AI use case

AI for litigation and recovery triage

An AI system that scores open insurance claims and overdue accounts for the risk that they end up in litigation or formal legal recovery, and ranks the attorneys, law firms or recovery agencies likely to get the best result, so a claims handler, legal panel manager or collections specialist can act early instead of after the case has already escalated.

By Len Debets · Last verified 29 September 2026 · 2 public deployments

USD 14 million to USD 67 million
Indicative value per year
A workers' compensation insurer or claims administrator handling 50,000 lost time claims a year. Worked example, see how it is calculated.

What problem does it solve?

Litigation is a small share of workers' compensation claims, but it changes the economics of the case completely. Sedgwick's Chief Claims Officer describes three drivers behind it: adversarial relationships, where employees seek legal representation out of frustration or anger toward their employer or the claims administrator; a claims process confusing enough that people turn to an attorney for guidance; and a belief that hiring an attorney is simply part of the process, especially when a claim is denied.

By the time a claim has an attorney, or an overdue account is in a recovery agency's queue, the organization has usually lost the cheapest options: an early, well handled conversation, a fair settlement offer, a workable payment plan. Sedgwick reports that litigated workers' compensation claims cost over three and a half times more than claims that never escalate, and CLARA Analytics' research on casualty claims with attorney involvement found a claim duration 295% higher than unrepresented claims.

CLARA Analytics describes QBE Australia's earlier claims triage as "a rudimentary triage system based on a single criterion." In this page's view, a single criterion, whether that is claim severity or days past due, misses the mix of case, claimant and prior outcome signals that actually predict whether a case will escalate and how it is likely to resolve. Litigation and recovery triage aims to replace that single signal with that fuller picture.

How does it work?

  1. Score continuously, not once. The model rescores open claims and overdue accounts as new information arrives, such as a new medical bill, a missed payment or a demand letter, not only at intake, because litigation and legal escalation risk changes over the life of the case.
  2. Surface the drivers, not just a number. The score comes with the specific signals behind it (injury type, days past due, prior attorney involvement in similar cases, claimant or debtor history) so a handler can act on the reason, not just the rank.
  3. Score the counsel and agencies too. For cases that are already litigated or in legal recovery, a second model ranks the attorneys, firms or recovery agencies available by their track record on cost and outcome for similar cases, so the choice of who handles it is evidence based as well.
  4. Recommend the next step. The system proposes a specific action, such as engaging a named attorney, making a settlement offer now, or referring an account to legal recovery, with the reasoning attached, for a person to approve.
  5. Feed outcomes back. Every closed case, whether it litigated or not and however it resolved, becomes a labelled example that keeps the score and the attorney or agency rankings current.
Audience
Employee facing
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: Lower cost to serve, Risk and loss reduction, Speed and cycle time.

Indicative value

A workers' compensation insurer or claims administrator handling 50,000 lost time claims a year

USD 14 million to USD 67 million

Annual indemnity cost avoided by reducing unnecessary legal involvement per year

How this is calculated

Formula: claimsPerYear * legalInvolvementShare * reductionAchieved * (costPerLegalClaim - costPerNonLegalClaim). The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Lost time claims per year claimsPerYear, claims per year50,00050,000The reference insurer or claims administrator.
Share of lost time claims with attorney involvement legalInvolvementShare, fraction of claims0.10.144Sedgwick reports litigation at less than 4% of workers' compensation claims overall but about 14.4% for indemnity (lost time) claims. Litigation and attorney involvement are not the same thing, but no source splits them out for lost time claims, so this range uses the litigation rate as a proxy for attorney involvement, capped at Sedgwick's reported 14.4% rather than above it. Source
Relative reduction in legal involvement from earlier triage reductionAchieved, fraction reduction in legal involvement0.050.15Gradient AI's study of over 200,000 lost time claims across more than 60 carriers found AI enabled claims management cut legal involvement by 15%; the low end is conservative against that figure. Source
Average indemnity cost of a claim with attorney involvement costPerLegalClaim, USD per claim70,00080,000CLARA Analytics reports average indemnity costs of 77,807 US dollars for casualty claims with attorney involvement, across workers' compensation, commercial auto and general liability lines combined; applied here to the workers' compensation lost time reference book as the closest available figure. Source
Average indemnity cost of a claim without attorney involvement costPerNonLegalClaim, USD per claim14,00018,000CLARA Analytics reports average indemnity costs of 15,936 US dollars for unrepresented casualty claims, across the same combined lines; applied here to the same workers' compensation lost time reference book. Source

What it leaves out: Gross indemnity cost avoided only. It leaves out the cost of running the triage system and the legal panel, the effect on claimant experience and settlement quality, and debt collection litigation triage, which follows the same logic on different unit economics and would need its own inputs.

