Claims leakage: what the term actually counts, and what it quietly cannot
In short
Claims leakage is the gap between what a claim actually cost and what a reviewer, applying a stated standard, concludes it should have cost. One of those 2 numbers was never observed, so the review standard is part of the figure. A duplicate payment or an unapplied deductible can be evidenced from records alone; a settlement value cannot, and merging the two produces a number nobody can defend.
Key takeaways
- Leakage compares a real payment against a counterfactual nobody observed, so the review standard is part of the number.
- 2 categories, not 1: findings a record can evidence, and findings that are a reviewer's opinion about a settlement.
- Duplicate payments, unapplied deductibles and off-schedule rates are checkable against source records.
- Indemnity and expense findings have different owners, so one combined percentage hides both fixes.
Claims leakage is the difference between what a file was paid and what a reviewer, applying a defined standard, concludes it should have been paid. It is a subtraction, and only 1 of its 2 numbers exists in the ledger. The other is a reconstruction of a claim that was never handled that way.
That is not an argument for abandoning the measure. It is an argument for publishing the standard beside the figure: 2 review teams on different standards read one file and report different leakage, and neither has erred.
The second number in the subtraction never happened
A leakage review reconstructs how a file should have been handled and prices the gap. Everything contentious follows from that move, because the reconstruction rests on 4 choices made before a file is opened.
- Hindsight or contemporaneous. Judging a decision against what was knowable on the day, or against what emerged later, flips the result on most disputed files.
- Whose standard. The carrier's handling manual, a client service instruction, or a view of what a competent adjuster would do: 3 standards, 3 numbers.
- How the sample was drawn. Stratifying by severity finds different leakage from a random draw, because large findings sit in large files.
- Whether the file is coded as what happened. Rule checks key off loss type, so a mis-typed loss is measured against the wrong population — the gap in the coded cause of loss versus the caller's story.
The findings a record can prove on its own
One group of findings needs no opinion. Each compares 2 records that already exist, so it runs over the whole book rather than a sample.
| Category | The comparison | The proof |
|---|---|---|
| Duplicate or overlapping payment | 2 payments, same invoice or period | Both payment records and their references |
| Deductible unapplied or applied twice | Policy deductible against the deduction | The policy term in force on the loss date |
| Recovery found and never pursued | Third party on file against recovery activity | The liability note and an empty ledger |
| Rate or fee outside the schedule | Amount billed against the schedule that day | The vendor schedule, versioned by date |
| Payment outside the period of cover | Loss dates against inception and expiry | The policy record and the dated line item |
| Released above the approver's authority | Amount against that approver's level | The approval record and authority matrix |
Row 3 is most often miscounted. An unpursued recovery is leakage only if the file was referable at all, which turns on the 4 conditions in what makes a file referable to subrogation.
The findings that stay contestable however complete the data is
- Settlement value. That a claim settled higher than it needed to is an opinion about a negotiation that ended, tested against files never quite comparable.
- Negotiation posture. Whether to hold, split or move early turns on facts about the other side nobody wrote down.
- Timing. Pricing delay asserts what the claim would have cost had it moved. The stalling itself is measurable, in the files in the middle of the caseload.
- Whether more investigation was warranted. Leakage review gets asked to double as fraud detection, a different question with its own precision problem, in why fraud indicators fire on a third of the book.
Indemnity and expense findings answer to different people
Indemnity leakage sits in what was paid to or for the claimant: settlement, repair, benefit. Expense leakage sits in what was paid to run the file — adjusting, experts, counsel, vendor charges. Indemnity findings go back to handling standards and authority levels; expense findings usually go back to a vendor schedule or a panel rule, fixable by a rule rather than by retraining anyone.
A leakage figure is only as good as the standard it was measured against, and a figure published without that standard is an assertion wearing a decimal point.
What has to travel with the figure before anyone quotes it
- The standard, dated and versioned. If the handling manual changed in March, findings from January were measured against a different rule and cannot be trended against later ones.
