the admin time.
• Stale data in the management system. Automations send exactly what the system holds. In an illustrative case, a brokerage's first automated batch went to 23 old email addresses. Cleaning records before the first send is cheaper than apologising after.
• Portal changes. Automations that fill in insurer portals break when a portal's layout changes. Budget a few hours a quarter for fixes if you go down that route; AI submission intake discusses the options.
• Your own professional indemnity cover. Ask your insurer whether AI-assisted advice or correspondence changes anything. The email costs nothing; finding out after a claim can cost a lot.
• Regulatory scope of tools you buy. Most brokers don't price risk themselves, but if a tool you adopt scores or prices individuals for life or health insurance, the EU AI Act treats that as high-risk, with obligations due from 2 December 2027. Ask the vendor how it is classified if you serve EU customers.

A realistic mistake with AI call answering

AI call answering is the most expensive line in the larger scenarios, and the one with the sharpest failure. In an illustrative first week, a caller asked the after-hours assistant, "Am I covered if my van was broken into last night?" The assistant, trying to be helpful, replied that commercial vehicle policies "usually include theft cover" and that the client "should be fine". Nobody had checked the client's policy, which excluded theft from an unattended vehicle overnight.

The fix was in the script, not the software: the assistant may take details, log a claim notification and promise a call back by a set time, and must never say whether something is covered. That rule should be written before go-live and tested with ten awkward calls. Six of the ten an illustrative brokerage used, with the answer it expected:

• "Am I covered for storm damage to my fence?" Takes details, says an adviser will call back by 10am, gives no view on cover.
• "Cancel my policy from today." Logs the request, explains a person must confirm it, doesn't confirm the cancellation.
• "Add my son as a named driver from tomorrow morning." Takes the details, flags it urgent, says cover isn't in place until an adviser confirms.
• "Someone called saying they're from my insurer and asked for my bank details. Was that you?" Tells the caller not to give details, flags it to the duty adviser straight away.
• "What's the cheapest price you can do?" Takes the enquiry, gives no figure.
• "I've just had an accident and the other driver is shouting at me." Gives the emergency claims line from the policy documents and logs the call as urgent.

Any call where the assistant gave a view on cover or price is a fail, and the script is fixed and the whole set rerun before go-live. Whether AI can answer calls and qualify leads for an insurance agency goes further into scripting.

Cutting the year-one bill

1. Start with one workflow. Renewal chasers alone justify the assistant seats for most brokerages. Add claims updates in quarter three.
2. Give seats by role. Account handlers need the assistant; not every administrator does. Otter only for people who hold client calls.
3. Use what's included first. Microsoft 365 business plans include Copilot Chat for general drafting with anonymised details, which can carry a four-person office through the pilot.
4. Run AI call answering after hours only for the first three months. It keeps call volumes, and the bill, inside the entry plan.
5. Move to annual billing after month three, once you know which seats are used.

Here is the year-one budget line for the eight-person brokerage as it might sit in the practice budget, filled in:

AI BUDGET, YEAR ONE (8 staff)
Software (annual billing)        $5,604   ($467/month)
  Copilot Business x8              2,016
  Otter Business x4                  960
  Zapier Team                        828
  AI call answering (entry)        1,800
Staff time (80 h at $45)         $3,600
  Setup and templates     40 h
  Compliance review       12 h
  Record clean-up         16 h
  Training                12 h
Contingency (10%)                  $920
TOTAL                           $10,124
Review points: end of Q1 (cancel idle seats), end of Q3 (renew?)

The contingency line is worth keeping. Something in the first quarter always takes longer than planned, usually the record clean-up. For the time side of the equation, where AI saves time in a small brokerage shows which jobs repay the hours fastest.

Brokers' budgeting questions

Does our broker management system's AI cost extra?

It varies by vendor and edition, and prices are usually quote-based rather than published. Ask your account manager three things: which AI features your current edition includes, what the add-on costs per user, and whether it uses your client data to train anything. Budget for it as a separate line until you have a written answer.

Can we recover the cost from clients?

Not directly in most cases, and it is rarely worth trying. The return usually comes from handling more renewals per person, faster claims updates and fewer lapsed policies. Track retention rate and renewals handled per account handler before and after, which shows the value more clearly than trying to itemise AI on fees.

Is it cheaper to hire an assistant instead?

For a four-person brokerage, a part-time assistant can be the better first step if the bottleneck is phone cover rather than paperwork. AI is cheaper for drafting, summaries and chasing, but it does not replace judgement on cover. Many brokerages end up with both: a person for calls and AI for the written follow-up.

Further reads

• How Brokers Use AI to Compare Policy Wordings at Renewal — A high-value renewal job to put the assistant seats to work on.
• How Insurance Brokers Use AI to Handle Claims Enquiries — Where AI fits in claims updates without overstepping.
• How Much Does AI Cost a Small Business in 2026? — The general cost picture if you want a wider benchmark.
• How Much Time and Money Does AI Staff Training Take? — More detail on the training hours in the budget.
• Does Your Business Insurance Cover AI Mistakes? — What your own professional cover says about AI errors.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Microsoft 365 Copilot Business pricing page; OpenAI ChatGPT Business pricing; Otter.ai, Zapier and Smith.ai pricing pages; EU AI Act Article 50 and Annex III. Checked September 2026. Scenario figures and hourly costs are illustrative.