Buyer hub

Reinsurer operations AI: inbound packs, XoL, claims, accumulation

Inbound facultative files, inbound bordereaux, and exposure rolls. The failure mode is posting a limit that never appeared on the page. Source spans first; technical price later.

Reinsurer operations are inbound. Facultative packs land in a mailbox. Treaty bordereaux land on a shared drive or as an attachment to a quarterly email. Claims notices arrive with incomplete loss runs. None of that work is a chatbot problem. The failure mode is posting a limit, a TIV, or an attachment that never appeared on the page.

This hub is the operations map for inbound files. Source spans first. Technical price later. Quoted price last, and only after a human has seen the evidence. The marketing site does not take live client packs. Production ingest is tenant-scoped and revocable. The public walkthrough is the sample demo.

Inbound facultative packs

A facultative file is optional on both sides. The cedent or broker chooses to offer the risk. The underwriter chooses whether to look. What arrives is rarely a complete pack. It is a slip PDF, an SOV workbook, a loss run with no as-at date, a forwarded thread with three versions of the same schedule, and a covering email that restates the limit in a number that does not match page 2 of the slip.

The job of inbound facultative is not to summarise the file. The job is to make a pack an underwriter can triage against guidance without re-keying. Named insured, period, interest, territory, occurrence limit, deductible, TIV, and the schedule of locations should each carry a document, a page or cell, and a span. If a field has no span, it is a gap. If two spans disagree, it is a conflict. Neither is a rounding error.

Take the fictional pack used across this site. ACME Construction Ltd, acme.example, property facultative. Slip page 2 states an occurrence limit of USD 10,000,000. Slip TIV is USD 42,000,000. The SOV totals USD 47,100,000 because a warehouse was added after the slip was typed. The hours clause is not in the pack. The SOV has no as-at date. A useful inbound system records the limit with a span on page 2, surfaces the TIV conflict with both spans, and puts hours clause and as-at date on the chase list. It does not average the TIVs. It does not invent a hours clause from a similar risk last year.

Triage is the first human decision, not the first model decision. Appetite, class, territory, and capacity sit in underwriting guidance. Software can route a pack toward the right desk when the extracted class and territory are cited. Software should not auto-decline a construction risk because a model felt confident. The Triage Agent is the routing step. The facultative hub is the topic map for slips, certificates, and chase lists. This page is the reinsurer side of the same file: inbound, not outbound placing.

A complete inbound facultative pack, for an operator, looks like this.

• A slip or MRC with named insured, period, limits, deductibles, interest, and territory, each with a source span.
• An SOV or schedule that can be summed, with conflicts against the slip left visible.
• Loss history with an as-at date. Stale runs are chase items, not silent inputs to price.
• Wording excerpts that actually govern — hours clause, exclusions, occupancy — not a marketing brochure.
• A chase list the underwriter can send back without rewriting the file.

Speed without that file is a faster wrong quote. The quoted price is a later artefact. The inbound job is evidence.

Inbound bordereaux

Treaty inbound is a different mailbox and a different failure mode. A bordereau is a schedule of premium, claims, or commission that must map to a treaty year and a wording. Cedents send Excel with column names that change every quarter. Some send PDFs. Some send a scan of a printout. The accountant's job is to book rows that belong on this treaty, in this period, in this currency, against this attachment and this limit.

The failure mode is posting a total that cannot be traced to a sheet and a cell, or posting a risk that the wording excludes. Template extraction that maps "Pol Ref" on last quarter's file will silently map the wrong column when this quarter's file inserts a new field. A chat summary of the workbook is worse: it will give you a confident total with no row list.

Inbound bordereaux work is validation versus wording before you book. Period. Class. Territory. Currency. Attachment. Limit. Ceding commission basis. Claims bordereaux need event date, report date, paid, outstanding, and a policy reference that exists on the premium bordereau. If the wording says the treaty is 1 January to 31 December and the file is labelled "Q4" with rows dated into January, that is a gap or a conflict, not a helpful roll-forward.

The Bordereaux Agent is the inbound extraction and exception loop. The long-form operations note is the bordereaux automation guide. Treaty structure — quota share, surplus, excess of loss — lives on the treaty hub. This hub does not re-teach those structures. It says: inbound files must cite the wording they claim to settle.

Premium, claims, and commission files are three jobs that people collapse into one spreadsheet. Keep them separate in the pack even when the cedent combined them.

• Premium bordereaux: what was written, what was ceded, what commission was taken, against which treaty share.
• Claims bordereaux: what was paid, what is outstanding, what event, what policy, what year of account.
• Commission and profit-commission statements: the formula in the wording, not the formula in the covering email.

If a row cannot be mapped to a treaty term with a span in the wording or a cell in the file, it does not get booked. It goes on an exception list. That list is the reinsurer's chase list for treaty inbound. It is the same object as a facultative chase list: missing evidence, not a narrative.

Pricing: technical price versus quoted

Technical price is an actuarial construct assembled from cited inputs. Quoted price is a market decision. Mixing them is how a model output becomes a binder without anyone noticing the attachment was never on the page.

Inputs to technical price are documents. Subject premium by year. Loss experience with large-loss detail. Exposure and TIV distributions. Development on long-tail classes. The current structure: attachment, limit, reinstatement, hours clause, event definition. Those inputs have to be extracted with spans, or the burning-cost sheet is a story.

