A third party administrator is not allowed to be paid for the money it saves its client. The NAIC model guideline on third party administrators bars any fee contingent on savings in the payment of losses, and expressly permits compensation based on the number of claims paid or processed. That one rule decides how claims automation for TPAs gets bought. A carrier can fund a leakage tool out of the loss ratio it owns. A TPA cannot. It funds it out of the per-file fee, or not at all.
Which is the more honest arithmetic anyway. When the fee is fixed per claim and the handling duration is not, every hour of file-building work taken off an adjuster is margin the TPA can count without asking a client to agree on a counterfactual. No baseline to negotiate, no argument about what would have leaked regardless.
That constraint is why the buying conversation at a TPA looks nothing like the one at a carrier. A carrier's claims department is a cost center inside a loss ratio. A TPA's claims department is the product: sold per file, audited by the client that bought it, and clocked by state regulators, not by the client.
Three threads carry the whole decision: margin per file, the file the client's auditor opens, and how many files each adjuster can carry. What follows works each one against primary sources: the NAIC TPA guideline and MGA Act, two public workers' compensation TPA solicitations, the Lloyd's delegated claims administrator regime, three state prompt-pay statutes and the Bureau of Labor Statistics adjuster outlook. The stage map underneath all of it is the complete guide to claims automation.
The fee is per file, and the NAIC model guideline is why
Claims automation for TPAs is software that removes the file-building work - intake, coverage reading, evidence gathering, investigation, document drafting, subrogation screening - from the adjuster, so the fixed per-claim fee covers fewer labor hours. At a TPA it is a margin decision, not a loss-ratio decision, because the fee is set before the handling effort is known.
Start with the rule that closes the door most buyers try first. NAIC Guideline GDL-1090, section 9A: a TPA "shall not enter into an agreement or understanding with a payor ... in which the effect is to make the amount of the TPA's commissions, fees, or charges contingent upon savings effected in the payment of losses covered by the payor's obligations." A share of the leakage stopped, or of the fraud avoided, is out. That is not a negotiating position a TPA can win on; it is the shape of the rule.
Section 9C states what stays open: "This section shall not prevent the compensation of a TPA from being based on premiums or charges collected or the number of claims paid or processed." Section 9A carves out three more - hospital and other auditing services, managed care, and compensation for subrogation expenses. Hold that last one; it returns at the end of this post. Counting the NAIC state adoption chart (Fall 2024), 47 of 50 states have a TPA statute, regulation or bulletin, so most administration agreements are drafted against some adaptation of this text. The expense-line mechanics sit in four ways to charge back AI investigation.
The structures were catalogued thirty years ago and have not changed. A 1996 Casualty Actuarial Society discussion paper on TPA service pricing and incentive contracts groups them as Per Claim, Dedicated Office/Unit, Percent of Incurred, Percent of Paid and Percent of Premium, each sold with a service length: 12 months, 24 months, life of partnership, or life of claim. It records that "Per Claim has been the most popular choice among self-insureds," that "12-month handling is the predominant choice," and that "the pricing of TPA service contracts is extremely competitive." Its 1995 dollar examples are worthless now; its structures are still on the table.
Both shapes show up in current solicitations. A California transit district priced workers' compensation TPA services as a fixed price per contract year. The City of Pleasanton's April 2026 solicitation asks for "monthly fixed price payments to the TPA" and requires bidders to disclose their assumptions, including "the number of claims expected to open and close." The revenue line is set before anyone has read a file.
The tail is the part that was never priced
A TPA's tail is the population of claims that stay open past the contract year that priced them. Under life-of-claim and life-of-partnership service lengths the fee is fixed when the claim is received, while the handling duration runs until settlement. Every drifting or reopened file is therefore worked at a price set years earlier, against effort nobody forecast.
The tail does not always arrive gradually. The Santa Barbara MTD workers' compensation TPA solicitation, issued in September 2020, makes it a condition of award: "The TPA shall be entrusted to manage both new claims and assume any 'tail claims' from an existing administrator." The incoming TPA inherits another administrator's open inventory in whatever state it was left, and prices the program before reading any of it.
