See pricing
Why AI Could Turn Small Law Firms Into Billion-Dollar Businesses
The a16z ShowAI

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

This episode is a sharp case study in where AI creates new markets, not just lower costs. Jay Madheswaran explains why automating plaintiff law can turn previously uneconomic cases into viable business and let some firms increase attorney capacity from about 100 clients to 500.

October 30, 2026 · 53 min listen · 15 min read · Jay Madheswaran
0:00
–:––

Context

Jay Madheswaran is the co-founder and CEO of Eve, an AI company focused on plaintiff law. His argument is that the biggest opportunity in legal AI is not helping already well-resourced institutions write documents faster, but making legal representation economically possible for ordinary people whose cases are too small or too costly for lawyers to take today. That matters to PMs because it is a strong example of AI changing market structure, workflow design, and pricing, not just internal productivity. The episode is also a useful lens on vertical AI: why raw model intelligence is not enough, and why the real product advantage comes from owning the end-to-end workflow and the hard-to-collect data inside it.

The Big Idea

AI matters most when it removes a labor bottleneck that was suppressing an entire market, not just when it trims internal costs. In plaintiff law, that means making many subscale cases economically worth taking, which can expand access to representation and turn small firms into much larger businesses.

Eve is the concrete example. Jay is not pitching a chatbot for lawyers. He is pitching workflow automation from intake through settlement, with the business value showing up as more clients served, better use of attorney time, and a much larger addressable market than legacy legal software ever suggested.

Key Insights

Demand was never the issue

  • Plaintiff firms often have more inbound demand than they can serve.
  • The real constraint is legal labor, not lead volume. Jay says many lawyers will not take a case unless they can make at least $5,000 in fees, and even then many cases still lose money after staff time.
  • That is why many ordinary problems never get representation. Jay gives lower-value car accident and employment issues as examples, and says 96% of labor and employment cases are rejected.
  • The PM lesson is that some markets look small only because human service costs screen out huge amounts of real demand.

AI makes small cases viable

  • Eve compresses work that previously required a 30 to 45 minute call plus roughly 10 hours of legal research into an immediate first pass.
  • That changes more than efficiency. It changes which cases are worth touching at all.
  • When the minimum viable case size falls, firms can serve people who were previously below the economic threshold, even when the underlying harm was real.
  • This is the same pattern PMs should look for in other service markets: AI can unlock demand that was already there but could not be served profitably.

Full workflow automation matters

  • Jay's argument is that a point tool does not fix plaintiff law because the work stretches from the first client call to settlement and can run for six months to a year.
  • He later describes a single case as involving roughly 10,000 back and forths across calls, records requests, drafts, negotiations, and scheduling.
  • That is why Eve is built as an end-to-end system. If AI only drafts a document but humans still do all routing, follow-up, and status tracking, the main bottleneck remains.
  • For PMs, this is a reminder that the user does not buy a model output. The user buys progress across the whole job.

Revenue expansion beats cost cutting

  • Jay is explicit that the best customers do not buy Eve mainly to cut headcount. They buy it to serve more cases and make more money.
  • One Los Angeles firm used Eve from intake through settlement preparation and went from about 100 clients per attorney to 500.
  • That is a stronger ROI story than generic automation savings because it maps directly to the firm's main economic goal.
  • The broader product lesson is that when a buyer sits under a hard throughput ceiling, the clearest AI value proposition is often new revenue capacity.

Raw models need workflow context

Jay's Albert Einstein analogy captures a common AI product mistake. A powerful model is still unproductive if it lacks the local playbook, the right tools, the case history, and the exact handoff points inside the workflow. Eve's value is not that it invented intelligence. Eve's value is that it makes intelligence productive inside plaintiff law.

Proprietary data compounds advantage

  • Jay says less than 1% of cases turn into case opinions, which means the most useful plaintiff-side data is not publicly scrapable.
  • Most cases settle before litigation or before any published opinion exists, so the real signal lives in firm behavior, negotiation patterns, settlement history, and local tactics.
  • That creates a compounding product advantage. The system learns what works against a given insurer, in a given jurisdiction, for a given fact pattern.
  • For PMs, this is the core moat pattern in many vertical AI products: the most valuable data often never reaches the public internet.

