Summary

This JURIX paper is a strong direct legal workflow record because it keeps AI in a bounded role that legal services actually need: issue spotting, intake routing, and follow-up question generation. The FETCH classifier is already in use in phone or referral workflows tied to Virginia Legal Aid Society and the Oregon State Bar, and the paper reports measured classification performance rather than generic adoption claims.

Why It Matters

This matters because it shows a more defensible operational lane for legal AI than drafting briefs or giving unsupervised advice.

  • the system classifies problem type from an applicant's narrative so the person can be routed to the right legal resource faster
  • it automatically generates follow-up questions for ambiguous matters instead of pretending the first prompt is enough
  • it keeps the interface in a structured form metaphor rather than a free-form chatbot, which the paper argues can reduce user confusion and the uncanny-valley effect
  • it documents a low-risk, high-volume workflow where accuracy can be measured and sampling-based quality control is realistic

Investigator Workflow

The concrete investigator task is first-contact intake and case categorization. The maturity is `simple workflow` because the value is in collecting structured facts, spotting likely case type, and asking a small number of follow-up questions before a human investigator decides whether to open the matter. The intake-routing workflow is source-stated for legal aid and bar referral. The PI transfer is an internal inference, but it is straightforward and operational.

What the Source Says

The paper says Virginia Legal Aid Society receives about 18,000 calls per year and the Oregon State Bar fields about 100,000 inquiries through its referral service. It evaluates the FETCH classifier on 419 real-world queries to a nonprofit lawyer referral service and reports 97.37% hits@2 accuracy using an ensemble of smaller models plus keyword and traditional machine-learning components. The classifier is described as a REST-like API already in use by the Virginia Legal Aid Society phone intake system and the Oregon State Bar's online referral service. The paper also says the tool can automatically generate follow-up questions so an applicant can add clarifying facts without waiting for a human callback.