The 1996 CAS paper treats that as a pricing problem, not an operations problem. It describes life-of-claim service as handling "until settlement at a fixed cost," notes that "TPAs in general have difficulties in forecasting the costs and pricing their products," and adds that "the longer the service length, the more uncertainty in service pricing and revenue accrual." Its working assumption is that "the older the claims, the less time they need for service," which holds only while a file is stable. It stops the moment an old file reopens, a provider changes, an attorney appears or a subrogation target surfaces late, and the aged claim demands new-claim effort at an aged-claim price.
Where the program is delegated, that drift is also a reporting event. The NAIC Managing General Agents Act requires a copy of the claim file to go to the insurer as soon as the claim "Is open for more than six months." An aged file is a file the carrier can call for, so it has to be presentable, not merely open.
Reconstructing one stalled file by hand - pulling the medical set, reading the endorsements, re-interviewing, rebuilding the timeline - runs to 14+ days per case as a Hesper internal benchmark, which is why it gets rationed to the files somebody complains about. Agents run 15+ investigation phases in parallel (Hesper internal benchmarks) and return the result in hours, not weeks, so the question stops being which tail files are worth reconstructing and becomes how many can be brought to a known state. The workers' compensation version is on the workers' compensation claims page.
What the client's auditor actually opens
A TPA claims audit tests files, not process. The client or its carrier samples open and closed claims and asks whether the documentation in each file supports the decision, the reserve and the timeline. The NAIC TPA guideline requires a TPA to keep complete books and records of every transaction performed on the payor's behalf for at least five years.
Section 5A of GDL-1090 requires a TPA to "maintain and make available to the payor complete books and records of all transactions performed on behalf of the payor," retained "for a period of not less than five (5) years from the date of their creation." Section 7A requires a written agreement kept for the term and five years after. Section 7C is the sentence that should govern how a TPA thinks about file quality: an insurer using a TPA "is responsible for the acts of the TPA." The client cannot delegate away the consequence of a bad file, so the client audits the file.
Service agreements add their own machinery. The 2020 Santa Barbara MTD solicitation gives the client the right "to audit at any time with three (3) days' notice to TPA," and requires the TPA to be available "at its sole cost and expense" for an annual on-site audit. Appendix A sets out the grid it is graded against.
One solicitation is an example, not a benchmark
The 24-hour to 180-day grid above, the 120-file caseload cap and the 10% late-payment clause all come from a single public workers' compensation TPA solicitation issued in September 2020. They are what one real client wrote into one real contract, not an industry average. The transferable part is the structure: clients buy TPA services against dated obligations and audit the files that prove them.
Read the right-hand column as a documentation specification. Every row is a date the TPA can either evidence from inside the file or not, and it is the same surface that produces leakage findings, as set out in how claims leakage actually happens. Hesper inverts the usual build order: the audit trail is the primary output, not a log of it, with every agent action recorded against sources, reasoning and timestamps and attached to the claim record. Evidence behind every decision. Coverage matters as much as the trail, because carriers fully investigate roughly 25% of flagged claims manually and the model is built to take that to 100% (Hesper internal benchmarks). An audit sample does not restrict itself to the 25%.
Delegated authority raises the bar again
Delegated authority is an arrangement where a TPA or managing general agent determines claims on an insurer's behalf: accepting or denying, agreeing amounts, resolving disputes. Under the NAIC Managing General Agents Act the acts of the MGA are considered to be the acts of the insurer, and the insurer has to review the claims operation on site at least semi-annually.
Model #225 section 5C: "The insurer shall periodically (at least semi-annually) conduct an on-site review of the underwriting and claims processing operations of the MGA." Section 6 settles who owns the outcome: "The acts of the MGA are considered to be the acts of the insurer on whose behalf it is acting. An MGA may be examined as if it were the insurer." Section 4I(3) adds that "All claim files will be the joint property of the insurer and MGA." The file is not the administrator's to keep, to shape, or to explain after the fact.
Two semi-annual reviews, and only one of them is a P&C requirement
GDL-1090 section 7H also requires the insurer to review the TPA's operations at least semi-annually, with at least one review including an on-site audit. That provision is written for life, annuity, health and employee benefit stop-loss coverage administered for more than 100 certificate holders, subscribers, claimants or policyholders, so read it as the model's audit cadence, not as a property and casualty obligation. The P&C-side analogue is NAIC Model #225 section 5C, which applies where a managing general agent sits in the structure. Confirm which text your state adopted and which your program sits under.
