Claims leakage calculator
Claims leakage is the gap between what an insurer paid and what it should have paid had every process worked. EY puts it at roughly 7% to 14% of a carrier’s total spend. There is no agreed measurement standard, and published estimates run from 2% to 30%, so this tool returns a range built from named benchmarks rather than a single number.
Free, ungated, no email and no form. Every coefficient below carries its publisher, its year and a link. Where a number could not be sourced, the line is left blank and the reason is printed instead of a guess.
Estimate leakage in your book
Five numbers a claims leader can answer cold. Every coefficient below is a published benchmark with its source on screen - nothing here is our own estimate, and nothing you type is sent to us.
Presets are editable starting points, not benchmarks. Each implies an average paid claim between $8,000 and $15,000, which is the order of magnitude published US P&C severity runs at - the Insurance Information Institute puts the average homeowners claim at $17,059 across 2019 to 2023. Replace every figure with your own.
Estimated annual leakage $22.9M to $44.8M, 7.2% to 14% of claims paid.
On $320M paid across 40,000 claims
How: non-litigated paid ($304M) × 7%, plus litigated paid ($16M) × 10%, for the low end. Total paid × 14% for the high end. The 7% and 14% are EY’s published band; the 10% floor on litigated files is EY’s own closed-file audit result.
Read this as an order of magnitude, not a finding. Leakage varies by book, by line and by jurisdiction, and there is no agreed standard for measuring it - IRMI defines the term and publishes no percentage for it.The only way to know your real number is a closed-file audit of your own claims.
The spread between credible estimates is enormous
Published leakage percentages span an order of magnitude. Only the middle band drives the estimate above; the other two are shown so you can see how wide the disagreement is.
Four buckets, each sized on its own benchmark
These are independent estimates of overlapping things - a fraudulent claim that was also overvalued sits in two of them. Do not add them up, and do not expect them to reconcile to the headline range. A single bucket can come out larger than the whole range above, because each one rests on a different published base: the fraud figure is a share of incurred losses and loss adjustment expense, the leakage range is a share of total claims spend.
Reserves and settlements set against an incomplete file: coverage not verified against the policy as written, damages accepted without the underlying medical or repair evidence, and prior claims history never pulled.
Formulalitigated paid × 10% to 14% leakage × 85%. EY’s closed-file audit put at least 85% of the leakage it found in coverage determination, litigation prevention, and evaluation and resolution - so this is a floor on the litigated share only.
A third party is at fault and nobody spots it before the file closes. Roughly one in four property-liability insurers records no salvage or subrogation recovery at all.
Formulatotal claims paid × (6.2% − 4.5%) = 1.7% of paid. The gap between the mean recovery across all property-liability insurers, including the quarter that recover nothing, and the mean across only those that recover something. It is a difference between two samples rather than a gap measured inside one carrier, so read it as an upper bound. If you already recover above 6.2% of paid, this line is zero for you.
Claims that are flagged but never worked to a conclusion, and claims that are never flagged at all. Fraud that is detected but not resolved still gets paid.
Formulatotal claims paid × 10%. This is the industry’s conventional multiplier and it rests on Insurance Information Institute interviews with claims adjusters in the late 1980s. Read the provenance note before using it.
Every extra day a file stays open is another day of adjuster time, vendor time and reserve movement. Cycle time is the one leakage driver a claims operation can see on its own dashboard.
Formulatotal claims paid × 15.6% = implied loss adjustment expense; ÷ claims volume ÷ average days = handling cost per claim per day; × days above the 40.7-day benchmark × claims volume.
4.3 days above the published benchmark
How: $320M paid × 15.6% (the 2025 US P&C industry ratio of loss adjustment expense to losses incurred) = $49.9M of implied LAE; ÷ 40,000 claims ÷ 45 days = $27.71 per claim per day; × 4.3 excess days × 40,000 claims.
40.7 days is a US homeowners/property benchmark. There is no comparable published cross-line cycle-time figure, so on a casualty, commercial or workers’ comp book treat it as a proxy rather than a target.
What is claims leakage?
IRMI defines claims leakage as “dollars lost through claims management inefficiencies that ultimately result from failures in existing processes (manual and automated)”, or more plainly, the difference between what you did spend and what you should have spent on a claim. It is normally discovered through an audit of closed claim files, which is the only reliable way to measure it.
