What problem does it solve?
When an insurer pays a claim that someone else caused, it can recover the money from that party or its insurer. In practice many of those recoveries are never pursued. Handlers focus on settling the claim for the customer, the signs of third party liability sit in free text notes, police reports and photos, and the rules on comparative negligence and recovery differ by state or country.
Referrals to the subrogation team therefore depend on individual handlers spotting the opportunity, often too late, when evidence is gone or deadlines have passed. Recovery teams in turn spend time on referrals with little chance of success. Missed subrogation is a quiet form of claims leakage: no customer complains about it, so it rarely surfaces on its own.
How does it work?
- Read every claim early. The AI reads the claim notes, statements, police reports and other documents from the first days of the claim, not only when a handler refers it.
- Identify who else is responsible. It extracts the parties and facts and assesses whether a third party, product, contractor or other insurer may be liable.
- Apply the rules. It checks the applicable comparative negligence, recovery and limitation rules for the jurisdiction and line of business.
- Estimate and score. It estimates liability shares and the recoverable amount and scores the opportunity by expected recovery.
- Refer with reasons. Scored alerts with the supporting facts and rules go to the subrogation team, which decides whether to pursue, and outcomes feed back into the model.
- Audience
- Back office
- Autonomy
- Assist
- Adoption
- Early adopters
- Channels
- Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Risk and loss reduction, Employee productivity, Speed and cycle time.
Indicative value
An auto and property insurer paying USD 500 million in claims a year
USD 600,000 to USD 5.3 million
Additional recoveries collected per year
How this is calculated
Formula: claimsPaid * recoverableShare * missedShare * collectedShare. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Claims paid per year claimsPaid, USD per year | 500,000,000 | 500,000,000 | The reference insurer. |
| Share of claims paid that is recoverable from third parties recoverableShare, fraction of claims paid | 0.03 | 0.06 | Editorial assumption; depends heavily on the lines of business and jurisdictions. Replace with your own recovery history. |
| Share of recoverable amounts not pursued today missedShare, fraction of recoverable amounts | 0.1 | 0.25 | Editorial assumption; estimate it by auditing a sample of closed claims. |
| Share of newly found opportunities actually collected collectedShare, fraction of opportunities | 0.4 | 0.7 | Editorial assumption; not every liable party pays in full. |
What it leaves out: Gross recoveries only. It leaves out the cost of pursuing recoveries (staff, legal, arbitration fees), the time value of money, the reduction in customer excess where recoveries are shared with the policyholder, and the cost of the platform.
Who already uses it?
2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Elephant Insurance
United States · Insurance · 2024
Elephant Insurance, a Virginia based auto insurer owned by Admiral Group, added Shift Subrogation Detection in November 2024 after using Shift for underwriting fraud since 2021 and claims fraud since 2020. Elephant's head of claims says the model helps the insurer find subrogation opportunities at scale and recommends handler actions to improve recovery, alongside the fraud detection already in place. No recovery figures were published.
No outcome disclosed.
Central Insurance
United States · Insurance · 2023
Central Insurance, a US insurer, added Shift Subrogation Detection in 2023 after three years of using Shift to detect claims fraud. The system reviews claims for signs that a third party is wholly or partly responsible, often on the first day of the claim, and gives the recovery team alerts with the facts, comparative negligence rules and state recovery laws for PIP and medical payments. Handlers still refer claims to subrogation; the AI catches the ones they miss. Its claims recovery supervisor reports more referrals and hours saved every week, but no figures were published.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Claim notes, statements, police reports and photos linked to each claim
- Historical subrogation referrals with outcomes and amounts recovered
- Comparative negligence, recovery and limitation rules per jurisdiction and line
Systems to integrate
- Claims management system (read access to claims, notes and payments)
- Subrogation or recovery workflow and case management
- Document storage for police reports and evidence
- Inter company arbitration or recovery platforms where used
Complexity: Medium
The model needs claim notes and documents plus machine readable recovery rules per jurisdiction. Integration is mostly read only on the claims system plus a referral into the recovery workflow, so the operational risk is modest; the effort is in the rules content and in feedback from recovery outcomes.
