Best claims software for TPAs: four purchases, and one of them is a competitor

Best claims software for TPAs is a four-way question in 2026, and only three of the answers are software. The two loudest claims-AI launches aimed at a third-party administrator's book this year came from TPAs, not software companies: Sedgwick announced Omni on May 4, 2026, and Corgi Claims launched on June 29, 2026 as an AI-native administrator with more than 5,000 licensed adjusters nationwide.

That changes the shape of the shortlist. Comparing FileHandler Enterprise against Origami Risk against SpearSuite is still a real decision, and the matrix below works through it. But the entrant that moved fastest did not buy a claims system. It bought the whole claims function, and it now sells supplemental claims capacity to the carriers, MGAs and captives a traditional TPA sells to. So the software question has a prior question stacked on it: what does this TPA own that a funded AI-native competitor does not, and which purchase makes that thing stronger?

Three unglamorous assets: adjuster licences in the states the programs sit in, client relationships on paper the TPA does not own, and a claim file that holds up when a Department pulls a sample. The third is measurable. When the Delaware Department of Insurance expanded a market conduct examination of the Liberty Mutual Group to cover homeowner claims adjusted by third-party administrators, half of everything the Department told the carrier to fix traced back to those files. The fee mechanics underneath the purchase - why automation at a TPA pays for itself out of the per-file fee rather than out of savings - are worked through in claims automation for TPAs.

Sedgwick's Omni release describes a platform, not a product line. Its proprietary AI "has been expanded to assist with document and call summarization, digital triage, severity modeling, automated reserving, fraud detection, and quality oversight." Automated reserving and fraud detection are two of the capabilities a mid-market TPA is most likely shopping for, and the largest administrator in the market now ships both. The scale figures in the same release are Sedgwick's own - 33,000+ colleagues, 10,000 clients across 80 countries, a dataset it says is five times larger than its nearest competitors' - so read them as the company's numbers. Either way the structural point holds: Omni is Sedgwick's platform for Sedgwick's clients, not for sale to a TPA competing with Sedgwick.

Corgi is the sharper version of the same move because it started from zero. Its launch release puts $374 million raised since founding and a $2.6 billion valuation on the record as of June 29, 2026; it has raised again since, at figures this post will not print because they are not confirmed at primary source. Corgi's own site describes the company as "a third-party claims administrator for carriers, captives, and MGAs" with "5,000 licensed adjusters across the country" in all 50 states, handling claims "from first notice of loss to final payment" including subrogation and special investigations. Lines at launch: commercial liability, property, catastrophe, renters, trucking, workers' compensation and specialty programs.

The premise is correct and worth conceding. Busywork should happen at the moment of notice, and licensed adjusters should spend their time on judgment. The disagreement is about who runs that operating model. Corgi's buyers are carriers, MGAs, captives, program administrators and self-insured organizations wanting outsourced administration or supplemental capacity, which is the same list a traditional TPA sells to. There is no version of Corgi a TPA deploys to defend its own book.

Most TPA software evaluations go wrong at the first step, by putting all four in one spreadsheet and scoring them on the same features. They solve different constraints. If four client programs run on four systems and nobody can see the book, that is a system-of-record problem. If each adjuster carries more files than they can touch, moving where the file is stored changes that number by exactly zero.

Search results for claims software rank systems of record, so buyers shop systems of record. That is the quiet distortion in this category. A TPA administering a carrier's program often does not own the system of record at all, and a TPA running five programs may be working inside several at once. Anything it buys has to work across all of them. The by-stage version of this argument is in the 2026 AI claims processing shortlist: pick by job, not by logo.

The pressure forcing the decision is arithmetic. NCCI's 2026 State of the Line put the workers' compensation calendar-year 2025 combined ratio at 91% and the accident-year 2025 combined ratio at 102%, with the gap carried by reserve releases against a $14 billion redundant position. Lost-time frequency fell 2% while medical and indemnity severity each rose 4%. Fewer claims, each more expensive and each taking more work.

Every cell below traces to a vendor page or a primary release. The limitation column is the point of it: a roundup with no limitations is a directory, and each one is phrased as what the vendor's public materials describe, because that is what a buyer can verify before a demo.

Sedgwick and Corgi are not rows. Neither sells a product a competing TPA can license, so neither belongs in a matrix you can shortlist from. They sit in the fourth column of the previous table, which is a different conversation with a different owner.

