Claims leakage is not a single event. It is a category of outcomes — the aggregate gap between what a well-managed file should have cost and what the carrier actually paid. Industry estimates of leakage as a percentage of incurred losses vary by line of business and carrier sophistication, but for personal auto, figures in the 10–20% range are routinely cited in actuarial literature. The practical implication: a regional carrier writing $80 million in personal auto premiums and running a combined ratio near 100 may be sitting on $8–$16 million in potentially recoverable leakage. Most of it is not fraud. Most of it is process failure.
Detecting leakage requires understanding its taxonomy first. The four primary leakage types — overpayment, reserve inadequacy, missed subrogation, and fraud-not-flagged — each require a different detection mechanism and operate at a different point in the claim lifecycle.
Overpayment: The Largest Bucket, and the Hardest to See
Overpayment leakage is the excess between what a carrier paid and what the file — properly adjudicated — should have settled for. It is not fabricated losses or staged accidents. It is the ordinary friction of the settlement process: medical bill padding that wasn't audited, body shop supplements that weren't reviewed, total-loss ACV calculations that used the wrong comparable set, pain-and-suffering settlements that drifted beyond comparable reserve levels.
The detection challenge with overpayment is that individual file variances look reasonable in isolation. A $2,200 supplement on a $14,000 repair estimate draws less scrutiny than a $14,000 claim on a $8,000 vehicle. But across 600 files per month, systematic supplement patterns become visible: certain repair facilities routinely generate supplements averaging 18% of initial estimate, against a facility-average of 11%. That gap, undetected per-file, represents $70–$90 per file in unearned supplement payments at scale.
Leakage analytics on overpayment works by building distribution models at the file, coverage, geography, and vendor level simultaneously. The question is not "is this file suspicious?" — it is "does this file's settlement pattern deviate from the expected distribution for its cohort?" A file settling at the 85th percentile of its cohort by coverage type and geographic area is not automatically leakage. But a vendor or adjuster generating 85th-percentile outcomes at 3x the frequency of peers is generating a signal worth reviewing.
Reserve Inadequacy: Leakage That Develops Slowly
Inadequate reserves are a distinct leakage mechanism. When a reserve is set too low relative to ultimate file cost — a phenomenon sometimes called "adverse development" in loss development triangle analysis — the carrier's financial position is materially misstated until the reserve is corrected. For individual files, the consequence is administrative: an adjuster re-opening reserves and updating the development log. At portfolio scale, systematic reserve inadequacy inflates loss ratios and distorts the carrier's actuarial signal.
Reserve leakage is particularly prevalent in represented bodily injury claims that are initially reported as minor and develop upward over 6–18 months. The initial FNOL suggests a soft-tissue injury; the adjuster sets a $4,500 reserve. At month four, demand letters arrive for $45,000. The gap between initial reserve and ultimate settlement is not necessarily adjuster error — it reflects information asymmetry at reserve-setting time. But systematic patterns of BI reserve inadequacy (initial reserve consistently below 30% of ultimate settlement on represented files) indicate that the reserve-setting workflow needs additional input signals: representation flag at FNOL, injury severity scoring, geographic litigation rate adjustment.
Detection here requires monitoring the reserve-to-ultimate ratio over the life of closed files and comparing individual adjuster and coverage-segment outcomes against portfolio expectations. Carriers that run this analysis quarterly see reserve adequacy trends before they appear in DOI reserve-strengthening orders.
Missed Subrogation: Recovery Dollars Left on the Table
Subrogation — the carrier's right to recover from a liable third party what it paid to its own insured — is frequently the third-largest leakage category and the most systematically underworked. The identification problem is not complicated in principle: if a carrier paid its insured for losses caused by a third party's negligence, the carrier can seek recovery from that third party or their carrier.
In practice, subro identification fails at two points. First, the FNOL or early file notes may not capture the clear liability indicators that flag subro potential: a police report showing fault, a rear-end collision where liability is unambiguous, a work-related vehicle accident where a third-party employer may be liable. If the intake process does not systematically surface these indicators, the file proceeds to settlement without a subro hold.
