Jay Madheswaran on AI and the Economics of Plaintiff Law

Eve cofounder Jay Madheswaran argues that AI can change plaintiff law by reducing the work required to represent a client. He describes Eve as a tool for lawyers handling cases from an initial call through settlement, not simply a way to cut headcount.

Alex Rampell of a16z introduces Madheswaran and distinguishes distributed plaintiff practices from the large corporate firms familiar to Silicon Valley. Madheswaran says his company focuses on people who often cannot afford institutional legal resources.

Madheswaran explains the contingency model: a plaintiff lawyer is paid when a client wins rather than charging an open-ended hourly bill. Eve aims to help lawyers work through an entire matter and improve the outcomes they can provide.

Rampell describes the economics of a small claim. Even a lawyer paid a share of a settlement may decline a $1,000 matter if staff and research costs exceed the possible fee. He argues that cheaper work could make cases that are currently uneconomic worth taking.

Madheswaran agrees that legal labor limits access. He estimates that many firms need at least $5,000 in fees to justify a case and may still lose money on some matters; these are his descriptions of the market, not universal thresholds.

He gives the example of an employment complaint: a lawyer must hear the facts, research potential causes of action and ask follow-up questions before knowing whether the case is viable. He claims that many inquiries are rejected because that preliminary work itself is costly.

The work can include listening to a caller, identifying a legally relevant claim and reviewing case law. Madheswaran says the firm can spend hours investigating and still conclude that the expected fee is too small, leaving many people without representation.

Madheswaran says Eve can answer an intake call, apply the firm’s criteria and prepare preliminary work. Cases still run over months and involve many interactions; his longer-term goal is to handle more of that sequence as the technology improves.

Rampell recalls explaining Eve to a trial attorney who wanted to try it. His point is that AI’s value for a firm may be increased revenue from serving more clients, rather than merely reducing costs.

Asked about a firm’s economics, Madheswaran says demand is often less constraining than the staff available to handle incoming calls and cases. Different firms have different bottlenecks: a solo attorney may struggle to respond, while a larger intake team can be overwhelmed by marketing-driven demand.

He describes how the bottleneck shifts as a practice grows. A small office may rely on one lawyer and an office manager; building an intake team can help until an increase in calls overwhelms that team as well.

Rampell suggests that a low-cost artificial assistant could make it possible to serve more of the leads a firm receives. Madheswaran offers one customer example: a Los Angeles firm reportedly increased the number of clients managed per attorney from around 100 to 500.

He attributes that result to the firm’s work using Eve across intake, qualification, handoffs and settlement preparation. The system can surface prior tactics and comparable outcomes, while lawyers decide how to apply their own expertise.

Rampell asks why plaintiff practices historically spent little on software. He contrasts hourly billing tools at corporate firms with contingency practices and suggests that AI software may address a much larger labor market than prior software spending alone would imply.

The host argues that assessing a software market by current spending can miss the value of work that the software might eventually perform. He contrasts modest historical technology budgets with the much larger value of outcomes and the cost of legal labor.

Madheswaran says some small plaintiff practices once needed little more than a phone and email; traditional products did not solve their labor constraint. He argues that legal software companies followed the larger budgets available on the institutional and defense side.

For a small firm, Madheswaran says a product must produce useful results to be renewed. His proposition is to encode work as a process that can handle different facts while drawing an attorney’s attention to the decisions that require judgment.

Rampell asks how an application can compete when model providers supply increasingly capable general intelligence. Madheswaran welcomes better models but argues that raw ability alone does not supply the knowledge and context needed to work effectively inside a specific law firm.

Madheswaran says much plaintiff-side settlement information is not publicly searchable because cases often resolve before trial. Rampell illustrates the gap with an insurance settlement: the insurer sees many claims, whereas a local practice may see only a small slice of that history.

Rampell imagines an insurer offering little after a car accident and a lawyer eventually obtaining a settlement. The example illustrates why knowledge of a counterparty’s past behavior could be useful, though the participants are offering an illustration rather than a verified case study.

The relevant context, Madheswaran argues, includes the counterparty, local courts, opposing counsel and fact pattern. He says Eve helps firms use that multidimensional information to decide what action could improve a case, beginning with which clients they are equipped to represent.

Rampell asks about sharing knowledge when firms have private case data but face common counterparties. Madheswaran says Eve does not train on confidential case data and describes lawyers sharing tactics through professional education; this is the company’s account of its practices.

