Rehearsing the Verdict

For an insurer staring down a bodily-injury claim that could detonate into a nine-figure award, the cruelest fact about trial preparation has always been its timing. The most useful information, how a real jury reacts to your adjuster's tone, whether the claimant's story survives cross-examination, where a coverage argument quietly loses the room, arrives only after the verdict is read, when nothing can be changed. Litigation has been a discipline rehearsed once and performed once, with everything riding on the single live take.

That asymmetry is now under pressure from a new category of tooling: agentic systems that stand up believable AI personas, a claimant, an opposing expert, a panel of jurors, even a presiding judge, that defense teams can interrogate, provoke and learn from before anyone walks into a courtroom. The premise is borrowed from aviation and surgery: practice the high-stakes moment thousands of times in simulation, so the live performance is the least surprising thing you do all year. For the insurance industry, which absorbs the financial blast radius of the verdict economy, the appeal is not novelty. It is survival math.

The figures that frame the problem are stark. Nuclear verdicts, jury awards of $10 million or more, climbed to a record 135 against corporate defendants in 2024, a 52% jump over the prior year, with the total reaching $31.3 billion, according to a report by research firm Marathon Strategies as reported by Insurance Journal. A traditional live mock trial, meanwhile, has long run between $10,000 and $60,000 or higher per exercise according to IMS Legal Strategies, while only 1.8% of litigated claims ever reach a verdict, with 98.2% settling, per Sedgwick's analysis reported in Claims Journal. The mismatch is obvious: the events that hurt insurers most are rare, ruinous and almost impossible to rehearse affordably. Agentic panels are the industry's first serious attempt to fix that.

The Old Way: Expensive, Episodic and Late

For decades, the gold standard for testing a case before trial was the live mock trial or focus group. The mechanics are labor-intensive. A three-panel project requires recruiting roughly 30 mock jurors, with academic research suggesting groups of eight to ten produce the best discussion, according to IMS Legal Strategies. Practitioners screen and recruit participants, sometimes through classified ads, paying $50 to $60 for a three-to-four-hour session as described by trial firm Aitken Aitken Cohn, rent facilities, build demonstratives, script attorney presentations and staff an observation room. The project expenses alone, one consultancy notes, can approach or equal the billable consultant fee, per IMS Legal Strategies.

The economics forced rationing. Because a single exercise could consume tens of thousands of dollars and weeks of calendar, mock trials were reserved for catastrophic exposure, the case already deemed too dangerous to settle quietly. Routine claims, mid-size coverage disputes and the long tail of bad-faith allegations went to trial cold, with witness preparation amounting to a conference-room read-through and an adjuster's gut feel about how a deposition might land. Depositions themselves, the moments where claims most often unravel, were rehearsed by having a colleague lob a few predictable questions across the table.

That would matter less if juries behaved as they once did. They do not. The Swiss Re Institute found that social inflation drove up U.S. liability claims by 57% over the past decade, reaching an annual peak of 7% in 2023, when 27 court cases each awarded more than $100 million, according to Swiss Re Institute. The same analysis found U.S. commercial casualty losses grew at an average annual rate of 11% to reach $143 billion in 2023, per Swiss Re Institute. The whole U.S. tort system, by the U.S. Chamber's Institute for Legal Reform, cost $529 billion in 2022, about 2.1% of GDP and $4,207 per household, growing 7.1% a year, according to the Institute for Legal Reform. Preparing once, late and expensively was never ideal. Against a verdict curve bending this steeply, it became a liability of its own.

The Shift: Personas That Push Back

What changed is that generative systems became good enough to hold a role under pressure. An agentic panel is not a chatbot answering questions; it is a set of persona-driven agents instructed to stay in character, defend a position, react to evidence and behave with the stubbornness of a real adversary or the wandering attention of a real juror. A defense team can depose a synthetic claimant who has been briefed on the plaintiff's actual narrative, cross-examine an AI expert witness tuned to a hostile methodology, and then run the closing past a simulated jury that returns reactions, hesitations and a verdict, in an afternoon, repeatedly, at marginal cost.

The underlying enthusiasm is industry-wide. Generative AI adoption inside insurance roughly doubled year over year, with 55% of respondents reporting early or full adoption in 2025, and 90% somewhere in evaluation, according to research firm Conning. In Europe, roughly 65% of insurance undertakings are already actively using generative AI systems, with another 23% planning to within three years, per the European Insurance and Occupational Pensions Authority. On the legal side that defends these claims, the Thomson Reuters Institute found organizational use of generative AI nearly doubled to 22% in 2025 from 12% in 2024, with law-firm legal usage rising to 26% from 14%, and 95% of professionals expecting the technology to be central to their workflow within five years, according to the Thomson Reuters Institute.

The events that hurt insurers most are rare, ruinous and almost impossible to rehearse affordably. Agentic panels are the first serious attempt to fix that.

What It Looks Like Now

In practice, a defense team building a claims-trial strategy moves through a workflow that would have been unthinkable at scale three years ago. They feed the case file, pleadings, the claimant's deposition transcript, medical records, the policy language, into an agentic environment and stand up the relevant personas. The deposition rehearsal comes first: counsel questions a synthetic claimant or expert, and the system flags contradictions, evasions and openings, mirroring what AI-assisted deposition tools already do when they surface gaps and inconsistencies in the record, as documented by legal-technology practitioners writing for Clio.

