AI for insurance brokers and financial advisers: where to start

Where does the time go in a broker or adviser firm?

A broker's or adviser's expertise is judgement, yet much of the day goes elsewhere. A new client sends a fact find in one format, a policy schedule in another and an email chain with the rest. Someone reads each document, retypes details into the back-office system, chases what is missing and writes up the meeting afterwards. Renewals arrive in waves, and every one needs the same checks. None of this is advice, but all of it has to be done, and it is done by qualified people who could be speaking to clients.

This admin has a cost, and it is paid three times over. The first is time: hours each week go on reading and retyping. The second is quality, because manual copying creates small errors that surface later, when a renewal is wrong or a document is missing. The third is client experience, since replies slow down when the team is buried in paperwork. Reducing it does not change what advice is given. It gives advisers more time to give it.

General-purpose assistants can already help with some of this, and most staff have tried them. They can summarise a policy wording or tidy a draft email. But they do not know your forms, your naming rules or where your client records live, so staff end up copying text in and out. The result is a faster first draft and the same manual workflow around it, with the added question of where client information went.

Custom solutions start from the other end. They connect to the systems the firm already uses, follow its own checklists and write results back where people expect to find them. Because they are built for one firm, they can be limited to the right clients' files and the right staff. Each team can have its own set of tasks, such as onboarding for new business and renewals for account management, without re-explaining the firm each time.

The difference matters most for ownership. When a firm's processes live inside a generic tool, they are hard to inspect, export or move. When they are built into the firm's own systems, with clear documentation, the firm keeps control of how its work is done. That is the approach we take, and it is why we start by studying the work rather than choosing a product.

None of this argues against agents themselves, and our guide to adding agents explains how to start. What remains is a practical question: which work to automate first, and who keeps control. Our answer is to study the work, build around it and keep people in charge. The sections that follow take it step by step.

How can AI help with client onboarding?

Onboarding comes first. A new client usually means a fact find, identity and address documents, existing policies or pensions, and a stack of emails. An agent can check what has arrived against a checklist, extract the key details into your system and send the client a polite request for anything missing. The adviser opens a file that is already organised, instead of a folder of attachments. The checklist and the rules behind it live in one place that every part of the system reads.

We call it Memory, the firm's own record of how it works: its checklists, document types and wording preferences, held as plain notes that every agent reads through the same connector. The firm can open, edit and export it. Agents that handle onboarding, renewals and meeting notes then follow the same rules, instead of each being set up separately and drifting apart over time.

Policy comparison is the next gain. Advisers often compare wording, limits, exclusions and excesses across several quotes or existing policies. An agent can read each document, extract the same fields and lay them side by side, with the page reference for each so a person can check the source. The adviser then spends time on what matters: whether the cover fits the client's circumstances. The agent does not recommend, and it does not decide. It prepares the table that makes the decision faster and better informed.

Renewals follow the same pattern. The agent can list policies coming up, gather the latest information, flag changes since last year and draft the renewal email for the broker to edit and send. The routine is the same each time, which is what makes it reliable.

How does AI support compliance review and record keeping?

Here is what an ordinary client meeting becomes once an agent handles the follow-up. The client in the example is fictional, and the notes are illustrative only.

"On 2026-06-03, Atelier Vasseur
 chose Royal Mail for shipments
 within France. Chronopost was
 ruled out because the volume
 did not justify its price. The
 decision is reviewed every
 quarter."

        |
        |  chunking, extraction
        v

[Atelier Vasseur]
  |
  +--chose (2026-06-03)--> [Royal Mail]
  |                             |
  |                  [France shipments]
  |
  +--ruled out--> [Chronopost]
                        |
                     because
                        v
                  [low volume]

[Carrier decision]
  |
  +--reviewed--> [every quarter]

The agent takes the recording or notes from the meeting, writes a clear summary, lists the actions with owners and dates, and drafts the follow-up email to the client. It also updates the file note in your system in the firm's usual format. The adviser reads, corrects and approves it. What used to take a long evening of typing becomes a short check. Because every file note follows the same structure, anyone in the firm can pick up the client's file later and see what was agreed and why.

