Why AI Could Turn Small Law Firms Into Billion-Dollar Businesses
Eve CEO Jay Madheswaran joins a16z's Alex Rampell to explain how AI is automating plaintiff law end to end, expanding access to legal representation by making previously uneconomical cases viable, and why the economics point toward outcome-based pricing, agent-to-agent negotiation, and a fundamentally different law firm structure.
Plaintiff Law Has Been Starved of Labor, Not Software
Plaintiff law has historically run on little more than a phone and an email address. The core constraint was never a lack of software tools; it was labor. Every case requires intake calls, qualification, legal research, demand letters, and negotiation, and a firm’s capacity was capped by how many people it could hire and train. Eve’s thesis is that AI can absorb that labor end to end, from the first client call through settlement, and in doing so turn cases that were previously uneconomical into viable ones.
The Economics of Contingency Law
Contingency arrangements exist because most people cannot pay $500 to $600 an hour on an open-ended bill. Lawyers take a third of the outcome, which means they only pursue cases where the expected payout justifies the labor. In practice, many firms will not take a case unless it can plausibly generate around $5,000 in fees, and even then they lose money on many matters, banking on a few large payouts to cover the rest. The result is a market failure that resembles a supply curve with an inelastic lower tail: a person with $1,000 in damages has no realistic path to representation, and even a person with $3,000 to $4,000 in damages may find that the cost of carrying the case exceeds the recovery.
The labor-intensive reality is stark. When a potential client calls, a firm may spend 30 to 45 minutes on the intake call and then roughly 10 hours on research to determine whether a case has a viable cause of action. Employment law in the United States is largely at-will, so claims must be tied to a specific protected category, and that analysis is expensive. As a result, around 96% of labor and unemployment cases are rejected. Many people with legitimate grievances simply absorb the harm because no lawyer will touch their matter.
Revenue Expansion Rather Than Cost Cutting
The usual framing of AI in professional services is cost reduction, with firms shrinking headcount. Eve’s pitch is different: the same firm can take on more clients and generate more revenue. The clearest example is a Los Angeles firm that went from roughly 100 clients per attorney to 500. Achieving that required automating intake, qualification, the hand-offs between attorneys and case managers, and even pre-mediation briefs that recommend settlement targets and tactics drawn from comparable cases.
The bottleneck in a plaintiff firm also shifts as it grows. A solo attorney with a case manager becomes constrained by the attorney’s time. A firm that systematizes with an intake call center finds that the call center becomes the limit as marketing spend rises and inbound calls multiply. Each stage of this progression hits a labor ceiling, and the market for “hire more people to do this” is effectively nonexistent at the price points that would make these cases worthwhile. In that sense, software is functioning as a shadow labor market, and its potential size is far larger than the roughly tens of billions of dollars spent on legal software today would suggest, given that the outcomes these firms deliver run into the hundreds of billions.
Why Raw Intelligence Is Not Enough
The question of whether frontier models will simply absorb legal work gets a pointed answer: drop Albert Einstein into a plaintiff firm today and he would be unproductive for a long time. Much of the knowledge needed to win plaintiff cases is not published. Fewer than one percent of cases reach a written opinion, and most resolve through pre-litigation negotiation with insurers. Insurers operate with their own playbooks, such as Colossus, a tool that estimates the value of injuries like a broken thumb, and they repeat defenses across thousands of claims. Demand letters for soft-tissue injury claims, for example, are not documented anywhere a new paralegal can learn from, and local court systems and opposing counsel vary in ways that matter.
Eve’s approach is to convert this fragmented, context-dependent knowledge into a structured process. Each case involves thousands of decisions about what to do next, and a human cannot hold all of that in working memory across hundreds of matters. Eve encodes these decisions, tracks success rates across its own attorney-run cases, and runs an auditor that reviews every page of a file overnight to catch missed information, which the team says is nearly universal in plaintiff files.
Data, Privilege, and the Rules of Engagement
On data, Eve follows the norms the profession already observes while amplifying them with software. Client information is protected by attorney-client privilege, and Eve says it does not train on it. Lawyers have always shared tactics informally at CLE events and conferences, including how to counter a specific insurer’s recurring defense, and Eve sees its role as making that knowledge systematic.
The Adversarial System Gets Superpowers
The sharpest question is what happens when both sides of an adversarial process adopt AI. Insurers historically have been able to rely on the fact that individual claimants lacked the resources to fight back. If claimants suddenly have agents that can respond, demand, and document, insurers will need to adapt. The analogy to consumer agents is apt: if a restaurant reservation line can be flooded by agents making calls, the restaurant has to find ways to filter them. The likely outcome is a world in which routine back-and-forth between parties happens agent-to-agent, with humans reserved for judgment, strategy, and advocacy. Eve’s own agents already call insurers and healthcare providers to collect medical records, which used to take two weeks, and they have reached the point where agents talk to other agents.
The human role is not disappearing. The expectation is that AI handles something like 99 percent of the work, but the remaining share, the attorney’s understanding of the client, the facts, and the local adversary, is where outcomes are decided. Customers are already choosing to spend more time on trial work, which was previously prohibitively expensive because a trial consumed the whole firm. Lawyers are moving toward advocacy and away from paperwork.
Dispute Resolution and the Shape of the Legal System
The longer-term question is whether the court system remains the default venue for disputes. Parts of the legal system already route claims elsewhere, such as veterans’ affairs adjudication, the EEOC process for employment claims, and arbitration. The likely next step is automated dispute resolution, in which AI-supported mediators or neutral parties synthesize historical patterns to speed up judgments. The judgment itself would remain human, but it would be informed by a clearer picture of how comparable cases were resolved, along with an explicit effort to remove bias.
Firms as Software Companies
The structure of plaintiff firms may also change. A typical settlement might pay the attorney roughly a third of the recovery, with about half of that going to labor and a significant portion to marketing. That leaves little room to reinvest in better service. If AI reduces labor costs and lets firms build proprietary data on which tactics work for which fact patterns, a small firm could specialize deeply in a narrow area, such as a particular type of injury case, and advertise that specialty nationally more cost-effectively than a generalist giant. Some large firms are also acquiring smaller ones, using Eve to integrate them quickly. Whether a local firm can become a billion-dollar business is an open question, but the trajectory points toward firms that operate more like software companies.
Pricing Should Follow Outcomes
The conversation returns to pricing. Software has moved from per-seat subscriptions toward outcome-based models, and plaintiff law has been outcome-oriented from the start because contingency fees already tie payment to results. Other areas of law, such as estate planning, immigration, or corporate work, are billed by the hour largely because the scope is hard to predict in advance, not because clients prefer it. As agents make the work more predictable and better measured, the billable hour may persist for some matters but is likely to give way to amortized or outcome-based pricing for many others.
Eve started with per-seat pricing, found that customers cared about outcomes and client volume, and moved to outcome-based models early. The company’s current constraints are largely engineering: each case type must be learned as a distinct problem, and the team is building an agent that can carry a case from intake to settlement with minimal human touch. Early versions are live with some customers, and the team reports that the human touches required per case have fallen substantially month over month. The work is also beginning to change defense-side behavior. One firm found that insurers had been timing discovery requests to arrive just before holidays, when attorneys would be away, and had to respond when Eve answered on the same day.