Then comes the harder test: argument and reaction. Teams run competing framings of a coverage exclusion or a bad-faith allegation past a jury panel calibrated to the demographics and attitudes of a likely venue, and watch where sympathy moves. This matters because venue is destiny, Texas alone produced 23 nuclear verdicts in 2024, followed by California with 17 and Pennsylvania with 12, per Marathon Strategies via Insurance Journal. Jury-reaction modeling against this geography lets insurers price the tail risk of a case before deciding whether to settle or fight. And because behavioral research shows jurors anchor on injury severity rather than corporate culpability, a 2025 Swiss Re behavioral study found respondents awarded damages against small and midsize companies at nearly the same rate as large corporations when injuries were serious, as reported by MoneyGeek, testing the emotional arc of a defense, not just its legal logic, becomes the point of the exercise.

The recurring theme is iteration. A live focus group answers one question once; an agentic panel lets a team ask twenty variations of the same question and watch the distribution of outcomes. That maps neatly onto how settlement actually works. With the overwhelming majority of claims resolving before verdict, the deliverable is rarely a courtroom win, it is a sharper reserve estimate and a more defensible settlement range. Simulation evidence from adjacent fields supports the instinct that structured practice transfers: a randomized study published in Nature found AI-mediated simulation produced significantly greater skill improvement than human-led instruction in early training sessions, according to research in npj Digital Medicine.

Reading the Numbers Honestly

None of this works if the data behind it is treated loosely, and the nuclear-verdict statistics carry real caveats worth stating plainly. No independent federal dataset tracks nuclear verdicts; the most-cited figures come from industry-affiliated research, and the verdict totals do not reflect later reductions, remittiturs or reversals, as MoneyGeek notes. Median figures also vary by source and methodology, the Institute for Legal Reform's ten-year study of 1,288 personal-injury and wrongful-death verdicts found a median of $21 million against an $89 million average, per MoneyGeek, while the median rose from $19.3 million in 2010 to $24.6 million in 2019, a 27.5% climb that outpaced the 17.2% inflation of the period, according to Institute for Legal Reform data via Claims Journal. An agentic panel inherits whatever assumptions go into it; calibrating jury behavior to inflated or unrepresentative numbers simply produces a confident, wrong rehearsal.

The Next Few Years

Expect three shifts. First, normalization: as generative tooling moves from pilot to plumbing, the Thomson Reuters Institute found 95% of professionals expect it central within five years, and insurers are racing in at similar speed, persona rehearsal will stop being an exotic add-on for catastrophic cases and become a default step in claims handling, applied to mid-size disputes that never justified a live mock trial. Second, integration: deposition simulators, jury-reaction models and coverage-argument testing will converge into a single environment that follows a claim from first notice of loss through mediation, with each rehearsal feeding the reserve. Third, scrutiny: as outputs start influencing settlement decisions and reserves, regulators and courts will ask how the personas were built, what data trained them, and whether they encode bias.

That last point is the load-bearing risk. A jury persona calibrated on skewed data can launder demographic stereotypes into a "data-driven" recommendation; an over-confident synthetic claimant can lull a team into a cross-examination strategy that collapses against a real, unpredictable human. The European supervisory community has already flagged that most generative deployments in insurance remain at proof-of-concept stage, per EIOPA, a useful reminder that adoption enthusiasm is running ahead of validation. The realism limit is fundamental: an agentic panel models a plausible jury, not the actual twelve people who will decide the case, and treating the simulation as prophecy rather than practice is the failure mode to guard against.

The honest framing is the one borrowed from the cockpit. No pilot believes the simulator is the flight, but no airline would put a pilot in the air without thousands of simulated hours. For an insurance industry facing a verdict curve that bent 116% in a single year, the question is no longer whether to rehearse the courtroom, but how to do it often enough, cheaply enough and skeptically enough that the live performance holds. Agentic panels do not promise to tame the nuclear verdict. They promise something more modest and more useful: that insurers stop walking into the most expensive room in American commerce having never run the play.

Sources

1. Insurance Journal, "Corporate Nuclear Verdicts Surged to New Record High in 2024" (Marathon Strategies report)
2. MoneyGeek, "Nuclear Verdicts and Small Business Insurance"
3. Claims Journal, "Navigating The Challenging US Liability Litigation Environment" (Sedgwick)
4. Claims Journal, "The Impact of Nuclear Verdicts on Casualty Claims"
5. IMS Legal Strategies, "What Does a Mock Trial Cost?"
6. Aitken Aitken Cohn, "Focus Groups on a Shoestring Budget"
7. Swiss Re Institute, "Litigation costs drive US liability claims by 57% over past decade"
8. U.S. Chamber Institute for Legal Reform, "U.S. Households Pay the Price of Rising Tort Lawsuit Costs"
9. Conning, "2025 Survey on AI & Insurance Technology"
10. EIOPA, "Generative AI Market Survey: Outlook, Use Cases and Risk"
11. Thomson Reuters Institute, "2025 Generative AI in Professional Services Report"
12. LawSites, Thomson Reuters survey coverage (legal adoption detail)
13. Clio, "Deposition Prep in the Age of AI: From Overload to Clarity"
14. npj Digital Medicine (Nature), "Combining real-time AI and in-person expert instruction in simulated skills training"