Market estimates (analyst estimates, not deployments)

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.

Sedgwick

United States · Insurance · 2025

ProductionGrade B

Sedgwick's Chief Claims Officer, Max Koonce, described the claims administrator's use of predictive modelling to identify workers' compensation claims with a high propensity for litigation early in the process, so they receive tailored workflows and additional resources before legal escalation happens. Sedgwick has separately used outcome data, for about four years at the time of writing, to score and select defence attorneys by cost and case results, and adopted AI driven medical record summarisation for examiners and defence counsel around 12 to 15 months before the post. No litigation rate or cost figure specific to the AI tools is given.

No outcome disclosed.

QBE Insurance Group

Australia · Insurance · 2018

ProductionGrade C

QBE's Australian Pacific division first engaged CLARA Analytics in September 2017 for its workers' compensation claims, using CLARA's AI and machine learning products as an early warning system for frontline claims teams. In December 2018 QBE Australia expanded its adoption of CLARA's product suite across its Australian statutory claims, covering workers' compensation and CTP (auto liability) businesses, and as part of that expansion will deploy CLARA Litigation, a product CLARA describes as designed to avoid and reduce the overall cost of litigation. A related, undated CLARA blog post, describing the same QBE Australia Pacific claims team, attributes AI and machine learning powered processes to routing high risk claims to experienced adjusters, guiding claims managers toward medical providers likely to produce good outcomes, and scoring cases for overall litigation risk and attorney performance so claims managers know when promptly settling a case is likely to produce the best outcome.

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

  • Historical claims or account records labelled by whether the case litigated or went to legal recovery, at what cost and after how long
  • Legal bill review or outside counsel invoice data, mapped to case and attorney or firm
  • A definition, agreed with legal and compliance, of what counts as high litigation or recovery risk for this book

Systems to integrate

  • Claims or collections case management system
  • Legal bill review or electronic billing platform, for attorney or agency performance data
  • Document management system for demand letters, pleadings and settlement offers

Complexity: Medium

Scoring is the easy part. The hard part is joining claims or account data with legal bill review or outside counsel invoice data cleanly enough to trust an attorney or agency score, and building a workflow so handlers act on the flag before the case escalates rather than after.

  1. 1

    Baseline the current mix

    Measure how many cases litigate or go to legal recovery today, at what cost and after how long, split by case type, before building anything.

  2. 2

    Build and validate the risk score

    Train or configure the model on closed cases with a documented litigation or legal recovery outcome, and validate it on a holdout period before it touches live cases.

  3. 3

    Design the human decision, not just the score

    Define who reviews a flagged case, what they can do (settle, engage a specific attorney, escalate, hold), and within what limits, before the score reaches anyone.

  4. 4

    Add attorney or agency performance scoring where it applies

    Benchmark counsel and recovery agencies on cost and outcome for comparable cases, not only on speed, and require a minimum case count before a firm gets a score at all.

  5. 5

    Pilot on one line or portfolio, then expand

    Measure the litigation or legal escalation rate, cost per case and cycle time before and after on a comparable population, then widen to other lines.

Guardrails

  • A person decides every case flagged as high litigation or recovery risk; the system never files a claim into litigation or writes off a debt on its own
  • Explainable score drivers shown to the handler, not a single black box number
  • Fair treatment and vulnerable customer review of any collections use, including unfair, deceptive or abusive practice checks
  • Legal privilege and confidentiality controls around any data that touches active litigation

KPIs to instrument

  • Litigation or legal escalation rate, by case type
  • Average cost and cycle time of litigated versus non litigated cases
  • Recovery rate on accounts referred to legal action versus settled
  • Share of flagged cases where the handler followed the recommendation

Human in the loop

Adjusters, claims handlers, legal panel managers and collections specialists make every decision: whether to settle, which attorney or agency to use, and whether to write off a debt. The system's job is to bring the relevant history and a score in front of them earlier than a manual review would.