Lawyers move toward advocacy

  • Jay does not argue that lawyers disappear.
  • He argues that as routine work gets automated, lawyers can spend more time on the parts where judgment changes outcomes: client counseling, trial strategy, local courtroom dynamics, and deciding when a settlement is already good enough versus when to push further.
  • His forward view is that more routine back and forth will happen agent-to-agent, while human lawyers increasingly focus on the few interventions that carry real leverage.

Mental Models & Frameworks

Supply constrained market lens

  • Ask whether a market is actually demand constrained or labor constrained.
  • Plaintiff law looked like a small software market because firms bought little software. Jay's point is that the real market size was hidden inside unmet demand that firms could not serve profitably.
  • Use this lens when evaluating AI opportunities in services. If buyers already turn away willing customers because fulfillment is too labor intensive, AI can expand the market rather than just steal share.

Outcomes not hours or seats

  • Traditional legal pricing maps to inputs: billable hours on the defense side, or seat-based software on the vendor side.
  • Eve moved away from the standard SaaS 'tall, grande, venti' model because plaintiff firms care about outcomes and client volume, not user licenses.
  • Use this model when the customer can clearly connect product use to economic results. If value is measurable, input pricing often leaves money on the table and misaligns incentives.

Intelligence needs operating context

  • A powerful model without tools, workflows, and domain memory is like dropping a genius into an unfamiliar system.
  • The operating context includes:
  • firm-specific playbooks
  • local jurisdiction patterns
  • access to calls, medical records, and insurer history
  • rules for when a human attorney must step in
  • This model is useful for PMs building AI products in regulated or process-heavy industries. The product layer is the system that makes intelligence usable.

Long horizon agent design

  • Jay's core product bet is that many valuable tasks are not one-shot prompts. They are long sequences with dependencies, delays, and external actors.
  • A legal case spans months, touches many documents, and requires repeated follow-up. That is why Eve cares about end-to-end progress, not benchmark scores on isolated tasks.
  • Use this framework when a workflow has three traits:
  • the work stretches over time
  • the work depends on many external systems or counterparties
  • the value comes from finishing the whole job, not generating a single artifact

Decision Principles

Principle: Follow the bottleneck

  • When: evaluating where AI can create the biggest business impact in a service workflow.
  • Why: Jay argues plaintiff firms do not mainly suffer from lack of leads or weak demand. They suffer from too much human work per case. Fixing the real bottleneck changes case economics. Fixing a peripheral task does not.

Principle: Price on outcomes

  • When: the customer can connect product usage to revenue, recoveries, throughput, or another business result.
  • Why: Eve found that legal buyers care far more about serving more clients and improving settlements than about how many seats they have. Pricing on the customer's real value driver creates a stronger fit than copying legacy SaaS pricing.

Principle: Productize the workflow

  • When: general purpose AI looks impressive in demos but users still need heavy manual coordination to get real work done.
  • Why: Jay's view is that raw model intelligence is not the scarce thing. Product value comes from wiring intelligence into the exact steps, data, and handoffs that determine whether a case actually moves forward.

Trade-offs & Nuance

Access versus system overload

  • Cheaper case handling can expand representation, but it can also send more disputes into a legal system that already has limited capacity.
  • Jay expects more cases to be resolved outside the court system through mediation, arbitration, or some form of automated dispute resolution.
  • The broader lesson is that reducing one bottleneck often exposes the next one. Product teams should expect downstream institutions to become the new constraint.

Automation versus human judgment

  • Jay believes representation could become 99% AI in many workflows.
  • He still thinks the human layer matters because attorneys contribute client trust, local knowledge, trial instincts, and the judgment to know when software has reached a good outcome or missed a better one.
  • For PMs, this is the practical middle ground between full automation and human-in-the-loop theater. Automate the routine mass, then reserve humans for the moments where small judgments change outcomes.

