Why AI Could Turn Small Law Firms Into Billion-Dollar Businesses

Host Alex Rampell opens by contrasting the legal world familiar to Silicon Valley—large corporate firms such as Wilson Sonsini, where lawyers may charge up to $2,000 an hour—with plaintiff law, represented by highly visible firms such as Morgan & Morgan. Jay Madheswaran defines Eve as “legal AI for real people,” rather than for institutions or in-house legal departments that already have substantial resources. Jay explains that plaintiff lawyers generally work on contingency: they are paid only if they win, usually receiving a portion of the settlement. This model gives individuals some ability to challenge institutions, but it does not eliminate the underlying labor constraint. Jay says a lawyer typically will not accept a case unless it could generate at least $5,000 in fees, and even then the firm may lose money if the case consumes too much staff time. A small employment claim, for example, can require a 30-to-45-minute intake call, roughly 10 hours of legal research, fact gathering, and follow-up questions before the firm knows whether there is a viable cause of action. Jay says 96% of labor and employment cases are rejected, often because the potential recovery cannot support the required work.

Alex uses supply-and-demand economics to describe the opportunity. Someone with $1,000 in damages cannot pay an hourly lawyer, and a contingency lawyer may still decline because taking a third of the recovery will not cover paralegal and attorney costs. A fender-bender below a $3,000 or $4,000 deductible may be financially significant to an individual but economically unattractive to a law firm. The result is a market failure: people need representation, but firms cannot afford to supply it. Jay agrees that legal work is bottlenecked by labor, not by demand. Eve is designed to answer the initial call, use the firm’s own criteria and legal knowledge to decide whether the case is worth taking, and begin the preliminary work or demand immediately. Over a case that may last six months or a year and involve roughly a thousand actions and exchanges, Eve is intended to keep the matter moving from intake through settlement. Jay argues that this changes AI from a cost-cutting tool into a revenue-expansion tool: a firm can represent more clients and pursue cases that were previously too small to be viable. Alex illustrates the point with a story about meeting a trial attorney at Blue Bottle. After Alex described Eve as a way to make a firm more money rather than simply fire employees, the attorney approached him and said he wanted to try the product.

Alex points to the large number of plaintiff-law advertisements on highways such as California’s 101 and the East Coast’s I-95 and asks what happens after a firm pays for that marketing. Jay says the fundamental math is simple: representing more people generally produces more revenue, but firms are often forced to reject most inbound demand because they lack enough people to service it. He describes several stages of operational bottlenecks. A very small practice may have one attorney and an office or case manager taking calls, recording notes, and waiting for the lawyer to respond between other tasks. A more organized firm may create an intake team or call center, but then increased advertising simply overwhelms that team. Eve consolidates intake, qualification, case routing, and ongoing case work so lawyers are brought in for the judgments and interventions that require their expertise. Jay gives one concrete result: a Los Angeles firm increased the number of clients managed per attorney from about 100 to 500. The firm used Eve from the initial phone call and qualification through the removal of handoffs between attorney and case manager, and through settlement preparation. Eve can brief an attorney before mediation by identifying the issues to emphasize, likely settlement outcomes for comparable matters, and tactics that have worked previously. Jay says the attorney’s role increasingly resembles that of an engineer or data scientist improving a system, while still focusing on client outcomes.

Alex argues that investors could easily misunderstand plaintiff law by looking only at its historical software spending. A large corporate firm may need time tracking because it bills by the hour, while a plaintiff firm traditionally needed little more than a cell phone, an email address, and perhaps Microsoft Word. Because plaintiff lawyers do not bill clients for every hour, and because they do relatively little collaborative redlining, conventional legal software did not address their main constraint. Jay agrees that historically plaintiff law did not require much more than those basic tools. Legal software companies and large AI products largely concentrated on defense institutions because those organizations had larger technology budgets and were easier to sell to. Plaintiff practices are often small businesses whose owners watch monthly spending closely; if software does not produce useful outcomes immediately, they will not renew it. Eve’s premise is that AI can encode work that previously had to be performed by people. Every legal case is different enough that earlier attempts to systematize operations still required hiring smart employees who could handle exceptions. Jay says modern AI can let one person—or one firm—manage hundreds of cases with different facts and circumstances without holding every detail in a human brain. Eve therefore represents a “shadow labor market”: its opportunity is tied not to the historically small software budget, but to the much larger value of legal outcomes that firms could not previously deliver because labor was unavailable.

