By Jason Weber, CEO of Litmas AI and experienced litigator
For two decades, the corporate defense bar perfected one of the most potent structural defenses in modern American law: the class-action waiver, embedded in consumer and employment arbitration clauses. This potency largely derives from the incredible success of appellate advocacy at the United States Supreme Court that resulted in huge doctrinal wins like the rules that came out of from Concepcion and Epic Systems.1 The premise of this defense was straightforward: if a dispute had to be resolved in arbitration, one claimant at a time, then no plaintiff firm could afford the per-claim cost of bringing low-dollar consumer or employment claims. The math, on paper, was supposed to kill aggregation at the source.
What no one fully war-gamed was what happens when a plaintiff firm decides to actually do the math the other way. Twenty thousand individual arbitration demands, and potentially twenty thousand provider filing fees. Twenty thousand discrete claims a defendant has to evaluate, answer, and either fight or pay, before any of them have been adjudicated. The arbitration clause built to be a shield turned out to be a delivery mechanism. And the firms on both sides of this dynamic, plaintiff and defense, are now operating in a procedural environment that almost no piece of legal technology was built to handle.
Mass arbitration has its own version of the brutal math that defines mass litigation, but the math runs differently. In MDLs and class actions, the cost pressure sits with whichever side is doing the most factual-development work. In mass arbitration, the cost pressure sits with whichever side is currently writing the larger check to the arbitration provider. Filing fees are not a theoretical concern. They are often paid up-front before merits, discovery, or anything resembling adjudication, and they scale linearly with claim volume. A respondent facing ten thousand demands is looking at filing-fee exposure that, in many AAA matters, can run into eight or nine figures before the first hearing is calendared. A claimant firm staking the capital to put those ten thousand demands on file is committing real numbers on the opposite side of the ledger. Both sides are looking at line items that change what a settlement conversation actually looks like.
The plaintiff side has its own brutal math, just running the other direction. Filing those ten thousand demands requires actually preparing them. Each one is technically an individual proceeding. Each one has its own claimant, its own facts, its own statute-of-limitations clock, its own opt-out validity question, and its own statutory-damages calculation. A plaintiff firm running a mass arbitration is not running a class action; it is running ten thousand individual cases simultaneously, with no aggregation device to compress the operational load. The whole leverage thesis works only if the firm can credibly produce, file, and prosecute that volume at a per-claim cost low enough for the economics to hold.
Whoever moves first on collapsing that per-claim cost, whether it is a plaintiff firm, defense organization, or third-party administrator, is going to define the next decade of mass arbitration practice. That calculation is about to change.
Where Mass Arbitration Is Headed
Anyone who has actually worked a mass arbitration on the plaintiff side already knows the structural problem. The firm with five thousand wage-and-hour claims is not running five thousand cases. It is running one case five thousand times, with the parts that should distinguish one claimant from another flattened into spreadsheets because no one has the bandwidth to do otherwise. The same compression happens on the defense side, where the analytical work that would actually win individual claims at the margin gets sacrificed to volume. Defense organizations facing high-volume arbitration dockets are running the same calculation in reverse — deciding which incoming demands get individualized review and which get processed against templates that opposing counsel built to be processed against.
What AI changes is the unit economics, and the privacy-statute wave that arrived in the last three years has made that change urgent rather than theoretical. BIPA in Illinois,2 VPPA at the federal level,3 CIPA in California,4 and a growing pile of state mirror statutes have produced a type of claim of claim that is tailor-made for mass arbitrations: one providing for high statutory damages applicable to subject matter often controlled by a consumer or employment contract containing an arbitration clause. Unsurprisingly, these claims have driven a wave of new mass arbitrations, or at least a litany of threatened ones. Moreover, plaintiff firms have begun building dedicated practices around the opportunity. Defendant companies that previously thought of arbitration clauses as a tail-risk hedge are increasingly finding them mobilized as the primary exposure vector for entire product lines. Both observations describe the same market; the parties just sit on opposite sides of it.
If the marginal cost of preparing an individualized claimant file collapses by an order of magnitude, then the firm that previously could afford to file a thousand of a particular kind of demand can file ten thousand. The firm that could afford ten thousand can file a hundred thousand. And with advanced AI litigation workflow systems like Litmas AI coming online, who is willing to definitively say where the true ceiling sits? The same math runs on the defense side. A respondent that previously had to triage which incoming demands got real analytical treatment can give every demand that treatment when the marginal cost of doing so collapses. Settlement leverage moves toward whoever can credibly evaluate every file, on every side, at every stage. The firms doing this work in 2030 will not look like the firms doing it today.
The first wave of that change is already visible. Plaintiff firms in consumer protection, employment, and privacy-statute practice have started deploying AI tools for claimant intake, opt-out-clause analysis, statutory-damages modeling, and pattern recognition across claim cohorts. Defense organizations have responded with their own deployments, often centered on per-claim merits triage, settlement-matrix valuation, and process-arbitrator strategy under the AAA’s 2024 Mass Arbitration Supplementary Rules. Third-party administrators have begun reorganizing around the same workflows.
