Choosing outside counsel is one of the highest-stakes decisions a litigant makes—and one of the noisiest. Buyers of legal services mostly rely on familiar signals like prestige, firm size, and revenue, captured in popular law firm rankings, to narrow the field. Those signals are abundant and easy to cite, but our new study in Nature Computational Science shows they tell us surprisingly little about who actually wins in court. When we compared popular reputation/size/revenue lists to empirical litigation results, the correlation with courtroom success was near zero. We introduce an outcome-based, data-driven approach built from over 60,000 case results which predicts the winning side far better—about 60% accuracy on balanced test sets—offering a sharper lens on advocacy performance than brand gravity alone.

What Changed: AI Can Read the Record

This research was made possible by the new reality of AI-enabled legal data. Until recently, extracting reliable, structured information from judicial opinions or other legal data at scale was prohibitively labor-intensive. We can now use AI information extraction systems to parse unstructured text with enough fidelity to support serious empirical analysis. In our work, we use empirical data to model each lawsuit as a competitive, asymmetric game between plaintiff and defense counsel, adjusting for case-type differences and the well-documented baseline advantage defendants often enjoy. From those observed head-to-head results, we infer each firm’s latent skill—an outcome-anchored signal that can be updated as new cases resolve.

Implications for Corporate Counsel

Why does this matter in practice? For in-house teams, it means counsel selection can be grounded in evidence that’s truly fit-for-purpose. Instead of defaulting to the largest or most storied brand, a GC can ask: which firms have the strongest demonstrated track record for this kind of matter, on this side of the “v,” in this forum? Outcome-based analytics don’t replace the nuanced considerations—strategy, conflicts, cost, chemistry—but they inject a much-needed empirical prior into the decision. That, in turn, can improve both selection and settlement dynamics in the shadow of expected results.

Labor-Market Signals for Lawyers

The benefits extend to lawyers themselves. For laterals and newer attorneys, reputation proxies tend to amplify incumbency: big names get bigger, while under-the-radar excellence is hard to surface. A transparent, practice-area-level performance signal helps articulate real value, identify the right platform, and assemble teams where complementary strengths compound. Firms can spot overlooked talent pools, and lawyers can demonstrate impact beyond the soft glow of a marquee brand.

Law Firm Rankings for Access-to-Justice 

There’s also an access-to-justice dimension. Repeat players with deep market memory already know who’s effective where; first-time or infrequent litigants usually don’t. Making evidence-based indicators widely available levels that informational playing field. When a small business or individual can see which firms actually deliver results in comparable matters, they are better positioned to choose capable counsel and to negotiate credibly—outcomes that matter even if a case never reaches a verdict.

One Metric Isn’t Everything

It’s important to say what we are not claiming. Litigation performance is broader than trial outcomes. Clients care about settlement quality, total cost and speed, business-savvy advice, ethical practice, and client experience. Our empirical point is narrower: revenue, size, and reputation are poor predictors of courtroom results, so they’re unlikely to be the right proxies for other dimensions of excellence either. The remedy isn’t to idolize one metric, but to systematically measure more of what matters. We hope this work helps catalyze that broader agenda: outcome-based measures for different stages of a case, richer models of settlement, time-to-resolution and cost, team-composition effects, judicial factors, and ultimately client value.

Toward a science of law

Stepping back, the larger story is methodological. AI is beginning to unlock legal text at scale, letting us ask questions that were previously infeasible. That enables a more empirical, cumulative inquiry into advocacy and institutions—approaching the Holmesian ideal of law as a science. With better data and transparent methods, we can move beyond lore and toward accountable, testable claims about what works, where, and why.

What’s next

To make that progress collective, we’re releasing our data and code so others can audit, stress-test, and extend the approach. If you’re a GC exploring evidence-based counsel selection, a firm leader interested in building outcome-aware teams, or a researcher probing settlement dynamics and forum effects, we’d love to compare notes. Reputation will always play a role in the legal market—but it shouldn’t be mistaken for results. Outcome-based analytics give us a clearer baseline. From there, we can start measuring and improving what really counts. At CodeX, we’re deeply excited about how AI can structure and enrich legal data to unlock new insights. Legal analytics is one of our core initiatives, and we invite collaborators from industry, academia, and the public sector.