Workers' compensation TPA claims handling: 120 claims per examiner, and the TPA pays for being late

A workers' compensation TPA does not sell claim handling. It sells provable timeliness against clocks it did not set, and when it misses one it often writes the check itself. That last part is in writing. The Judicial Council of California publishes the service guidelines for its own Judicial Branch Workers' Compensation Program, dated March 2023, and the document is unusually direct. On caseload: "the average monthly caseload of 120 claims per Examiner is desired. At no time shall the average caseload exceed 120 claims." On late benefits: "If SC is responsible for the penalty/increase, they shall issue a check to the JBWCP for the penalties/increases."

Read those two clauses together and the business comes into focus. The administrator is capped on how many files one examiner may carry, measured against that cap by an annual audit, and financially responsible for the statutory increase when its own lateness caused it. Nothing in that arrangement is about being fast. All of it is about being provably on time on a file somebody else will read later.

Most writing on this topic says a TPA makes claims handling faster and cheaper. The actual deliverable is a file that proves, claim by claim, which clock was met and why each decision was taken. Cycle time is a byproduct. Defensibility is the product, and a client audit, a state audit and a penalty assessment all read it. What follows is the decision clock in six states, what a miss costs, where caseload caps come from, and the three places the work concentrates: compensability, the medical record and subrogation. For the economics of automating inside that model, see automation inside a TPA's operating model; for the stages in order, the full claim path from first notice to settlement.

Two rule sets govern every file, and only one of them is public law. The first is the jurisdiction: forms, payment deadlines, notice requirements and presumptions, different in every state. The second is the client contract, which is almost always stricter. The Judicial Branch guidelines require the examiner to review every first report and contact the injured worker, the treating physician and the Member within one business day, a triage nurse within three business days, initial reserves within five business days, and a Team Lead to review the diary within ten calendar days of claim entry to confirm that benefits, notices and subrogation were addressed. None of that is in the Labor Code. All of it is auditable.

Scale makes the arithmetic unforgiving. Sedgwick states it handles over 8 million claims annually. Crawford and Company's FY2025 results put Broadspire, its TPA segment, at $401.9M of revenues before reimbursements, up 3.6% from $388.1M, with growth attributed in part to increased medical management usage. Public TPA scale reads more reliably off filings like those than off market-study estimates, most of which trace back to paywalls. On the regulator's side of the same market, California's DWC Audit Unit monitors 362 active claims-adjusting locations and logged 814 complaints in 2022, each of which an administrator has to answer within 30 days.

For a TPA claims operations leader, files per examiner, margin per file and client audit score are not three metrics. They are one constraint seen from three directions: a program renews on the audit, the audit reads the file, and the file is built by an examiner whose caseload the contract caps. Anything a TPA buys has to move inside that triangle, which is the subject of claims software for TPAs.

Start with one state, because even one state is not one clock. In California the employer must give the injured worker a DWC-1 claim form and authorise up to $10,000 of medical treatment within one working day of receiving the claim. The first temporary disability payment is then due no later than 14 days after knowledge of the injury and disability, under Labor Code 4650, the deadline California's administrative penalty schedule is keyed to. And a claim not denied within 90 days of the DWC-1 is presumed compensable under Labor Code 5402(b), which means an undecided file decides itself against the payer. All three dates run from events the administrator learns about secondhand, through an employer report it does not control.

Across states the deadlines are structurally similar and numerically different, which is the operational problem for any multi-state program. The table below is the decision clock, not the full filing calendar.

Two rows deserve emphasis. Florida's is the only rule in the set that converts delay into a permanent loss of rights: Florida Statute 440.20 lets a carrier pay and investigate, then waives the right to deny if it has not denied within 120 days of first providing benefits. New York's is a compound deadline, built from the 10-day employer notice and the 18-day payment rule the Workers' Compensation Board sets out, and Subject No. 046-760 on payor compliance applies that 18-and-10 test to three filings: the first report of injury, the initial payment of compensation and the Notice of Controversy, with a $300 penalty under WCL 25(2)(a) for a late controversy.

