// Delegated authority and ceded reporting
One book of record, twenty bordereau layouts.
A coverholder with twenty binding authority relationships produces twenty materially different bordereau layouts for the same period. Premium, claims, technical account and risk bordereaux run 30 to 60-plus fields each, against Lloyd's deadlines of 15 to 20 business days or whatever a US program contract happens to say.
Dearborn Labs is an AI-native software development firm built specifically for insurance. There is no Dearborn Labs product in this — you hire the engineers, and the mapping layer is yours to edit when a carrier changes its template.
// The situation
This year's reporting is next year's capacity.
AM Best reports capacity providers now demanding long-term underwriting quality over growth, with enhanced due diligence before renewal, while reinsurers impose shorter durations and expanded data requirements on delegated business.
AM Best · 2026.
That turns a reporting obligation into a renewal exposure, so the first thing worth counting is the cycle itself: days from period close to last bordereau delivered, per relationship; how many were rejected or re-requested last quarter and why; and the standing variance between what you reported and what the carrier booked. Those three numbers are usually not written down anywhere, which is itself the finding.
There is a further structural problem. Lloyd’s Coverholder Reporting Standards v5.2 gives you a specification if you are coverholder-backed. There is no US-domestic equivalent — every fronting and program carrier sets its own, contract by contract — so the format count is not a temporary condition you can standardize your way out of.
// The stack you actually run
The systems, the schemas, and the spreadsheet.
The spreadsheet is not laziness. It is frequently the correct answer.
A program administrator writing a manuscript program has a rating model no policy-administration vendor supports out of the box, a carrier that changes the rate structure at each treaty renewal, and a book too small to justify a six-figure configuration project. The named failure mode in this segment is a startup MGA overbuying platform capacity before it has generated any operating data. Where the spreadsheet is still right, the work is to automate around it — keep the rating model where the underwriter can edit it, and put the intake, the reconciliation and the reporting on rails.
On the ceded side the same shape appears from the other direction. A fronting carrier holds statutory reporting obligations and regulatory accountability for business it did not underwrite and whose data it did not originate, assembled from inbound bordereaux in N formats, on N cadences, at N quality levels. And treaty terms live in contract PDFs while claims data lives in a database, which is the whole structural cause of ceded recovery leakage: nobody joins them systematically.
// The build
Four builds, and what each one is measured on.
N bordereaux from one canonical record
Premium, claims, technical account and risk bordereaux generated for every relationship from one book of record, with per-carrier field mapping held as configuration rather than as a person's memory. Currency, date format and premium definition are declared per counterparty instead of inferred per file.
Surface: your policy and claims records into each counterparty's layout, with the mapping as an editable artifact you own.
Reported-versus-booked reconciliation
The gap between what you reported to the carrier and what the carrier booked. Today nobody owns it, it surfaces at audit, and it damages the relationship that renews your capacity. Least glamorous, highest value: unavoidable, currently manual, and measurable in dollars.
Surface: your ledger against the carrier's statement, reconciled per period and per section, with the variance and its cause named.
Exposure schema translation with a validation report
CEDE to EDM to OED across the versions actually in circulation, with a report on every conversion stating what was lost, coerced or defaulted rather than silently absorbing it. A conversion that does not report its losses is how a portfolio gets modeled on assumptions nobody chose.
Surface: cedent and program exposure files into your modeling schema, with the validation report attached to the run.
The delegated-authority audit file, assembled continuously
NAIC Model 225 requires quarterly financial accounts in a format that lets the carrier complete its annual statement, annual CPA reports, and at least semi-annual audits. We assemble the evidence as it accrues instead of in a scramble — and the file you build for the audit is the file that answers the reinsurer's expanded data request.
Surface: the same canonical record, exported as the quarterly account and the audit evidence pack.
What we measure
Days from period close to last bordereau delivered, per relationship, against the deadline in that contract. Reconciliation variance between reported and booked premium, as a dollar figure and a trend rather than a single quarter. Rejected or re-requested bordereaux per quarter, with the field that caused each rejection named. Hours of manual keying per reporting cycle, baselined by observation rather than estimate. On the ceded side: percentage of locations geocoded at rooftop against ZIP centroid, unrecovered ceded dollars identified, and days from period close to a filing-ready aggregate. Baselines are set before anything is built.
// We ran one of these
The deadline on this work is not an abstraction. The file goes out on the date, or the conversation with the capacity panel changes.
The operating record →// What's hard about this
Two limits, and what we do about each.
Column-name drift is a governance problem, not a parsing problem.
GWP against Gross Written Premium. DD/MM/YYYY colliding with MM/DD/YYYY. Currency stated in a different place, or not at all. All of it is normalizable. What no build can do is stop a carrier changing its template next quarter without telling you.
So the durable answer is a mapping layer your own team owns and can edit — not a model that has memorized this quarter’s file. Each counterparty’s mapping is a versioned artifact with an owner, a new or changed column fails loudly at ingest instead of being coerced into the nearest match, and the failure names the field and the file so the fix takes minutes rather than a reprocessing cycle.
Exposure schema versioning is a live business blocker.
Moody’s documents the case in full: a reinsurer received Florida exposure in CEDE 8.0, needed it in EDM to model it, and its conversion tool supported only CEDE 9.0 and 10.0. That left three options — a months-long tool update with revalidation, declining the business, or surrendering analytical control to somebody else’s conversion.
Moody’s · exposure data management · publication date not stated in our source.
None of the three is a technology decision. Each is an underwriting decision made by a file format, which is why the translator belongs in your repository rather than on a vendor’s release calendar. We build it against the versions actually in circulation on your inbound, with a validation report per conversion, and we widen version coverage on evidence of what arrives rather than on a specification’s version list.
// What ships with it
The evidence file, scoped to the delegated-authority audit.
The quarterly financial accounts in the format Model 225 contemplates, the reconciliation record with variances and their causes, the per-counterparty mapping artifacts with their version history, and the validation report on every exposure conversion. Where a model touches the reporting — classification, gap filling, geocoding — it arrives with a model inventory entry, lineage to source file and row, and a human-override specification. Roughly half the states have adopted the NAIC AI Model Bulletin, and delegated business is where the reporting obligation and the model risk meet.
NAIC Model 225 and state AI bulletins · current at September 2026.
What the governance file contains →Bring us one reporting cycle.
We will tell you where the reported-versus-booked variance is, and what the deadline actually costs you.