Lloyd's Bordereaux Are Not a Data-Integration Detail

The word bordereau makes an important thing sound clerical. It arrives late in the process, usually as a spreadsheet or a feed. It holds what we expect to integrate. Policy numbers. Premiums. Dates. Classes. Locations. Codes. So we file it in reporting, after underwriting and before finance. That is the wrong mental model.

The bordereau is not the spreadsheet after the decision. It is the record that makes the decision delegable in the first place.

An MGA or Lloyd’s coverholder binds a risk before the capacity provider has seen that account. Which is the point of delegated underwriting. The managing agent granted a pen. It has not surrendered responsibility. The gap between binding and first bordereau is not a latency annoyance. It is the interval where a promise exists and oversight depends on a record that has not arrived. Treat the bordereau as a periodic export and we acquire a very modern way to lose the reason a risk was written. The rows reconcile. The file loads. And nobody can show whether the location was inside territory, whether a referral was approved, which endorsement attached, or whether the reported premium relates to the authority granted.

Not an integration defect.

A control failure with good column names.

The authority is already a data model

The binding authority agreement is the operative rulebook.

It specifies delegated classes and territories, the maximum limit, an aggregate that requires referral, policy-period rules, prohibited occupancies, rates and rating basis, approved wordings, permitted endorsements, and the triggers that send a file back to us. Not paragraphs to reduce to a programme name and product code. They are the conditions under which we allowed somebody else to create our contractual obligations. A good agreement permits a twelve-month term plus a stated tolerance.

It may prohibit a risk outright.

It may allow it with a particular endorsement.

It may allow it after written special acceptance.

It also bars sub-delegation to a producing broker without our written consent.

Each distinction changes what a valid row represents.

Two documents called a binder make this easy to muddle. A binder agreement is the standing contract granting the pen. A binder issued to a broker is interim evidence of coverage, good for thirty to ninety days.

One establishes authority.

The other shows coverage is in force.

A model calling both binder opens its audit trail with an ambiguity. Lloyd’s adds a control before the first risk binds. A firm needs coverholder approval before it can bind Lloyd’s paper. The approval is recorded centrally, on prescribed market wording.

The managing agent is separately authorised. Syndicates supply the capacity.

“Lloyd’s approved it” is no answer if we cannot name coverholder, authority, and capacity.

Canada does not simplify the perimeter.

Lloyd’s is licensed to write here, and an MGA holds licences province by province, its underwriters individually.

Ontario’s P&C side registers with RIBO; life and accident business goes to FSRA.

Quebec uses the AMF, and a designated responsible officer.

Four western provinces use councils alongside a superintendent.

We keep the licence and jurisdiction in the operating record.

Not in a compliance slide.

The row has to carry the judgment

A risk and premium bordereau is typically due monthly, within thirty days of month end.

The claims bordereau follows on that cycle or tighter.

That deadline is a control boundary.

Late or malformed reporting is a breach, because we cannot reserve, report, or reinsure what we can’t see.

At minimum the row carries policy number, insured name, dates, class, location, sums insured, premium, commission, tax, transaction type.

The same data loads policy administration and actuarial, and drives treaty cessions.

A missing location is not a quality blemish.

It means our reserving, reporting, and reinsurance sit on the wrong portfolio.

But the commercial fields are half the record.

A bind also needs its link to the authority schedule in force at the time.

The policy-number block. The approved wording version. The endorsement library. The underwriter’s licence.

Outside the grid, it needs the acceptance request, our written response, and the condition that made it acceptable.

Verbal special acceptances fail because they feel efficient.

An underwriter phones us, gets a yes, binds.

The policyholder has coverage.

Our file has nothing.

And whoever said yes may have moved on.

“The carrier was comfortable” is not a testable field.

Here is the counterintuitive part.

An MGA binding outside authority usually still binds us, where the broker acted in good faith.

We may ratify the bind.