- The sample frame. Which period, which severity bands, how files were selected and how many were excluded.
- Per-finding evidence links. Each finding should resolve to the transaction, the policy term and the schedule behind it, so a disputing manager sees the record, not the conclusion.
- A reviewer identifier and an inter-rater check. Put 2 reviewers over the same files and publish how far apart they landed.
Steps 1 and 3 are what make a leakage programme actionable rather than arguable, and both are ordinary software problems: version the schedules and the authority matrix by effective date, and keep every finding linked to source. That is MVP and product build work. This page sits in claims handling, fraud flags and recovery, part of insurance and claims software.
Frequently asked questions
Short answers to the follow-ups this page tends to raise.
What is claims leakage in plain terms?
It is the difference between what a claim was paid and what a reviewer says it should have been paid under a stated handling standard. The first number comes from the ledger; the second is a reconstruction. That is why the same claim yields different leakage under 2 review standards, and why the standard must be published with the figure.
How is claims leakage measured?
By file review against a written standard, usually on a sample rather than the whole book: a reviewer reads the file, decides what should have happened, and prices the difference. The method choices — hindsight or contemporaneous judgement, whose standard applies, how the sample was stratified — move the result as much as the files do.
What is the difference between indemnity leakage and expense leakage?
Indemnity leakage is money paid to or for the claimant that a reviewer says was not owed; expense leakage is money paid to handle the file — adjusters, experts, counsel, vendors — beyond what the handling required. The fixes differ: indemnity points at handling standards and authority levels, expense usually at a vendor schedule or a panel rule.
Can software identify claims leakage without a reviewer?
For 1 group of findings, yes. Duplicate payments, unapplied or double-applied deductibles, amounts outside a contracted schedule, payments outside the period of cover and releases above an approver's authority all compare records that already exist, so they run over the whole book and evidence back to source. Settlement value is not in that group.
- claims leakage
- file review
- claims quality
- measurement
The work behind this page
Builds from our portfolio that this page draws on.
AP Copilot
An AI accounts-payable copilot that reads invoices, matches them to POs, and routes clean approvals
FintechPriorPilot
An AI prior-authorization and denial-management platform that auto-assembles and submits auths, predicts denials before submission, and drafts the appeals to recover revenue.
Healthcare AIQuoteForge
An AI CPQ and proposal platform that builds enterprise quotes from your catalog, guards every discount against the margin floor, routes approvals, and generates the proposal.
Sales AIRead next
- Subrogation referral: the facts a file needs before recovery is possibleRecovery teams argue about which files deserve attention. The argument disappears once referability is written as 4 conditions, each pointing at a field.definition
- Fraud indicators fire on a third of the book and nobody reads them any moreInvestigators ignoring alerts is the symptom. The finding underneath is almost always that nobody can say what any single indicator's firing rate buys in accepted referrals.diagnostic
- The oldest and newest files move; the ones in the middle of a caseload stopNew files move because a clock is running and old files move because someone escalated them. The band in between has no forcing function at all, and that is where inventory quietly accumulates.diagnostic
- Reserves get set in the first week and never move until the file closesA reserve that has not moved in nine months is not a stable claim. It is an unasked question, and the repair is a set of file-event triggers rather than a better first estimate.diagnostic
- The claim diary: a scheduling primitive, not a to-do listA diary entry is a dated obligation with an owner, an escalation target and a condition that says when it is met. Without that condition it is a reminder, and reminders leave files untouched.definition
- Cause of loss: the coded field, and why the caller's story is not itThe loss narrative and the coded cause of loss are two fields with two jobs. Conflating them gives reporting nobody trusts and routing nobody can explain.definition
Working on something in this space?
Tell us where you are in a sentence or two. We'll tell you honestly whether we're the right team, and what a sensible first slice of the work looks like.
Start the conversation