The Lloyd's market runs the strictest written version. Its Delegated Claims Administrator Register guidance (May 2025) defines a DCA by the authority it holds - a firm authorised to "accept or deny a claim in whole or in part; agree any amount payable" - requires that "All DCAs are required to be registered with Lloyd's prior to appointment," and expects managing agents to keep "continued regular ongoing oversight of the third party's claims performance," evidenced "and readily available to Lloyd's if requested." The test is not size or scope. It is whether you decide.
The coordinated audit guidance (May 2026) targets an end-to-end audit assignment "within a maximum of six months, from scoping to post-audit action issuance," and recommends new binders be audited "within the first 12 months of inception." Add an annual on-site at the client's request and triggers that can pull any claim into the carrier's hands at any time, and the constraint for a lean team is not decision quality. It is whether the record can be assembled often enough without pulling adjusters off files. A trail produced as the work happens costs nothing to produce again.
The clocks belong to the state, not to the client
A prompt-pay clock is a statutory deadline for acknowledging, investigating and deciding a claim. It binds the insurer, and it reaches the TPA through the definitions: California's regulations apply to any claims agent authorized by an insurer to conduct an investigation of a claim on the insurer's behalf. The service agreement sits on top of that clock, never underneath it.
California Code of Regulations title 10 section 2695.2 defines a "claims agent" as "any person employed or authorized by an insurer, to conduct an investigation of a claim on behalf of an insurer ..." That is a description of a TPA. The obligations in section 2695.7 therefore land inside the TPA's own workflow, not at the client's door.
Pennsylvania and Texas price failure differently. 31 Pa. Code section 146.6 buys time with paperwork. Texas Insurance Code section 542.060 attaches interest at 18 percent a year plus fees and does not ask why the file was late. California also sets a bar on the work itself, not only its timing: section 2695.7(d) requires every insurer to "conduct and diligently pursue a thorough, fair and objective investigation." Speed alone does not satisfy that, and neither does thoroughness arriving on day 60. Speed and leakage are the same problem, and this is one of the few places a regulator wrote that down as a single duty.
Contracts then push the consequence onto the TPA. The 2020 Santa Barbara solicitation requires late payments to carry "the self-imposed 10% penalty ... at the TPA's cost" unless the client caused the delay, and requires an arising-out-of-employment investigation to be "initiated within three (3) days of the decision to delay." The work that triggers a delay decision is also the work whose own deadline starts running from it. That is the argument for pointing agents at the investigation instead of at the notifications: automating the 30-day status letter removes a clerical task and leaves the investigation as slow as it was, while finishing the investigation in hours, not weeks, removes the reason the letter was needed. The stage mapping is in the end-to-end claims process walkthrough, and the routing layer is on the claims triage page.
Caseload is the only variable left
Adjuster caseload is the number of open files one adjuster carries at a time, and at a TPA it is frequently a contract term, not an internal target. One public solicitation caps it at 120 open indemnity claims with one assistant per adjuster, and restricts indemnity files to adjusters with at least three years of experience.
The 2020 Santa Barbara language is blunt: "Adjuster Caseloads shall be no more than one hundred twenty (120) open indemnity claims; one assistant per adjuster." A caseload ceiling in a service agreement is a cost structure in a service agreement. It sets minimum headcount for a given claim count and removes the TPA's ability to absorb a volume swing on the people it already has. The 1996 CAS paper frames the ceiling as a purchase: a lower caseload per examiner buys better claimant service, while "a higher caseload per examiner" saves adjustment expense. The client's RFP usually decides where a TPA sits on that line.
There are three ways to carry more files: hire, raise the cap, or move the work off the adjuster. The first is getting harder. The US Bureau of Labor Statistics projects employment of claims adjusters, appraisers, examiners and investigators to "decline 6 percent from 2025 to 2035" from a 2025 base of 389,700 jobs, roughly 23,400 fewer positions, with about 21,600 openings a year attributed largely to replacement rather than growth. The 2025 median adjuster wage is $78,020. The second route is a renegotiation with a client who wrote the cap in for a reason. That leaves the third.