Alex asks why Eve would be defensible if general-purpose models such as ChatGPT or Claude continue improving. Jay says Eve is happy to use better underlying models, but raw intelligence is not enough. Even if a law firm could hire “a thousand Albert Einsteins,” Einstein would not immediately be productive because he would lack the firm’s workflows, local context, case history, tools, and knowledge of how a particular opponent behaves. Plaintiff-law data is also unusually difficult to obtain from public sources. Jay estimates that less than 1% of cases settle and become published case opinions; most resolve before litigation, before a demand, or through informal negotiation. A car-accident example shows why that information matters: an insurer such as Geico may know that it paid $29,000 in a particular fact pattern, but that settlement will not appear in a searchable court opinion. Insurers use systems such as Colossus to estimate injury values and determine what they might get away with paying, while individual lawyers usually lack comparable collective information. Eve is meant to organize that multidimensional knowledge: the facts of the case, local courts and opposing counsel, insurer tactics, and the action most likely to improve the result at each point. Jay says Eve does not train on a firm’s privileged case data. Instead, Eve follows existing industry norms: firms keep their own case information private, while attorneys share general tactics through professional education and discussions. Eve’s Auditor product reviews cases page by page and looks for information a perfect attorney should have caught.

Alex asks how insurers and other institutional defendants will respond once individuals have AI-assisted representation. Jay says defense institutions still have more money, time, and centralized information, and they can exploit an individual’s need to pay fees or settle quickly. But if the plaintiff’s side can perform much more work, cases will have more “teeth,” and routine exchanges may increasingly occur between agents. Jay describes a current example. A defense firm repeatedly sent discovery requests just before Thanksgiving or another major holiday, knowing that the plaintiff’s lawyer would be away or overloaded and might respond poorly. After seeing this tactic repeatedly, an Eve customer created an agent that responded the same day. The defense side then reacted with surprise because the timing tactic no longer created the intended advantage. Jay expects lawyers to spend less time on repetitive exchanges and more time on advocacy: understanding the client, the facts, the local legal environment, and the particular opposing party. He estimates that 99% or even 99.9% of representation work could eventually be performed by AI, while the remaining human involvement still materially affects the outcome. Eve customers are already using the time saved to prepare for trial, interview jurors, and decide how strongly to pursue a case instead of treating trial preparation as something that would consume the entire firm. The result, in his view, is not the disappearance of lawyers but a shift toward higher-value judgment and advocacy.

Alex connects Eve’s data to decisions that lawyers and litigation-finance companies already try to make: whether to accept a $12,000 offer from Geico, push for $15,000, or risk trial for a potentially much larger recovery. Historically, settlement amounts and tactical information were scattered among individual firms. Litigation financiers have begun using judge, venue, and case data to decide which matters to fund, but Eve could make that analysis available during everyday legal practice. Jay says Eve can help determine when an optimal outcome has been reached, while leaving responsibility with the attorney to decide whether to push further. He believes law will increasingly use automated dispute resolution, especially in areas already handled outside ordinary courts through administrative processes, mediation, or arbitration. AI-enabled neutral parties could assemble relevant historical facts and accelerate decisions, while humans remain responsible for removing bias and exercising judgment. The conversation then turns to pricing. Alex notes that plaintiff law has always been outcome-oriented, whereas many other legal fields still bill by the hour because neither side knows in advance how much work a matter will require. Jay says Eve originally used a conventional seat-based pricing model, but within roughly a year and a half the company moved toward pricing around outcomes because firms care about serving more clients and winning more cases. He does not expect hourly billing to disappear entirely, but believes many legal services can eventually be priced according to the value delivered once firms can estimate the required computational and legal work.

Alex asks whether a firm that can grow from 100 to 500 clients per attorney could eventually handle 5,000, expand beyond its local market, and become a very large institution. Jay predicts that many law firms will begin to resemble software companies because they will finally participate in the software revolution that has largely passed them by. A $25,000 settlement might produce roughly $8,000 for the attorney after a one-third contingency fee; Jay estimates that about half of that fee goes to labor, 30% to marketing, and the remainder to profit. That leaves little cash to reinvest in better service. With AI, a firm can build a proprietary dataset about a narrow category of case and achieve better results than a generalist competitor. Jay expects more specialization—for example, firms becoming particularly effective at soft-tissue injury cases with specific fact patterns—followed by geographic expansion because the expertise can be applied nationally. Smaller firms could therefore become much larger outcome-driven businesses without simply adding proportional headcount. At the same time, large firms may use Eve to acquire and consolidate smaller practices. Jay says Eve can reduce the painful integration of clients, workflows, and operations after an acquisition, allowing larger groups to realize value more quickly. He describes the long-term product challenge as building a “mini attorney brain” for each case type. A personal-injury case can require around 10,000 back-and-forth actions, including obtaining medical records, contacting insurers and healthcare providers, creating documents, and negotiating. Eve already handles many of these steps, including voice agents calling other agents, but still needs human intervention in some live cases. Jay says engineering capacity is the main constraint, while the potential value of even one case type is in the hundreds of billions of dollars.