The important thing about that definition is what it does not contain: a number. IRMI publishes no percentage for how large leakage is, and no regulator publishes one either. Every figure you have seen quoted comes from a consulting practice describing its own file reviews, or from industry convention with no study behind it. That is why this page shows you the spread.
The figure most often quoted in the trade press is 5% to 10% of claims spend. It is a convention rather than a finding, and it sits below the 7% to 14% band EY reports from its own casualty file reviews, which is the band this calculator runs on. Both are averages across books that look nothing like each other, which is the argument for auditing your own closed files rather than adopting anybody’s percentage.
For the definition itself and how leakage relates to loss adjustment expense, indemnity and recovery, see the claims leakage glossary entry. For where the money actually goes and what closes the gap, see how uninvestigated claims drain profitability.
How this calculator works
Five inputs, four outputs, and every coefficient is a published figure. Here is the arithmetic in plain language, followed by every benchmark with its source.
Split your total claims paid into a litigated share and a non-litigated share using the litigation percentage you set.
Low estimate = (non-litigated paid × 7%) + (litigated paid × 10%).
High estimate = total paid × 14%.
The 7% and 14% are the two ends of EY’s published band. The 10% floor on litigated files is EY’s own closed-file audit result on a litigated book, which is how the litigated share ends up carrying the higher published percentage without us inventing a weighting.
One judgement call, stated plainly: EY expresses the band against “carriers’ total spend”, and this calculator applies it to the claims paid figure you enter. Total spend is the larger denominator, so applying the same percentage to indemnity alone is the conservative reading. It is still a reading, and it is the single place where this tool interprets a source rather than quoting it.
At least = litigated paid × 10% × 85%. Upper = litigated paid × 14% × 85%.
EY’s audit of a litigated book found more than 85% of the leakage sat in three areas: coverage determination, litigation prevention, and evaluation and resolution. Because the audit gives a combined figure rather than a split across the three, this is a floor on the litigated portion only, and it is not sized at all on a book with no litigation.
Up to = total claims paid × (6.2% − 4.5%) = 1.7% of paid.
NAIC Schedule P data across 1996 to 2021 puts salvage and subrogation at a mean of 4.5% of net claims paid across the full sample of property-liability insurers, and 6.2% across only those that recover anything at all. Roughly one in four insurers records no recovery in a given year. Both are unweighted means across firm-years, winsorised at the 1st and 99th percentiles, so they are averages across firms rather than a dollar-weighted industry rate, and the medians are much lower at 1.85% and 3.15%. The 1.7-point gap is a difference between two samples, not a gap measured inside one carrier, which is why this line is only ever an upper bound. If your book already recovers above 6.2% of paid, it is zero for you.
For auto liability / bodily injury: total claims paid × 15% to 17%, from the Insurance Research Council’s 2015 study on 2012 data. The IRC measured that share against auto bodily injury payments alone, and an auto liability book also carries property damage liability, so on a whole book read the result as a ceiling rather than an estimate.
For every other line: total claims paid × 10%. This is the industry’s conventional multiplier and it is the weakest number on this page. See the provenance section below before you rely on it.
Implied loss adjustment expense = total claims paid × 15.6%, the 2025 NAIC industry ratio of loss expenses incurred to net losses incurred.
Handling cost per claim per day = implied LAE ÷ claims volume ÷ your average days to closure.
Annual cost of the excess = handling cost per claim per day × (your average days − 40.7) × claims volume, charged only on days above the benchmark.
40.7 days is the J.D. Power 2026 average from claim to final payment for US homeowners claims. There is no comparable published cross-line cycle-time benchmark, so on a casualty or workers’ comp book treat it as a proxy. If your cycle is already at or inside it, the figure is left blank rather than guessed.
Every benchmark, with its source
Each URL below was fetched and checked against the figure quoted.
Claims leakage as a share of carriers’ total spend, from EY’s own claims quality assessments in casualty and litigated claims. Both ends of the headline range come from here.
EY (Ernst & Young LLP), “Property and casualty insurers tackle indemnity in litigated claims”. Source
EY’s denominator is total spend. This calculator applies the band to indemnity paid alone, which is the smaller base, so the dollar result is the conservative reading.
