---
title: "Claims shared services and TPAs: four ways to charge back AI investigation, and only one survives"
description: "For a shared-services or TPA claims leader, AI investigation stalls on allocation, not capability. Share-of-savings is barred, per-case approval is a compliance risk, and the admin fee buries the cost."
date: "2026-08-18"
lastModified: "2026-11-16"
author: "Pankaj Dhariwal"
tags: ["Guides"]
canonical: "https://gethesperai.com/blog/head-of-claims-shared-services-ai-investigation/"
---

# Claims shared services and TPAs: four ways to charge back AI investigation, and only one survives

> **TL;DR** For a claims shared-services or TPA leader, AI investigation is an allocation problem before it is a technology problem. Share-of-savings pricing is barred by NAIC Guideline GL-1090 section 9.A, per-case client approval runs into California's disincentive-to-investigate prohibition, and burying the cost in the administration fee forfeits allocability. A flat per-claim charge with standing authority is the one model left standing.
>
> - GL-1090 section 9.A bars savings-contingent fees; 9.C permits per-claim
> - The same fraud investigator lands in DCC or A&O by capacity
> - One workflow across N books replicates one defect N times

- **$86.0B** - US P&C loss adjustment expense, 2025 (About 9% of net premiums earned (NAIC 2025 full year results))
- **~50%** - Lower expense per investigation at national vs regional carriers (The shared-services thesis in the industry's own benchmark (CAIF and Aon, September 2024))
- **0.12%** - Of premium spent on SIU (Roughly 1.3% of the entire loss-adjustment expense pool (CAIF and Aon 2024; NAIC 2025))
- **28** - Jurisdictions that adopted NAIC Model #668 (Plus 1 pending and 3 with their own data-security provisions (NAIC map, April 1, 2026))

For a claims shared-services leader or the claims head at a third-party administrator, the blocker on AI investigation is not whether it works. The cost has to land on somebody's expense line, and the three models a claims leader reaches for first are each closed off by something already written down.

The multi-book structure that looks like the hardest possible place to deploy AI investigation is actually the easiest place to prove it. What is left is one model: a flat per-claim investigation charge on every eligible claim, disclosed in the administration agreement before the first claim is referred. For what automates across a claim stage by stage, start with [the complete guide to claims automation](/blog/claims-automation-pillar/).

## A shared-services claims operation cannot buy investigation, it can only allocate it

Claims shared services is one claims organization administering several books, legal entities or client carriers under separate agreements. It runs as a cost center, so it owns no loss ratio it can improve by spending more. Every purchase is an allocation question first: which book pays, on which expense line, under which contract.

The pool is large and the fraud slice is small. The [NAIC full year property and casualty results](https://content.naic.org/sites/default/files/2025-annual-property-and-casualty-and-title-insurance-industries-analysis-report.pdf) put 2025 US net premiums earned at $958,726 million and loss expenses incurred at $86,003 million, about 9.0% of earned premium. The [2024 Insurer SIU Benchmarking Study](https://insurancefraud.org/wp-content/uploads/2024-Insurer-SIU-Benchmarking-Study.pdf) from the Coalition Against Insurance Fraud and Aon, covering 35 participating property and casualty carriers, puts SIU budgets at about 0.12% of premium, down from 0.13% two years earlier.

**Figure 1 (The pool it competes in) - SIU spend is about 1.3% of what the industry spends adjusting claims.** Divide the SIU share of premium by the loss-adjustment share: the line you are asked to grow is among the smallest on the sheet.

0.12% divided by 9.0% equals 1.3%

- SIU budget, share of premium: 0.12% (CAIF and Aon 2024, down from 0.13%)
- Loss adjustment expense, share: about 9.0% (of earned premium, NAIC full year 2025)
- SIU share of all LAE: about 1.3% (An order of magnitude, not a decimal)

**Context:** US P&C loss expenses incurred were $86,003 million in 2025, about nine cents of every earned premium dollar.

Sources: NAIC 2025 full year property and casualty results (net premiums earned $958,726 million, loss expenses incurred $86,003 million); Coalition Against Insurance Fraud and Aon, 2024 Insurer SIU Benchmarking Study (35 participating P&C carriers). The two ratios come from different years, so the result is derived and an order of magnitude.

The unit cost is harder to pin down. The Coalition and Aon put budget expense per investigation just above $1,200 in 2024. That figure excludes adjuster hours, outside vendor invoices and counsel time, so the fully loaded number runs well above it.

