Guide · AI for insurance brokers
AI can safely take the admin around insurance advice: claims intake and document reading, data entry into your broking system, routine client questions, renewal chasing and draft file notes. For one anonymised New Zealand broker, automating claims, data entry and routine queries meant claims processed 72% faster, data entry errors down 63% and manual hours cut from 180 a week to 60. Advice, suitability and anything a client relies on stays with a licensed adviser, because the financial advice rules and the Privacy Act 2020 hold the adviser and the brokerage responsible whichever tool touched the work.
Let AI do the admin around insurance advice, and keep the advice, and the judgement that it suits the client, with a licensed adviser.
No New Zealand rule bans AI, and neither the Code of Professional Conduct for Financial Advice Services nor the FMA's standard licence conditions mention it. They hold the adviser and the licensed provider responsible for the advice, the records and the client's information, whichever tool touched them. So every build asks: does this step give advice, who checks it, and where does the data go?
Prices here are as of October 2026 and exclude GST. This is general information, not legal or compliance advice.
An anonymised New Zealand insurance broker went from 180 hours a week of manual work to 60, with claims processed 72% faster. The details are in the published case study.
| Area | Before | What was built | Result |
|---|---|---|---|
| Claims processing | Manual verification and document review | Automated claims verification and data extraction | Claims processed 72% faster |
| Data entry | Claim details keyed by hand, frequent corrections | AI-powered data extraction and validation | Data entry errors down 63% |
| Client support | Routine queries swamping staff | Chatbots for routine queries | Support load down 60% |
| Fraud | Manual review, limited analytics | Models that flag suspicious claims for further investigation | Fraud-related losses down 50% |
Overall, 95% of claims were processed straight through, the team got back more than 120 hours a week, customer satisfaction improved 30%, and the saving is put at $240K a year as a conservative estimate. It took about eight weeks: discovery and planning in weeks 1 to 4, building in weeks 4 to 7, testing and training in weeks 7 to 8, then full operation.
The safe work is anything that moves information from documents and inboxes into your systems, or drafts something a person then checks.
| Task | What the automation does | Where a person stays in |
|---|---|---|
| Claims intake | Reads claim forms, photos, invoices and assessor reports, and checks nothing is missing | Exceptions go to a claims handler. The insurer decides cover. |
| Data entry into the broking system | Writes policy, renewal and claim details from the documents into the client record | Anything it isn't sure of is held for checking, not guessed. |
| Routine client queries | Answers claim status, renewal dates and document requests from your records | Anything that asks "should I" or "am I covered" goes to an adviser. |
| Renewal and follow-up admin | Sends reminders and chases missing or updated details | The adviser reviews the cover and makes any recommendation. |
| File notes | Drafts a note of a call or email thread | The adviser approves it, and the approved note is the record. |
Claims intake is usually where I'd start. Under the Insurance Council's Fair Insurance Code, member insurers decide on a general insurance claim from an individual or a business with 19 or fewer employees within 10 business days of having all the information they need, so a claim that arrives complete starts that clock sooner. Most claims come in by email, which the email automation guide covers.
Advice, suitability, and anything a client will rely on when deciding about cover stay with a licensed person.
On any client-facing assistant, I draw the line there: facts from your records, and anything that sounds like advice handed to an adviser. Software that recommends cover is a digital advice service, which the FMA's standard licence conditions give as an example of a material change to notify within 10 working days. Talk to your compliance adviser first.
A NZ financial adviser can use AI, and the existing duties decide how: suitable advice, protected client information, adequate records and secure systems. Since 15 March 2021, under the Financial Markets Conduct Act 2013 as amended by the Financial Services Legislation Amendment Act 2019, advice to retail clients is given under a financial advice provider (FAP) licence, with duties that include care, diligence and skill and complying with the Code (FMA summary). A revised version of the Code has applied since 1 November 2025.
| Rule | What it means for an AI build |
|---|---|
| Code standard 5: protect client information | Business-grade tools, access by role, no client details in personal accounts. |
| Code standards 6 to 9: competence | The adviser still meets the bar, and staff learn where the tool can be wrong before go-live. |
| FAP licence condition 1: records kept at least 7 years | Every extraction, draft and message logged, with who approved it. |
| Licence condition 4: outsourcing | If an AI system becomes necessary to your advice service, be satisfied the provider can perform to the standard your obligations need. |
| Licence condition 5: technology systems | Keep critical systems secure, and tell the FMA within 10 working days of discovering an event that materially affects their security. |
Insurers have their own regime: the conduct of financial institutions (CoFI) regime came into full effect on 31 March 2025, requiring licensed insurers to treat consumers fairly and run a fair conduct programme. FMA guidance calls fair treatment a shared responsibility of insurers and intermediaries, so automation that messages an insurer's customers should be able to show it treats them fairly.
As of October 2026 the FMA hasn't issued AI-specific rules. In August 2026 it opened an exploratory thematic review of AI in financial advice.
The Privacy Act doesn't name AI either, but it keeps the brokerage responsible for client information whichever tool or provider handles it.
Claims files hold bank details and sometimes health information. If advisers paste that into their own ChatGPT accounts, read this guide on personal AI accounts, then which AI tool a NZ business should use.
Start with the step that eats the most hours, usually claims intake or keying documents into the broking system. Both sit outside advice, so they're the lowest-risk place to prove it works.
| The job | Fixed price, as of October 2026 |
|---|---|
| Renewal reminders, chasers and data moving between the tools you already run | NZ$500 to $5,000 |
| Claims intake, and reading documents and email into your broking system | NZ$1,000 to $8,000 |
| An assistant that answers routine questions from your own records | NZ$3,000 to $12,000 |
| A client portal for claim status and documents | NZ$4,000 to $15,000 |
| Hosting, monitoring and support | From NZ$99 a month |
Discovery is always free, and nothing is billed until a fixed price is agreed. A typical build takes 4 to 6 weeks; the case study, covering claims, support and fraud, took about eight. Training, written documentation and a test mode come before go-live, so you see it work before it touches live client data. Full ranges are on the pricing page, and what a fixed-price AI project includes covers what's in and out.
Yes. Nothing in the financial advice regime bans it, and neither the Code nor the FMA's standard licence conditions mention AI. The existing duties still apply: suitable advice the client understands, protected client information, and proper records. As of October 2026 the FMA is running an exploratory review of AI in financial advice rather than issuing AI rules.
Yes, the administrative side: reading claim forms and reports, keying the details into the broking system and chasing what's missing. In the published NZ broker case study, claims were processed 72% faster and 95% went straight through. Exceptions go to a person, and the insurer still decides whether a claim is accepted.
Routine ones, yes: claim status, renewal dates and policy documents. In the case study, chatbots cut support load by 60%. Questions about what cover to take, or whether a claim is covered, go to an adviser. Software that recommends cover is a digital advice service, which the FMA treats as a material change to notify within 10 working days.
No. A provider that only stores or processes data on your behalf is generally treated as holding it for you, so the brokerage stays responsible. You still need reasonable security, to know where the data sits, and to report any breach likely to cause serious harm to the Privacy Commissioner as soon as practicable.
As of October 2026, automating claims intake, documents and email with Ez-AI is NZ$1,000 to $8,000 as a fixed price for most jobs, and renewal and follow-up automation is NZ$500 to $5,000. Hosting and support start at NZ$99 a month. Prices exclude GST and discovery is free.
Walk me through one claim or one renewal, from first email to closed file, on a free 30-minute call. I'll tell you which steps are worth automating, which should stay with an adviser, and the fixed price. Discovery is always free.
Book a free call