The four the search results surface

Aclaimant is the strongest front end here for what happens before a claim exists: field incident intake, digitised forms, OSHA compliance automation, return-to-work tracking. Its verticals read as incident-driven workers' comp and general liability, from staffing and construction to transportation, manufacturing and the public sector. It publishes three tiers, RMIS Basics, Core and Enterprise, with no dollar figures. Be clear-eyed about the category: a TPA adjudicating on a carrier's paper needs an adjudication system underneath an RMIS, not instead of one.

Spear Technologies is the most interesting bet for a TPA already standardised on Microsoft. SpearSuite runs on the low-code Power Platform, so configuration can plausibly be done by people the TPA already employs, and Spear is explicit that the deployment is customer managed and that "you own your data." It names the Virginia Auto Dealers Association, Utah Business Insurance Co., Synergy Comp and Pitman Family Farms as customers, and claims SOC 2 Type 2 certification. The tradeoff is that sentence backwards: what makes low-code fast to change makes it yours to maintain.

JW Software has been in business since 1989 and its integration list is the most TPA-literate here: state EDI, medical and pharmacy bill review, HR and payroll, ISO ClaimSearch, other TPA platforms, policy systems, banking ACH. Compliance posture is unusually explicit for a mid-market vendor - SOC 1, SOC 2, SOC 2 plus HITRUST, HIPAA, GLBA, PCI-DSS, NIST - alongside a 99.9% uptime guarantee. JW says it has over 200 clients and a 100% implementation success rate; those are its own figures. The notable part is its AI position published on August 10, 2026: "AI should operate within the organization's security and governance framework. The security and governance framework should not be redesigned around the AI."

That sentence is right. What is too narrow is the tradeoff implied around it, between automation depth and governance. It holds for configurable rules and diaries. It does not hold for an agent that logs every source it read, its reasoning and its timestamp, because that record is more reconstructable than a human adjuster's note, not less.

VCA Software is the only vendor here whose core design assumption is that the work gets assigned out, which is how property TPAs, CAT operations and independent adjuster firms actually run: automated assignment, real-time status, Lloyd's-compliant reporting for delegated authority programs, the InsuredConnect policyholder app, ClaimPay payments. VCA says onboarding runs two to three weeks for Premier clients and that the platform reduces the cost of the claims journey by as much as 30%; its own figures. Its published AI is claim file review and document summarisation, which makes reading the file faster. The assignment still waits for a human panel, and that panel is the ceiling.

The systems of record underneath

Origami Risk is the only vendor here that sells the client-facing half of a TPA's job as a product. Client-branded portals, self-service configuration to onboard new clients, subrogation tracking and bordereau generation are revenue-protection features, and its TPA page names Harford Mutual Insurance Group, Mission Underwriters and Cities and Villages Mutual Insurance Company. It also partners publicly with CLARA Analytics and Gradient AI, which is a useful tell: the platform is the distribution, the intelligence is brought in on top. Sapiens ClaimsPro took a Celent 2024 Luminary designation out of 32 North American core claims solutions, with Celent calling it a top choice for midsize and small insurers. Both are profiled in the top 10 claims management systems for P&C carriers.

Guidewire ClaimCenter usually is not the TPA's decision. It is the client's system, and a TPA administering a program for a ClaimCenter carrier inherits it as a constraint. Five Sigma is the closest thing here to an AI-native core platform that will sell to a TPA, across 11 P&C lines, with Clive as a multi-agent layer that can run on top of an existing claims system rather than replacing it. Snapsheet names TPAs as a segment and puts payments and vendor management in the same platform as the file; its one published customer result is a 15% decrease in operation costs at Clearcover, Snapsheet's figure. Head-to-head detail on the systems of record is in the claims management systems comparison.

Feature sets that look interchangeable on a grid stop being interchangeable the moment you name the book. Five shapes cover most of the US market, and each has a different thing that disqualifies a vendor outright.

• Workers' compensation TPAs. State EDI filing, medical and pharmacy bill review integration, caseload caps written into client contracts, and a tail that runs for years. JW Software, Origami Risk and Spear cluster here. A platform without state EDI is not a candidate at any price.
• Property TPAs and independent adjuster firms. CAT surge capacity, licensed adjusters by state, photo estimating, and assignment logistics as the core workflow. VCA and Snapsheet cluster here, and Corgi lists catastrophe explicitly among its own lines.
• Liability and casualty TPAs. Medical records, demand packages and litigation exposure drive the work, and the incumbents are document and analytics point tools sitting beside the claims system rather than core platforms.
• Warranty and specialty TPAs. High volume, low severity, and straight-through resolution is the whole game. Five Sigma's multi-line posture fits this shape; anything requiring an adjuster touch per file does not.
• MGA-delegated claims. The capacity partner's claims audit is the gate on renewal, so the audit trail beats the feature list. This is the one segment where accepting fewer features for a cleaner record is the right trade.