A closed-file leakage study of one US carrier’s litigated book. Nearly two-thirds of the files reviewed had some leakage, and more than 85% of the leakage sat in coverage determination, litigation prevention, and evaluation and resolution. This sets the floor for the litigated share and the weight on the overpayment bucket.
EY (Ernst & Young LLP), case study in the same paper. Source
One carrier, one book. Used as a floor, never as an industry average.
Salvage and subrogation recovery as a share of net claims paid: a mean of 4.5% across the full sample of property-liability insurers, 6.2% across only those with any positive recovery. The 1.7-point gap is the missed-recovery estimate.
NAIC, Journal of Insurance Regulation, “How’s the Recovery? Salvage and Subrogation”, on NAIC Schedule P data 1996-2021. Source
Unweighted means across firm-years, winsorised at the 1st and 99th percentiles. Medians are 1.85% and 3.15%, so this is an average across firms and not a dollar-weighted industry rate.
The share of property-liability insurers recording any salvage or subrogation recovery in a given year. Roughly one in four records none.
NAIC, Journal of Insurance Regulation, same paper. Source
US P&C industry loss adjustment expense for the year ended 31 December 2025: $86,003M of loss expenses incurred against $551,755M of net losses incurred, against 15.3% the year before. Converts your indemnity into an implied handling cost.
NAIC, U.S. Property & Casualty and Title Insurance Industries - 2025 Full Year Results. Source
A statutory net figure. It excludes claims-handling cost carriers book as other underwriting expense, so it understates handling cost rather than overstating it.
Average time from claim to final payment for US homeowners claims, with an average repair cycle of 29.6 days. The cost-of-delay line is only charged on days above this.
J.D. Power 2026 U.S. Property Claims Satisfaction Study, as reported by Claims Journal, 18 March 2026. Source
A property benchmark. There is no comparable published cross-line figure, so on a casualty or workers’ comp book it is a proxy.
The fraud multiplier used for every line except auto bodily injury. Read the provenance section before you use it.
Insurance Information Institute, quoted at page 8 of the Coalition Against Insurance Fraud / Colorado State University Global study. Source
The weakest number in this calculator. It traces to interviews with claims adjusters conducted in the late 1980s.
Casualty fraud as a share of total auto bodily injury claims payments, worth $5.6bn to $7.7bn. Used only when auto liability is the primary line.
Insurance Research Council study, cited at page 9 of the Coalition Against Insurance Fraud study. Source
IRMI defines claims leakage and publishes no figure for how large it is. That absence is the reason this tool shows a range and a spread rather than one number.
IRMI, Glossary of Insurance and Risk Management Terms. Source
The two ends of the published spread, shown as context only. Neither drives any number in the calculator.
The Lab Consulting, which reports 2-4% as the conventional industry answer and 20-30% as what its own audits document. Source
The low end is industry convention with no study named behind it. The high end is one consultancy’s own client findings.
What we left out, and why
- A split of overpayment leakage by cause. EY reports that more than 85% of the leakage it found sat in three areas, but not how that 85% divides between them. So the calculator shows the combined floor and no per-cause breakdown.
- A leakage rate that varies by line of business. No published band does. Only the fraud coefficient changes with line, because auto bodily injury has its own study.
- A cost-of-delay figure for books already inside the benchmark. There is no published basis for costing a cycle that is already at benchmark, so that line is blank.
- Most of the provenance section below. Six widely quoted numbers whose own footnotes do not hold. Five of them drive nothing here. The sixth, the 10% fraud multiplier, is used because there is no better published figure, and it is labelled as weak everywhere it appears.
What the industry’s most-quoted numbers actually rest on
Building this meant chasing every figure back to a primary source. Several of the most repeated numbers in claims do not have one. They are printed here so you can stop quoting them, and so you can see exactly which parts of the calculator above are load-bearing and which are folklore.
“Around 10% of P&C claims involve fraud.”
This traces to the Insurance Information Institute interviewing claims adjusters in the late 1980s and concluding that fraud accounted for about 10 percent of the industry’s incurred losses and loss adjustment expenses. The 2022 Coalition Against Insurance Fraud study adopts it as its own multiplier and adds a footnote: the III “did not specify exactly how they derived at this figure.” The study’s researchers then reverse-engineered the arithmetic themselves. It is a convention the industry has agreed to keep using, roughly forty years old, not a measurement. This calculator uses it because there is nothing better, and labels it every time it appears.
