Negotiating When Your Comp Database Is Trial-Biased

TL;DR — Stop using public verdict databases to set claims reserves. Verdicts suffer from selection bias and represent the failure of negotiation. Calibrate your day-one reserves and defense strategy using the actual settlement value of comparable resolved cases.

Walk into any mediation, and you will see binders full of jury verdicts. Both sides bring them. The plaintiff attorney has a stack of nuclear verdicts from the past eighteen months. Defense counsel has a curated list of zero-dollar defense verdicts. They slide these numbers across the table as if they represent the market value of the claim. They do not. A jury verdict is the definition of an outlier. It is the artifact of a case where the settlement process failed. Basing your settlement strategy on trial verdicts is a fundamental error in claims management. It distorts day-one reserves, misdirects defense spend, and creates false anchors during negotiation.

The industry relies on verdict databases because verdict data is public. Settlement data is proprietary, highly guarded, and difficult to standardize. Carriers and their defense firms build their negotiation models on the data they can easily buy, rather than the data that actually reflects how disputes are resolved.

The Selection Bias Problem

This is a well-documented phenomenon in legal economics, often referred to as Priest-Klein selection bias. Cases do not proceed to trial randomly. They go to trial when the plaintiff and the defendant have mutually exclusive, highly confident predictions about what a jury will do. If both sides agree a case is worth fifty thousand dollars, they settle. The cases that reach a verdict are the ones defined by extreme uncertainty, novel legal theories, unbridgeable gaps in liability assessment, or profound miscalculations by one side. When a claims professional uses a verdict database to set a reserve, they are pricing the routine claim based on the most volatile, unpredictable subset of litigation.

You are taking the temperature of the sickest patients to determine the average body temperature of the general public. The result is a reserve that is either wildly over-inflated to account for a phantom nuclear verdict or dangerously under-funded because you anchored to a string of defense wins. This forces claims teams into a reactive posture. They wait for discovery to close before adjusting the reserve, missing the window for early resolution.

The distortion compounds when you factor in the difference between a jury award and the actual amount paid. A headline-grabbing fifty-million-dollar verdict is rarely the final check cut by the carrier. Appeals, high-low agreements, collectability issues, and policy limits drastically alter the final payout. Yet, the unadjusted verdict number is what gets recorded in the public database. It is what gets brought to the mediation table. If you negotiate against that public number, you concede ground before the conversation even begins.

Negotiating From the Dark Data

To negotiate effectively, claims teams need to anchor to reality. Reality is the large majority of cases that settle. The challenge is that settlement data is messy. It lives in unstructured case files, scattered across claims systems, buried in thousands of pages of pleadings, medical records, and correspondence. Extracting the variables that actually drive settlement value requires a level of file review that human adjusters simply do not have the time to perform on every single claim.

At Canotera, we built a forecasting platform to solve this exact structural deficit. Generative AI reads the case file. It does the heavy lifting of reading and structuring the unstructured text across those thousands of pages, extracting the critical facts. The AI does not make the prediction. It translates the messy reality of a claim into structured inputs.

Separate mathematical machine-learning models then produce the forecast. These models are trained on large volumes of resolved cases with known, actual outcomes. We deliberately separate verdict-heavy comps from settlement-anchored data to avoid selection bias. The system produces a calibrated settlement range, an escalation probability, and a reserve delta compared to the current manual reserve. It surfaces comparable resolved cases and the specific drivers behind the numbers. Every output is entirely traceable to the source documents. You are not asking a black box for a point estimate. You are using a geometric model to define the honest uncertainty bands of a settlement.

Changing the Operating Model

This approach fundamentally changes the operating model for claims and litigation leaders. When you know the realistic settlement range on day one, you stop reacting to the plaintiff’s anchors. You stop paying defense counsel to litigate cases that should be resolved in month three. We see the financial drain of misallocated defense spend every day. A carrier will spend sixty thousand dollars fighting liability on a claim, only to settle on the courthouse steps for the exact amount the plaintiff demanded a year earlier. They do this because the initial reserve was set based on an optimistic reading of a verdict database, rather than a dispassionate analysis of comparable settlements.

Social inflation and third-party litigation funding thrive on this inefficiency. Litigation funders use sophisticated data models to identify which cases to back. They know the statistical value of a settlement, and they know how to apply pressure to force a carrier into a bad decision. If carriers respond by relying on gut instinct and skewed public verdict data, they will continue to lose ground. Funders operate with clinical precision. Carriers must do the same. Early detection of escalation allows the claims team to deploy senior adjusters and specialized defense counsel where they are actually needed, rather than spreading resources thin across the entire portfolio.

The goal is not to eliminate human judgment from the claims process. The goal is to arm the claims professional with a baseline that is mathematically sound and grounded in actual dispute resolution. When you bring a calibrated settlement range to mediation, backed by traceable drivers from comparable resolved cases, the conversation changes. You force the other side to argue the merits of the specific claim, rather than trading competing stacks of irrelevant trial anomalies. You negotiate from data, not from fear.

A jury verdict is what happens when negotiation fails. You cannot use the failure condition to price the cure.