---
title: "One in four insurers recovers nothing. Missed subrogation opportunities are lost at intake, not at closure"
description: "One in four US property-liability insurers records no salvage or subrogation recovery. Missed subrogation opportunities are an evidence-timing problem created at intake, not a detection problem at closure."
date: "2026-09-22"
lastModified: "2026-09-22"
author: "Pankaj Dhariwal"
tags: ["Guides"]
canonical: "https://gethesperai.com/blog/subrogation-recovery-ai/"
---

# One in four insurers recovers nothing. Missed subrogation opportunities are lost at intake, not at closure

> **TL;DR** Roughly one in four US property-liability insurers records no salvage or subrogation recovery in a given year, and physical damage subrogation recovers about ten times what liability subrogation does. The gap is evidence, not intent. An arbitration panel decides on the documents attached at filing, and those documents entered the claim file at FNOL, coverage and investigation, upstream of any recovery queue.
>
> - 1 in 4 US insurers record no subrogation recovery at all
> - Liability subro recovers 2% of claims paid; physical damage 20%
> - Screen at intake, because panels see only what you attached

- **1 in 4** - US property-liability insurers that record no salvage or subrogation recovery (Bisco & Fier, Journal of Insurance Regulation (NAIC), 2023)
- **2.3M** - Subrogation demands filed through Arbitration Forums in 2025 (Arbitration Forums, alongside 1.1M arbitration disputes)
- **2% vs 20%** - Net claims paid recovered on personal auto liability vs physical damage (Bisco & Fier, Journal of Insurance Regulation (NAIC), 2023)
- **28%** - Personal auto liability recovery collected in year one after the loss (Bisco & Fier, Journal of Insurance Regulation (NAIC), 2023)

Roughly one in four US property-liability insurers records no salvage or subrogation recovery in a given year. The finding is from Jill Bisco and Stephen Fier, writing in the NAIC's [Journal of Insurance Regulation](https://content.naic.org/sites/default/files/cipr-jir-2023-2.pdf), built on 26 years of NAIC annual statement data. It is not a statement about small carriers or odd books. It is a statement about operating practice.

Missed subrogation opportunities get described as a detection problem: the flag nobody raised. That description points the fix at the wrong stage. A recovery is not lost when a file closes without a flag. It is lost months earlier, when the police report was never pulled, the recorded statement was never transcribed, the endorsement schedule was never read, and the failed part went back to the shop. By the time a claim reaches the recovery queue the identification question is easy and the evidence question is already settled.

What follows walks that argument through the data: why physical damage subrogation recovers ten times what liability subrogation does, why the spread between carriers is 37-fold and therefore not a book-mix accident, how the recovery clock runs from the loss date across six states, and what an intercompany arbitration panel is actually permitted to look at. Recovery is stage 8 of the eight stages mapped in the [complete guide to claims automation](/blog/claims-automation-pillar/), and it is the one stage where automation returns cash instead of avoiding cost.

## Recovery is two businesses wearing one name

Subrogation splits into two businesses with recovery rates an order of magnitude apart. In 2021 US insurers recovered about 20% of net claims paid on auto physical damage. Across the 1996-2021 sample the same ratio averaged roughly 2% on personal auto liability and 1.1% on commercial auto liability. The difference is not effort or intent. It is how much evidence each demand has to carry.

Those ratios come from the Bisco and Fier study, which reconstructs salvage and subrogation from NAIC annual statement filings between 1996 and 2021. Across the three auto lines combined, insurers recovered nearly $51.6 billion in 2021. The authors put the physical damage result plainly: for every five dollars paid in claims, one dollar was recouped.