“Missed subrogation costs the industry $15 billion a year.”
The NAIC’s Journal of Insurance Regulation does repeat this, which is why it gets quoted as regulator-grade. Its footnote points to a September 2021 PropertyCasualty360 trade article, not to a study or a filing. The related claim that 15% of files close with a missed subrogation opportunity has the same lineage. This calculator does not use either. It uses the NAIC Schedule P recovery ratios from the same paper instead, because those are filed data.
“Claims leakage is X% of claims paid.”
There is no agreed measurement standard. IRMI defines the term and publishes no percentage. Published estimates run from 2-4%, offered as conventional industry wisdom with no study named, through EY’s 7-14% from its own file assessments, to 20-30% documented by one consultancy in its own engagements. That is an order of magnitude of disagreement about the same quantity. Any tool that gives you one number is hiding this from you.
“The FBI estimates non-health insurance fraud exceeds $40 billion a year”, and “fraud costs the average family $400 to $700 a year.”
The Coalition Against Insurance Fraud’s own reference list cites this as “Federal Bureau of Investigation. (n.d.). Insurance Fraud”, retrieved 15 November 2021 from fbi.gov/stats-services/publications/insurance-fraud. Note the “n.d.”: no date. The page carried no study, no methodology and no year when it existed, and it is no longer served. We do not use it.
“The FBI estimates fraud costs the average family $4,000 to $7,000 over ten years.”
The NAIC’s own insurance fraud topic page (content.naic.org/insurance-topics/insurance-fraud) attributes this estimate to the FBI. The hyperlink on that sentence resolves to a National Insurance Crime Bureau blog post, not to an FBI publication. We do not use it.
“Auto insurers lose $29 billion a year to leakage.”
This is Verisk’s estimate, from its own 2016 Auto Insurance Premium Leakage Survey, reported in March 2017 (Claims Journal, “Misinformation Leads to Auto Premium Leakage of $29B Per Year”, 15 March 2017). Two problems: it is a vendor estimate of its own addressable problem, and, more importantly, it measures premium leakage at underwriting - omitted or misstated rating information - not claims leakage. It is a different quantity and does not belong in a claims model.
One more, for context rather than criticism: Arbitration Forums, the largest inter-carrier recovery venue in the US, reports that in 2025 its members filed 1.1 million arbitration disputes and 2.3 million subrogation demands collectively worth almost $27 billion. That is throughput through one venue, not a measure of what was missed, so it drives nothing here.
Where leakage actually comes from
Overpayment and valuation error
Reserves and settlements set against an incomplete file. Coverage is not verified line by line against the policy as written, so exclusions and limits that would have applied never get raised. Damages are accepted on the strength of a demand package rather than the underlying medical records, repair estimates or wage documentation. Prior claims history is never pulled, so a pre-existing condition or a previously paid loss goes unnoticed. In EY's audit of one litigated book this accounted for more than 85% of the leakage found, and it is the hardest failure to see from the inside, because nothing about an overpaid file looks wrong at the time.
Missed recovery: subrogation and salvage
A third party is at fault and nobody identifies it before the file closes. Liability facts that would support a recovery sit in an adjuster note, a police report or a photograph that no one reads with recovery in mind. Statutes of limitation run. NAIC Schedule P data shows roughly one in four property-liability insurers records no salvage or subrogation recovery at all in a given year, and the carriers that do pursue it recover meaningfully more as a share of claims paid than the industry average.
Fraud that gets paid
Two failure modes, and the second is the larger one. Claims that are never flagged, because the indicators are spread across documents nobody reads together. And claims that are flagged and then never worked to a conclusion, because manual SIU investigation takes 14 or more days per case and one investigator carries 200 or more cases. Detection without resolution still ends in a payment. A referral that expires on a queue is indistinguishable, financially, from a claim that was never suspected.
Handling cost driven by cycle time
Every extra day a file stays open is another day of adjuster attention, vendor invoices, rental or additional living expense, and reserve movement. Cycle time is the one leakage driver a claims operation can already see on its own dashboard, which is why it tends to be the first one attacked. It is also the one most directly tied to the others: files stay open because information is missing, and the same missing information is what produces the valuation and recovery errors above.
