01 — The Overview

Litigators spend significant non-billable time reading long, dense complaints to orient themselves to a case. Bloomberg Law's docket sheet is the most-used page on the platform. The opportunity was clear: surface a one-click AI summary exactly where attorneys were already working, without asking them to go anywhere new.

I was the sole designer on this feature, working with a Principal PM, three engineers, and the analytics team to define scope, validate the approach with legal experts, and ship the experience directly onto the docket sheet.

Team

• Principal Product Manager (licensed lawyer)
• 3 Engineers

02 — The Problem

Litigators can't bill for reading. But they have to.

Litigators rely on complaints to orient themselves to a case, scanning for legal precedents, opposing arguments, possible outcomes, and supporting evidence. These documents are long, dense, and time-consuming to read. That time is non-billable. The faster an attorney can get oriented, the sooner they can do the work that actually counts.

03 — The Constraints

The summary couldn't ask attorneys to go somewhere new. It had to appear where they already were, surface only what they needed, and get out of the way.

Image-based PDFs took too long to process

The AI model generated summaries from text-based PDFs in under 5 seconds. Image-based files averaged 30 minutes. Keeping a user on the page for 30 minutes while a summary generated wasn't a UX problem. It was a broken product. The solution was a non-disruptive fallback state for image-based files, paired with a shift to pre-generation for live summaries. Pre-generation kept attorneys on the page and eliminated the wait entirely for the most common document types.

The docket sheet wasn't ours to redesign

The docket sheet organizes case events in a rigid table with action buttons awkwardly positioned in the middle of each row. My recommendation was to move the button column to the end of the row, which would have aligned with standard UI patterns and made summary integration cleaner. Engineering flagged the rebuild as high-effort, low-priority. So we designed around the constraint instead of fighting it, integrating the feature within the existing table structure.

04 — The Outcome

The feature launched to production and the adoption numbers were clear.

Summary views tell you people clicked. Copy actions tell you people found it useful enough to use in their work. That's the difference between adoption and engagement.

Screen 01 · The Start

Screen 02 · The Preloader

Screen 03 · The Summary