Insurance November 18, 2026  ·  11 min read

Subrogation Automation: Recovering What the Claims Process Already Paid Out

Verisk’s analysis of 2025 claims data makes a quiet point that every claims leader already half-knows: for a carrier managing high claim…

Pawel Scheffler
Head of Marketing
Insurance

Verisk’s analysis of 2025 claims data makes a quiet point that every claims leader already half-knows: for a carrier managing high claim volumes, lifting subrogation recoveries by even one percent adds up to millions of dollars a year. The money is already owed. Another party was liable and the claim was paid, so the right to recover existed the moment the file was opened. Whether the carrier actually collects comes down to a back-office process that, at most insurers, still runs on an adjuster remembering to look.

Why subrogation is the recovery the lifecycle forgets

Subrogation is the step where an insurer reclaims what it paid out from the party that was actually at fault. It sits at the very end of the claims lifecycle, after intake, after adjudication, after the cheque has gone out. Solomon Partners, in its December 2025 walk through the lifecycle of a claim from first notice of loss to subrogation, puts recovery at the far end of that chain for a reason: it depends on everything that happened before it. By the time a file reaches subrogation, the facts that determine whether recovery is even possible were recorded, or missed, weeks earlier.

This is where the Manual Wall shows up in a form most carriers never name. A subrogation unit falls behind for a structural reason rather than a staffing one: the volume of paid claims grows faster than the number of adjusters who can comb through each file looking for a liable third party. Recovery that depends on manual review scales with headcount, and headcount does not keep pace with claim count. The work that identifies a recoverable claim is the same Human API problem that runs through the rest of the back office: a person bridging what the intake system knows, what the adjudication notes say, and what the recovery team needs, by reading and re-keying it by hand.

What weak subrogation actually costs

The cost of a missed recovery does not show up as a line item. It shows up as a number that never arrives, which is why it is easy to under-manage. A paid claim with an unpursued third-party liability is a dollar that left the business and was never clawed back, and it lands directly on the loss ratio. Multiply that by the share of a book where the liability signal was there to be found and nobody found it, and the leak is material rather than rounding.

Verisk’s 2025 ClaimSearch work shows how thin the signals can be. Looking across claims from 2021 to 2025, only a small fraction of workers’ compensation claims, around a quarter of one percent, matched a general liability claim with the same claimant and date of loss. That sounds negligible until you sit with what it means: the overlap that makes a recovery possible is often a single data match buried across two separate systems, and a human reviewer working a full caseload will not see it. The recoveries hide in exactly the places manual review is worst at searching.

Carriers feel this as a recovery rate that sits lower than it should and never quite explains why. The explanation is rarely dramatic, just an accumulation of small misses: a police report that noted a second vehicle but never made it to the subro queue, a subrogation referral an adjuster meant to flag and then buried under the next twenty files, a liability clause in a contract that nobody read before the claim closed. Each one is a few thousand dollars. Together they are a structural drain on the book.

The recovery window closes faster than the backlog clears

Subrogation also runs against a clock, which is what makes a late signal so expensive. Every recovery sits under a statute of limitations, and in casualty lines the window to act can lapse while a file is still working its way through a manual subro queue. Inter-company recovery adds its own deadlines on top of that: in the United States, the bulk of auto subrogation between insurers is resolved through Arbitration Forums, whose processes run on filing windows rather than on whenever a recovery team happens to reach the file. A recoverable claim identified six months after the loss is not worth the same as one identified in the first week, because some share of the late ones are now unrecoverable on timing grounds alone.

This is the part a detection engine at the end of the pipeline cannot fix. By the time a closed file is scored for recovery potential, the clock has already run for the entire adjudication period, and for some cases it has run out. Capturing the liability signal at intake helps on two fronts at once: the signal is cleaner, and it is earlier. An opportunity flagged while the claim is still live has the full statutory window available to pursue it, while one surfaced after close may have weeks left or none. Speed of identification is itself a recovery lever, and manual review is the slowest way a carrier can identify. Raising a recovery rate means shortening the distance between the loss happening and the recovery being flagged, and that distance is set upstream, in intake and adjudication, not in the recovery department.

What “subrogation automation” actually automates

Here is where most of the market and most of DF’s view part company. The vendors in this space, and there are capable ones, largely sell detection engines that sit at the end of the pipeline and score closed or near-closed files for recovery potential. That catches some of what manual review misses, and it is better than nothing. The deeper problem is that a detection tool bolted onto the end is still working from whatever the upstream steps happened to capture.

A subrogation case is only as good as the evidence recorded when the loss was fresh. The liability signal, the second party, the police report, the contract clause, the photographs, exists at first notice of loss and during claims adjudication. If intake captures those fields as structured data and adjudication preserves them, subrogation starts with a clean, searchable file. If intake takes them as free text in a notes field, or not at all, the recovery team is reconstructing the case from fragments months later, when memories have faded and evidence has gone cold.

