Insights Sources included

Where AI actually helps a South African brokerage.

Not the advice. The forty emails a day around it. What an AI agent can take off a broker’s desk, what FAIS keeps on it, and how to start without betting the brokerage.

By Mathieu Sroussi, founder of Syrup 4 min read

Most of what’s written about AI in insurance is about insurers. Underwriting models, claims triage, fraud scores. That makes sense: they have the data and the budgets. When the FSCA and the Prudential Authority published their study of AI in the financial sector in November 2025, banks led adoption at 52%. Brokers don’t get a figure of their own.1

That’s odd, because a brokerage is where AI can pay back fastest. Not on the advice. On everything around it.

Start with the work that isn’t advice

The FAIS Act defines advice as a recommendation, guidance or proposal of a financial nature. It also spells out what advice is not: factual information on the procedure for a transaction, a description of a financial product, an answer to a routine administrative query, objective information about a product.2

Read that list again with your inbox open. “Where’s my certificate?” is a routine administrative query. “What’s my excess on the bakkie?” is a description of a product. “Has the insurer got my claim form?” is procedure.

That’s a large share of a brokerage’s email. It’s also exactly the work an AI agent does well, because the answer already exists somewhere in your files. The agent’s job is to find it, attach it and draft the reply.

Three jobs worth automating first

Certificates and status emails. The agent reads the request, finds the client and the policy, attaches the document and drafts the reply. A person sends it. In renewal season that’s dozens of emails a day, each one a two-minute job that costs ten minutes of focus.

Quote comparison. Three insurers, three formats, three wordings. The agent reads each quote and lays out premium, excess, limits, extensions and exclusions in one table, and flags where they differ. This is where it earns its keep. A cheaper quote that quietly drops cover the client has today is a claim waiting to land on your desk. A tired person at 9 pm can miss it. A table doesn’t.

Renewal packs. Last year’s terms, the claims history, this year’s quotes, your template. The agent assembles the pack and drafts the commentary. Your broker edits it and signs it off.

Notice what’s missing from that list. The recommendation. The call where you explain to a client why the more expensive quote is the right one. That stays with your licensed people, and it should.

Where the lines are

Three rules we build into every agent that works for a brokerage.

It never gives advice. Under FAIS, advice comes from an authorised FSP or its representatives. Where advice is automated, the fit and proper requirements expect the provider to monitor, review and test the algorithms behind it.1 A drafting agent stays out of that by design: it prepares, a person recommends.

It never binds, changes or cancels cover. It drafts the instruction, and a person sends it to the insurer.

It never handles bank details by email. A “please update our banking details” email is one of the oldest frauds there is, and an agent that reads email is exactly who a fraudster would like to talk to. Those messages go to a person, every time.

What POPIA asks of you

Client data is personal information, and some of it, like claims and medical history on life cover, is sensitive. Three practical points.

Keep the data where it lives. The agent should work inside your own email and document accounts, not copy everything into someone else’s system.

Ask your provider, in writing, whether they train AI models on your clients’ data. The answer should be no.

Keep a person in any decision that substantially affects a client. Section 71 of POPIA limits decisions based solely on automated processing.3 An agent that drafts, and a person who decides, keeps you well clear of it.

None of this is exotic. It’s what a careful brokerage already does with a new staff member.

How to start without betting the brokerage

Pick one job. Not “AI for the brokerage”. The one that eats the most hours and has the clearest right answer. For most brokerages that’s certificates and status emails, or quote comparison in renewal season.

Measure it before you start: how many a week, and how long each one takes. If the job eats less than about 50 hours a month, the maths probably doesn’t work yet, and you should hear that before you spend anything.

Then run it with a person approving every draft. After a few weeks you’ll know which emails the agent gets right every time, and you can decide, with evidence, what it may send on its own.

That’s the whole method. One workflow, measured, with a person in charge. Then the next one.

Sources

  1. Financial Sector Conduct Authority and Prudential Authority, Artificial Intelligence in the South African Financial Sector, November 2025. resbank.co.za ↑
  2. Financial Advisory and Intermediary Services Act 37 of 2002, section 1, definition of “advice”. lawlibrary.org.za ↑
  3. Protection of Personal Information Act 4 of 2013, section 71. popia.co.za ↑

Got a workflow in mind? Let’s look at it.

Thirty minutes, free, no deck.

Book a call