The authorised representative, turned toward underwriting

Broking solved its licence problem thirty years ago. Underwriting is solving it now — and nearly everyone doing so has chosen the same shape. A note on the three models, why the market landed where it did, and why the system underneath matters more than the ownership above.

The problem the licence creates

Every year a number of experienced Australian underwriters reach the same conclusion: they could run a better book than the one they are running for someone else. Most of them never act on it, and the reason is rarely capacity, appetite, or nerve. It is the licence.

Starting an underwriting agency in Australia means holding, or being covered by, an Australian financial services licence before the first policy is issued. Obtaining one is a months-long exercise with a six-figure first-year cost once responsible managers, professional indemnity cover, dispute-resolution membership, compliance arrangements, and the application itself are counted. Capacity providers, reasonably, want to see the licence before they commit. The underwriter, reasonably, cannot justify the licence before capacity is committed. The result is a queue of good businesses that never start, and a market that keeps consolidating into the hands of those who already hold paper.

What broking worked out

Broking met the same wall in the early 2000s and went through it. One licensee, many businesses: the licensee holds the AFSL and carries regulatory responsibility; each broking business is appointed as an authorised representative, owns its clients, and trades under its own name. Networks built on this structure now account for a large share of Australian SME broking. The model is well understood by ASIC, by professional indemnity insurers, and by the people who join it.

Underwriting is now receiving the equivalent, and it is arriving in a particular form. Every organisation that has built a licence-hosting structure for underwriting agencies — in Australia, in London, in Bermuda — has taken equity in the agencies it hosts. That is not a coincidence, and the reason is worth understanding before deciding whether it matters.

Three models

Whoever builds an authorised-representative network for underwriters chooses, whether they know it or not, between three shapes. The choice is about what the host’s business actually is.

Model A — the pure host

The licensee supervises and nothing else. Each representative owns 100% of its business, its clients and its renewals. The host takes a published fee for supervision, holds no equity, and never competes with the agencies it supervises. When a representative wants out, the host helps it find a buyer but never buys.

Its virtue is structural: no conflict between supervision and ownership, no capital requirement, no reason for the regulator to suspect the host’s judgement. Its problem is commercial. Supervision is a cost centre with a fee attached; the margin is thin, and the people best placed to run such a host — experienced underwriters — can earn far more doing almost anything else with the same licence. A representative also gets no funding at inception and no floor at exit, which narrows the pool of founders to those who can fund their own first year.

Model B — the host with a floor

As model A, but the host commits to buy a representative’s business at a published valuation formula if no one else will, after a minimum term, and holds what it buys in a separate entity for onward sale. The representative gets an exit floor; the host stays neutral in day-to-day operation.

This solves the founder’s exit problem and preserves most of model A’s integrity, at the cost of a capital requirement that grows with the network and a discipline — sell on, always — that has to survive the first profitable book the host would rather keep. Nobody has built it.

Model C — the equity network

The host funds the representative’s start, holds a majority or significant minority from day one, and provides a mandatory back office. The founder gets a salary bridge, a professional platform, and a defined exit — usually to the host. The host gets the thing that actually pays: a share of the agency’s value and, eventually, its sale.

This is where the market has landed. Rhodian in Australia, 360 Underwriting and the broker-owned agency groups, Pine Walk inside The Fidelis Partnership, even Accelerant’s incubator — every working example takes equity. The appendix sets them out. The reasons are consistent across all of them: good underwriters will not leave secure employment without funding; capacity providers are more comfortable with a group that has a balance sheet behind it; and a host that is not paid through equity has no economic reason to exist.

Everyone has chosen model C. The honest reading is not that models A and B are wrong, but that supervision alone does not pay, and the equity is how the host gets paid.

What model C has to get right

If equity is the norm, the interesting question stops being whether the host should own its representatives and becomes how it owns them. Five conditions separate an equity network that regulators, capacity providers and founders can trust from one that is a roll-up with a licence attached. Each exists because the obvious version fails without it.

1. 01
   The founder’s share is real.

   Whatever the split, the representative’s client relationships, renewal rights and goodwill are recorded as belonging to the agency, not the host, and the founder’s equity is in that agency. Valuation on exit follows a formula published in advance and applied identically to every representative. The founder should be able to reconstruct what their stake is worth from numbers they can see.

   Without this, the network attracts people who want to be bought rather than people who want to build, and the second population is the durable one.

2. 02
   Ownership and supervision are kept apart.