The trend lines separate further. Inflation-adjusted, auto physical damage recovery grew from $12.79 billion in 1996 to $31.2 billion in 2021, up 144%. Commercial auto liability recovery over the same 25 years moved from just over $1.4 billion to $1.6 billion, up 13%. Personal auto liability rose from under $11 billion to about $19 billion by 2020 and slipped in 2021. One business compounded. The other stood still.

| Line of business | Recovered as a share of net claims paid | Collected in year one | Recovery trend, 1996-2021 |
| --- | --- | --- | --- |
| Auto physical damage | ~20% (2021), up from ~11% (1996) | Over 55% | $12.79B to $31.2B, +144% |
| Personal auto liability | ~2% (sample average) | ~28% | Under $11B to $19B (2020), slight 2021 decline |
| Commercial auto liability | ~1.1% (sample average) | ~35% | $1.4B to $1.6B, +13% |

Every cell is from Bisco and Fier, inflation-adjusted, drawn from NAIC annual statement data. Read the table as a statement about proof rather than about motivation. A physical damage demand is close to mechanical: an adverse vehicle, a police report, a repair estimate, a photograph of the impact. Those facts are in the file because the claim cannot be adjusted without them. A liability demand needs a reconstruction of who caused the loss, assembled from statements, scene documentation, maintenance records, contractor invoices and expert opinion. None of that is a byproduct of paying the claim. Somebody has to decide to collect it while the evidence still exists.

NAIC annual statement practice assumes total recovery lands after two years on physical damage claims and after nine on liability claims. Nine years is a long time for an operating practice to drift unnoticed, and liability recovery is where that drift sits.

## The 37x spread between carriers is an operating choice

Among insurers that record any subrogation recovery, the top quintile recovers 20.61% of net claims paid and the bottom quintile 0.54%, a 37-fold gap. Across the full sample, including firms with no recovery at all, the bottom quintile recovers 0.00%. A range that wide is not explained by what each firm chooses to write.

| Recovery as a share of net claims paid, by quintile (firms with any subrogation recovery) | Value | Share |
| --- | --- | --- |
| Quintile 1 (lowest) | 0.54% | 2.6% |
| Quintile 2 | 1.77% | 8.6% |
| Quintile 3 | 3.14% | 15.2% |
| Quintile 4 | 4.86% | 23.6% |
| Quintile 5 (highest) | 20.61% | 100% |

Bisco and Fier report salvage and subrogation at 4.5% of net claims paid across the full sample, 6.2% among firms with any positive recovery, and a maximum near 90% of net claims paid at the top of the distribution. Book mix moves that number. It does not move it 37-fold between two groups of US property-liability insurers writing broadly comparable business over 26 years. What separates the quintiles is whether recovery is a standing step in the process or something an adjuster gets to.

There is a balance-sheet tell for finance. Under Statutory Accounting Principles an insurer may accrue for anticipated salvage and subrogation; declining to accrue means reporting the claim liability at its gross amount. Research cited in the same study (Ames, Graden and Sankara, 2019) finds mutual insurers, firms with a weaker AM Best financial strength rating and firms with greater premium growth are less likely to accrue, while publicly traded insurers are more likely to. An accrual decision is a stated expectation about pursuit, made in the carrier's own filings.

> Recovery rate is the one claims metric that reads like a market outcome and behaves like an operating choice. The distribution is not a bell curve: a quarter of the industry at zero, a top quintile at a fifth of claims paid. Nothing about loss geography produces that shape.
>
> - Hesper AI product research

Missed recovery is a leakage category, and it is the one that behaves differently from the rest. Every other form of [claims leakage](/blog/insurance-claims-leakage-reduce-losses/) is money that should not have been paid. Missed subrogation is money the carrier was entitled to collect and did not. Closing it returns cash rather than avoiding cost, which is why it is the line a CFO engages with before any cycle-time argument.

## The clock runs from the loss, not from the closure

A subrogating insurer stands in its insured's shoes, so the limitation period is inherited rather than reset and it runs from the date of loss. Across six states the window on comparable property damage runs from two years to six. One state changed its window in 2024. One state runs two clocks on the same loss.