The two rows without a penalty figure are not the lenient ones. Pennsylvania's 21 days runs from the later of employer knowledge or the first date disability began, and the state publishes its own guidance on calculating 21-day compliance because the start date is the part payers get wrong. Illinois has no fixed accept-or-deny clock at all, which sounds easier until you read where the exposure actually sits: 820 ILCS 305 section 19(l) attaches additional compensation of $30 a day up to $10,000 once payment is withheld beyond 14 days after a written demand, and section 19(k) adds 50% of the unpaid amount where the delay was in bad faith. A state with no deadline is a state where the claimant's attorney sets one.

Texas shows what a filing calendar looks like under one decision clock

The 15th-day decision deadline in Texas sits on top of a form cascade that has nothing to do with the accept-or-deny call. The Texas Department of Insurance workers' compensation primer lays out the employer side: DWC Form-001 within 8 days of a lost-time injury, DWC Form-006 within 3 days when the employee loses time or returns to work and within 10 days of any status change, and a DWC Form-003 wage statement within 30 days, or within 7 days if the division asks for it. The employee has 30 days to notify the employer and a year to file the claim. Count the distinct deadlines in one state and the answer is six before anybody decides anything.

Per file, the statutory penalty is small. In aggregate, and in front of an auditor, it is structural. The California DWC Audit and Enforcement Unit's 2022 report to the Legislature completed 48 audits, reviewed 2,506 claim files and cited 3,053 violations carrying $739,519 in administrative penalties. 16 of the 42 routine audit subjects were third-party administrators, alongside seven combined insurer and TPA locations. Divide violations by files and the rate is roughly 1.2 per audited file, a derivation from the report's own two figures rather than something it states.

The sample findings are the part a client reads. The Audit Unit found unpaid indemnity on 10.81% of audited claims, averaging $117.26 of unpaid compensation on a randomly sampled file, and $30,813.71 of the 2022 unpaid total, 10.48% of it, was the 10% self-imposed increase owed on late payments. That last figure is the one to sit with: a tenth of what went unpaid was the penalty for paying late rather than the benefit itself.

Most administrators passed. The report records that 46 of the 48 audit subjects met the profile audit review standard, which is why most of the cited penalties were never assessed. Passing is not the same as clean, and the ceiling behind the schedule is real: Labor Code 129.5(e) allows a civil penalty of up to $100,000 where an employer, insurer or third-party administrator has made a violation a general business practice.

The per-payment schedule itself is published. 8 CCR 10111.1 sets $25 for a first temporary disability payment 1 to 7 days late, $50 at 8 to 14 days, $75 at 15 to 21 days and $100 beyond 21 days, with $50 or $75 for a delay-in-decision notice issued late. Other states express the same idea as a share of the payment, and the shares are not small.

Those are three different legal instruments drawn on one axis, and the dispersion is the point rather than the ranking. Illinois adds a second layer under 820 ILCS 305, where the Commission can award $30 a day up to $10,000 for unreasonable delay once payment is withheld beyond 14 days after a written demand. Florida separately holds medical bill payment to a 95% timeliness standard, with per-bill penalties of $25 at 90% to 99% compliance and $50 below 90%. Four states, four definitions of late, one book.

This is where most explainers go wrong, by implying an industry standard exists. It does not. Every workers' compensation statute reviewed for this post is silent on caseload. What exists instead is a set of privately negotiated numbers, and they cluster tightly enough to be useful.

Bill Zachry summarises a 2019 national claims-management research study that found indemnity caseloads from 80 to more than 300 per adjuster, with most administrators reporting 110 to 140, and an informal consensus that under 110 improves closures while above 125 does not achieve optimum results. Risk managers interviewed by Business Insurance described contractual caps of 88 to 125 open claims, set lower in California because litigation frequency is higher there.

Run the cap against the contract's own recurring standards and the day fills up before a single new claim arrives. The Judicial Branch guidelines require injured-worker contact at intervals not exceeding 90 calendar days, a plan of action updated on an indemnity claim every 90 calendar days, and an ISO index refresh every 180 days until the claim closes. At 120 open files those three cycles work out to about 1.3 contacts, 1.3 plan updates and 0.7 index refreshes a day, roughly 3.3 standing obligations per calendar day. Each of those is a derivation from the cap divided by the interval, not a published figure. And all of it lands before any statutory clock, and before the one-business-day three-point contact and ten-calendar-day diary review that every new claim triggers on top.

