A delegated claims authority audit almost never finds a wrong decision. It finds a decision with nothing recorded behind it. The reviewer is not re-adjudicating your claims. It is testing whether the programme can be trusted with somebody else's balance sheet, and the only evidence is the file you kept.
Capacity renews on that record. A capacity partner reads a sample as a proxy for the thousands of files it never opens. What one file has to survive on its own is in the claim file review checklist; the internal version of the exercise is the stage-by-stage claims leakage audit.
Three reviewers, three rulebooks, one claim file
Answer
What is a delegated claims authority audit?
A delegated claims authority audit is a capacity partner's test of a claims operation that settles claims on someone else's paper. It examines the contract, the authority grants, the reporting and the money, then a built sample of claim files. Up to three reviewers test the same file against three different rulebooks.
Three parties can audit a delegated claims operation, and only one of them is reading your contract. The carrier tests the contract. Under the NAIC Managing General Agents Act, Model #225, the insurer "shall periodically (at least semi-annually) conduct an on-site review of the underwriting and claims processing operations of the MGA," and Section 6 states the exposure: "An MGA may be examined as if it were the insurer."
The state regulator tests the statute. The NAIC Unfair Claims Settlement Practices Act, Model #900, defines "insurer" to include "agents, brokers, adjusters and third party administrators," and its trigger is frequency, not a single bad file.
The managing agent tests the binding authority. On Lloyd's business the audit runs against the standardised LMA coverholder and DCA audit scope embedded in DAM. Different reviewer, different instrument, different consequence. The same claim file.
For an MGA the consequence lands on capacity, which is why audit readiness decides whether delegated-authority claims programmes get renewed or repriced. For a TPA it lands on a client programme, which is why TPA claims operations live on client audit scores.
Whoever arrives, a finding is scored against the contract. The question is not whether a payment was generous, but whether it sat inside a documented grant of authority.
A finding is not feedback. It is a dated obligation.
The reviewer picks the scope, builds the file sample by rule, then rates each finding and attaches a deadline to it.
- Scope12The managing agent selects scope sections and the number of files to review
- Claims sections3 of 12Claims controls with authority, claims controls without authority, claims testing
- Findings4 ratingsEach finding rated immediate, high, medium or low by the managing agent
- Deadline7 to 90 daysHighest severity and shortest deadline lead where ratings conflict
The file you keep for the auditor is the same file that defends the decision.
The contract is the first exhibit in a delegated claims authority audit
Answer
What does a reviewer check before opening a claim file?
The contract. Every file-level finding is scored against what the contract allowed, so the reviewer reads the agreement first. Model #225 Section 4 lists the minimum provisions a written MGA contract must contain, and Section 4.I sets out what has to happen whenever that contract lets the MGA settle claims.
Model #225 opens Section 4 with a prohibition: "No person acting in the capacity of an MGA shall place business with an insurer unless there is in force a written contract between the parties which sets forth the responsibilities of each party," containing the minimum provisions that follow. The claims block is Section 4.I, and it bites only where the contract permits the MGA to settle claims.
The London market reading is more operational. Its first claims risk is that the department "is inappropriately structured, managed, resourced or experienced," which is a resourcing test as much as a process test. No published rulebook sets a caseload number, so what counts as adequate is the reviewer's judgement; the version of that argument with statutory deadlines attached runs through workers' compensation TPA claims handling.
What the reviewer samples, and why none of it is random
Answer
How are claim files selected for a delegated authority audit?
By rule, not at random. The bordereaux are reconciled to the claims system first, then files are pulled on six published criteria: a spread across period and authority, small reserves with large costs, nominal reserves, keyword hits in the loss description, denials, and claims with recoveries identified.
Claims testing starts with the report you already send. The auditor reconciles the bordereaux against your claims system and looks for unreported claims or movements. Unreported claims are rare. Unreported movements are common and quiet: a reserve that moves after the bordereau is cut leaves the carrier's view of the book stale.
Then the files. The LMA Coverholder Audit Scope v3.0 of October 2019 names six selection rules:
- A representative spread across the period, authority and activity.
- Small reserves with large costs.
- Nominal reserves.
