AI Document Validation
AI Document Validation – Improving Trust in AI-Powered Legal Review
Lead UX Designer | Spring 2023
Overview
Over six weeks, I led design on Filevine’s Document Validation project — an upgrade to our AI-powered document analysis feature known as “AI Fields.” I partnered with the Docs team (1 PM, 2 engineers, 1 QA), though due to my PM's onboarding, I took more initiative driving strategy and discovery than usual. We followed the Shape Up methodology.
The Problem
AI Fields generated document insights instantly — but users didn’t trust them. Why?
Because the AI summaries appeared as long blocks of text, disconnected from the source document. To verify accuracy, users had to open the original document in another tab and manually cross-reference — a clunky, inefficient process that slowed them down and undermined confidence in the tool.
Why this mattered
Trust is make-or-break for AI in legal workflows. If users can’t verify AI results quickly, they’ll abandon the tool — or worse, make errors. We needed to reduce review time and help users feel confident in what they were seeing.
Design Process
After conducting user interviews and a quick competitive analysis, I proposed a new solution: a side-by-side validation experience with interactive highlighting. This would allow users to see the AI output and the source content together — making it easier to cross-reference without switching tabs.
It was initially out of scope, but after aligning with engineering, we agreed to de-prioritize three less critical features (edit AI output, validation status tracking, and a smart table of contents) and push them to the next cycle in order to ship the validation flow.
Team Collaboration
I worked closely with engineers to balance trade-offs, created interactive prototypes for usability testing, and handled most of the product decision-making due to the PM’s ramp-up time.
Outcomes & Impact
60% reduction in document validation time
80% adoption of the new flow in the first month
Increased user trust in AI-generated summaries
(Estimated impact: time saved per user translates to ~$150K–200K in annual productivity gain)