   The host owns equity; the host also supervises. Those are different jobs with opposite incentives, and they must not be done by the same people on the same reporting line. Compliance decisions — to narrow, suspend or terminate an authority — sit outside the commercial line, and the people making them are not paid on representative income or agency valuation. The same supervision applies to an agency the host owns 80% of as to one it owns 20% of.

   This is the condition regulators care about most. An equity holder supervising its own investment has every reason to look away from its largest one.

3. 03
   Supervision is complete, not sampled.

   The traditional AR framework supervises by sampling files after the fact. That was the only option when representatives transacted on their own systems. A network can do better by requiring that all appointed business is transacted on a single platform the compliance function observes in full. Underwriting authority — classes, limits, territories, wordings — is enforced by the system at the point of every transaction, so the authority schedule in the appointment document and the authority enforced at bind are the same object. Every regulated act is written to a ledger that cannot be edited afterwards. If the platform is unavailable, binding stops; it does not move to email.

   This is the condition that makes the other four affordable, and it is the subject of the next section.

4. 04
   Terms are published and uniform.

   The supervision fee, the services charge, the valuation formula and the exit terms are set out in a single schedule that applies identically to every representative. No bilateral deals, no volume tiers, no side letters. Where the host’s income includes a percentage of representative revenue, that percentage is flat.

   A fee that varies with volume gives the host a reason to favour its largest representative; a term negotiated privately gives the network a reason to favour whoever negotiated best. Both are conduct risks, and both are removed by publishing the number.

5. 05
   Growth is gated by supervision capacity.

   The host states, in advance, the maximum number of representatives it will supervise per full-time compliance resource, and does not appoint beyond it regardless of how attractive the next founder is. Networks rarely fail from a single bad appointment; they erode because recruitment outran supervision.

The case for one system underneath every representative

Condition three asks that all appointed business be transacted on a single platform the host’s compliance function observes in full. That is a large ask of a representative who may already have tools it likes, and it deserves a proper argument rather than a rule. The argument has two halves. The first is about supervision. The second is about what happens when a representative is bought, merged, or graduates — and in an equity network that is not an edge case, it is the business model.

Supervision: observation instead of inspection

The traditional AR framework was built for a world where each representative kept its own records on its own systems. Supervision therefore meant inspection: a compliance officer visits, pulls a sample of files, reads them, and forms a view about the whole from the part. It is slow, it is expensive per representative, and it catches problems months after they occurred. Every scaling constraint in a conventional network — the ratio of compliance staff to representatives, the length of the appointment process, the narrowness of what a new representative is trusted with — follows from that method.

A common system replaces inspection with observation. Four things change.

Authority becomes a property of the transaction, not the file. The classes, limits, territories and wordings a representative is permitted to bind are configured once and enforced at every quote and every bind. A transaction outside authority does not proceed to be discovered later; it does not proceed at all. The authority schedule in the appointment deed and the authority enforced at bind are the same object, so there is no drift between what was agreed and what is happening.

Monitoring coverage becomes complete. Because every regulated act occurs on the platform and is written to a ledger that cannot be edited afterwards, the compliance function reviews the population, not a sample. Exceptions — a limit approached, a wording substituted, a referral overridden, a premium outside the rating band — are surfaced as they occur. Risk-weighted file review still happens, but it is directed by what the data shows rather than by a random draw.

There is no unsupervised path. If the platform is unavailable, binding stops. That sounds like a cost, and it is, but the alternative — an email or a spreadsheet as the fallback — is precisely the channel in which most conduct failures in delegated authority have historically occurred.

The evidence writes itself. A licensee asked by the regulator, a capacity provider, or a professional indemnity insurer to demonstrate that its representatives are supervised can produce the record rather than reconstruct it. Every representative is supervised on identical terms with identical evidence, which is what a capacity provider means when it asks whether controls are consistent across the network.

The practical consequence is that the supervision-capacity ratio in condition five can be materially higher than a conventional network could defend, initial authorities can be narrower and widened faster on evidence, and a second responsible manager or a full-time compliance manager is a scaling decision rather than a rescue.

Transactions: a book that can move without breaking

In an equity network representatives will be bought by the host, merged with each other, or occasionally leave to take their own licence. In a conventional network each of those events is a data migration, and data migrations are where books lose value.

Diligence from the record, not the representation. A buyer of an underwriting book normally relies on bordereaux the seller prepared, a loss run the capacity provider produced on a different basis, and a management summary that reconciles to neither. When the whole book has been transacted on one system, the buyer can be given read access to the actual policy, premium, endorsement, referral and claims-notification record, on the same data model the buyer already uses if it is also on the network. Diligence shortens from months to weeks, the discount buyers apply for uncertainty narrows, and the seller is not asked to warrant numbers it cannot fully verify.