- Key words from the description which indicate unusual claims activity.
- Any examples of claims denials.
- Any examples of claims with recoveries, salvage or subrogation identified.
Rules two and three expose reserving discipline, which is a file-notes test rather than a number test: did anybody write down why the number moved, or why it never did? Rule four is a keyword search, which is how a file with a third party in the narrative and no recovery entry surfaces. Lloyd's says the common scope now sits inside the Delegated Audit Manager and has been widened since.
How many files is a scoping decision, not a published standard. US market conduct examiners sample by claim status instead and calculate error ratios. One bad file is a complaint, a ratio is a practice. The per-file version is in the claim file review checklist.
The seventeen tests applied inside one claim file
Answer
What do carriers check inside an MGA or coverholder claim file?
Seventeen tests, scored across eleven disciplines. The LMA claims testing section runs from coverage adequacy and investigation appropriateness through reserving, recoveries, authority limits, denials and complaints. US market conduct examiners cover the same ground through eleven claims standards, seven of them marked Essential.
The claims file review is a published list, stable for years, so a delegated programme can build against it rather than guess. The table maps the tests that matter most to the US examiner's equivalent and to the absence a reviewer usually finds.
Results do not come back as a pass or a fail. Each scored discipline gets a percentage. The section also asks one closed question that no amount of file quality can answer: "Was any unauthorised sub-delegation noted?"
Items 1, 2 and 4 are separately scored rows that most claims operations run as one workflow. A programme that does coverage well and subrogation late loses points on a file it handled correctly.
A published order shows what those tests actually find
Answer
What does a real claims audit finding look like?
A Texas Department of Insurance market conduct examination of Tesla Property and Casualty, closed by consent order in July 2026, is the clearest published picture: real sample sizes, an error ratio for each test, findings on a missing MGA contract and unlicensed adjusters, and a $100,000 administrative penalty.
The Commissioner's Order No. 2026-10045 of 29 July 2026 is a state examination of a carrier, not a capacity partner's audit of an MGA. Read it as evidence of what reviewers test, not a verdict on any company.
Three conclusions were available before a single claim file was opened: business accepted "from an MGA without a written contract," an MGA not appointed on time, and "using adjusters who were not licensed."
One 110-payment sample carried findings on five separate tests
Market conduct examiners report an error ratio per test, because the statutes turn on whether something is a general business practice.
The file-level findings in Order No. 2026-10045 come back as ratios. In 24% of the claim payments reviewed, 26 of 110, the claim was not paid by the fifth business day, and "The delays in payment ranged from six days to 96 days."
In 7%, 8 of 110, the claimant did not receive the required Notice of Rights Regarding Repair of Motor Vehicle. In 5%, 5 of 110, the adjusters used were not licensed. In 1%, 1 of 110, there was no good-faith attempt at a prompt, fair and equitable settlement where liability had become reasonably clear. Of 40 complaints reviewed, 17, or 43%, were confirmed.
Then the scale finding. The same Texas order records that TDI "reviewed 39,481 claims" and that "In approximately 10% of the claims reviewed, the claims were not paid timely after accepting liability."
The order records $117,384.65 in statutory interest paid and a $100,000 administrative penalty. The chain is the point: a procedure gap produced a payment-timing practice, the practice produced a ratio, the ratio produced an order. Nobody had to make a wrong coverage decision.
What fails is the absence of a recorded decision
Answer
What fails a delegated claims authority audit?
Absences, mostly. A reserve moved with no recorded reason. A referral check that left no trace when the exposure crossed the limit. A denial with no stated basis. A third-party loss closed with no recovery entry. A claim open more than six months with no file sent to the insurer.
Start with the most mechanical failure. Model #225 Section 4.I(2)(d) requires a copy of the claim file to go to the insurer as soon as it is known that the claim "Is open for more than six months." There is no decision to get wrong, and compliance cannot be reconstructed afterwards.
Nothing in a claims system raises its hand on day 181. Every other lane on an open claim writes a record as it goes; the reporting duty comes due on a date with no event behind it, and nothing can be put back on a day that has already passed.
These are the findings that recur:
- No written contract with the MGA, or an MGA appointed late.