The supply side is not coming to the rescue. The US Bureau of Labor Statistics counts 389,700 claims adjusters, appraisers, examiners and investigators employed in 2025 and projects employment to fall 6% through 2035, about 21,800 fewer jobs, against roughly 21,600 openings a year that are nearly all replacement need. A contractual cap plus a shrinking bench leaves one variable: how much work one file takes. That is a different problem from headcount, and we separated the two in capacity is not the same problem as headcount.

For context from the investigation side of the same house, a manual SIU investigator carries 200+ cases and a manual investigation runs 14+ days per case, both Hesper internal benchmarks. Those are not examiner caseloads and should not be compared with 120 directly. What they share is the shape: a fixed human attention budget against a file count somebody else set.

The mechanics are an investigation on a timer. Arising out of and in the course of employment has to be established from an employer report, a recorded statement, the treating physician's first report and a prior-claim history. The Judicial Branch guidelines put dates on that too: indemnity claims are indexed at inception and every 180 days until closed, an investigation need must be identified and referred within five business days of claim receipt or of learning of a questionable issue, and a vendor referral goes out within five business days of client authorisation.

Most files never become litigated, and the ones that do are not random. Claims Journal, reporting a WCRI study of more than 950,000 claims with more than seven days of lost time, found attorney representation at 34% overall, 14% on temporary-disability-only claims, 64% where permanent partial disability or a lump sum was paid and 79% where both were, and associated attorney involvement with $7,700 to $12,400 in additional payments.

The volume side comes from Illinois, which publishes its whole funnel. The Illinois Workers' Compensation Commission handbook reports 200,000 to 250,000 injuries reported to employers a year, 65,000 to 80,000 reported to the Commission, 50,000 to 55,000 Applications for Adjustment of Claim filed and 3,000 to 4,000 arbitrator decisions. Roughly two in three claims reaching the Commission end up with a formal application filed, a ratio derived from the Commission's own published ranges. Settlements are approved on average about two years after filing, and unpaid medical bills accrue interest at 1% a month.

Fast denials are where this gets contested, and both sides are worth stating. Business Insurance, reporting WCIRB data, puts cumulative trauma at about a quarter of California indemnity claims in 2024 and on pace to exceed 30% in 2025, with med-legal service use on those claims up 38% between 2020 and 2024 against 2% for other claims, and 22% still open after five years. Allocated loss adjustment expense reached $13,860 per California indemnity claim in 2025 after roughly 9% annual growth since 2021. In the same piece, a Sedgwick vice president defends early denials.

That is the correct standard, and it is also the hard part. In the same article, plaintiff attorney Adam Dombchik argues that cumulative trauma claims are often denied within 14 to 21 days and without a proper investigation. Both statements can be true across a book of business, and the difference is not the speed of the denial. It is whether the documented reason exists in the file on the day the denial goes out, which is the same evidentiary bar a questionable-claim referral has to clear: workers' compensation fraud investigation.

NCCI reports the medical share of workers' compensation losses at 53% for accident year 2025 on a preliminary basis, unchanged from accident year 2024 and down from a 56.6% peak in accident year 2016. The share drifted down because indemnity grew faster, not because the records got shorter.

The evidence that record volume has outgrown the reader sits in the fee schedule. A California Workers' Compensation Institute study of the 2021 medical-legal fee schedule, covering January 2015 to October 2023, found that 43.3% of comprehensive medical-legal evaluations carried excess record-review charges, adding an average of $1,817 on top of the $2,015 flat fee for the evaluation itself. When a schedule has to price reading the record separately from examining the patient, the record has stopped being an input and become the work.

Treatment disputes add another reading layer with its own clock. California's independent medical review program received 201,037 applications in 2025 and issued 152,351 final determination letters, with about 91% of unique filings eligible for review, decisions issued 6 to 7 days after the required records arrived, 10.2% of utilization-review denials overturned, down from 12.7% in 2024, and pharmaceuticals accounting for 31% of requests. Nearly nine denials in ten survive review, which means the expensive part of that process is assembling and reading the record, not changing the answer.