Transfer is a change of authority, not a change of system. When one representative acquires another on the same platform, the acquired policies do not move. The authority schedule is re-pointed, the ledger records the change, renewals continue on the same rating tables and wordings, and the insured never receives a letter about a new system. Where a representative graduates to its own licence, the same mechanism applies in reverse: the book is re-assigned to the new licensee’s tenancy, and the compliance history travels with it. The migration risk that ordinarily makes capacity providers reluctant to consent to a transfer is largely absent, because from their side nothing about the controls has changed.

Valuation becomes comparable. A published buy-out formula — if the network has one — depends on numbers that must mean the same thing for every representative. Net commission, retention, loss ratio and expense allocation computed by one system on one basis are comparable in a way that figures assembled from a dozen spreadsheets never are. The same property makes third-party sales easier to price and makes the network’s own representatives credible buyers of one another.

Mergers between representatives become ordinary. Two small agencies that would each struggle to reach the scale a capacity provider wants can combine without a systems project, because there is no systems project. The network’s non-exclusivity means such combinations are driven by the parties’ judgement rather than by the host’s design, and the common system means they are cheap to execute and reverse.

The objections, and where they hold

It is lock-in. Partly. A representative on a common system has a real cost of leaving the system, and any network that requires one should say so plainly. The mitigants are that the representative owns its data absolutely and can take a complete export at any time, that graduation to an own licence is a tenancy re-assignment rather than a migration, and that the platform’s fees are published. The test is whether leaving the network is easy — which it must be — not whether leaving the system is free, which it is not.

It concentrates operational risk. Yes. One system down is every representative down. That argues for the platform’s resilience and its outsourcing terms being treated as licence-critical — unconditional access to supervision data surviving any dispute, seven-year retention, fail-closed rather than fail-open — and disclosed to the regulator as such, not for abandoning the model.

It flattens what makes each representative distinctive. Less than it appears. The system enforces authority and records transactions; it does not decide appetite, pricing philosophy, or how an underwriter talks to a broker. Distinctiveness in underwriting has never lived in the policy administration system.

The platform provider has an interest in saying all of this. Correct, and it is stated at the end of this paper. The arguments above stand or fall on their own; the reader should weigh them knowing who is making them.

What this looks like from each seat

The trade-offs, stated

Model C is the norm because it works. It still carries costs that a network should name rather than hide.

The host competes, in effect, with anyone it does not own. A network that owns its agencies and also hosts independent ones has an interest in the former. Most avoid the problem by not hosting independents at all — which is why models A and B remain untested.

The host carries the regulatory responsibility. Under Australian law the licensee answers for its representatives’ conduct. A host that appoints badly, or supervises its own investment gently, is answerable for it.

Retail is harder than wholesale. The design above is straightforward for specialty commercial classes sold through brokers. Once retail-classified business enters, design and distribution obligations, disclosure and dispute-resolution standards rise sharply, and a network should not activate retail until its compliance function is staffed for it.

Equity aligns and distorts at once. The host’s stake gives it every reason to want its agencies to write more. That conflict cannot be removed; it can be managed, through conditions two and four, and through putting supervision actions and agency valuations in front of the same board on the same page.

Where Cuttleflow sits in this

Cuttleflow builds the operating system that condition three depends on: authority enforced at every transaction, a ledger that does not forget, supervision that reads the whole book rather than a sample, and a data model that lets a book be bought, merged or valued without a migration. That infrastructure is what makes the authorised-representative model work well for underwriting — under any of the three models, and whoever owns the agencies.

Cuttleflow does not operate an authorised-representative network and does not hold an Australian financial services licence. Whether it ever does is not the subject of this paper. The subject is that the market has chosen its model, that the model is only as good as the supervision beneath it, and that supervision of this kind is now a systems problem with a systems answer.

Appendix — What the market has already built

Several organisations have built structures that solve parts of the problem described in this paper. Each is described from public sources, without judgement, and then mapped onto the three model types. The broking networks that established the AR model in Australia — Insurance Advisernet, PSC Connect and others — are the template; the five below are the underwriting-side variations.

The Lloyd’s coverholder — the ancestor

Delegated underwriting is not new. For decades Lloyd’s syndicates have granted binding authorities to coverholders — independent agencies that write in the syndicate’s name, within an authority schedule, and report by bordereau. The coverholder owns its business; the syndicate owns the risk; the managing agent supervises. Every model below is a variation on this arrangement. What Lloyd’s never solved is the licence: a coverholder still needs its own local authorisation, and the supervision remains sample-based and after the fact.