Automated investigation changes the unit. Hesper runs 15+ investigation phases in parallel on a flagged claim and returns an audit-ready file in minutes rather than weeks. For a shared-services operation that means a unit price small enough to put on a rate card and charge per claim.

## ALAE or ULAE: the line item that decides the business case

Allocated loss adjustment expense is claim-specific. Unallocated is everything else, the administrator's overhead. Under the NAIC definitions effective January 1, 1998, the same fraud investigator falls on either side of that line depending on the capacity they are working in.

[Texas Department of Insurance Bulletin B-0002-98](https://tdi.texas.gov/bulletins/1998/b-0002-8.html) reproduces the NAIC lists. The only words that move the expense are "if working in defense of a claim" versus "if working in the capacity of an adjuster." AI investigation spend inherits the same fork.

In the annual statement these buckets are reported as defense and cost containment, or DCC, and adjusting and other, or A&O. This post uses DCC and A&O from here.

Now the consequence. AI investigation bought as a departmental subscription and booked to A&O is fixed overhead the administrator absorbs. The identical capability, scoped and invoiced per claim as investigative work in defense of a claim, is allocable to that claim. In deductible, self-insured retention and program structures, that reaches the client or insured layer.

*Figure: The same investigation, two expense buckets. Booked to A&O it is the administrator's overhead, absorbed once and spread across every book. Scoped as claim-specific defense and cost containment it lands in DCC on the individual claim and follows that claim's expense treatment out to the client or deductible layer, book by book.*

An allocation survives a claim file audit when it ties to a specific claim, with a record of what was done, by what method, and when. Hesper logs every step with sources, reasoning and timestamps and attaches it to the claim record.

## The chargeback models that survive contact with the contract

There are four ways to recover the cost of AI investigation from the books or clients that consume it. One is prohibited outright for administrators inside GL-1090's scope. One creates a compliance problem most buyers do not see coming. Two survive, and only one of those preserves allocability.

| Chargeback model | Accounting bucket | Allowed under GL-1090 section 9 | Risk under CA section 2698.33(c)(2) | Verdict |
| --- | --- | --- | --- | --- |
| Share of fraud savings | Neither cleanly | No. Section 9.A bars fees contingent on savings in loss payments | High | Do not build |
| Per-case charge, client approves each case | DCC (allocated) | Yes. Section 9.C permits per-claim compensation | High. Approval friction is a disincentive to investigate | Rework before launch |
| Per-case charge, standing authority for all flagged claims | DCC (allocated) | Yes | Low | Cleanest |
| Rolled into the base per-claim administration fee | A&O (unallocated) | Yes | Low | Clean, but absorbs the cost and forfeits allocability |

Share of savings is closed. [NAIC Guideline GL-1090](https://content.naic.org/sites/default/files/GL1090.pdf) section 9.A bars an arrangement whose effect is to make a TPA's compensation "contingent upon savings effected in the payment of losses." Section 9.C leaves compensation based on "the number of claims paid or processed" open. Hesper prices a flat per-case rate for that reason, explained in [the Hesper pricing guide](/blog/hesper-ai-pricing-guide/).

Per-case client approval fails somewhere less obvious. [California 10 CCR section 2698.33(c)(2)](https://www.law.cornell.edu/regulations/california/10-CCR-2698.33) requires that an SIU contract "Not include provisions that could provide disincentives to the referral and/or investigation of suspected insurance fraud." Claim-by-claim approval makes declining the investigation the cheapest path for a client's claims staff.

The third row is the design that holds: standing authority to investigate every flagged claim, a flat per-claim charge, and disclosure up front. Rolling the cost into the base administration fee is also clean, but it lands in A&O and the administrator absorbs it.

The rate itself matters. A per-case charge at manual cost levels invites a conversation about every case, and each one is a moment where an investigation does not happen. Bring the charge down far enough and it becomes a rate, not a decision. That holds coverage at 100% of flagged claims instead of the ~25% a manual operation reaches.

> **Scope caveat worth stating to your own compliance team**
>
> GL-1090 is an NAIC Guideline rather than a model law, and its scope covers life, annuity, health, stop-loss and, in one version, workers' compensation administration. It does not reach general property, auto or liability claims administration. Treat it as the clearest written statement of the structure regulators expect, not as the statute governing your auto book. The California SIU regulations carry no such limit.

## Who actually owns the SIU obligation when a TPA administers the book

Functions delegate. Obligations do not. An insurer may contract out the performance of SIU work, and remains answerable for it. The carrier client will want visibility into every investigation run on its book.