The Pricing Agent assembles those inputs. It does not pretend to be the treaty pricing committee. It does not silently bind a layer. A human sets technical price. A human, usually a different human, sets the quote. Software that emits a rate-on-line without a cited attachment is not pricing software. It is a completion engine.

On inbound facultative, the same split applies. The pack can show TIV, limit, occupancy, and loss history with spans. An underwriter may still decline, or quote a different share, or wait for the warehouse that appeared on the SOV. The quoted number is not a field the model is allowed to fill from a similar risk. If the hours clause is missing, technical price for a period-of-indemnity sensitive occupancy is incomplete. The chase list is the work remaining, not an appendix.

Do not ask the model for a "market indication" as a substitute for a cited attachment. Market colour is a conversation. Attachment is a span. See source-grounded extraction for the field contract. Pricing essays without a cited attachment are not ops.

Claims, IBNR, and leakage

Reinsurer claims work is not a retail FNOL bot. Notices arrive as emails, bordereaux rows, and scanned loss runs. The questions are operational.

Does this loss attach to this treaty or this facultative certificate? What is the event date versus the report date versus the period? Is the policy on the premium bordereau? Has the hours clause been applied the way the wording states, or the way the loss adjuster's narrative prefers? Is there a second notice for the same event under a different reference?

IBNR is a reserve language problem as much as a triangle problem. The triangle is only as good as the paid and outstanding that went into it. If claims bordereaux drop references, change currencies without a rate, or restate prior quarters without a bridge, the IBNR movement is not a signal. It is a data defect. Flag the defect. Do not smooth it.

Leakage, in this house, means money that moved without a cited contract basis: a recovery booked against the wrong year, a claim paid below attachment because someone used the slip TIV instead of the certificate limit, a facultative share applied to a treaty loss. The Claims Agent flags anomalies in notices and bordereaux against terms and history. It does not settle. It does not "estimate IBNR" as a single confident number with no triangle and no as-at. Operators review exceptions. Actuaries own the reserve.

A useful claims exception looks like a chase item.

• Missing event date or report date, so attachment to period cannot be shown.
• Policy reference that does not appear on the premium bordereau for that treaty year.
• Paid plus outstanding that exceeds the cited limit without a reinstatement row.
• Duplicate notices for the same insured and event with different references.
• Currency movement with no rate and no as-at.

Those are ops. A paragraph that says "IBNR appears adequate" is not.

Accumulation and clash

Accumulation is the same exposure counted more than once across inbound files, or counted in the wrong zone, or counted on a TIV that was never sourced. Clash is one event hitting more than one contract — proportional and non-proportional, facultative and treaty, two cedents with the same original insured — in a way the underwriting view did not add together.

Neither view is safe on invented TIV. If the SOV and the slip disagree, the accumulation engine that picks the larger number "to be conservative" has hidden a conflict. If it picks the smaller number, it has hidden a location. Show both. Let the accumulation desk choose, with the spans in front of them.

The Accumulation Agent is the inbound exposure roll: locations, values, occupancy, geocodes when the schedule actually contains an address. The Clash Agent is the multi-contract view for a single loss event. Both sit on extracted fields. They are not catastrophe models. They do not replace RMS or AIR. They tell you whether the file you just accepted is already in the pile, and whether two treaties will see the same original loss.

For ACME Construction Ltd, the extra warehouse on the SOV is an accumulation question before it is a pricing question. If that location is already on another facultative certificate, or already in a cedent's quota-share bordereau, the inbound underwriter needs to see the overlap. If the hours clause is missing, a weather event that aggregates over 168 hours versus 72 hours is not a modelling debate. It is a missing span.

Do not feed unsourced TIV into an accumulation dashboard and call it a peak-zone view. The dashboard will look complete. The clash at the loss will not.

Retrocession is a child of this hub

Retrocession is reinsurance bought by a reinsurer. It is the same document types — wordings, bordereaux, facultative packs, claims notices — with a different counterparty and a credit-risk overlay. It is not a new media pillar. It is not a second product. It is inbound and outbound on the same operations spine.

If you treat retrocession as a separate website section, you will duplicate the treaty and facultative maps and drift the definitions. Keep the definitions in one place. What is retrocession? is the short answer. This hub is the parent. The same rule applies: no span, no posted limit. The retrocessionaire who pays when you have a large loss needs the same cited attachment you needed from the original cedent.

Credit risk, collateral, and special-acceptances language still belong to the contract and the credit desk. Software that extracts a retrocession limit without a span has not reduced retrocession risk. It has created a second copy of an unsourced number.

What this hub is not

This is not a strategy essay about transforming reinsurance. It is not a public model that you paste a treaty into. It does not replace underwriting judgement, catastrophe models, or the pricing committee. It does not quote a savings figure. It does not publish an accuracy percentage on a buyer page.

It is inbound facultative, inbound bordereaux, pricing inputs versus quoted price, claims exceptions, accumulation and clash on cited exposure, and retrocession as a child. If you want the field contract, read source-grounded extraction. If you want the facultative placing view, read the facultative hub. If you want treaty structures as ops problems, read the treaty hub. The demo is a walkthrough of a sample pack, not a request that you upload a live cedent file to this website.