Last week’s blog examined whether pretrial discovery methods — and depositions in particular — have been responsible for the dramatic decline in civil trials. While the available evidence suggests that liberal discovery rules clearly contribute to pretrial resolutions, nobody has measured exactly how much. Meanwhile, two powerful forces are quietly reshaping how civil claims are valued and resolved: the insurance industry’s evolving dependence on deposition transcripts, and the arrival of artificial intelligence in litigation strategy. Both deserve closer examination because understanding them may reveal where pretrial settlement rates in civil litigation are headed next.
Predictive analytics may reshape settlement decisions, but they cannot replace the human insight captured in a deposition.
The Adjuster’s Dilemma
In cases involving insurance coverage — which account for a substantial share of civil litigation in America — no settlement happens without the insurance adjuster’s approval. The adjuster, armed with a claims file and a settlement authority granted by the insurer, stands between the litigants and the resolution of the underlying dispute. Understanding how adjusters reach their valuations may shed light on why certain kinds of pretrial discovery matter more than others.
Claims adjusters investigate claims, assess damages, and negotiate settlements — but they cannot authorize payment unilaterally. Adjusters are assisted and supervised to some extent by a claims examiner who reviews the adjuster’s work and approves the claim before the adjuster can extend an offer. If the claimant contests the settlement, adjusters coordinate with attorneys and expert witnesses to defend the insurer’s position.
The adjuster’s investigation draws on every available source: police reports, medical records, bills, witness statements, and background checks. A thorough adjuster will not make a settlement offer without having first reviewed everything necessary to value the claim.
When a case enters litigation, deposition transcripts become a critical component of the claims file. Transcripts shed light on witness credibility, expose inconsistencies in testimony, and allow adjusters to assess the risk that a jury might return a large verdict. For the adjuster, a deposition transcript is vital to answering the two questions that drive every settlement decision: What are the claimant’s chances of winning at trial? And how much might a jury award?
In this light, depositions function as more than a discovery tool for lawyers. They serve as the raw material that adjusters need to justify settlement authority — to their examiners, to their managers, and increasingly, to the software systems that help them calculate what a claim is worth.
When Algorithms Set the Price
Recently, the insurance industry’s embrace of claims valuation software has fundamentally changed how adjusters do their work. The most widely known platform, Colossus, was developed by Computer Services Corporation and adopted by many major insurers. Adjusters input injury data — diagnosis codes, treatment types, duration of care, geographic location — and the software calculates a settlement value based on data collected from thousands of previous claims and settlements in comparable situations.
Colossus and its successors employ sophisticated algorithms that track numerous variables believed to affect claim value: the county where the accident occurred, recent settlement figures for similar injuries, and recent trial verdicts in local courts. Colossus evaluates both claim data and contextual information to generate a recommended settlement range. Colossus is not without its critics, however, and some members of the plaintiff’s bar believe that it is biased against claimants.
The implications for deposition practice are significant. If an adjuster’s settlement authority derives in part from an algorithm’s output, then the information captured in deposition transcripts matters only to the extent that it feeds — or overrides — the variables driving the Colossus algorithm. A devastating deposition that exposes a key witness as unreliable may prompt an adjuster to seek higher settlement authority. But that request must often compete with the number that Colossus has already generated.
This tension between human judgment and algorithmic output defines much of modern claims handling. The adjuster’s role has evolved from independent evaluator to something closer to a data operator managing inputs for an increasingly automated valuation process.
Predictive Analytics Enters the Courtroom
The algorithmic transformation visible in insurance claims handling has a parallel on the litigation side. AI-powered predictive analytics tools now offer lawyers something previously available only through years of trial experience: data-driven forecasts of how contested civil matters will be decided if, and when, they are tried.
These tools analyze court filings, reported jury verdicts, settlement data, and judge-specific ruling patterns to predict likely outcomes. Artificial intelligence will analyze court filings and jury verdicts, reports of settlements, and other data to draw predictions on how a judge may rule or to predict the lawsuit’s likely settlement value.
According to one litigation technology vendor, current predictive AI technology has 80-90% accuracy in forecasting legal case outcomes.
The Convergence No One Predicted
Here is where the two threads — insurance valuation and legal AI — begin to converge in ways that may accelerate pretrial resolution even further.
On the insurance side, Colossus-type software already functions as an early-stage predictive analytics engine. It ingests historical settlement and verdict data, applies it to current claims, and generates recommended valuations. On the litigation side, tools like Lex Machina perform a similar function for attorneys — analyzing judicial behavior, verdict patterns, and comparable outcomes to guide strategy.
When both the adjuster’s software and the attorney’s analytics tools draw from the same underlying data — historical verdicts and settlements — the parties’ outcome predictions should converge more quickly than ever before. And convergent predictions, as the information exchange theory discussed in last week’s blog suggests, promote settlement.
If true, AI may function as a powerful accelerant for the vanishing trial, narrowing the expectation gap between parties earlier in the litigation cycle and reducing the informational role that depositions have traditionally played. A 2025 paper from CanLII, “The Impact of Artificial Intelligence on Access to Justice: Predictive Analytics and the Legal Services Market,” begins to explore these dynamics, examining how AI-driven prediction affects the broader legal services market.
What We Don’t Know
For all the promise of predictive analytics, the honest assessment is sobering: no comprehensive study has yet measured whether AI tools actually increase settlement rates in civil litigation. The technology is young. Adoption remains uneven. And the most important question — whether algorithmic prediction substitutes for, or merely supplements, the information that depositions provide — remains unanswered.
There are reasons for caution. AI predictions depend on historical data, and historical data reflects the legal landscape that produced it. As courts, rules, and social conditions change, backward-looking models may miss forward-looking risks. AI hallucinations and bias remain documented concerns. A recent Stanford University study found troubling hallucination rates in general-purpose AI chatbots, and legal-specific tools were also found to carry their own accuracy risks. A growing number of state bar associations have released guidance on AI use, and federal and state courts are demonstrating no patience at all with AI-generated misstatements of law and fact.
Perhaps most importantly, depositions capture something that algorithms cannot: the human element. A witness who crumbles under cross-examination, an expert whose confidence exceeds the data, a plaintiff whose sincerity resonates — these intangibles shape settlement calculations in ways that no historical dataset can fully replicate. The adjuster who reads a deposition transcript gains insight that no software output can provide on its own.