**Figure 2 (Delegation · California) - Functions delegate. Obligations do not.** Follow a contracted SIU function through the regulation: the work moves to the administrator, the answerability stays with the carrier.

**Where the obligation lands**

1. The carrier may staff its SIU by contracting with other entities, 2698.30(p)
2. A TPA handling claims is a contracted entity, down to sub-subcontractors, 2698.30(f)
3. The contract shall not relieve the insurer of any obligation, 2698.33(a)
4. The plan says how the carrier monitors the third party, Guideline #1690 4.C(7)

**The same answer, in other words**

- New York DFS: Contractor allowed
- Florida 626.9891: Contract filed
- NAIC GL-1090 section 7.C: Insurer answers

**The record is the payor's.** Whatever an investigation layer produces joins the payor's claim file, kept not less than five years under GL-1090 section 5.A.

Sources: California 10 CCR sections 2698.30(p), 2698.30(f) and 2698.33(a); NAIC Antifraud Plan Guideline #1690, adopted October 29, 2020, section 4.C(7); New York DFS; Florida Statutes section 626.9891; NAIC Guideline GL-1090 sections 5.A and 7.C.

California is the most explicit. 10 CCR section 2698.33(a) says any contract for the performance of SIU duties "shall not relieve the insurer of any obligation under these regulations or the IFPA." A TPA administering claims sits squarely inside the regulation's definition of a contracted entity. See [the guide to California 10 CCR 2698 SIU compliance](/blog/california-10-ccr-2698-siu-compliance/).

Other states land in the same place. [Florida Statutes section 626.9891](https://www.flsenate.gov/Laws/Statutes/2024/626.9891) lets an insurer contract with others to investigate possible fraud, treats that contractor as part of the anti-fraud investigative unit, and requires the insurer to file a copy of the executed contract.

The [NAIC Antifraud Plan Guideline #1690](https://insurancefraud.org/wp-content/uploads/Antifraud-Plan-Guideline-GDL1690-Clean-Adopted-by-ATWG-10-29-20.pdf), adopted October 29, 2020, adds a monitoring duty. Section 4.C(7) requires an insurer using an external SIU to describe "how they will monitor and/or gauge the external / third party's compliance with insurer antifraud mandates." That requirement decides which vendors are viable in a delegated structure. A scored output does not satisfy it. A complete investigative record does.

## The SLA floor is a regulation, not a negotiation

The service levels in a claims administration agreement are negotiated. The service levels underneath them are not. [Model #900](https://content.naic.org/sites/default/files/model-law-900.pdf) section 2.C defines an insurer to include third party administrators, which lands unfair claims settlement standards on the administrator directly. [Model Regulation #902](https://www.propertyinsurancecoveragelaw.com/wp-content/uploads/2017/08/MDL-902-1.pdf) sets the clock:

- Acknowledge receipt of a claim within fifteen days.
- Accept or deny within twenty-one days of properly executed proofs of loss.
- If more time is needed, say so with reasons within those twenty-one days, then send a letter every forty-five days while the investigation remains incomplete.

Manual SIU investigation takes 14+ days per case, so a file routinely crosses the twenty-one day mark and drops into the 45-day letter cadence. An investigation that closes in minutes rather than weeks resolves inside the window and never generates the first letter.

## One workflow across N books means one defect across N books

Shared services concentrates the upside and the exposure in the same place. Model #900 section 3 makes an act an improper claims practice when it is "committed with such frequency to indicate a general business practice." A single workflow running across every book is the fastest available way to establish frequency.

A shared operation runs one intake rule, one referral threshold and one investigative workflow across every book it administers. A defect in any of the three is replicated N times before anyone notices.

Turn it around and the same mechanic runs in your favor. A workflow that is right is right N times. Manual SIU teams fully investigate roughly 25% of flagged claims. Moving that book by book to 100% is the concentration you want: one control, applied and evidenced identically across every entity you administer.