The industry-level version of the arithmetic sits on the NAIC results page. For full year 2025 the NAIC property and casualty industry report shows loss expenses incurred of $86,003 million against net losses incurred of $551,755 million: about 15.6 cents spent handling every dollar of loss paid, against 15.3 cents the year before, and about 9.0% of net premiums earned inside a 92.9% combined ratio. A TPA fee is the client-facing version of that line, and it is the line a client benchmarks at renewal.
Moving the work does not mean moving the judgment. The adjuster still decides compensability, reserve, settlement and denial. What changes is who assembles the material underneath: the policy and endorsement read, the record retrieval, the timeline reconstruction, the open-source checks, the subrogation screen, the draft correspondence. Those phases run 15+ at a time on a single file (Hesper internal benchmarks), which is why coverage moves before speed does.
The throughput benchmarks behind that chart are 200+ cases per investigator and 800+ cases per investigator per month, against roughly 10 per investigator per month manually (Hesper internal benchmarks). Read them as capacity, not as a staffing plan. This is not a headcount argument and a TPA should not run it as one: clients audit the staffing model, and a bid that wins on thinner staffing loses at the next claims audit. The caseload ceiling simply stops being the binding constraint, because the work behind each file shrank. Adjusters review evidence-backed files instead of building them. Which files earn the full workup is a triage decision, covered in the five-lane triage framework, and the decisions that stay with people are set out in which decisions stay human.
A TPA is not allowed to be paid for the money it saves its client. That one rule closes the left column, and everything a claims automation business case can be at a TPA has to be built out of the right one.
A summary compresses reading time, not the file
A claim summary and a claim file are different artifacts. A summary shortens the time an examiner spends reading what is already there. A file is what a client auditor, a capacity partner or a state examiner opens, and it has to carry the sourced documents, the timestamps and the reasoning behind each decision, including decisions that produced no payment.
The distinction is easiest to see at the top of the market. Sedgwick describes itself as having 33,000 or more colleagues across 80 countries, serving 59% of the Fortune 500. In February 2025 it launched a generative AI feature in viaOne that, in its own words, "allows examiners and claims professionals to quickly view a summary of the past 12 months of information for a particular claim at the click of a button."
That is a sensible first build and it deserves respect. It also marks the boundary. A summary compresses reading time for the person already handling the claim. It does not run the coverage analysis, order the records, complete the investigation, close the subrogation loop, or produce the artifact the client's auditor opens. Every TPA that is not Sedgwick has to buy that second capability rather than build it.
The rest of the field clusters at two ends. At the front door, Bevaya (formerly Roots Automation) reports 115 or more workflows in production on a claims page describing intake, indexing, extraction and invoice payment, and counts three of the top 20 TPAs among the deployments listed on its homepage. Avallon sells AI agents for TPA claims operations covering intake, status calls, extraction and outreach. At the container end sit the claims administration systems. In between are scoring layers: Five Sigma's Clive describes a Risk agent that "flags suspicious activity and recommends further investigation," which is a handoff rather than a finding.
Those are vendor descriptions, and the gap is a gap in what the public materials describe, not proof of what any product cannot do. Read as a map it is consistent: the front door is crowded, the container is crowded, and the middle is thin. The middle is completing the investigation and drafting the recovery demand inside the same flow that pays the claim, with sourced and timestamped evidence attached, which is precisely the file the auditor opens. Hesper is the work-and-evidence layer on top of whichever claims system the TPA's client already runs. Clean claims resolve straight through; suspicious claims get an investigation-grade workup.
Drop-in, because the system of record is the client's
At a TPA the claims system of record is frequently the client's rather than the TPA's. A TPA running programs across Guidewire ClaimCenter, Duck Creek Claims, FileHandler Enterprise, Origami Risk and Snapsheet cannot make a deployment decision that requires every client's IT department to agree first. The only deployable shape is additive.
Those systems are good at what they are for. JW Software describes FileHandler Enterprise as claims management software "designed to simplify complex processes and improve detailed tracking for TPAs, self-insured groups, workers' comp administrators," with "over 50 built-in business automation rules." Origami Risk sells cloud-based claims administration "with embedded AI features" to carriers, MGAs, TPAs and risk pools. Snapsheet markets a "complete claims system" to four segments including TPAs. All three are vendor statements, and all three describe a system of record doing its job.