What to do about it
Most of what works here is not software, and none of it requires a vendor. In rough order of return per unit of effort:
- Run a closed-file audit and get your own number. Pull a stratified sample of closed files across line, severity and adjuster, and have someone who did not handle them assess what should have been paid. This is the only way to replace the estimate above with a fact. Everything else is guesswork until you have done it.
- Measure leakage as a standing metric, not a project. The reason the industry has no agreed percentage is that almost nobody measures it continuously. Put it on the same reporting cadence as your loss ratio.
- Put a recovery checkpoint before closure, not after. A mandatory subrogation and salvage question at the point of closing catches far more than a retrospective sweep, and the NAIC ratios suggest the gap between average and active recovery is real money.
- Separate coverage verification from damages evaluation. The two failures have different causes and different fixes, and adjusters under caseload pressure collapse them into one judgement.
- Track referral outcomes, not referral volume. A fraud programme that reports referrals made rather than referrals resolved is measuring its own activity, not its effect on paid losses.
- Attack cycle time on information, not on pressure. Files stay open because something is missing. Shortening the cycle by pushing adjusters harder trades cycle-time leakage for valuation leakage.
Where an AI claims platform fits, honestly. Nothing above needs software to start, and a carrier that has not run a closed-file audit should do that first. What software changes is throughput on the work that is already understood: reading every document in a file against the policy rather than a sample, checking every claim for recovery and fraud indicators rather than the ones an adjuster had time to flag, and finishing an investigation in hours rather than the 14 or more days a manual SIU case takes. That matters most in the buckets above where the failure is “nobody had time to look” rather than “somebody judged it wrong”. Hesper AI is an AI claims resolution platform covering every claim from first notice to final recovery, with investigation-grade evidence behind each decision. The two parts nearest to this page are claim file review and subrogation recovery. We are not going to put a number on what that is worth to your book, because the honest answer is that it depends on your audit, not on our model.
Questions about claims leakage
What is claims leakage?
Claims leakage is the difference between what an insurer paid on a claim and what it should have paid if every process had worked correctly. IRMI defines it as dollars lost through claims management inefficiencies resulting from failures in existing processes, manual and automated. It covers overpayment and valuation error, missed subrogation and salvage recovery, fraud that gets paid, and handling cost driven by files staying open too long.
How much claims leakage is typical?
There is no agreed answer, and anyone who gives you one number is overstating what is known. EY puts leakage at approximately 7% to 14% of carriers' total spend based on its own claims quality assessments. A conventional industry figure of 2% to 4% circulates without a study behind it. One consultancy says its own audits routinely document 20% to 30%. This calculator uses EY's band, because it is the only published range tied to a named firm's documented file-review practice, and shows the others alongside it.
Is this calculator free, and do I have to give you my email?
It is free and there is no form. The calculator runs in your browser, has no back end, and sends nothing you type to us. If you share a link to your numbers, they ride in the URL fragment after the hash, which is not part of the HTTP request. One honest caveat: this site loads Google Analytics on every page, as most sites do, and analytics scripts read the address bar. So do not treat a shared link as confidential, even though we never receive your figures ourselves.
Why is the result a range instead of a number?
Because the underlying benchmarks are ranges and the spread between credible published estimates is an order of magnitude. A point estimate would imply a precision that the source data does not support. Leakage also varies materially by book, by line of business and by jurisdiction, so the only way to establish your own figure is a closed-file audit of your own claims.
Why do the buckets not add up to the total?
They are independent estimates of overlapping things. A fraudulent claim that was also overvalued and also closed without a subrogation check appears in three buckets. Each bucket is computed from its own published benchmark rather than being carved out of the headline number, because no published source splits leakage into these four categories with weights. Adding them would double count.
Where does the cost-of-delay figure come from?
It converts your indemnity into an implied loss adjustment expense using the 2025 NAIC industry ratio of loss expenses incurred to net losses incurred, which was 15.6%. That gives a handling cost per claim, which divided by your average cycle gives a cost per claim per day. Only days above the published 40.7-day benchmark for US property claims are charged. If your cycle is already at or inside the benchmark, the line is left blank rather than estimated.
Replace the estimate with your own number
Bring a sample of closed files and we will walk through what a reviewer would have caught in them. No slideware.