The table below puts five claims structures against the same four questions. The last column is the thing that fails first in each.

| Claims structure | Who owns the SIU obligation | Where investigation cost lands | Who approves the spend | What breaks first |
| --- | --- | --- | --- | --- |
| In-house carrier SIU, single book | The carrier's SIU, directly | Mostly A&O, absorbed | SIU director inside an annual budget | Investigator capacity: 174 accepted referrals per investigator per year, headcount growing 0.5% |
| Carrier SIU with contracted investigative vendors | Still the carrier; the vendor is a contracted entity | Claim-specific vendor invoices to DCC | Adjuster or SIU lead, per case | Per-case approval friction becomes a de facto ceiling on investigations |
| TPA administering one carrier's book | The carrier retains it; the TPA performs it and is an insurer under Model #900 | Split: the administration fee is A&O, claim-specific investigation is DCC | The carrier client, under the master services agreement | Ambiguity over what is inside the per-claim fee and what is billable |
| TPA administering several carriers' books | Each carrier separately; the TPA carries unfair-claims exposure across all of them | Split, and split per client under that client's terms | Each carrier client, under a different agreement | One workflow, N contracts: a defect becomes a general business practice under Model #900 section 3.B |
| Carrier-group shared services across multiple books | The group, potentially under one antifraud plan per Guideline #1690 section 4.B | Internal allocation between books; A&O by default unless structured otherwise | Group claims leadership plus each book's P&L owner | Books with low fraud incidence refuse to subsidize books with high incidence |

## Data segregation is architectural, not contractual

A confidentiality clause is a promise. Tenant isolation is a control. Under the NAIC Insurance Data Security Model Law, a registered administrator is itself a Licensee, and any vendor it gives access to nonpublic information is a Third-Party Service Provider.

[Model #668](https://content.naic.org/sites/default/files/model-law-668.pdf) section 4.F requires a licensee to "exercise due diligence in selecting its Third-Party Service Provider." Per the [NAIC Cybersecurity Working Group adoption map dated April 1, 2026](https://content.naic.org/sites/default/files/cmte-h-cybersecurity-wg-state-adoption-map-model-668.pdf), 28 jurisdictions have adopted the model. An administrator serving client carriers across them has to answer the segregation question once, in the architecture.

Do not collapse two things. Contributory data sharing is a consented, opt-in industry utility, covered in [what cross-carrier fraud data networks can and cannot do](/blog/cross-carrier-fraud-data-networks-limits/). Incidental commingling is one client's claim data influencing another client's outcome without a consent model behind it. A TPA's master services agreements typically prohibit that outright.

The controls that answer it are tenant isolation, per-client key separation and no cross-tenant model training. Hesper holds SOC 2 Type I and does not train on customer data. The review a carrier client will run on you is walked through in [the procurement and AI vendor risk playbook](/blog/procurement-vendor-risk-ai-claims-investigation/).

## The rollout sequence: what book one has to prove before book two starts

Sequence the rollout by contract and consent homogeneity, not by claim volume. Book one should be the book where standing authority to investigate flagged claims already exists, where the agreement does not require client sign-off on a new subprocessor, and where the antifraud plan amendment is a form rather than a negotiation.

The plan amendment is the item most often discovered late. Guideline #1690 section 3.E requires the insurer to amend and submit its plan after a material change, which section 2.D defines to include any change in the "methods, staffing or outsourcing" used to investigate fraud. That is a per-book filing event.

Contract length decides the order more than fraud incidence does. A book on a twelve-month agreement can adopt a new per-claim charge at the next renewal. A run-off book administered on a life-of-claim basis needs a repricing negotiation. Sequence renewals first, multi-year agreements second, run-off last.

The buying center is also duplicated once per client, so one business case has to survive N separate client-side reviews. The seat-by-seat map of one review is in [the carrier buying center map](/blog/buying-center-mapping-ai-investigation/).

Four numbers should be fixed on book one before book two opens, each measured under a definition you have written down.

1. Flagged-claim coverage, per book. The variance between books is the finding worth reporting.
2. Cycle time from referral to closed investigative report, measured against the 21-day and 45-day marks in Model Regulation #902.
3. Fully loaded cost per investigated case, split into DCC and A&O. If you cannot split it, you cannot charge it back.
4. Referral acceptance ratio, tracked separately for adjuster referrals and automated referrals.

| SIU referral acceptance ratio by source (Coalition Against Insurance Fraud and Aon, 2024) | Value | Share |
| --- | --- | --- |
| Adjuster referrals accepted | 70% | 70% |
| Automated-tool referrals accepted | 36% | 36% |

That last number is the one to put in front of a skeptical client. The Coalition and Aon found adjuster referrals were accepted 70% of the time in 2024, and referrals from automated fraud detection tools only 36% of the time. More detection alone does not produce more resolved files. For adoption at each site, see [the claims operations manager transition guide](/blog/claims-operations-manager-ai-transition/).

## What shared is actually worth: the scale number the industry already published

The Coalition and Aon found national carriers report expense per investigation roughly 50% below regional carriers. The open question is how to capture that gap without the national carrier's headcount, because headcount is the one input that is not growing.

