Subrogation recovery represents money the carrier has already paid and has a legal right to reclaim from a third party whose negligence caused the loss. A vehicle damaged by a driver who ran a red light. A property damaged by a neighbouring property owner’s negligence. A workers’ compensation claim where a third-party contractor was responsible for the incident. In each case the carrier paid the claim and has a recovery right against a party who was not the policyholder.
The industry recovers between 40 and 60 percent of theoretically recoverable subrogation. The gap is not primarily a legal problem or a negotiation problem. Both of those improve when the opportunity is identified and worked effectively. The gap is an identification problem — the claims where the third-party liability exists and the recovery is viable are not being found because the identification process does not operate at the scale required to review the full settled claims population.
Where the identification decision breaks down
At most carriers, subrogation identification relies on handlers flagging claims for recovery review at the point of closure. A handler who recognises a third-party liability indicator — a police report establishing fault, a subrogation demand from a claimant’s attorney, a clear cause of loss pointing to a third party — routes the claim to the recovery team. A handler who does not recognise the indicator, or who closes the file under time pressure without a thorough liability review, does not.
The claims that enter the recovery queue through handler identification are the clearest cases. The claims where liability is present but embedded in the claim record rather than surfaced at closure are not reviewed at all. Police report analysis, accident reconstruction notes, liability assessments in the claim narrative, and the cause-of-loss characteristics that correlate with third-party liability across similar claims — these are all present in the claims data but require systematic review to extract.
A model trained on the settled claims population — specifically on the characteristics of claims where subrogation was subsequently identified and recovered — applies that pattern recognition to the full settled claims inventory rather than to the fraction that handlers flag. The model does not determine whether a recovery is legally viable. It identifies which claims warrant review by a recovery specialist who can make that determination.
The prioritisation dimension
Even when a subrogation opportunity is identified, recovery resources are finite. A recovery queue of 2,000 referred claims, worked in arrival order, produces a different economic outcome than the same queue worked in order of expected recovery yield. The $75,000 case with clear liability and a financially solvent defendant should not wait behind thirty $2,000 cases with uncertain liability and limited recovery prospects.
A prioritisation model that scores each referral by estimated recovery probability, estimated recovery amount, statute of limitations urgency, and defendant solvency directs recovery effort to the cases where it will produce the highest return. The improvement in recovery per investigator does not require additional headcount. It requires a different sequencing of the caseload, applied consistently.
The technology dimension
Subrogation identification models require access to the full settled claims record — cause of loss, liability notes, police report data, claimant and third-party information, and payment details. For carriers running their claims infrastructure on IBM Z, deploying the identification model via IBM Machine Learning for z/OS accesses the complete settled claims population on the same platform, scoring each closed claim against the liability indicator pattern without requiring data extraction. The scored output is a prioritised referral list that flows directly into the recovery workflow.
What success looks like
The primary metrics are subrogation identification rate as a percentage of eligible settled claims, recovery yield per referred claim, cost per dollar recovered, and statute of limitations expiry rate. The identification rate is the upstream metric that determines the ceiling on everything else. A programme that doubles the identification rate from 30 to 60 percent of eligible claims doubles the recoverable population before any improvement in negotiation or legal execution has occurred. That upstream improvement is where most of the industry’s 40 to 60 percent recovery gap sits.
