The State of
PI Case Valuation
Jurisdiction-by-jurisdiction data on the asymmetry between plaintiff attorneys and the AI tools insurance carriers have used for years.
A free annual report from predict.law, the jurisdiction-tuned settlement-prediction model built for plaintiff attorneys. This page previews the first edition. The full report is free when it ships.
90–92% accuracy (MdAPE) on a held-out 2024–2025 test set, 90% confidence interval via quantile regression. How we measure accuracy →
The plaintiff bar didn't lose to better lawyers. It lost to better data.
For ten years, insurance carriers have been pricing claims with proprietary AI trained on the entirety of US settlement and verdict history. Plaintiff attorneys have been pricing the same cases with Verdict Search lookups and a partner's intuition. This is the asymmetry that defines a generation of personal-injury practice.
In 2014, the largest US property-casualty insurers began running claims-side machine-learning models over their internal settlement datasets. Those datasets, by virtue of being defense-side, already incorporated jury outcomes, judge composition, plaintiff-counsel reputation, and per-county settlement medians. Over the years that followed, carrier-side claims models moved from pilot to standard practice, setting reserves on a large share of US MVA claims and a growing share of premises liability matters. The full report documents how we sized that adoption and what it left for the plaintiff side to recover.
The plaintiff side, meanwhile, was operating with the same tools it had in 1995: Verdict Search lookups, Jury Verdict Reporter subscriptions, partner-level institutional memory, and the occasional outsourced settlement-evaluator engagement at $1,500 per case.
The result is the most consequential information asymmetry in modern personal-injury practice. It manifests at every settlement-negotiation table, in every demand-letter exchange, and — most expensively — at the intake call where the case-selection decision is made before any pricing data exists.
This is the gap Predict closes. We have spent the last 24 months assembling the largest plaintiff-side case-outcome dataset ever compiled: 312,000 jury verdicts and reported settlements across 50 jurisdictions, 2018–2025, with case-history attributes structured for predictive modeling. Trained on this data, the Predict model produces jurisdiction-tuned settlement value estimates with explicit confidence bands and full methodology disclosure.
This first edition of the State of PI Case Valuation reports what we've learned. Where the carrier-vs-plaintiff asymmetry is widest. Where settlements diverge most from jury verdicts. Where the data is dense enough to predict reliably, and where it isn't yet. And, for the first time, how far first offers sit below the value our model assigns the same case profile, jurisdiction by jurisdiction.
Every figure in the report is shown with its 90% confidence band and a full methodology disclosure, the same way the model reports a number in-product.
Median MVA settlement, low-speed rear-end, by jurisdiction
Carrier first offer vs. Predict median settlement value, by jurisdiction
Reserve your copy. Free for plaintiff attorneys.
~30 pages. Jurisdiction-by-jurisdiction breakouts and the carrier-vs-plaintiff asymmetry numbers, every figure shown with its 90% confidence band. We'll email the full report the day it ships, December 2026. No trial, no card.
Partners: get the co-branded edition →
Want a jurisdiction-tuned number on your own cases before December? Start a 14-day free trial of the prediction model → The trial gives you the live model today, no card charged. The report gives you the jurisdiction-by-jurisdiction analysis when it ships.