An MGA and coverholder audit prepared in a day, with every finding tied to its evidence
A delegate audit prepared in a day, with every finding tied to its evidence.
See one case, screen by screen ↓An audit solution for carriers that tests each delegate's risks and claims against the binder. It tests every record on the bordereaux, draws a risk-weighted sample of files with the reason for each, runs each file's test steps against its evidence and drafts the audit report with rated findings. The DA auditor reviews, rates and signs.
How to choose AI solutions for insurance →
Updated
Auditing an MGA or coverholder means sending the information request, waiting for the bordereaux and files, picking a sample of policy and claim files by hand, ticking test steps in a spreadsheet workpaper and writing the findings up in Word. For a DA auditor with 14 delegates, each reviewed on site twice a year, that is weeks per delegate.
And a sample picked by hand rarely holds the handful of risks bound over the per-location limit or the one claim paid over settlement authority. Once the report is out, the agreed actions live in email: the delegate closes an action with a procedure update, and six months later the same exception is back on the bordereau.
Estimated split for one semi-annual delegate audit, by hand and with the solution.
Six specialist agents test the whole book, pick the files most likely to fail and draft the report; the auditor rates and signs.
Harrowgate Specialty audits Bayline Coastal Underwriters, its commercial property MGA in Florida, Alabama, Mississippi and Louisiana, for April to September 2026: 2,318 risks and 412 claims under binding authority HSI-BA-26-014. Here is how the audit is prepared, screen by screen, in the working solution.
Information request IR-26-014-02 goes to Celia Marchetti, Bayline's Chief Compliance Officer: 41 items before the visit — the binder with Schedules A–D and both endorsements, six risk and premium bordereaux, six claims bordereaux, trust account statements, procedures and licences — plus the 36 sampled files once the sample is drawn. Each item is marked received only once it has been read.
40 of 41 items in. The business continuity test report is “Not yet — test booked 22 Oct”, tied to open action A-26-15.
Fourteen tests run on all 2,318 risks and 412 claims: territory, class, TIV per location, the Tier 1 wind aggregate, sanctions screening, rating credits, issue dates, remittance, settlement authority and reserve reviews. They find 61 exceptions in 9 tests. B5 alone has 37 risks screened after bind, and the wind aggregate stands at $348.2M — 87% of the cap.
Exceptions are never cleared by an agent. The referred bar/tavern risk and the corrected bordereau line are shown as cleared, with the reason beside each.
Every record is scored on the plan's weights: a whole-book exception that needs the file, within 10% of a binder limit, a process raised at the last audit, a Tier 1 wind county or parish, size, month-end binding, new staff and linked records. The 14 files with an exception that needs the file go in first, then the highest scores up to 30, plus 6 at random with seed 4471. The sample locks when fieldwork starts.
“60 of 2,730 records carry an exception (2.2%), so a random 36 would expect 0.8.”
The file tester worked overnight: 29 of 36 files tested, 232 of 288 test steps, 11 files with exceptions. On the right, the audit report assembles itself as the tests finish — opinion “Requires improvement”, 2 High, 1 Medium and 3 Low findings, with a placeholder for the one finding still waiting on its file.
Palmetto Shores Condominium Association, Gulf Shores, scores 70 — rank 4 of 2,318 risks: TIV $17.4M is 16% over the $15M per-location limit, Baldwin County is Tier 1, the premium is top 5% of the book. Step U2 fails, and the evidence is the referral log itself, page 3.
“No entry for Palmetto Shores Condominium Association between 1 Jun and 30 Jun 2026.”
Rachel presses one button. The tester reads the documents for each remaining file and ticks its eight steps one by one, each against a cited page — loss inside the policy period, acknowledgement within 3 business days, reserve reviewed, payment within authority, payee screened, the claims bordereau matching the file.
The file tester never passes a step without a cited passage, and never guesses a missing document.
CL-26-0611, Delacroix Hospitality Group: hail damage to the hotel roof and rooftop units, paid $62,400 on 24 Sep against a $50,000 settlement authority under claims agreement §5.1 — $12,400 over, with no Harrowgate approval in the file or the claims referral inbox. The claim note gives the reason.
Claim note, O. Pryce, 23 Sep: “Paying in full to keep the hotel open — will let Harrowgate know.”
36 of 36 files tested, 288 steps, 12 with exceptions. Finding F-04, “Claim paid above the settlement authority without approval”, is written into the draft as Medium. Each finding carries the condition, the criteria with the clause, the effect, the recommendation and the files it rests on — and F-02 is marked REPEAT ×3.
Ratings are proposals from the rating matrix. The auditor sets them.
Rachel reads the draft and sends it to Martin Achebe, Head of Delegated Authority, because two High findings need his approval before anything is shared with Bayline. Her note explains the call: F-02 is the third sanctions-timing finding in four audits, raised to High under the repeat rule.
“F-01 needs a decision on the three risks before the renewal discussion.”
Every action has an owner and a due date, 30, 60 or 90 days after issue by rating. The tracker chases owners 14 and 3 days before each due date, reviews the closure evidence Bayline sends, and reopens an action when the exception comes back — as A-26-13, the April sanctions action Bayline closed on 28 Jun, has.
Five audits since October 2024 on one line per theme: sanctions screening timing, “third time in four audits”.
The same Bayline audit, seen by the four people it runs through — what their audit season looked like, and what it looks like now.
Six specialist agents test the whole book, pick the files most likely to fail and draft the report; the auditor rates and signs.
Builds the information request from the audit scope and the binder, sends it to the delegate's named contact, chases and checks receipt.