What has not changed, yet, is the architecture of the tools doing this work. Most of what is being marketed as legal AI for mass arbitration is a general-purpose language model with a thin practice-area wrapper, sometimes layered onto a generalist legal practice-management system that was originally built for personal injury intake or insurance defense billing. That works for some workflows. It does not work for arbitration at mass arbitration’s leviathan scale, and the firms that have tried to run mass arbitration dockets through general-purpose tools are probably learning that the hard way.
Why Generic AI Will Not Get You There
A general-purpose language model is, by design, a probabilistic system. It produces the most plausible next text given the context it has been given. In conversational use, that produces output that is fluent and usually right. For lawyers running a mass arbitration, usually right can be a category-five error. A demand letter that misstates a statutory damages provision in one filing out of a thousand is not a tool that works 99.9 percent of the time; it is a tool that has materially misstated the claim across an entire cohort. A defense response that hallucinates an opt-out provision in one out of a thousand answers is not 99.9 percent reliable; it is a tool that has waived a defense across an unknown subset of the docket.
That problem becomes existential at mass-arbitration scale, because mass arbitration is run not on a single master pleading but on thousands of individualized filings, each of which is procedurally independent. The verification burden is precisely what makes mass arbitration so operationally expensive in the first place. The whole point of bringing AI into this work is to dissolve that burden. A tool that creates new verification problems while solving old ones is not, on balance, helping.
There is a second problem more specific to mass arbitration than to mass litigation: the procedural mechanics are now genuinely complex. AAA’s 2024 supplementary rules introduced a process-arbitrator track that handles preliminary administrative matters across batches of demands.5 JAMS published a parallel set of mass arbitration procedures.6 SB 707 in California, preserved against FAA preemption by the Supreme Court of California’s 2025 decision in Hohenshelt v. Superior Court, installed a no-cure fee-default regime at the front of the arbitration timeline: automatic material breach if the drafting party “willful[ly], grossly negligent[ly], or fraudulent[ly]” misses the 30-day payment deadline.7 Similar legislation is likely to be moving soon, if not already moving, in other states. A generalist legal AI does not know any of this, because it was not trained on it and because it has no procedural memory across cases. A purpose-built mass-arbitration workflow system can know all of it, encode it as procedural state, and run the procedural calendar as a first-class object across thousands of sub-cases at once.
The firms that figure out mass arbitration in the AI era will be the firms that adopt purpose-built tools like Litmas AI, not general assistants. They will use systems constrained to verified statutes and arbitration-provider rules, that produce traceable per-claim records, that track procedural state at the master-case and sub-case level, that adapt to provider-specific and state-specific procedural overlays without being told, and that fail in ways litigators can see coming. The architecture is fundamentally different from what most of the market is currently selling, and the gap is widening.
What This Means for Firms Right Now
The honest position, for any firm doing mass-arbitration work on either side of the v., is that the next five years will reward early experimentation and punish late adoption in roughly the same way the last fifteen years rewarded early e-discovery adoption and punished firms that stayed on paper.
Whether AI is ready is not the useful question. The useful question is which tools, deployed where in the workflow, change your firm’s capacity in ways that compound. Intake automation that turns a week of non-attorney staff review into ninety minutes of attorney review is not a feature improvement. It is a business-model change. A plaintiff firm that can credibly file and prosecute a docket five times larger than its current capacity, without proportional growth in overhead, is competing for different cases at different settlement leverage points. A defense firm that can run individualized merits review on every demand in a five-thousand-claim docket, not just the bellwethers, is no longer negotiating from a position of triage. The leverage equilibrium of an entire mass arbitration shifts when either side gains that capability before the other.
Some firms will figure this out inside the next eighteen months. Others will spend the next three years buying tools that promise transformation and deliver friction. The differentiator will not be which firms moved fastest. It will be which firms thought hardest about what they actually needed, and executed effectively on that deliberation.
The Million Case Question
The central question is not what can AI do for mass arbitration. It is what does mass arbitration become when AI is doing it well? The leverage thesis underlying the mass arbitration model, plaintiff or defense, depends on per-claim operational cost. When that cost collapses, the strategic premises change. Claims that were never economically viable to file become viable. Defenses that were never economically viable to assert across an entire docket become assertable. The settlement-matrix valuation that used to be a back-of-envelope exercise becomes a portfolio-level analytical product. The supervision burden under ABA Formal Opinion 512 becomes a manageable workflow rather than an open compliance question.8
Ten thousand demands at once is not science fiction. It is a Tuesday in a firm that has built its workflow around the right tools. The frontier is closer than most of the profession has admitted to itself. The firms that will define what mass arbitration looks like in 2030 are already building toward it. So are the platforms that will serve them, and Litmas AI is one of them.