Our remedy runs against the MGA: indemnity, clawback, a reduced class, termination.

The record cannot be a retrospective permission check.

It preserves the difference between a valid bind, an approved exception, and a breach that still binds us.

Lloyd’s subscription mechanics make record identity familiar.

The Market Reform Contract wants terms agreed before inception, and its Unique Market Reference travels with premium, endorsements, and claims for the life of the liability.

A bordereau is not a Market Reform Contract with an identifier attached.

The lesson is narrower than that.

The join key is not a convenience.

It is how a market proves every later event belongs to the same original authority.

Capacity is not operational underwriting

Delegation does not mean the capacity provider stopped underwriting.

It means we decided which judgments another firm may make, under what rules, on what evidence.

They are on risk the moment the MGA issues the binder.

Before the account reaches its own systems.

That separation matters at claim time.

Claims authority is a separate grant, often a separate agreement, with its own thresholds and reportable events.

A third-party administrator may set a reserve and pay from a carrier-funded loss fund within authority.

We normally keep the coverage decision.

Being closest to the file grants nobody authority to decline coverage.

The claims bordereau reports paid, outstanding, and closed.

Our actuaries set IBNR from that history.

One reason a premium file, a claims file, and an authority file can’t be unrelated feeds.

Three views of one delegated judgment, three financial consequences.

Premium follows a separate clock under account-current terms.

The broker collects, remits net of commission to the MGA, and the MGA remits net to us.

Provincial law requires client premiums held in trust.

The report reconciles to that chain.

A schedule matching the MGA system but not the trust account establishes nothing.

OSFI treats delegated underwriting as a third-party arrangement.

Proportionate diligence. A written agreement. Monitoring. Access rights.

That isn’t a request for a nicer portal.

It asks us to keep evidence that a third party used a continuing authority we can inspect.

Audit consumes lineage, not screens

The delegated-authority audit tells us whether our record is real.

It samples bound files against the schedule, checks rating against the agreed basis, verifies the wording version, tests referrals, reconciles the bordereau to the MGA system, inspects the trust account.

Annual for a new binder, biennial once things settle.

In a sample of fifty policies, nobody wants a dashboard confirming the integration completed.

The auditor wants to know the class was allowed.

Whether insured value sat within the limit.

Whether the Quebec location had written special acceptance.

Whether the sprinkler warranty attached.

Whether the underwriter was licensed there.

And whether the row matches the issued policy.

I do not think a prettier portal is the answer.

A portal can make data look near real time and still erase the authority schedule, overwrite an endorsement version, flatten an exception into free text, or read a cancellation as a revised premium.

The design task isn’t an attractive shared view.

It is a chain from authority to transaction to evidence.

Lloyd’s makes this particularly clear.

It runs common audit and reporting standards, so one audit serves several syndicates on a binder.

That works only because the record is meant to be inspected across a chain of security.

Not merely consumed by two systems.

Findings are graded and tracked to remediation.

Repeat findings end relationships.

First findings rarely do.

The fastest sanction is a suspended pen.

The slowest system is irrelevant once the authority goes.

The question before the spend

The board conversation is not about integration polish.

It is whether we can audit what the MGA bound before a loss or a regulator asks.

Which produces a more useful investment test.

For one transaction, can we recover the governing binder version?

The class, territory, and limit test. The licence. The referral evidence. The wording and endorsement version. Policy number, location, premium, commission, tax. Later claim movements. Trust reconciliation.

And can it tell a routine transaction from an out-of-authority one, without asking whoever worked the file?

AI helps here. It retrieves a referral email, compares wording editions, classifies an exception, spots a missing location.

It shouldn’t infer authorisation because the class looks similar.

Nor decide that a verbal approval probably existed.

A fluent reconstruction is not evidence of authority.

The durable asset in a delegated programme is not the portal, and it is not the monthly spreadsheet.

It is the evidence that judgment travelled with the business.