Crawford's FY2025 results, published March 2, 2026, are the cleanest public read on what claims administration earns. Broadspire delivered $401.9 million of revenue, a record annual high for the segment, at a 13.6% operating margin against 13.5% the year before. The group reported $1,265.7 million of consolidated revenue, $82.3 million of operating earnings and a 6.5% consolidated operating margin. CorVel's fiscal 2026, for the year ended March 31, 2026, gives a second data point: $959 million of revenue, up 7%, with earnings per share of $2.14, up 17%.

That is the number a software quote has to clear. At a 13.6% segment margin the licence line is not what decides the purchase; the hours an examiner spends assembling a file are. And the hours are not getting cheaper. The Bureau of Labor Statistics counts 389,700 claims adjuster, appraiser, examiner and investigator jobs in 2025, projects a 6% decline through 2035, and expects roughly 21,600 openings a year. The sentence worth reading twice is what it says about those openings: "All of those openings are expected to result from the need to replace workers who transfer to other occupations or exit the labor force, such as to retire." A replacement market, at a median wage of $78,020.

That distinction is the capacity argument in craft terms. A TPA does not lack people who can read a file. It lacks people willing to spend the first six hours of a file being a processor so they can spend the seventh being an adjuster. Casaleggio's other line in the same Claims Journal piece: "Claims professionals should not spend their careers performing tasks that technology can perform faster and more consistently."

This is the slot the automation layer gets bought for, and the investigation work is where the manual baseline is starkest. Hesper's internal benchmarks put manual investigation at 14+ days per case against 200+ cases per investigator, which is why roughly 25% of flagged claims get a real workup; with agents running 15+ investigation phases in parallel, coverage goes to 100% of flagged claims and the file arrives in minutes, not weeks. Those are Hesper internal benchmarks, not industry figures. What matters to a TPA is the shape of the change: adjusters review evidence-backed files instead of building them.

The Delaware Department of Insurance market conduct examination report on Liberty Insurance Corporation, Liberty Mutual Personal Insurance Company and LM Insurance Corporation, as of July 31, 2023 and filed March 18, 2025, is the most specific public document on how TPA-handled files read to a regulator. The report records that "the examination was expanded on March 27, 2024, to include the adjusting and processing of Homeowner Claims for the Liberty Mutual Group by Third-Party Adjusters (TPAs)." The companies identified ALLCAT and Sedgwick as the administrators in use, produced a universe of 386 paid and denied claims, and a random sample of 84 was reviewed.

Those five counts are the report's own: nine under 902-3.1.5 (coverage not affirmed or denied within 30 days of proof of loss), seven under 903-5.0 (claim not paid within 30 days), two under 18 Del. C. 1703 (adjusters not properly licensed), and one each under 902-3.1.1 (pertinent facts misrepresented on a denial letter) and 902-3.1.2 (communication not acknowledged within 15 working days). Because they are exceptions rather than files, converting them into a percentage of the sample would overstate it. The figure that carries the commercial weight sits in the conclusion instead. Of the ten recommendations the Department made, five are tagged "(Third-Party Adjusters)".

And the exceptions were cited against the companies, not against ALLCAT or Sedgwick. That is the mechanism. The carrier carries the regulatory exposure for work its administrators performed, which is why a client's claims audit is not a formality and why a TPA's file quality is a commercial asset rather than an overhead line.

Four clocks a TPA's software has to satisfy

California's file and record documentation rule reads like a software specification once you take it literally. The file has to carry enough detail that "pertinent events and the dates of the events can be reconstructed," and the licensee has to record the date it received, the dates it processed and the date it transmitted every material and relevant document. Texas keys its administrator clock differently: records held until the fifth anniversary of the end of the contract term. California keys retention to the claim, Texas to the contract, so a TPA that loses a program still owes five years of records on it.

Texas 4151.1042 removes the question of whether a TPA gets audited. Subsection (b) states that "an insurer shall ensure competent administration of its programs," and subsection (c) requires a review of the administrator's operations at least semiannually and an on-site audit at least biennially above 100 participants. The fraud obligation behaves differently again: California 10 CCR 2698.33 provides that contracting out SIU or integral anti-fraud duties "shall not relieve the insurer of any obligation" under the regulations or the IFPA, so a TPA on a carrier's paper is operating the carrier's non-delegable obligation. Chargeback mechanics and who owns delegated SIU are worked out in the shared-services investigation post.