Shared learning versus privacy

  • Eve does not train on privileged client data. Jay says attorney-client privacy is non-negotiable.
  • At the same time, plaintiff firms already share tactics through conferences and continuing legal education, because insurers repeat the same defenses across firms.
  • The product opportunity is to capture cross-firm learning at the tactic level without violating case-level confidentiality. That is a useful pattern for any AI product in a regulated market.

Common Mistakes

Mistake: Selling cost cuts only

  • Jay is explicit that plaintiff firms buy Eve to make more money, not mainly to reduce payroll.
  • A cost-saving pitch undersells the product because the real win is serving more viable cases and getting more value from marketing spend already in motion.
  • The better approach is to show how AI changes unit economics and throughput.

Mistake: Shipping a raw model

  • A standalone model or drafting assistant does not solve a months-long workflow with records requests, phone calls, negotiations, and local legal tactics.
  • Jay's Albert Einstein analogy is the warning sign. Intelligence without process, memory, and tooling is still unproductive.
  • The better approach is to own the operating system around the model, not just the model output.

Mistake: Misreading market size

  • Plaintiff law looked under-softwared, which made it easy to assume the software opportunity was small.
  • Jay and Alex argue the opposite. The small software spend reflected a weak fit between legacy tools and the real bottleneck, not a small economic base.
  • The better approach is to size the market by unlocked outcomes, not by historic software budgets.

Practical Application

Find your labor bottleneck

  • List the customer requests, cases, tickets, or deals your product team currently ignores because the manual work is too expensive.
  • For each one, estimate the minimum revenue or value needed to justify human handling today.
  • Then ask what would happen if AI cut the labor cost of the first 80% of the workflow. The best AI opportunities often hide in work that currently looks too small to matter.

Model newly viable demand

  • Build a simple table with three columns:
  • jobs customers want done
  • current manual cost to serve each job
  • new cost to serve each job with AI assistance
  • Use that table to spot the segment that becomes viable only after automation. That is the segment most likely to create new market volume rather than incremental efficiency.

Automate a full journey

  • Pick one customer journey that currently requires repeated handoffs across teams, tools, or time.
  • Design the AI product around the full sequence, not a single artifact. Include intake, qualification, follow-up, routing, exception handling, and completion.
  • Jay's core lesson is that end-to-end ownership is what changes economics. A clever single-step assistant rarely does.

Store hard won outcomes

  • Capture the non-public details that experts use to make better decisions, such as which objections recur, which counterarguments work, and which conditions predict settlement or escalation.
  • Structure that information so future AI workflows can retrieve it.
  • Jay's point about plaintiff law is general: the most valuable operational data often never appears on the public internet.

Move experts to edge cases

  • Use AI to absorb the repetitive work that experts do out of habit, then redesign the expert role around exceptions and leverage points.
  • In a PM organization, that could mean letting AI prepare recurring analyses or first drafts, while product leaders focus on the ambiguous decisions that actually shape outcomes.
  • The gain is not just speed. It is better use of scarce judgment.

Questions to Consider

  • Where does your team currently reject customer requests, support cases, or sales opportunities because the service work is too labor intensive to justify the revenue?
  • What part of your product still behaves like a point tool, even though the user's real problem spans intake, coordination, follow-up, and completion over weeks or months?
  • Which valuable decisions in your business rely on non-public operational history, the equivalent of settlement tactics or insurer behavior, that a general model would never learn from the open web?
  • If AI made it dramatically cheaper for customers to press a claim, submit a request, or contest a decision in your market, which downstream institution would become the next bottleneck?

Bottom Line

Eve's story is not mainly about replacing lawyers. It is about using AI to remove the labor bottleneck that kept huge amounts of real demand from ever being served. For PMs, the broader lesson is that the biggest AI products may come from owning an end-to-end workflow, capturing the non-public data inside it, and pricing against the outcome that workflow unlocks.