So the useful definition of subrogation automation points upstream rather than at the recovery scorer: automating subrogation means capturing the liability signals during intake and adjudication so recovery potential is flagged while the claim is still live, then routing identified cases into recovery with the evidence already attached. The detection step matters, and the choice between rules-based automation and AI for that detection is a real one. But detection only improves things at the margin, while capturing the signal upstream improves the base, which is why a recovery team that receives structured, complete files recovers more than one re-keying free text, regardless of how clever the scoring model at the end is.

Where subrogation automation should start

The instinct, when a carrier decides to take subrogation seriously, is to buy the recovery platform and point it at the back of the pipeline. That is the step a vendor demo makes look easy, and it is usually the wrong place to begin. The highest-recoverable-cost gap in most subrogation operations is upstream, at the point where the liability signal is either captured as data or lost as a note.

This is the logic Digital Forms builds its engagements around, and it is why we sequence every initiative in ROI order, biggest recoverable cost first. For subrogation that ordering almost always puts signal capture at intake and adjudication ahead of a detection engine at the end, because fixing what the file contains lifts every recovery downstream, while a better scorer only re-reads files that are already incomplete. The question worth answering before any tool is bought is a measurement one: of the recoveries the book is missing, how many were missed because no system flagged a liable party, versus missed because a flagged case was worked badly. The first failure is a data-capture problem. The second is a recovery-operations problem. They have different fixes, and buying for the second when you have the first is how carriers end up with an expensive platform and a recovery rate that barely moves.

What good subrogation operations look like

Picture two casualty carriers of similar size and book. The first runs subrogation as a separate recovery unit at the end of the line. Adjusters close files and move on, a subro team pulls what it can from closed claims weeks later, and recovery depends on which cases happen to still have usable evidence. Its recovery rate is whatever its most diligent people can salvage from an incomplete record, and it rises and falls with staffing.

The second carrier treats recovery as the final step of one connected lifecycle. At intake, the fields that signal third-party liability are captured as structured data, not prose. During adjudication, those signals travel with the file instead of being summarised away. By the time a claim is paid, the system already knows whether a recovery opportunity exists and has the evidence attached to pursue it. The subro team spends its time negotiating and collecting rather than hunting through records for cases that went cold. The recovery rate stops depending on individual memory and starts depending on the process, which means it holds as volume grows instead of sagging under it.

In the operations we have looked at, the gap between those two carriers shows up less in the recovery team and more in the shape of the file itself. The fields that mark a liable third party are either present as structured data on the day the loss is reported or they are not, and that single difference tends to move the recovery rate more than any change made downstream of it. The same book of business, run the second way, simply returns more of the money it is owed, without adding a single adjuster to the subro desk.

What to do about it

The first move is a measurement rather than a purchase, because you cannot sequence a fix for a leak you have not sized. Before committing to a recovery platform, a carrier should quantify where its subrogation is actually leaking: how much recoverable cost is lost to signals never captured, how much to referrals never made, how much to cases worked too late. Surfacing and sizing that recoverable cost is what we built our Profit Leak Diagnostic to do, and in subrogation it usually reframes the problem from “we need a better recovery tool” to “we are losing most of this before the file ever reaches recovery.”

From there the build is narrow and fast. Rather than a multi-quarter platform programme, the higher-return first step is usually to make one signal-capture improvement real at intake or adjudication and prove the recovery lift it produces, which is the shape of an Operations Sprint: one working change live in weeks, measured against the recovery rate it moves. A proven lift on one claim type is both the business case and the template for the next.

See how much recoverable cost is leaking before your files ever reach the subro team.

Our claims automation work maps where the liability signal is lost across intake, adjudication and recovery, sizes it in recovered dollars, and makes the lift provable from clean data.

Explore how Digital Forms approaches claims recovery →

The recovery was always owed

Most conversations about subrogation start at the end of the pipeline, with the recovery team and the tools that score their files. The money, though, is decided much earlier, in whether a liable party was recorded as data the moment the loss was reported. A carrier that gets that right is not so much running a better recovery department as running a claims process that stops losing the recoveries it was always entitled to collect, which is a quieter advantage and a more durable one. The question for any claims leader is simple enough to answer with a week of data: how much of what you are owed never makes it to the team whose job is to collect it?

Written by
Pawel Scheffler
Head of Marketing

Pawel Scheffler leads B2B marketing at Digital Forms. He writes for mid-market service-company CEOs on what actually moves the P&L — breaking through the Manual Wall, turning digital transformation into measurable ROI, and scaling operations without simply hiring more people.

Get started

Let's talk about
your business.

30 minutes. No slides. Just an honest conversation about what's holding your operation back — and what a realistic fix looks like.

Free call. You leave with a clear read on your biggest operational bottleneck, whatever you decide next.
No obligation. No sales deck. No pressure to sign anything.
Direct access. You speak to a founder, not an account manager.
cezary.bielecki@digitalforms.pl
+48 509 103 244

30 minutes · No obligation · No sales deck · You keep the diagnostic