Compliance teams read files looking for gaps: a missing signature, an absent suitability note, a disclosure that was not recorded. An agent can run those checks across every file rather than a sample, and produce a list of exceptions for a person to look at. It does not decide whether something is compliant. It finds what needs attention, so the compliance officer's time goes on the cases that matter. Rules differ by jurisdiction and by regulator, so the checks should be defined by your own compliance function.

This is where consistency pays. Advisers, paraplanners and administrators work in different ways, and a checklist applied the same way every time makes file quality less dependent on who handled the case. An onboarding agent, a review agent and a renewals agent can all use the same checklist, because the firm's rules are written down once. When a rule changes, it changes in one place. We also keep a clear log of what the agents did and what a person approved, so the firm can show how a file was handled. It is worth saying plainly what AI does not do here. It does not give regulated advice, it does not replace a qualified adviser's judgement, and it does not take responsibility for a recommendation. It supports the people who do. We give no legal, regulatory or financial advice, and the rules that apply to your firm depend on where you operate.

Client knowledge should stay with the firm

Control of data comes next. Client files are among the most sensitive records a firm holds, so where they go matters as much as what the agent does. Decide which systems the agent can read, where processing happens, how long outputs are kept and who can see them. Keep the firm's own documents in systems the firm controls, and check the terms of any supplier before sharing client information. If the answer to any of those questions is unclear, the project is not ready to go live.

This is the part we would put first. A firm's value is its relationships and its judgement, and both depend on trust. If an agent makes one careless mistake with a client's information, no time saving makes up for it. So we design the privacy and the review steps before we design the automation, and we test with dummy data before any real client file is used.

Good results also depend on good inputs, and three habits cover most of it. Name things consistently, so that a client is not one person in one document and a different household in another. Give each document a clear purpose, with a title and date at the top. And agree what complete means for each file type, so the agent can tell what is missing. Firms that do this find that people benefit too, because a new colleague can read the same files and understand the work.

Who reviews what the AI produces, and who can see what?

Review is a design choice, not an afterthought. In a small firm, one adviser may check everything an agent drafts. In a larger one, the paraplanner reviews summaries and a compliance officer reviews exceptions. Either way, the rule should be written down: which outputs are sent automatically, which need approval and who can approve them.

Access follows the same logic. Each agent should see only what its job needs. An onboarding agent does not need the firm's financial results, and a renewals agent does not need the files of clients who have left. We tie each agent to the permissions of the person using it, so two people asking the same agent do not see the same things. For a small firm, simple rules are enough, such as separate folders for each team and read-only access wherever writing is not needed. Larger firms can add approval stages and audit reports as they grow.

What comes next takes more. Measuring accuracy on your own documents, handling exceptions properly and giving each team its own agent all need an extra layer of design. This is the work of the Embedded Week, where we study how your team actually works before anything is built.

How do you start without a big project?

The habits above form the core of a firm's starter kit automation-lab, which should hold three templates, for a client file, a renewal and a meeting note, the writing conventions that make them easy for an agent to read, and a check that runs on each file before it is used. Plain-text templates work with any tool, and they make clear that the records belong to the firm rather than to a vendor. Starting from a kit like this saves weeks of debate about format.

Treat any starter kit as just that. It should be written for an ordinary firm, with a client onboarding, a renewal and a set of meeting notes, and tested on real examples before anyone relies on it. Feedback from people applying it to other trades is welcome, as it is for our other published guides.

At Arclight, we start with an Embedded Week

Nothing described here is specific to one kind of firm. Any broker, adviser or planner with clients, policies, forms and recurring tasks can have an agent that handles the routine work, on the firm's own systems or its usual hosting provider, using the tools it already has. The firm stays the owner from end to end, and its people stay responsible for the advice.

What this requires is mapping the work before anything is built, deciding what the agent may do alone and what needs approval, and giving staff a simple way to correct it. That is what the Embedded Week is for. We sit with your team for a week, watch the tools, hand-offs, copy-pasting and waiting, map where the time goes, then build a custom solution for your firm and train your people to run it. It is a project with a clear start and finish, and the firm can understand and develop the result with or without us. As with our guide to AI privacy and security, the advice here is the advice we give a firm that asks us to build it.