This is where the automation-layer argument stops being about speed. If every agent action is logged with its source, its reasoning and its timestamp, the reconstruction the California rule asks for is a byproduct of the work rather than a project after it. A disciplined examiner with good diary habits produces the same artifact; the difference is that the discipline does not degrade on the four hundredth file of the month.

Licensing is the most concrete. Texas Insurance Code Chapter 4151 provides that an entity may not act as or hold itself out as an administrator without a certificate of authority, and that before it issues the applicant must obtain and maintain a fidelity bond against an act of fraud or dishonesty in exercising its duties as administrator. Adjuster licensing bites separately, per state. Two of the Delaware exceptions were cited under 18 Del. C. 1703, the report finding that "the Companies transacted insurance in Delaware with adjusters that were not properly licensed." Even incumbents trip on it, which is exactly why Corgi leads with 5,000 licensed adjusters across all 50 states: the licensed panel is the asset being competed for, not the model.

The second asset is the one software decisions quietly damage. A TPA's client relationship rests on paper it does not own, inside systems the client often controls, under an audit the client is required to run. Every migration project puts that relationship at risk for two or three quarters, and every new point solution adds a second audit trail to reconcile before the biennial on-site. The purchases worth making improve what the client's auditor sees without asking the client's IT function for anything.

The third lever is the one TPAs most often leave alone, because it pays the client rather than the TPA and so never makes the business case. Recovery. Subrogation and salvage get screened late, by exception, or not at all, and a program where every file is screened produces a number the TPA can put in front of a client at renewal without touching the fee schedule. That is the argument for carrying evidence across the whole path rather than buying a tool per stage: every claim, first notice to final recovery. The stage-by-stage map is in the guide to automating the claims lifecycle.

The honest limit is the part vendors skip. Rami Hashish, writing in Claims Journal on April 29, 2026, put it well: "Algorithms are very good at the 80% of a file that looks like every other file, and consistently unreliable on the 20% that actually matters." The right response is not to argue the percentage. It is to notice that the 20% is where a TPA earns its fee, and that the only way to buy adjusters more time on it is to take the other 80% off their desk with a record good enough to defend.

None of the ten publishes pricing, which means every one is a quoted deal and cheapest is not a shoppable attribute. The comparison runs on fit and on what the file looks like after the system has touched it. Five questions get there faster than a 200-row requirements matrix.

1. Does it work across my clients' systems of record, or does it assume I own one? A TPA running five programs may be working inside a different core system for each, none of which it controls. Any answer that starts with a migration plan is answering a carrier's question, not a TPA's.
2. What does the file look like after it runs? Ask for a sample file, not a dashboard. Can an examiner reconstruct the pertinent events and the dates of the events from what the system wrote, with nobody narrating it? If the answer involves exporting a log and interpreting it, the answer is no.
3. Which lines, and which states, including adjuster licensing? State EDI for workers' comp, Lloyd's-compliant reporting for delegated authority, licensed capacity by state for property and CAT. This question eliminates more vendors than any other and it costs one email.
4. Does it remove hours, or only reading time? Summarisation makes an adjuster faster at reading a file somebody still has to build. Extraction, coverage reading, evidence assembly, investigation and recovery screening are the hours. Ask which of those run unattended and what happens when confidence is low.
5. What happens on the 20% of files that do not look like the others? Ask for the escalation path, the human decision rights, and what the system logs when it declines to decide. A platform with no answer here will quietly route its hardest files back to the people you were trying to free up.

Where Hesper lands on those five, limitations included. Hesper AI is an AI claims resolution platform: agents take every claim from first notice to final recovery, with investigation-grade evidence behind every decision and fraud detection built in. It sits on top of the system of record rather than replacing it, integrates with Guidewire, Duck Creek, Majesco and others, and picks a claim up at any stage, which is the specific reason it suits a TPA running several client systems. Security posture is SOC 2 Type I. It runs standalone or alongside FRISS, Shift and Verisk where a client already has them. The limitation is categorical: Hesper is not a claims administration system, so a TPA without a system of record still needs one, and both Spear and JW publish stronger certification claims than Hesper does today. The TPA version is at Hesper for TPAs; capability detail is on the product page.