Rampell asks whether the benefits of shared learning can coexist with client confidentiality. Madheswaran draws a distinction between private case material and the tactics that attorneys already discuss with one another at continuing legal education events.

Madheswaran says Eve hires attorneys and tracks case success. He describes an Auditor product that reviews case pages for potentially missed information, presenting it as a way to surface items a lawyer may want to inspect.

Rampell asks what happens when both plaintiffs and their institutional opponents gain AI tools. Lower costs could change the balance between a small claim and the resources available to respond, while raising questions about how the adversarial process itself evolves.

Madheswaran predicts that agents will handle more routine exchanges, but says human lawyers will continue to matter for understanding a client, the facts and local opponents. He describes some customers using time saved on routine work to focus more on trial advocacy.

He says an institution can use time and financial resources to wear down an individual. In his proposed future, lawyers use their understanding of particular clients and local opponents to direct the work, even when routine interactions are automated.

Rampell connects the discussion to litigation finance, where funders analyze judges and case data before backing a claim. More systematic information, he suggests, could also help lawyers choose among a settlement, further negotiation and trial.

Madheswaran says software can help identify possible outcomes, while the lawyer remains responsible for deciding how to pursue the client’s case. Rampell then recounts using an AI coding assistant to help initiate a claim against a bank, illustrating how filing work might become easier.

Rampell recounts his own experience using a coding assistant to initiate a dispute with Wells Fargo and arrange service. He says a court date is still months away, emphasizing that preparing a filing and obtaining an adjudication are different steps.

The participants consider a second bottleneck: courts and agencies still have limited capacity even if preparing a claim becomes cheaper. Madheswaran expects more matters to use mediation or other alternatives to the court system.

He speculates that AI-assisted mediators or neutral decision makers could assemble relevant facts more quickly. He does not expect human judgment to disappear, and says those decision makers would need to address possible bias.

Rampell asks whether a small firm that handles many more cases per attorney could grow into a much larger organization. Madheswaran predicts that some firms will come to resemble software companies, investing in their own specialized data and processes.

Madheswaran argues that a firm’s accumulated information about a particular kind of matter can become a competitive asset. Historically, he says, plaintiff practices have moved from general local practice toward more specialized areas of law.

Using a hypothetical $25,000 settlement, Madheswaran sketches a fee of roughly one third and says a significant portion of that fee may go to labor and marketing. He argues that lower servicing costs could free resources for a firm to improve its next case.

He expects specialization to narrow further, with a firm becoming particularly effective at a specific fact pattern and expanding beyond its local market. He also describes a competing trend in which larger firms acquire and integrate smaller practices.

Rampell contrasts per-seat software subscriptions with outcome-based pricing. Plaintiff firms already depend on outcomes, while in other legal work predicting the scope of a matter can be difficult, making a fixed price risky for the provider.

Rampell observes that clients may prefer a price tied to an outcome, but the lawyer may not know at the outset whether a matter will require ten hours or hundreds. That uncertainty can make a flat fee difficult even when both parties want one.

Madheswaran says Eve began with conventional subscription tiers and moved toward outcome-based pricing after learning what firms valued. He expects the billable hour to persist in some work, while more predictable matters may increasingly be priced around outcomes.

Asked what limits Eve’s growth, Madheswaran says the company needs engineers to build and evaluate long-running workflows for each type of case. He distinguishes isolated benchmark tasks from the much longer chain of work required for a real legal outcome.

Madheswaran describes learning how to resolve each case type and building a specialized workflow that can use available tools. He says demand is present, but the difficult engineering work limits how quickly the company can extend its coverage.

Rampell asks whether Eve has completed a whole case without human intervention. Madheswaran says not yet, though the company is working toward that goal. He estimates that a case can involve thousands of interactions, including obtaining records and communicating with other parties.

As one example, he says outbound voice agents can request records from healthcare and insurance organizations, a process that may otherwise take weeks. He describes individual tasks operating end to end while acknowledging that human touches still improve the overall case.

On the defense response, Madheswaran recounts a firm receiving discovery requests just before holidays, when the team would be under pressure. He says the firm configured an Eve agent to respond quickly to a repeated tactic, changing that interaction with opposing counsel.

Rampell closes by highlighting the amount of coordination involved in a single case. Madheswaran says Eve has helped hundreds of thousands of clients and hopes to reach many more; the figure is his company’s claim.