How AI Identifies Subrogation Recovery Opportunities Before Evidence Is Lost
Why recovery potential gets lost in the claim notes, and how to put the referral decision in front of the adjuster the moment the evidence arrives, so your team can pay the claim and open the recovery at the same time.
9 min read
Most subrogation programs are built to catch recoveries late: a line on the closing checklist, a closed-file audit, a vendor sweep of files that have already paid. By then, the evidence that proves liability has often been repaired or thrown away, and part of the filing window is gone.
The missed recovery potential was usually visible much earlier, written down somewhere in the file. It was in a police report attached on day 3, a contractor’s invoice, a note from the first call with the insured. The claim gets paid. The file closes. The recovery is never referred.
The best moment to act on a recovery is the moment the evidence arrives. Sometimes that’s at FNOL. It can just as well be a police report a week later, an inspection report, or an engineer’s findings a month in, when the claim is already in work. Whenever it lands, someone has to spot it then, while the evidence, the deadlines, and the negotiating position are still intact.
This is where agentic AI comes into the picture. It automates the identification: it reads every document as it lands on the file, flags the recovery indicators with their source, and prepares the referral, so your adjusters can pay the claim and open the recovery in parallel. The adjuster still decides whether to pursue. The system makes sure they know there’s something to decide.
What you will learn:
- Why subrogation recovery potential gets missed in the first place
- How AI flags recovery potential the moment the evidence arrives, by line of business
- How the AI-to-adjuster workflow runs
- Why running claim and recovery in parallel pays off
- The guardrails to set, and where to start
Why subrogation recovery potential gets missed in the first place
Recovery potential gets missed because the indicators arrive scattered and unstructured, each document is read for a different job, and the recovery question is only asked formally at closure.
The scale is well documented. Verisk, citing SubroIQ’s recovery benchmarks, reports that 15% of claims close with missed subrogation opportunities, roughly $15 billion in annual industry losses. A Journal of Insurance Regulation study cites the same $15 billion figure.
Trace a missed recovery back through the file and the same causes show up.
The indicators have no field. The fact that matters for recovery (a citation, a contractor’s name, a failed part) arrives later, as free text in a police report, a statement, or an invoice. Rules and reports built on structured fields can’t see it.
Every document is read for a different job. The police report confirms the parties, the invoice gets approved for payment, the engineer’s report sets the reserve. The recovery question sits outside all of them.
The referral depends on one person noticing. The adjuster owns the claim; the subrogation unit owns the recovery. The handoff happens only if the adjuster spots the indicator and takes an extra step, on top of the work they’re measured on. When files change hands, the early detail survives as a line in the notes the next handler skims.
Capacity runs out. Verisk also cites research finding that 58% of frontline claims professionals spend more than 20% of their time on manual data entry and compliance work. And files are getting heavier: CCC’s Crash Course 2026 report found average paid bodily injury severity up 32% over four years.
The formal check comes at the end. The closing checklist and the closed-file audit ask the recovery question once the file is complete, so by design they catch it late, often after remediation has removed the evidence. A subrogation manager writing in CLM Magazine named spoliation of evidence among the biggest obstacles subrogation professionals face.
58%
of frontline claims professionals spend 20%+ of their time on manual data entry and compliance
(Verisk)
Your adjusters know subrogation. What a full caseload takes away is the time to re-read every document for recovery while the file is moving. That reading is exactly the work agentic AI does well.
How AI identifies subrogation opportunities in claim files
Recovery indicators sit in police reports, invoices, statements, and adjuster notes, and they look different in each line of business. The AI has to know where to look in each one while processing every document the moment it lands on the file.
The AI reads every document type the file collects and looks for what recovery potential means in each line:
Line of business | Where it shows up | What the AI flags |
|---|---|---|
Personal and commercial auto | Police report narrative, recorded statements, photos, FNOL description | Citations, rear-end or failure-to-yield facts, the third-party carrier and policy, a commercial vehicle or rideshare driver involved |
Homeowners and commercial property | Contractor and plumber invoices, cause-and-origin reports, adjuster notes | Recent work on the property, a failed appliance or component, landlord or tenant responsibility, a utility involved |
Workers’ compensation | First report of injury, incident description, medical records | A third-party driver, premises owner, equipment manufacturer, or other contractor on site |
General and products liability | Incident reports, contracts, certificates of insurance | Contractual indemnity, additional insured status, an upstream manufacturer or supplier |
Every flag links back to the page it came from, so the handler can check it in one click. And the read keeps going: recovery potential often appears weeks in, with an engineer’s report or a second statement. It doesn’t matter whether the evidence arrives at FNOL or in week 5: the person responsible for the recovery decision hears about it the same day.
How the AI-to-adjuster recovery workflow runs, step by step
The AI does the reading, the checking, and the paperwork. The adjuster makes two calls: whether to pursue the recovery, and what goes out to the third party. Each handoff between them is prepared, so neither side waits on the other.
- AI: screens each new document for recovery potential and tests it against the insurer’s referral threshold for that line and value. Below the threshold, it’s noted on the file and the adjuster isn’t interrupted.
- AI to adjuster: sends a referral summary: what was found and on which page, who looks responsible, the estimated recovery against the cost to pursue, what to preserve today, and a recommended route.