| State | Statute | Window | What makes it a trap |
| --- | --- | --- | --- |
| Louisiana | La. Civ. Code art. 3493.1 | 2 years | Was 1 year until July 1, 2024; both windows live depending on loss date |
| Florida | Fla. Stat. § 95.11(5)(a) and (3)(g) | 2 years negligence / 4 years personal property | One loss, two clocks |
| California | Cal. Civ. Proc. Code § 338(b), (c) | 3 years | Property is 3 years; personal injury is 2 under § 335.1 |
| New York | N.Y. C.P.L.R. § 214(4), (5) | 3 years | Injury to property and personal injury both 3 years |
| Missouri | Mo. Rev. Stat. § 516.120(2)-(4) | 5 years | Covers liability created by statute and injury to goods or chattels |
| Maine | 14 M.R.S. § 752 | 6 years | General 6-year civil limitation running from accrual |

Louisiana moved delictual actions from one year to two effective July 1, 2024 under [Acts 2024, No. 423](https://www.legis.la.gov/Legis/Law.aspx?d=1386443), applying prospectively, so both windows remain live depending on when the loss occurred. [Florida](https://www.leg.state.fl.us/statutes/index.cfm?App_mode=Display_Statute&URL=0000-0099/0095/Sections/0095.11.html) allows two years for an action founded on negligence and four for taking, detaining or injuring personal property. [California](https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=CCP&sectionNum=338) and [New York](https://www.nysenate.gov/legislation/laws/CVP/214) allow three years on property damage, [Missouri](https://revisor.mo.gov/main/OneSection.aspx?section=516.120) five, and [Maine](https://legislature.maine.gov/statutes/14/title14sec752.html) six.

> **Two clocks, one loss**
>
> A Florida property loss can sit under a two-year negligence clock and a four-year personal property clock at the same time, depending on how the action is framed. A Louisiana loss sits under a one-year or a two-year clock depending on whether it arose before or after July 1, 2024. Static deadline tables decay fastest in the states a team handles least often. This is not legal advice.

Arbitration Forums advises members to [file at least 120 days before the statute of limitations](https://home.arbfile.org/about-us/frequently-asked-questions) where possible, so a jurisdictional challenge can be resolved before the deadline rather than after it. That guidance moves the operational deadline forward by four months from a date that was already fixed at the loss.

The collection curve tightens the window further. Over 55% of auto physical damage recovery lands in the first year after the loss. For liability, roughly 35% on commercial auto and 28% on personal auto arrives in year one, and about 76% of cumulative liability recovery is in by the end of year three. What has not been recovered by year three is largely never recovered. A file first reviewed for recovery potential at closure is competing against a curve that has already flattened.

This is why subrogation screening belongs at intake and triage rather than closure. The [end-to-end claims process](/blog/claims-process-end-to-end-fnol-to-settlement/) already has a decision point at first notice, when the adverse party, the scene and the physical evidence are all still reachable. Nothing about that decision gets easier by waiting.

## What an arbitration panel is allowed to look at

Intercompany arbitration is decided on the documents attached to the filing and nothing else. Arbitration Forums Rule 4-1 states that panels may not render default judgments and that decisions must be based on the evidence submitted. Rule 2-1 requires evidence to be attached when the filing is submitted, and neither amendments nor reschedules are allowed.

[Arbitration Forums](https://home.arbfile.org/about-us) is a not-for-profit founded by the insurance industry in 1943, and it is where most US intercompany recovery disputes are resolved. In 2025 its members filed 1.1 million arbitration disputes and 2.3 million subrogation demands, collectively worth almost $27 billion, across more than 5,100 member companies and a neutral panel of over 6,700 claims professionals.

The [rules](https://home.arbfile.org/training/reference-guides/arbitration-forums,-inc-rules), effective October 21, 2024, are unusually explicit about what does not count. Rule 3-2: a case will be heard even if the responding company fails to answer. Rule 2-5: arguments not properly raised will not be considered, a list that names causation and partial exclusions. Rule 3-7(c): an insured or witness may appear only if their written or transcribed recorded statement is listed as evidence, and even then "witness testimony is not evidential." Rule 3-6 says formal rules of evidence do not apply, which sounds permissive and is not. It means the documents carry the whole case, because nothing else is admitted to help them.