Second, even when subro potential is identified, manual pursuit requires time that adjuster queues rarely provide. A $3,800 property damage file with clear third-party liability has roughly $3,200 in subrogation potential after adjuster expense. If pursuing that demand through the adverse carrier's arbitration process takes four adjuster hours, the economics are marginal. Automated subro identification and demand generation change the economics: if identification is automated and initial demand letters are system-generated, the four-hour pursuit becomes a 45-minute review-and-approve workflow, making recovery economically viable on smaller files.
Fraud-Not-Flagged: The Detection Gap, Not the Fraud Gap
Fraud-not-flagged leakage is distinct from overt fraud. It is the class of claims that contained sufficient anomaly signals to warrant SIU referral but were processed to payment without review — because no one scored the file during the adjuster's active handling window.
We're not saying that all flagged claims are fraudulent or that high-volume SIU referral is the goal. Precision matters here (a subject addressed in detail in our fraud-flag precision article). The leakage issue with fraud-not-flagged is timing: fraud patterns detected during file handling, before payment, can be referred to SIU and may result in denial, reduction, or recovery. Fraud patterns detected post-payment require civil or criminal pursuit to recover, with far lower recovery rates and substantially higher pursuit costs.
The detection mechanism is an anomaly-scoring model applied during file enrichment — before the adjuster's first substantive action. Scores above defined thresholds trigger SIU referral notes in the adjuster queue; the adjuster does not proceed to payment authorization without adjudicating the flag. The key operational requirement is that scoring must run fast enough to be synchronous with adjuster queue population — scoring a file four days after the adjuster has already settled it does not prevent leakage. It generates a post-mortem report.
A Scenario: How the Four Types Appear Together
Consider a growing personal auto carrier operating in three Northeast states, processing approximately 800 files per month. A leakage analytics review of 12 months of closed files identified the following simultaneous patterns: (1) a cluster of 14 repair facilities generating supplement rates 22% above the state average, accounting for roughly $190,000 in excess payments over the period; (2) a bodily injury adjuster cohort whose initial reserves on represented files came in below 28% of ultimate settlement on 63% of files — a reserve adequacy gap that fed a $340,000 reserve strengthening event in Q3; (3) subro identification on files with police reports citing third-party fault running at a 41% capture rate against an industry-realistic benchmark of 65–75%; (4) fraud-not-flagged leakage estimated at 3.2% of total incurred losses based on post-payment pattern review, with the majority concentrated in glass claims and minor soft-tissue BI files under $6,000.
None of these patterns were visible in the carrier's monthly claims reports. Each was produced by a combination of file-level data that existed in the CMS but had not been aggregated across files, combined with external benchmarks (state supplement rates, subro capture benchmarks, fraud indicator distributions) that the carrier's internal reporting did not incorporate.
When to Run the Analysis
Leakage analytics is most actionable when run at two points: prospectively during active files (real-time enrichment scoring) and retrospectively on closed files (portfolio analysis). The prospective analysis drives operational changes — adjuster workflow, subro triggers, fraud flags. The retrospective analysis identifies systemic patterns that require structural correction: vendor management decisions, adjuster training needs, coverage-segment pricing adjustments.
The operational caution worth stating plainly: leakage analytics does not make adjuster decisions. It surfaces signals. An adjuster queue populated with anomaly-scored files still requires judgment on each individual file — the analytics layer provides the input; the adjuster owns the outcome. Carriers that treat leakage analytics as an automatic payment-reduction engine tend to produce both underpayment claims (a DOI unfair-settlement-practices exposure) and false-positive friction that degrades claim cycle time. The value is in directing attention, not replacing it.
The carriers that manage leakage best are the ones that treat it as a permanent operating discipline rather than a periodic audit event. A quarterly leakage review on closed files is valuable. A real-time leakage signal integrated into the active adjuster workflow changes the economics decisively — because it catches the dollar before it leaves the building.