- Bordereaux that do not reconcile to the claims system.
- A payment above the delegated limit with no record of the carrier's prior approval.
- An adjuster not licensed in the state of loss.
- A reserve movement with no recorded rationale, or a denial with no stated basis.
- A third-party loss closed with no recovery referral either way.
The cleanest statement of the standard is a California claims regulation. Title 10 CCR Section 2695.3(a) requires claim files kept "in such detail that pertinent events and the dates of the events can be reconstructed." Reconstructable by someone who was not there: that is the test.
The claims system proves a field changed. The audit asks why. That distinction is worked through in the post on claims decision evidence and, for recoveries, in missed subrogation opportunities.
The cadence, the records, and the clock a finding starts
Answer
How often does a capacity partner audit a delegated claims programme?
It depends on the rulebook, and a delegated programme runs several clocks at once. Model #225 sets an at-least-semi-annual on-site review of the MGA's underwriting and claims processing operations, Lloyd's frequency is risk-based with a two-year floor for its own company's coverholders, and every finding carries its own deadline.
Lloyd's frequency is a judgement with one hard rule inside it: "it is a Lloyd's requirement for Lloyd's Insurance Company (LIC) coverholders to be audited at least every two years." Anyone who says Lloyd's requires an annual audit of every coverholder is reading the coordination cycle, not the frequency rule.
Three rulebooks set three different clocks on one programme
Audit frequency and record retention come from separate instruments, and the shortest clock is the one that binds.
Retention catches programmes on a rule they never read, because the clock starts in a different place in each instrument. The NAIC guideline on third party administrators, GL-1090, counts five years from a record's creation. California counts the current year and the four preceding years.
The NAIC chart of state laws on records maintenance, MC-90 of Summer 2024, shows how little of this is uniform: Arizona Section 20-485.03 requires administrator records for the duration of the agreement and five years thereafter.
Then the finding clock. Under the May 2026 coordinated-audit guidance, managing agents confirm "the severity of each finding (immediate, high, medium or low), and setting an appropriate completion deadline (7, 30, 60 or 90 days)." Authority and reporting findings outrank file-management untidiness.
The contractual end state is Model #225 Section 4.I(4): settlement authority terminated for cause, or suspended while the cause is in dispute. What the same gap costs in dollars is in the claims leakage audit checklist.
None of this would bite if capacity were scarce. It is not. AM Best's market segment report, published on 30 June 2026, puts DUAE direct premium written at $108.7 billion in 2025 against $92.3 billion in 2024. Capacity providers can be selective, and the audit is where selection happens.
Building the file while the claim runs
Answer
How can a small claims team produce an audit-ready file on every claim?
Not by documenting harder in the four weeks before the auditor lands. The record has to be produced while the claim is running, on every file rather than on the ones somebody remembers, which is the constraint a lean delegated claims team cannot solve with effort alone.
Hesper's AI agents take every claim from first notice to final recovery, with investigation-grade evidence behind every decision. Every agent action is logged with its sources, its reasoning and its timestamp. Adjusters review evidence-backed files instead of building them.
That is why the record can exist on every file. By Hesper internal benchmarks, manual investigation runs 14+ days per case and one investigator carries 200+ cases, so roughly 25% of flagged claims ever get a real workup. With agents the workup lands in minutes, not weeks, on 100% of them. The human keeps every decision.
Hesper integrates by API into the claims system of record and does not replace it. An audit tests investigation adequacy rather than flag quality, a distinction that runs through the complete guide to claims automation.
Key takeaways
- A delegated claims authority audit is a programme audit first: the contract, the authority limits and the reporting triggers are tested before a claim file is opened.
- The file sample is built, not drawn: small reserves with large costs, nominal reserves, keyword hits, denials and files with recoveries.
- Seventeen published tests are scored across eleven disciplines, so coverage, investigation and recovery are judged separately on the same file.
- The Texas consent order of 29 July 2026 shows a real finding: a missing MGA contract, unlicensed adjusters, and 24% of a 110-payment sample paid late.
- Findings arrive with a severity rating and a completion deadline of 7, 30, 60 or 90 days.