- Adjuster: decides. Refer and confirm the route (in-house, inter-company arbitration, or a vendor), close the question with a reason such as a waiver of subrogation, or ask for more information first.
- AI: prepares the recovery. It drafts the notice to the responsible party, evidence preservation instructions (keep the failed water heater), and the demand package from documents it has already read, and sets up deductible recovery on the exposure.
- Adjuster: approves what goes out. Nothing reaches a third party until the adjuster signs it off.
- Subrogation specialist or vendor: pursues the recovery. If a new document changes the picture, the AI updates the summary and sends it to whoever owns the next decision.
See Five Sigma’s Clive AI in action
Request a live demo to see how Clive AI flags subrogation recovery potential on your own claim types.
Why running the claim and the recovery in parallel pays off.
In a late-referral model, recovery starts after the claim work ends and someone re-reads the file cold. Here, the adjuster keeps handling the claim and payment stays on its normal timeline, while the recovery runs alongside it from the day the evidence arrived, with evidence held, notice sent, and the filing window intact. The policyholder hears early that their deductible is being pursued, which often shapes how they remember the claim. And you gain a measure many programs lack: time from evidence arriving to referral.
Guardrails to set, and where to start
Keep the referral decision with a named adjuster, make every flag traceable, and write the thresholds down per line before switching anything on. Require a reason for every recovery closed without referral.
Recovery work touches third parties and sometimes litigation, so the design has to be accountable from the first flag. Set these four guardrails before anything is switched on:
- The adjuster decides. For each line, define who can refer, who can close without referral, and who signs off notices to third parties. AI flags and prepares. No notice goes to a third party without approval.
- Every flag is traceable to the page and line it came from.
- Thresholds are written down per line and jurisdiction: which indicators trigger a review, at what value, and which contract or policy terms stop a referral.
- “Closed without referral” carries a reason, so QA can review the decision itself without re-reading the file. Our post on designing the AI-to-human handoff covers how to package that decision.
A practical order for getting started:
- Sample closed files with no recovery, by line, and check them for indicators that were on the file. That gives you your own missed-recovery baseline.
- Agree the indicators, value thresholds, and referral routes for each line with the subrogation team that pursues them.
- Start with the line where indicators are clearest and volume is highest, then extend to property, workers’ compensation, and liability.
- Measure time from FNOL to referral alongside recovery rate from the first day.
Where Clive AI fits in your recovery workflow
Clive reads every document on the file in real time and surfaces subrogation recovery potential to the adjuster, within the insurer’s own rules.
In Clive™, Five Sigma’s multi-agent AI claims expert, Clive Claim Conductor orchestrates specialized AI agents working at each stage of the claim lifecycle, including subrogation and recoveries. It reads every document, email, and note as it arrives, so a liability indicator in a police report or an adjuster note surfaces without a manual referral.
Clive is configured by each insurer’s SOPs. Five Sigma configures what each agent does; the insurer sets the rules, guardrails, limits, and exceptions. Within them, Clive prepares the referral with its sources and brings the adjuster in to decide.
In Five Sigma’s AI-native claims platform, subrogation and salvage tracking is built into exposure management, with configurable deductible recovery, and litigation and jurisdiction data feed Clive’s context. Clive also runs on top of an existing claims system, so you can start without replacing your core platform.
Key takeaways
- 15% of claims close with missed subrogation (Verisk, citing SubroIQ), mostly because recovery indicators sit in unstructured documents that are read for other jobs.
- The best moment to act on a recovery is the moment the evidence arrives, at FNOL or weeks later. AI flags it then, so the person responsible can start right away.
- Agentic AI reads every document on arrival, flags recovery potential with its source, and prepares the referral; the adjuster decides and approves what goes out.
- Running the claim and the recovery in parallel keeps payment on time and starts the recovery weeks earlier. Track time from evidence to referral.
See how it would work on your claims book.
Find the subrogation recovery potential in your claim files while it’s still worth the most.
Frequently asked questions
How does AI identify subrogation recovery potential in claim files?
Agentic AI reads each document and note as it arrives and flags facts pointing to third-party responsibility, such as a police citation or a named contractor, with a link to the source page. The adjuster decides whether to refer.
When should subrogation potential be identified on a claim?
The moment the evidence arrives, whether at FNOL or weeks later in a police report or engineer’s report. Flagging it then keeps the evidence, notice, and filing window intact.
How does early subrogation identification speed up the claim?
Payment goes out on its normal timeline while the recovery runs in parallel, prepared from documents the AI has already read, so the recovery file isn’t rebuilt at closure.
How much subrogation do insurers miss?
Verisk, citing SubroIQ benchmarks, reports that 15% of claims close with missed subrogation opportunities, roughly $15 billion in annual industry losses.
What is the average subrogation recovery rate?
Across US property-liability insurers, salvage and subrogation averaged 4.5% of net claims paid from 1996 to 2021, according to a Journal of Insurance Regulation study. Auto physical damage reached about 20% in 2021; auto liability lines averaged 1–2%.
Which AI claims platform identifies subrogation recovery potential in real-time?
Insurers can use agentic AI solutions like Five Sigma’s Clive™ to continuously screen claim documents for subrogation indicators, identify potential recoveries, and automate referral preparation. Clive operates within insurer-defined rules, keeping adjusters in control of recovery decisions while reducing manual reviews and missed opportunities.