Common failure modes

A score without context
A number with no explanation gets ignored or overridden without a record. Show the driving factors and require a reason when a handler departs from the recommendation.
Attorney or agency scores built on too few cases
A firm with five cases can look as reliable as one with five hundred. Show a confidence measure and require a minimum case count before scoring a firm at all.
Vulnerable claimants or debtors pushed toward faster settlement
A model tuned purely for cost can recommend against people who would benefit from a fair hearing or hardship support. Test recommendations against vulnerability signals, not only cost.

What are the risks and rules?

EU AI Act

Depends on design

Litigation and recovery triage is not itself listed in Annex III. It becomes high risk when the same system evaluates a natural person's creditworthiness or credit score (Annex III point 5(b)), for example when a debt write off or pursue decision is based on such an assessment; scoring which attorney to instruct or whether an insurance claim is likely to litigate is outside that point on its own. Automated decisions with a legal or similarly significant effect on an individual are also subject to Article 22 of the GDPR.

Guidance

Controls to put in place

  • Every litigation, settlement or legal recovery decision is made and recorded by a person, not the model
  • Attorney, firm and recovery agency scores are shown with a confidence measure and a minimum case count before they drive a recommendation
  • Fair treatment and vulnerable customer review of any recommendation that touches an individual debtor or claimant
  • Decision log linking the score, the recommendation and the human's final action, for every case

Frequently asked questions

How is litigation risk triage different from claims triage?
Claims triage sorts every new claim by complexity and fraud signals so it reaches the right handler. Litigation and recovery triage is a narrower, later step: scoring which open claims or overdue accounts are heading toward a lawyer or a court, and which attorney, firm or recovery agency is likely to get the best result, so a person can act before costs escalate.
Does this replace legal judgment?
No. Sedgwick describes flagging claims with a high propensity for litigation so they receive tailored workflows and additional resources, and using outcome data to select the attorneys who consistently deliver the best results. CLARA Analytics describes its litigation risk scores for QBE Australia as alerting claims managers when promptly settling a case is likely to produce the best outcome. Neither source claims the AI decides; this page's own guardrails require a person to make every settlement, attorney or write off decision.
What KPI shows this is working?
Track the litigation or legal escalation rate by case type, and the average cost and cycle time of litigated versus non litigated cases, before and after. Gradient AI, a vendor, studied more than 60 workers' compensation carriers and found AI enabled claims management cut legal involvement in lost time claims by 15%; treat that as one vendor's study of its own customers, not an independent benchmark or a guarantee for any one book.
Does this apply to debt collections as well as insurance claims?
The same logic, scoring whether a case is worth escalating to legal action and who should handle it, applies to a lender or collections agency deciding whether to litigate or write off a delinquent account. The named evidence on this page is from insurance claims administration, so treat the collections application as emerging until named deployments confirm it.

How to cite this page

Blits.ai AI Use Case Library, "AI for litigation and recovery triage", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/litigation-and-recovery-triage. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 29 September 2026: First published

Related use cases

Insurance

AI for claims triage and straight through processing

AI that reads each new insurance claim and its documents, scores its complexity, cover questions, fraud and recovery signals, sends it to the right handling path and handler, and settles simple, low risk claims end to end within set limits without a person touching them.

Deployments
8 public, best grade B
Reported automation rate
about 55%
Lemonade, organization claim
Insurance

AI for subrogation opportunity detection

AI that reads open and closed claims to find cases where a third party is wholly or partly liable, estimates liability and the recoverable amount under the applicable negligence and recovery rules, and sends scored recovery opportunities with their reasons to the subrogation team.

Deployments
2 public, best grade B
Autonomy
Assist
Cross industryBanking

AI agent for early collections and hardship support

A voice and messaging agent that contacts customers in early arrears and answers their inbound calls, takes payments and sets up payment arrangements within preapproved rules, and recognises signs of hardship or vulnerability so those customers go straight to a trained person.

Deployments
2 public, best grade C
Reported cost reduction
75%
SameDay Auto Finance, vendor claim
Banking

AI recommendations for loan restructuring and hardship arrangements

An assistant that assembles a stressed borrower's position, tests restructuring options such as a term extension, rate relief, payment holiday or due date change against policy and affordability, and recommends the best fit with a written rationale for a person to approve.

Deployments
1 public, best grade B
Autonomy
Copilot