- 1
Audit a sample of closed claims
Have experienced recovery staff review a few hundred closed claims to estimate how much was missed and why. It sizes the prize and creates the first labelled data.
- 2
Start with one line and one jurisdiction set
Auto physical damage and the personal injury protection and medical payment rules of a few states or countries are typical starting points, because volume is high and the recovery rules for these exposures are written down per state. The top 25 US insurer on this page started with its auto subrogation team and added property later.
- 3
Run next to handler referrals
Keep manual referrals and let the AI add its own, so you can measure how many opportunities it finds that people missed and how early.
- 4
Tune to what the team accepts
Track which alerts the recovery team accepts and what is collected, and set thresholds to the team's capacity rather than the model's recall.
- 5
Keep the rules current
Recovery law and limitation periods change. Give the rules an owner and a review cycle, and version them with the model.
Guardrails
- The AI refers opportunities; people decide whether to pursue and what to demand
- Every alert shows the facts, liability reasoning and the rule applied
- Limitation deadlines tracked from the alert so late referrals are visible
- The policyholder's claim is never delayed or reduced because of a recovery opportunity
KPIs to instrument
- Share of alerts accepted by the recovery team
- Amount recovered from AI raised opportunities per month
- Days from first notice of loss to subrogation referral
- Opportunities found by the AI that handlers had not referred
- Recoveries lost to limitation deadlines
Human in the loop
The subrogation team reviews every alert, decides on pursuit and negotiates recoveries. Claims handlers can still refer claims manually, and recovery leads approve changes to rules and thresholds.
Common failure modes
- Alerts nobody pursues
- The model raises more opportunities than the team can work, and value is lost anyway. Match thresholds to capacity and prioritise by expected recovery.
- Wrong law, wrong demand
- Outdated or misapplied negligence rules lead to demands that fail. Version the rules and review them on a schedule.
- Handlers stop referring
- Teams assume the AI will catch everything. Keep manual referral and measure both sources.
What are the risks and rules?
EU AI Act
Minimal risk
Detecting recovery opportunities against third parties and other insurers is not listed in Annex III: point 5(c) covers only risk assessment and pricing of natural persons in life and health insurance, and the system does not decide on a natural person's access to a service. It is an internal tool that does not converse with the public or publish generated content, so the deployer transparency duties of Article 50 do not apply. Personal data in claim files, including data about the third party, is still subject to GDPR.
Rules that apply
Guidance
- Opinion on Artificial Intelligence governance and risk management (European Insurance and Occupational Pensions Authority, Europe). Sets risk based, proportionate expectations (data governance, explainability, human oversight) for insurers' AI systems that are neither prohibited nor high risk under the AI Act, which includes back office claims models like this one.
Controls to put in place
- Versioned recovery rules with an owner and review dates
- Log of alerts, decisions and recovery outcomes for model monitoring
- Access controls on claim files used by the model
Frequently asked questions
- How much can AI add to subrogation recoveries?
- Published results are vendor figures for unnamed insurers. Shift Technology reports a recurring average recovery of over USD 1 million per month for a top 25 US property and casualty insurer, and an acceptance rate of 60% or more for a small regional insurer. Size it on your own book by auditing closed claims first.
- Does AI replace handler referrals to subrogation?
- No, it adds to them. Central Insurance still asks its handlers to refer claims to its recovery team and uses the AI to catch the ones they miss, often on the first day of the claim.
- Which lines of business are the best starting point?
- Auto claims, including personal injury protection and medical payments recoveries, because volumes are high and the state recovery rules can be encoded. The top 25 US insurer on this page started with auto and later extended the system to property; the system also draws on external data such as product recall lists.
How to cite this page
Blits.ai AI Use Case Library, "AI for subrogation opportunity detection", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/subrogation-opportunity-detection. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published