Stack those rules together and a specific failure mode appears. A case is heard whether or not the other carrier answers, so a thin file does not lose an argument. It loses to silence. Nobody talks the panel through what the adjuster remembers. The burden is preponderance of the evidence, which Arbitration Forums describes as showing that a factual explanation is more likely true than the alternatives rather than proving it indisputably. That is a low bar, and it still has to be cleared in one submission, at filing.

Now look at what that one submission has to contain on a liability file: the police or fire report, the recorded statement transcribed rather than summarized, the scene photographs, the repair or contractor invoice, the maintenance record, the failed part itself. Every one of those documents entered the claim file at first notice, during the coverage review, or during the investigation. None of them are produced by the recovery stage. A recovery tool that starts when a file reaches the recovery queue can tell you an opportunity exists; it cannot go back and take the photograph. Whoever read those documents when they arrived is the only party who can assemble them now, which is why the same sourced evidence file that supported the coverage position and the fraud finding is the file that supports the demand.

*Figure: An arbitration panel cannot enter a default judgment. It reads what you attached, and you attached it months earlier or you did not.*

The rules also decide what a mistake costs. Compulsory arbitration covers total company-paid damages up to $100,000 in the Automobile, Medical Payment, Property and Uninsured Motorists forums, and a Contribution Sought Amount up to $250,000 in Special Arbitration. Appeals carry a substantial non-refundable fee and, under Rule 2-12, allow no additional documentation or evidence. Under Rule 6-2, physical evidence such as an allegedly defective part is returned only on request with prepaid postage, and everything else is destroyed after the hearing. There is no second submission anywhere in the process.

> **Identification and proof are different problems**
>
> Verisk pushes adverse-carrier identification close to FNOL: its [ClaimSearch](https://www.verisk.com/products/claimsearch/) timeline dates the Claims Coverage Identifier, launched in 2019 as the Policy Insights Report, to the job of matching a claim against the other carrier's policy. That is the right instinct at the right stage, and it is the industry's own data utility agreeing that end-of-file identification is too late. What identification establishes is who the other carrier is. What Rule 4-1 requires is evidence that they are liable. Only the second one wins a demand.

Several vendors sell recovery identification and they are good at it. CCC's [subrogation page](https://www.cccis.com/insurance-carriers/subrogation) describes detection models trained by subrogation professionals that identify and prioritize opportunities across auto, property and workers' compensation, and CCC reported in its second-quarter 2026 results that a top-five insurer adopted its AI subrogation. Shift's [subrogation page](https://www.shift-technology.com/en-gb/solutions/subrogation) claims a 2x increase in recovery rates and 33% faster time to recovery, and describes next best actions to secure time-sensitive evidence. The evidence is time-sensitive, which is the point: each of these products reads a claim file it did not build, so all it can do is prompt a person to go and get what should already be in it.

## The deductible is somebody's money

In California the pursue-or-waive decision is not a discretionary internal workflow step. The Fair Claims Settlement Practices Regulations require every insurer to notify the first-party claimant in writing whether it intends to pursue subrogation, and to include that claimant's deductible in every subrogation demand it makes.

Cal. Code Regs. tit. 10, § 2695.7(p) and (q), published by the [California Department of Insurance](https://www.insurance.ca.gov/01-consumers/130-laws-regs-hearings/05-CCR/fair-claims-regs.cfm), set both obligations. Subsection (p) carves out four situations where no notification is owed: the deductible was waived, the coverage carries no deductible, the loss did not exceed the deductible, or there is no legal basis for subrogation. Where an insurer elects not to pursue, the notice must say that any recovery is the claimant's own responsibility. Subsection (q) also requires insurers to share recoveries with the claimant proportionately unless the claimant has already recovered the whole deductible, and permits legal or collection expenses to be netted against the deductible recovery only where an outside attorney or collection agency was retained, and then only for a pro rata share of allocated loss adjustment expense. Requirements vary by state, so read California as one example rather than a national rule.

The operational consequence is that a California file which closes without anyone reaching the pursue-or-waive decision is not only leakage. It is a required decision that was never made, and an insured who never got the letter. Arbitration Forums carries the same money through its own rules: Rule 1-3(d) excludes the deductible from total company-paid damages when testing the compulsory threshold, and Rule 5-1 requires award payments to include deductible interest where applicable.

