// Excess and surplus
Softening makes the submission flood worse, not better.
One retail agent, three wholesalers, six markets each. That is 18 carrier submissions for one risk, and the same risk lands on your desk under three slightly different insured names. Inbound grows while rate falls, and the compliance file behind every bound policy does not get any shorter.
The clerical half of this line is the clearest genuine return in commercial insurance. The judgment half is where the form comparison stops and a human starts.
// The situation
More transactions, less premium each.
In the first half of 2026, stamping-office premium grew 2.8% against 13.2% a year earlier — while item filings rose 16.9% to 4.28 million.
WSIA stamping office data · H1 2026.
That shape is rate softening, not demand contraction: business is still flowing to surplus lines, it is priced lower, and there is more of it to process per dollar earned. Which is why the first build is triage that runs before an underwriter opens the file, and the second is a compliance file that assembles itself as the placement happens.
// The compliance machinery you actually run
Every bound policy triggers five obligations.
Excess and surplus is now 25.7% of US commercial lines direct premium, up from 7.1% in 2000, and 2024 closed at $129.8B, up 12.3% for a seventh consecutive double-digit year. AM Best · 2025. Each of those placements carries diligent-search documentation, an affidavit, a stamping-office filing, a tax calculation and remittance, and eligible-carrier verification.
Some states require different platforms for policy reporting and for tax payment, and some require filing with both the insurance department and the revenue department. Penalties run $500 to $5,000 per violation for incomplete diligent search and $100 to $1,000 for late filing, plus license risk and, in the worst case, a voided placement.
Vendor compliance research, 2026 — rates, deadlines and platform assignments change by legislative session; the governing sources are the Excess and Surplus Lines Laws Manual and the state offices, and the build tracks them there rather than in a cached table.
On submission volume: brokers are reported to submit 30% to 40% more policies per placement in soft markets. Vendor research, 2026 — internally inconsistent across the same source, so we use it as illustration and never as a benchmark. The arithmetic at the top of this page needs no survey.
// The build
Four builds, and what each one is measured on.
Intake triage with duplicate detection
Classify inbound email, extract ACORD and schedule data, deduplicate the same risk arriving from multiple wholesalers under slightly different insured names, and score against appetite before an underwriter opens the file. Duplicates also corrupt hit-ratio measurement, so the fix improves the metric and the metric's accuracy at the same time.
Surface: the submission mailbox and document store, into the clearance and appetite records in your underwriting workstation.
Diligent-search file assembly
Capture carrier name, contact date and declination reason as they occur, generate the affidavit, and check the risk against the relevant state export list and the export-eligible commercial purchaser thresholds before anyone runs a search that was never required. The compliance file becomes a byproduct of the placement rather than a separate scramble at bind.
Surface: the placement record and the document store, out to the affidavit template and the state export list.
Filing and tax automation across four platforms
Route each placement to the right platform, calculate home-state tax under the Nonadmitted and Reinsurance Reform Act, respect per-state deadlines, and produce the remittance record. Split-department states are handled explicitly rather than as an exception someone remembers. Rates change by legislative session, so the update path is part of the build.
Surface: OPTins, SLIP, SLAS or InsCipher, driven from the bound-policy record with the remittance written back.
Manuscript form diff
Compare this year's manuscript wording against last year's and surface every difference with its location — sublimits, exclusions, definitions, endorsement stacking. The build produces the diff and the flag. It does not produce the conclusion, for the reason set out below.
Surface: the policy document store, with the diff attached to the renewal file and the binding-authority record.
What we measure
Share of inbound triaged and scored before an underwriter opens the file. Duplicate rate across wholesalers, and hit ratio recomputed with duplicates removed. Share of bound placements where the diligent-search file was complete at bind rather than assembled afterward. Days from bind to filing-complete, measured against each state’s own statutory deadline rather than an average. Share of renewals where a manuscript diff was reviewed by a named person before quote. Definitions and baselines, set in the first week.
// We ran one of these
We built the filing and remittance side of a regulated carrier — per-state deadlines, per-state platforms, and an examination that asks for the file rather than the intention — which is the closest operating analog to what a surplus lines placement carries on every single policy.
The operating record →// What's hard about this
Two limits, and what we do about each.
Manuscript forms are the reason this market exists and the reason extraction models fail in it.
Freedom of rate and form means the wording is bespoke per risk, which is exactly what defeats models trained on standardized documents. Finding the sublimit that quietly moved between last year’s manuscript form and this year’s is still human work, and it is where errors-and-omissions exposure in this line actually lives. So the build is a reliable diff that surfaces every change with its location, and it stops there: it does not rule on which changes are material, because a wrong answer on materiality is worse than no answer. The diff makes the human review complete instead of replacing it — every change reaches a named reviewer, which is more than the current process guarantees.
A model cannot read why a risk was declined by the admitted market.
That reason is frequently relationship or tacit knowledge, communicated in a phone call, and it is often the most informative fact in the file. Nothing extracts it, because it was never written down. The answer is to capture it at the moment it exists rather than reconstruct it later: the declination reason is a required field on the diligent-search record, captured as the call happens, in the words of the person who took the call. That turns the least machine-readable fact in the placement into structured data and the compliance artifact into the same object.
// What ships with it
The compliance file, and the governance file behind it.
Per placement: the declination log with dates and reasons, the diligent-search affidavit, the export-list and eligible-purchaser check, the filing confirmation and the remittance record, each tied to the governing state deadline. Per build: model inventory entry, data lineage, pre-deployment testing results, drift thresholds with remediation triggers, and a named human decision-maker specification — because roughly half the states have adopted the NAIC AI Model Bulletin and regulatory responsibility does not transfer to a vendor.
NAIC and state bulletins · current at September 2026.
Bring us one week of inbound.
We will tell you which placements are carrying the most filing risk, and what to build first.