Building Trust in Machine-Assisted Review

Context

Relativity is an enterprise eDiscovery platform where legal teams rely on email threading to make high-stakes decisions about relevance, inclusion, and coding consistency. While the system generated threaded conversations through analytical logic, users struggled to interpret and trust those outputs during review—undermining confidence in decision-making. UX research was brought in after early stakeholder interviews and an initial prototype, requiring a reset of problem framing and success criteria.

When users cannot understand why emails are grouped or classified, confidence in both the interface and the underlying analytics breaks down—slowing review, increasing risk, and driving work outside the platform.

Leadership

As the Lead UX Researcher, I owned research strategy and execution, partnering with product and design to reframe the work from feature usability to decision-support quality. I designed a multi-iteration evaluative research program under active development constraints and led synthesis and alignment across teams.

Findings

While email threading was analytically sound, users struggled to understand why messages were grouped or classified. That lack of clarity undermined trust in high-stakes review decisions.

• nterpretability
  Users needed clear signals explaining why emails were included, excluded, or grouped.
• Context
  Dense, text-heavy views obscured conversation flow and user position within a thread.
• Validation
  Reviewers treated consistency checks as a core task, not a secondary action.

Impact

The research reshaped how legal teams understood and trusted machine-assisted email threading during review. By improving interpretability and validation cues, the work helped users better understand why emails were grouped or classified, increasing confidence in system-assisted decisions during high-stakes review workflows. At the time of delivery, the redesigned threading experience became one of the most widely adopted capabilities in the platform, reducing reliance on workarounds and external validation. Review workflows became more efficient, and UX research gained credibility as a contributor to analytical feature direction. In-person client validation confirmed that the improvements generalized across organizations and review practices.