Some recoveries are gone before the loss happens. Waiver-of-subrogation endorsements remove the right contractually: WC 00 03 13 on the workers' compensation and employers' liability form, and ISO CA 04 44 10 13 on the Business Auto, Motor Carrier and Auto Dealers coverage forms. A screening process that reads the claim but not the endorsement schedule produces demands that cannot be sent, and burns the recovery team's credibility on the ones that can.

## Who owns recovery at a TPA

At a TPA or MGA the recovery decision is made on one entity's file and audited by another entity's auditor. Recovery dollars flow to the client program while the screening labor sits with the TPA, so the incentive to review every file is weakest exactly where the file volume is highest. That is structural, not a diligence failure.

Two controls change its shape. The first is to screen for subrogation and salvage as a standing step at intake and triage rather than a discretionary step at closure, so no file depends on one adjuster remembering. The second is to document waive decisions with the same evidence as pursue decisions, because in a closed file a reasoned waiver and an oversight look identical: both are an absence. A client auditor cannot tell them apart without a record of what was checked and why the file closed without a demand. At a TPA that ambiguity costs a renewal conversation, not a fee credit.

Workers' compensation makes the ownership question concrete. Comp subrogation against the responsible party routes through Arbitration Forums' Special Arbitration forum, capped at $250,000 in Contribution Sought Amount, alongside contribution, concurrent coverage and construction defect disputes. CCC's outbound subrogation similarly spans auto, property and workers' compensation. The recovery route exists and is well defined. What decides whether a comp file uses it is whether anyone built the liability case while the employer, the third-party premises owner and the equipment maintenance records were all still in front of them.

## Screening every file instead of sampling it

The fix for missed subrogation is not a better recovery queue. It is moving identification to intake and carrying the liability evidence forward, so that when a file reaches stage 8 the demand is assembled rather than reconstructed from whatever survived. Recovery stops being a discovery exercise and becomes a filing exercise.

That is the design Hesper runs. Every claim, first notice to final recovery: the same agents that extract the FNOL, read the policy and endorsement schedule for the coverage position, and run the investigation phases on a suspicious file carry that evidence into stage 8. Every file is screened for subrogation and salvage, not the subset an adjuster flagged. The parallel is measurable elsewhere in the platform, where carriers fully investigate roughly 25% of flagged claims manually and the model is built to take that to 100% (Hesper internal benchmarks).

Evidence behind every decision is the operative part. Assembling a liability file by hand runs to 14+ days per case as a Hesper internal benchmark, which is why the work gets rationed, and rationing is what produces a one-in-four industry. The agents work in hours, not weeks, which changes how many files can carry a built evidence trail rather than how fast any single demand goes out. The pursue-or-waive decision stays with the recovery team; what changes is that they review an evidence-backed file instead of building one.

Be fair about the field. Verisk has already moved adverse-carrier identification to FNOL. CCC and Shift both rank opportunities and package demands. What none of their public materials describe is one sourced evidence file that serves the coverage position, the fraud finding and the subrogation demand at once, because each product starts at its own stage. For the mechanics of how recovery agents screen a file and draft a demand, see the [subrogation and recovery](/platform/subrogation-recovery/) page; the lifecycle around it sits on the [product](/product/) page.

## Key takeaways

- Roughly one in four US property-liability insurers records no salvage or subrogation recovery in a given year, according to Bisco and Fier in the NAIC's Journal of Insurance Regulation.
- Physical damage subrogation recovered about 20% of net claims paid in 2021 while personal auto liability averaged about 2% across the 1996-2021 sample, and the difference is how much evidence each demand has to carry.
- The 37-fold spread between the top and bottom recovery quintiles, 0.54% against 20.61% of net claims paid, is too wide to be book mix and is better read as an operating choice.
- Arbitration Forums Rule 4-1 requires panels to decide on the evidence submitted and Rule 2-1 requires that evidence to be attached at filing, so a recovery flag without documents is not a recovery.
- The documents that win a liability demand enter the file at FNOL, coverage and investigation, which is why subrogation screening belongs at intake rather than at closure.

