We study the economic and productivity effects of AI on the legal sector, drawing on data from our platform and working alongside academic partners. Our research spans four areas:
A longitudinal case study of time card entries at an AmLaw 200 firm from 2016 to 2026, spanning the period before and after LLM deployment. Each entry records a narrative description of the task, the hours worked and billed, and the billing rate. Together, these records let us quantify how the composition of legal work, and its unit economics, shift as LLMs are adopted.
We calibrate a model of firm economics on real time card data from a partner law firm. The central question is what happens to revenue, by matter type and fee arrangement, when the hours required for a given class of task compress. By tracking AI usage to estimate that compression, we derive economic elasticities and use them to ask whether meaningful sources of revenue exist beyond the billable hour.
Leads: Sophia Fend, Chijioke Mgbahurike, Riya RanjanBuilding on the empirical economics project, we examine what value law firms deliver, and can be paid for, when fewer hours are needed to do the work.
Routing tasks between models of different sizes is a common way to reduce cost, but it has rarely been tested in domains where output quality is difficult to verify. We study whether legal tasks can be routed to smaller models without loss of quality, and how such routing can lower costs for clients.
Leads: Riya RanjanLegal frameworks, developed through legitimate and deliberative institutions, have been proposed as a basis for aligning AI systems, but the idea has seen little direct experimentation. We run a benchmarked ablation study testing whether post-training on legal text measurably improves model alignment.
Leads: Riya Ranjan, with Stanford LiftLabLaw firm data is rarely centralized or standardized, which makes building extract-transform-load (ETL) pipelines a recurring bottleneck. We evaluate agent-based approaches to constructing these pipelines automatically across the many forms law firm data takes.