Bloomberg Law's AI Workspaces Research Study
Determined a definitive product direction as to how AI can play a role in legal tech, saving the company spending the time and money from building the wrong tool.
Overview
Company: Bloomberg Industry Group (Bloomberg Law)
Timeline: March – October 2025
Scope: End‑to‑end ownership across research, synthesis, and product direction
Outcome: Identified search and analysis are pain-points across multiple attorney types share, and determined a definitive product direction how AI can play in a role in legal tech.
Team members:
Senior Design Manager as Research Planner
Principal Product Manager (and licensed lawyer) as Co-Moderator,
Product Delivery Manager as Recruitment Planner
My Role
In 2024, Bloomberg invested heavily in conversational AI experiences, with plans to further leverage that technology to support additional use cases such as litigation drafting. However, a question-and-answer format does not reflect how lawyers actually think. Legal work is not linear; it is iterative.
I raised this concern with the Chief Product Officer and Head of UX, emphasizing the need to support lawyers beyond standalone chat interactions. By mid-2025, we were given approval to explore an AI-driven, generalized workspace designed specifically for legal workflows.
I owned discovery through evaluation and strategic recommendation, moderating 40+ qualitative interviews, synthesizing findings into product direction, designing and testing lo-fi prototypes, and partnering with Product Leadership to pressure-test scope against business constraints. The research reframed the concept multiple times.
That was the point.
The Starting Point
At the start of 2025, Bloomberg Law began exploring how generative AI could support legal work beyond isolated point solutions. The internal assumption was that litigation drafting was the lion’s share of work and if attorneys could write faster, everything else would follow.
Based on this idea, engineering built an internal tool called Drafting Slate. It guided attorneys through drafting a Motion to Dismiss. It launched in Bloomberg Law's Innovation Studio but nobody used it. Rather than asking why, the team refactored it for UX quality and kept moving.
That's the moment I want to flag because it shaped everything that followed. We were iterating on a solution without confirming it addressed a real problem. My job was to find out whether it did.
Finding the Real Problem
Getting attorneys to talk to us wasn't easy. A proposed research partnership with five law firms fell through immediately. It was evident that participation would reduce billable hours, with no clear benefit to the firms. Every firm declined.
I advocated for incentivized 45-minute interviews instead. Once approved, 14 attorneys signed up within days.
What they told us upended the drafting-first assumption. Attorneys weren't struggling to write. They were struggling to know whether their research was complete and whether prior work was reusable. The hard part wasn't writing nor assembling arguments. It was confidence.
That signal was consistent enough to warrant deeper investment.
Research question: What role should AI play in legal workflows and where would it create value?
First Pivot - Refactoring Drafting Slate
Based on early interviews, the product team decided to refactor an existing collaborative workspace to support litigators' multi-phase, iterative drafting workflows. I distilled the macro steps into a lo-fi prototype and concept-tested it with 6 attorneys and legal professionals to validate whether the concept resonated and where AI felt like a natural fit.
Refactored Drafting Slate: What worked?
Source management (top image) and GenAI analysis (bottom image) interfaces received positive feedback. Attorneys envisioned how research organization and complaint analysis could support their drafting workflows and feel more confident about their work.
What didn't resonate with users?
AI as a thinker (top image): Participants showed little interest in an interface dedicated to building legal arguments and bucket sources to arguments.
Organizing their notes (top image): We assumed attorneys would want help structuring or cleaning up their working documents. Feedback suggested the opposite.
Full draft generation (bottom image): Generating the full Motion to Dismiss proved misaligned. Composing the document itself was not the hardest phase of their workflow, and delegating this step to AI introduced risk without meaningful time savings.
Second Pivot: Generalized Workspace
Before synthesis on new prototype concluded, product leadership made a call: a solution this narrow couldn't be built exclusively for litigators. The scope needed to expand across legal domains.
This killed the litigator-specific direction and opened a broader research question: could a generalized legal workspace address high-friction workflows shared across practice types?
I led 18 additional foundational interviews to discover where AI could play a role in a broader legal context (e.g. litigation, transactional, in-house legal teams).
What the Research Revealed
All attorneys types struggle finding precedent: Existing document repositories (e.g. iManage, NetDocs, Box, research libraries) have poor search retrieval.
Prior work doesn't travel. Once an attorney finds relevant prior work, the research trail behind it is invisible. So the next attorney starts from scratch.
Document review creates anxiety. Tracking versions, clauses, and changes across large document sets consumes significant time and generates persistent anxiety about missing something important.
New problem statement: Attorneys spend significant time locating precedent and exemplar documents, only to redo the underlying analysis because prior work contains no visibility into research trails, legal reasoning, or rejected paths.
What the Product Became
Based on this research, I narrowed the direction to a multi-document analysis workspace. The AI's role was deliberately constrained.
What the AI did:
Highlight material changes between documents
Surface what could be skipped versus reviewed
Enable natural-language queries across document sets
Provide traceable outputs linked back to source text
What the AI did not do:
Draft final legal language
Replace attorney judgment
Validated by Testing
Lo-fi prototype testing with attorneys confirmed the direction. The workspace created clear value when it:
Reduced review time for version comparison and bulk document analysis
Lowered anxiety about missing important changes
Improved collaboration compared to download–edit–upload workflows
Felt easy to learn, with clear AI prompts and fast document access
Attorneys were consistent about what they needed in return for their trust: built-in verification, traceable outputs, role-aware AI behavior, and seamless integration with the tools they already used.
The Legal AI Experience I Recommend
If taken forward, I'd recommend starting with public or low-sensitivity documents to build trust before asking attorneys to upload client files. From there: interfaces for source organization, a writing tool that addresses collaboration friction, and deep integration with the document systems attorneys already live in
The Outcome
By the end of the engagement, the research had delivered something more valuable than a shipped feature; it had prevented the wrong thing from being built.
Disproved the drafting-first AI hypothesis with direct user evidence
Validated a shared, high-friction problem space across attorney types
Identified analysis as the highest-value AI opportunity
Surfaced real adoption blockers before significant engineering investment
Influenced how a workspace model could increase user retention across multiple Bloomberg intelligence platforms
The most important decision I made on this project was the one I made early: don't validate the existing direction, investigate it. Research that confirms what you already believe isn't research. It's documentation.
That meant being willing to bring back findings that made the team uncomfortable and being clear about what those findings meant for the roadmap.
See how these research insights translated into design