## Frequently asked questions

### What are missed subrogation opportunities?

A missed subrogation opportunity is a paid claim where another party was legally responsible for the loss but the insurer never pursued recovery. It happens one of three ways: nobody identified the liable third party while the file was open, the liability evidence was never assembled well enough to support a demand, or the limitation period lapsed first. The scale shows in the gap between lines. Bisco and Fier, in the NAIC's Journal of Insurance Regulation, found US insurers recovered about 20% of net claims paid on auto physical damage in 2021, while across their 1996-2021 sample the ratio averaged only about 2% on personal auto liability and 1.1% on commercial auto liability. Roughly one in four property-liability insurers records no recovery at all in a given year.

### How much do insurers actually recover through subrogation?

Across all lines, salvage and subrogation averaged 4.5% of net claims paid in the Bisco and Fier sample, rising to 6.2% among firms with any positive recovery, with a maximum near 90%. On the auto lines specifically, insurers recovered nearly $51.6 billion in 2021 across auto physical damage, commercial auto liability and personal auto liability combined. The spread between carriers is wide. Among firms with any recovery, the top quintile recovered 20.61% of net claims paid and the bottom quintile 0.54%, a 37-fold difference. That range is too large to be explained by book mix alone, which is why recovery rate is better read as an operating choice than a market outcome.

### What is the statute of limitations on a subrogation claim?

The subrogating insurer stands in its insured's shoes, so it inherits the insured's limitation period, which runs from the loss and varies by state and cause of action. Louisiana moved delictual actions from one year to two effective July 1, 2024 under La. Civ. Code art. 3493.1. Florida allows two years for negligence but four for taking, detaining or injuring personal property. California and New York allow three years for property damage. Missouri allows five and Maine six. Arbitration Forums advises filing at least 120 days before the statute of limitations where possible, so jurisdictional challenges can be resolved before the deadline. Confirm the period for the loss state and the cause of action on every file. This is not legal advice.

### Why do claims teams miss subrogation opportunities?

Because identification usually depends on an adjuster noticing a liable third party while working a file, or on rules keyed to a handful of coded fields, and the evidence that proves liability does not live in those fields. It lives in police and fire reports, repair invoices, contractor agreements, photographs and recorded statements. Timing compounds it. Bisco and Fier found only about 28% of personal auto liability recovery is collected in the first year after the loss and roughly 76% by the end of year three, so files reviewed late are competing with a decaying curve. Some recoveries are also waived before the loss happens, through endorsements such as WC 00 03 13 or ISO CA 04 44 10 13.

### How does intercompany arbitration work for subrogation?

Arbitration Forums, a not-for-profit founded by the industry in 1943, runs the main intercompany forums. Its members filed 1.1 million arbitration disputes and 2.3 million subrogation demands in 2025, collectively worth almost $27 billion. Compulsory arbitration applies up to $100,000 in the Automobile, Medical Payment, Property and Uninsured Motorists forums and $250,000 in Special Arbitration, which covers workers' compensation subrogation and contribution disputes. The decisive rules are procedural: evidence must be attached when the filing is submitted, no amendments are allowed, and under Rule 4-1 panels may not render default judgments because decisions must be based on the evidence submitted. A case is heard even if the other carrier never answers, so a thin file loses to silence.

### Can AI identify subrogation opportunities in claim files?

Yes, and several vendors sell it. CCC describes detection models trained by subrogation professionals that identify and prioritize opportunities across auto, property and workers' compensation. Shift's subrogation page claims a 2x increase in recovery rates and 33% faster time to recovery. Verisk's ClaimSearch pushes adverse-carrier identification close to FNOL through its Claims Coverage Identifier, which launched in 2019 as the Policy Insights Report. The harder problem is what happens after identification. An arbitration panel decides on documents attached at filing, so the value of an identified opportunity depends on whether the liability evidence was captured while the file was open. Identification and proof are different problems, and only the second one wins a demand.