**Figure 3 (Capacity · CAIF and Aon 2024) - Referrals per investigator went up as investigators per premium dollar fell.** The supply side of the industry's own benchmark: capacity is the constraint, and it does not resolve by hiring at 0.5% a year.

**Per investigator, a year**

- Accepted referrals, previous study: 162
- Accepted referrals, 2024 study: **174**

**Supply side**

- **+0.5%** - SIU headcount growth, 2023 to 2024, against 1.4% in the prior study
- **0.67** - SIU FTEs per $100 million of gross written premium, down from 0.9

**~50%** - Lower expense per investigation at national carriers than at regional ones. Scale is already the industry's own explanation for unit cost, and a shared operation is the structure built to capture it.

Source: Coalition Against Insurance Fraud and Aon, 2024 Insurer SIU Benchmarking Study, September 2024 (sixth iteration, 35 participating P&C carriers; prior figures from the previous study).

The same study puts the supply side in numbers. SIU headcount grew 0.5% from 2023 to 2024, against 1.4% in the prior study. Staffing runs at 0.67 SIU full-time equivalents per $100 million of gross written premium, down from 0.9 two years earlier. Each investigator carries 174 accepted referrals a year, up from 162. That is a capacity problem, and it does not resolve by hiring at 0.5% a year.

| Where accepted SIU referrals come from (Coalition Against Insurance Fraud and Aon, 2024) | Value | Share |
| --- | --- | --- |
| Adjusters | 60% | 60% |
| Automated fraud detection tools | 29% | 29% |
| All other sources | 10% | 10% |

Sixty percent of accepted referrals still come from adjusters and twenty-nine percent from automated detection tools. An operation that has bought detection and not investigation has spent money upstream of its actual constraint.

What closes the gap is throughput per investigator, not headcount. With automated investigation running 15+ phases in parallel on every flagged claim, the investigator's role shifts from execution to decision-making: which findings hold, which need a human interview, which go to referral.

Carrier groups are formalizing this structure. Insurance Journal reported on [April 13, 2026 that Old Republic launched Lodestar as an independent TPA brand](https://www.insurancejournal.com/news/midwest/2026/04/13/865643.htm). Workers compensation is where administered claims and SIU obligations overlap most heavily, which is the line covered in [the workers compensation use case](/use-cases/workers-comp/).

Make every flagged claim investigable. In a shared-services or TPA structure that is a product: a per-claim service, priced on a rate card, evidenced by a record, allocable to the claim that consumed it, and identical across every book you administer.

## Key takeaways

- Share-of-savings pricing is closed inside GL-1090's scope by section 9.A. Section 9.C expressly permits per-claim compensation.
- Per-case client approval risks California 10 CCR section 2698.33(c)(2), because the cheapest path for the client becomes declining the investigation.
- The same fraud investigator is allocated expense if working in defense of a claim and unallocated if working in the capacity of an adjuster. That fork decides who pays.
- Model #900 section 2.C defines insurer to include third party administrators, and one shared workflow across N books replicates any defect N times.
- National carriers report expense per investigation roughly 50% below regional carriers. A shared operation can capture that scale at 100% flagged-claim coverage rather than ~25%.

## Frequently asked questions

### Is AI claims investigation spend ALAE or ULAE?

It depends on how the work is scoped. Under the NAIC definitions effective January 1, 1998, fraud investigators working in defense of a claim fall in allocated expense, now reported as defense and cost containment. The identical role working in the capacity of an adjuster falls in unallocated, the administrator's own overhead.

### How do you charge multiple carrier clients for a shared AI investigation platform?

Per flagged claim, with standing authority, booked as claim-specific expense. GL-1090 section 9.A rules out share-of-savings pricing, while section 9.C permits compensation based on the number of claims paid or processed. California 10 CCR section 2698.33(c)(2) is a problem for any design where a client approves each investigation and can decline on cost.

### Who owns the claim file when a TPA administers claims for a carrier?

The payor. GL-1090 section 5.A requires the administrator to maintain and make available to the payor complete books and records, retained not less than five years. The evidence, reasoning and timestamps an AI investigation layer produces are part of that file: portable at termination and producible to a regulator.

### Does adding an AI investigation vendor require amending the antifraud plan?

Usually yes, and it is easy to miss. NAIC Antifraud Plan Guideline #1690 treats a change in the methods, staffing or outsourcing used to investigate fraud as material, and section 3.E requires the insurer to amend and submit the plan. Build the filing into the rollout plan for each book.