Rules route and track. They do not reach outside the system to gather the evidence a client auditor reads, and they were never built to. That is the additive slot. A head-to-head is in Hesper versus Snapsheet, and the wider field is mapped in the claims management systems comparison.
For the technology owner the questions are narrow and should be answered narrowly. Hesper integrates with the claims system of record by API and can pick a claim up at any stage instead of owning it from intake. It holds SOC 2 Type I. It does not train on customer data. Claim data is retained for the subscription term. Fraud detection is built in, so it runs standalone at a TPA with no detection vendor, and alongside FRISS, Shift or Verisk where a client carrier already has one. The integration bill, not the software, is what usually kills these projects, and that arithmetic is in the hidden integration costs of legacy claims AI.
Where the margin actually comes back
Margin per file at a TPA moves on four lines: the labor hours spent building each file, the penalties and rework avoided on missed deadlines, the audit findings that decide whether a program renews, and the recovery work the NAIC guideline permits a TPA to be paid for. Only the first is a cost line. The other three decide revenue.
Size the first against audited numbers, not market-research estimates. In its full year 2025 results, Crawford and Company reported revenues before reimbursements of $1.266 billion, down 2% from $1.293 billion. Its Broadspire TPA segment posted record segment revenues before reimbursements of $401.9 million, up 3.6% from $388.1 million, with operating earnings of $54.6 million and an operating margin of 13.6%. North America Loss Adjusting ran at 6.9%, up from 5.8% in 2024. At Broadspire's scale, one point of operating margin is about $4.0 million a year.
The top line is not fully in a TPA's hands. Crawford attributed softer volume to "a lack of severe storms throughout the year" that "drove lower claims activity industry wide." CorVel reported revenues of $959 million for the fiscal year ended 31 March 2026, up 7% from $896 million, with earnings per share of $2.14. When the claim count is weather-dependent, margin has to come from the file.
The fourth line is the one most TPAs under-run. GDL-1090 section 9A bars savings-contingent compensation but expressly preserves a TPA "being compensated for subrogation expenses." Recovery is the one place a TPA is paid for producing a result rather than processing a count, and the obligation usually already sits in the contract: the 2020 Santa Barbara solicitation requires that "When subrogation is to be pursued, the third party shall be contacted within ten (10) days." Ten days is not a recovery-queue timeline. It is an intake timeline, which is the argument in why missed subrogation is lost at intake, not at closure.
Hesper screens every file for subrogation and salvage and drafts the demand instead of screening the subset an adjuster flagged; the mechanics are on the subrogation and recovery page. On cost, the internal benchmark is roughly $2,500 per manually investigated case against roughly $150 per case run by agents (Hesper internal benchmarks), which turns full coverage into a pricing option instead of a staffing fantasy. Notice what that does to a bid: the Pleasanton RFP asks bidders to list services "provided at an additional cost to the basic fee quote, and what the cost will be (e.g. field investigation ...)." A TPA that can run field investigation inside the base fee is answering a different question on the scorecard. The variables are on the ROI page.
Key takeaways
- NAIC Guideline GDL-1090 permits TPA compensation based on the number of claims paid or processed and bars fees contingent on savings in the payment of losses, so automation has to show up as margin per file rather than as a shared-savings line.
- Life-of-claim and tail pricing fix the fee while leaving the handling duration open, so an hour saved on a file compounds and an hour added to an aged file is charged against a price set in a prior contract year.
- The client's claims audit tests the file rather than the process, and the NAIC Managing General Agents Act puts an on-site review of claims processing operations on at least a semi-annual cadence under delegated authority.
- State prompt-pay clocks bind the TPA on the client's behalf - 40 calendar days in California, 30 then every 45 in Pennsylvania, 15 business days in Texas with 18% a year attached - and one public TPA contract puts the late-payment penalty on the TPA itself.
- US claims adjuster employment is projected to decline 6% between 2025 and 2035, roughly 23,400 jobs, so a TPA that wants more files per adjuster has to move the file-building work instead of raising the caseload cap.