Resources Mentioned

ResourceTypeWhy it was mentioned
Jay Madheswaran on LinkedInProfileLinked in the episode description as a way to follow Jay's thinking on legal AI and Eve.
Alex Rampell on XProfileLinked in the episode description for listeners who want more of Alex's views on software markets and AI businesses.

Concepts to Explore

Contingency fee economics

Plaintiff law works on contingency because most individuals cannot pay open-ended hourly legal fees. That makes it a strong case study for outcome-based pricing, because the provider already wins only when the client wins. PMs working on AI products for services can learn a lot from markets where pricing is already tied to customer results.

Automated dispute resolution

Jay predicts that if AI makes it much cheaper to bring claims, more conflicts will need to be resolved outside overloaded courts. Automated dispute resolution is the idea that AI can help mediators or neutral third parties assemble the facts, surface relevant precedent, and speed up judgment without fully replacing human decision-making.

Case Studies Mentioned

Los Angeles throughput jump

One firm in Los Angeles used Eve across intake, qualification, internal routing, and settlement preparation. Jay says the result was a jump from about 100 clients per attorney to 500. The lesson is that end-to-end workflow automation can change the throughput ceiling of a service business, not just trim a few minutes from isolated tasks.

Holiday discovery countermeasure

Jay described a defense tactic where discovery requests would be sent right before Thanksgiving or another major holiday, increasing the chance that plaintiff counsel would respond late or do rushed work. A firm using Eve built an agent to respond the same day, which neutralized a tactic that depended on human overload. The lesson is that AI can erase opponent strategies that only work because the other side is operationally understaffed.

Law firm roll up playbook

Jay says some of the larger plaintiff firms are buying smaller firms and using Eve to reduce the usual integration pain. Because Eve is centered on the client outcome rather than on each firm's old processes, it can absorb some of the change management that normally slows acquisitions. The lesson is that workflow software can become an integration layer, not just a productivity tool.

Tools & Products

Tool / ProductWhat it doesWhy it was mentioned
EveAI platform for plaintiff law workflowsThe core product discussed, built to handle work from intake through settlement.
ColossusInsurance claims valuation softwareJay mentions it as an example of defense-side software that tries to standardize what injuries are worth.
HarveyLegal AI productMentioned as a reference point for how most legal software still serves the defense side or institutional budgets.
ChatGPT CodexCoding agent from OpenAIAlex uses Codex as an example of how AI can already lower the friction of filing a lawsuit.
Microsoft WordDocument editing softwareUsed as a symbol of the minimal legacy stack many plaintiff firms relied on before AI-native workflow tools.

People to Follow

Jay Madheswaran

Jay is the co-founder and CEO of Eve. He is a useful person to follow if you want to understand AI workflow products, outcome-based pricing, and how vertical AI changes the economics of service businesses rather than just their internal efficiency.

Experiments to Try

Experiment: Intake shadow mode

  • Hypothesis: AI can identify viable work your team rejects today because the first round of qualification is too expensive.
  • Change: Run an AI system in parallel on inbound requests that humans decline or ignore.
  • Test: Compare the AI's recommendations with later human review on a sample of cases or requests.
  • Measure: track how many previously rejected items would have become economically viable if the first-pass work were automated.

Experiment: Outcome pricing pilot

  • Hypothesis: customers will buy faster when pricing maps to the business result they want rather than to seats, hours, or usage alone.
  • Change: offer one customer segment a pilot price tied to throughput, recoveries, or another observable outcome.
  • Test: compare close rate, expansion, and retention against a comparable segment on traditional pricing.
  • Measure: watch whether the pilot changes willingness to buy, product usage depth, and the size of the unlocked opportunity.

Notable Quotes

"Historically plaintiff law really didn't need more than a cell phone email address." (Jay Madheswaran)

"If you drop Albert Einstein into a law firm today, he will not be productive for quite some time." (Jay Madheswaran)

"What they care about is outcomes. What they care about is how do I serve more clients?" (Jay Madheswaran)

AI PM course

Everyone hears the same episodes.
Few can do what they describe.

Start for free