Runs 14 tests on every risk and claim in the period against the binder limits, and lists every exception.
Scores every record on risk signals, picks the files most likely to fail plus a random set, and writes the reason for each pick.
Reads each sampled file and runs its eight test steps, attaching the document, page and passage to each result.
Drafts each finding — condition, criteria, effect, recommendation — proposes a rating from the matrix and marks repeats.
Tracks agreed actions, reviews the closure evidence the delegate sends, chases owners and reopens an action when the exception comes back.
Rates and signs; Head of DA approves High findings. The agents propose; a named person decides.
The auditor can ask about any file, test, finding or action in the audit — or ask for a change. Every answer points to its source.
Why were these 36 files picked?
Thirty were picked by risk weight and six at random. Every file with an exception that only the file can settle is in — among them the three risks over the $15M limit, four of the 37 late sanctions screens, the two biggest rating credits and the overpaid hail claim. The other 16 are the highest scores: near a limit, Tier 1 wind, bound at month end. A random 36 would expect 0.8 exceptions; this sample found 12.
Is the sanctions finding a repeat?
Yes — the third time in four audits. It was Low in April 2025 and Medium in April 2026; Bayline closed that action on 28 Jun with procedure BCU-CP-04 v3. Since then 37 risks were screened after bind, 22 of them bound in the last three days of a month. Your repeat rule raises it to High, so Martin Achebe approves it before the draft goes out.
Draft the agreed action for the overpaid hail claim
Drafted A-26-35 for F-04: block payments over $50,000 in Bayline's claims system unless an approval reference is entered, and ask Harrowgate to approve CL-26-0611 retrospectively. Owner Owen Pryce, due 27 Dec — 60 days for a Medium finding.
Which whole-book tests came back clean?
Five of the 14: territory (B1), the Tier 1 wind aggregate — $348.2M, 87% of the $400M cap (B4), minimum premium (B7), claims inside the period and cover (B10) and duplicate payments (B14). Two more were cleared on the file: the bar/tavern class was referred and approved, and a June bordereau line was corrected in July.
13 screens from the working solution, on its sample data. Pick one to see it large.
How audits are rated, sampled and approved is a setting for each carrier, and the binder limits are read from the delegate's own binding authority.
| Setting | Default | Choose from |
|---|---|---|
| High findings | Fix within 30 days of issue · Head of DA approves before issue | 15 to 180 days, in steps of 15 |
| Medium findings | Fix within 60 days | 15 to 180 days, in steps of 15 |
| Low findings | Fix within 90 days | 15 to 180 days, in steps of 15 |
| Repeat rule | On — the same theme at two audits running raises the rating one level | On · Off |
| Files in the sample | 30 risk-weighted · 6 random, drawn with a recorded seed | Set in the sampling plan |
| Exceptions that need the file | Up to 4 per test | All over an authority limit; otherwise up to 2, or 4 where the last audit raised it |
| Risks to claims | 2 : 1 | At least one file per state |
| Sampling weights | Over or near a limit, prior finding, Tier 1 wind, size, month end, new staff, linked records | Try other weights on the sample screen |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample audit data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published requirement (NAIC MGA Model Act, Model 225, and state adoptions — on-site review of the MGA at least semiannually). People, carriers, delegates and insureds named on this page are fictional — characters and sample data in the working solution.
An audit solution for carriers that tests each delegate's risks and claims against the binder. It tests every record on the bordereaux, draws a risk-weighted sample of files with the reason for each, runs each file's test steps against its evidence and drafts the audit report with rated findings. The DA auditor reviews, rates and signs.
The whole-book tester reads every risk, premium and claims bordereau for the period and the limits in the binding authority, then runs 14 tests on every record — territory, class, TIV per location, wind aggregate, sanctions screening, rating credits, issue and remittance dates, settlement authority and reserve reviews. In the working solution that is 2,318 risks and 412 claims in 11 minutes.
Every record is scored on the sampling plan's weights — an exception that needs the file, near a binder limit, a process raised at the last audit, Tier 1 wind, size, month-end binding, new staff, linked records. The highest scores are pulled, plus a set of random files drawn with a recorded seed, and the sampling memo records the weights and each file's reasons.
Yes. The action tracker chases owners 14 and 3 days before each due date, reviews the closure evidence the delegate sends and reopens an action when a later whole-book test finds the same exception. The repeat-finding history shows every theme across audits, and the repeat rule raises the rating of a theme raised at two audits running.
Yes. The report drafter only proposes ratings; the auditor sets them and signs the report, and High findings need the Head of Delegated Authority's approval. Agents never issue reports, close actions or clear exceptions, and clearing an exception or overriding a step needs a reason on the audit trail.
No. The audit request agent writes only to the delegate's named response contact, never outside the agreed scope, and the auditor approves the request before it is sent. The delegate's contact sees requests, draft findings and their own actions, and uploads evidence.
Yes. Fix-by days for High, Medium and Low findings, the repeat rule, the number of risk-weighted and random files, the cap on exceptions per test, the risks-to-claims ratio and the sampling weights are all settings. The binder limits are read from each delegate's binding authority.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your audit scope, rating matrix, report template and a delegate's binder and bordereaux — and it goes live once every quality gate has passed. We will walk you through it on one of your own delegates first.
Every bordereau in your layout on arrival, each rejected row with its reason.
1-min demoBinding Authority ComplianceBreaches flagged the day the risk is bound, not months later at audit.
1-min demoBinder Guidelines SearchFewer needless referrals, and every answer shows the clause that decides it.
1-min demoWe'll run MGA & Coverholder Audit on one of your own delegates' binders and bordereaux.