Rhodian — the incubator

Rhodian launched in Australia in early 2023, founded by a former head of Steadfast’s underwriting agencies with a minority investment from Amwins. New agencies trade under Rhodian’s financial services licence and receive funding and a complete back office — compliance, finance, HR, marketing, technology and capacity introductions. In return Rhodian takes a majority equity stake in each agency, with the founder’s holding sized on experience and market standing, and the services are compulsory rather than optional. Its founder has been candid that it is a funding model as much as a licensing one: good people will not leave corporate seats without security over income. It has launched several agencies across property and liability, accident and health, surety and marine.

Rhodian is model C, implemented well. It solves the licence problem and the salary bridge. It does not solve ownership — the founder starts as a minority holder in their own agency, and the natural buyer of that minority is the majority owner.

360 Underwriting and the group model

360 Underwriting Solutions, founded in 2017 and majority-owned by AUB Group since late 2020, is the mature form of the same answer: eighteen specialist agencies, more than $550 million in premium, growing by both organic launch and acquisition. Agency leaders typically hold minority equity in their unit; the group holds the licence, the platform, the capacity relationships and the brand. Steadfast Underwriting Agencies and Envest’s agency portfolio are built the same way. The model works because a large balance sheet behind the group gives capacity providers a comfort no start-up can offer, and because the group’s shared platform gives it consistent controls across every agency it owns.

This is the strongest version of model C, and the clearest statement of what an independent network has to replace: it must give capacity providers the same comfort about controls without the balance sheet, which is what condition three and the common-system argument are for.

The Fidelis Partnership and Pine Walk — the split

In January 2023 Fidelis, a specialty insurer founded in 2015, divided itself in two. Fidelis Insurance Group kept the capital, the licensed carriers and the obligation to pay claims. The Fidelis Partnership took the underwriters and became a privately-owned managing general underwriter, headquartered in Bermuda, with delegated authority to source and bind business on the insurer’s behalf under a long-term binder. The same people hold significant stakes in both halves.

Inside the Partnership sits Pine Walk, an incubator founded in 2017 that had established eighteen specialist agencies by mid-2026 with premium projected above US$1.2 billion. Each is a separately incorporated cell with its own founding underwriter; the Partnership funds it, owns the majority, and provides capacity and administration. Pine Walk obtained its own FCA authorisation in 2021 so that each cell can operate as an appointed representative — the UK equivalent of the Australian authorised representative.

Two features matter more than the equity. First, the capital provider is not locked in: when Pine Walk launched a casualty cell in 2025, Fidelis Insurance declined to participate because casualty was outside its appetite, and described that as the binder working as intended. Second, the Partnership has since arranged its own capacity — two Lloyd’s syndicates, one with Hampden in 2024 and one with Blackstone in 2025 — so that cells are not dependent on the original insurer. Fidelis proves at scale that underwriting judgement and capital can live in different companies, tied by a contract that gives the capital first refusal without exclusivity. That is condition two, applied one level up.

Accelerant — the exchange

Accelerant, founded in 2018 and listed in New York in 2025, is the closest existing structure to models A and B. It describes itself as a risk exchange: more than 290 member agencies, the large majority independently owned, connected to a panel of capacity providers through a single data platform. Accelerant charges fees to the capacity providers, not the agencies, and its central asset is that every member’s business is ingested, standardised and monitored on the same system — which is what lets capital providers deploy across hundreds of small agencies they have never met. It has an incubator (Mission) for underwriters leaving carriers, and holds equity in a minority of members, but ownership is the exception rather than the design. It entered Australia in 2023.

Accelerant validates the systems half of this paper: capital will back independent agencies at scale when it can see them all on one platform. Where it departs from the model here is on exclusivity — a member’s capacity is the exchange’s capacity, and that, rather than equity, is the lock. It is also a capacity platform rather than a licence host; in Australia the licence problem remains the member’s own.

What the precedents say

Three things follow. First, every organisation that hosts a licence also owns its agencies; nobody has built a pure host, and the likeliest reason is that a supervision-only business does not pay well enough to attract anyone to run it. Second, the two most successful structures at scale, Fidelis and Accelerant, both rest on a single system through which the capital sees every agency. That is the evidence for the common-system argument in the body of this paper. Third, the equity question has been answered the same way by everyone who has answered it, which means models A and B remain untested rather than disproven.

Cuttleflow Systems · Perspective 16 · 33°53′S · 151°16′E · Sydney