The economics follow the reading. Crawford attributed part of Broadspire's FY2025 revenue growth to increased medical management usage, which is the quiet version of a structural point: at TPA scale, reading and managing the medical is something clients pay for separately. Reading the record is not overhead around the claim. At TPA scale it is a business. What that reading has to catch is covered in medical records as evidence.

The industry-wide numbers come with a caveat that has to be stated. Bisco and Fier, writing in the NAIC's Journal of Insurance Regulation in 2023, report $51.6 billion recovered across three auto lines in 2021, a mean salvage-and-subrogation indicator of 0.7714 across property-liability insurers, which means roughly one insurer in four records no salvage or subrogation recovery at all in a given year, and recovery averaging 4.5% of net claims paid across the full sample and 6.2% among firms that recover anything. The paper also cites an estimate that missed subrogation costs the industry $15 billion a year. Those are firm-level, all-lines figures with auto-focused modelling, so none of them is a workers' compensation rate.

The workers' compensation argument is structural rather than statistical. The right is a statutory lien on the employee's recovery from the third party, which makes it a legal instrument with its own procedure. New York's WCL section 29 shows it plainly: a claimant who settles a third-party action without the carrier's written consent or a compromise order can forfeit future benefits. Subrogation in this line is not a collections exercise bolted on at closure. It is a position that has to be preserved while the claim is open.

Good contracts respond by making it a diary item rather than a recovery project. The Judicial Branch guidelines require the Team Lead to review the diary within ten calendar days of claim entry specifically to confirm that benefits, notices and subrogation were addressed, and require indemnity claims to be re-indexed every 180 days until closed. Those are two screening mechanisms, one human and one data. Neither works if the examiner carrying 120 files has not read the accident description closely enough to notice the delivery van. That is the argument for screening every file for subrogation rather than the ones somebody remembered to look at.

Compressing handling time does not help a TPA if the file cannot prove the clock was met. Every section above points the same way: the deliverable is a documented decision chain, and that chain is what an auditor, a judge and a client program manager each read for a different reason. So the useful question about any automation layer is not how fast it goes. It is what it leaves behind.

That is the design Hesper is built on. Hesper AI is the AI claims resolution platform: AI agents take every claim from first notice to final recovery, with investigation-grade evidence behind every decision. On a workers' compensation file that reads as extraction from the employer report, the claim form and the medical batch, a cited coverage position the examiner signs, the medical record summarised against the treatment history, 15+ investigation phases running in parallel where the file is questionable, and a subrogation screen on every file. Fraud detection is built in rather than bolted on. The reason coverage matters, on the same internal benchmarks cited above: manual investigation runs 14+ days per case and one investigator carries 200+ cases, so only about 25% of flagged claims get a real workup. Agents run those phases in parallel, in minutes, not weeks, which takes coverage of flagged claims to 100%. Examiners review evidence-backed files instead of building them, and they keep every decision.

The part that does not transfer to another vendor's page is the single file. The same evidence pass that documents a compensability decision also documents the utilization-review referral, the bill-review exception and the subrogation screen, because it is one file rather than eight tools, and every action in it is logged with its source, its reasoning and its timestamp. That is what makes it answerable to a DWC auditor asking why a payment was late, to a client's annual claims technical administration audit asking whether subrogation was addressed inside ten days, and to a third-party carrier asking what the lien is based on. Point tools answer one of those three questions well.

Two boundaries. Hesper is not a claims administration system: it integrates with the system of record by API and can pick a claim up at any stage, so a TPA running a different platform per client program migrates nothing. And the return is not headcount. Nobody on a 120-file cap is getting removed; the file count stays and the work per file falls, which is what lets a program absorb a new client. Fraud detection is built in, and it is complementary where a client already runs FRISS, Shift or Verisk. The line-specific version is workers' comp claims automation, and the wider category sits in claims automation.

If you are a self-insured risk manager judging your own administrator, the checklist is already above. Ask for the audit result, not the summary score. Ask what the contractual examiner cap is and how compliance with it is measured. Ask how many files in last year's random sample carried unpaid indemnity. Ask who paid the self-imposed increase on the last late payment. Every answer lives in the file, which is the point.