AI Design · Enterprise UX · Product Strategy
VTS is a commercial real estate platform managing lease data, deal workflows, and property portfolios for asset managers and lease administrators. Their lease abstraction capability was delivered through a third-party tool accessed via SSO redirect—bouncing users out of VTS to complete a core workflow.
I joined as Staff Designer and sole UX decision-maker across the company’s highest-priority AI initiative, and delivered an MVP of a fully native, end-to-end abstraction pipeline in approximately one month—after the initiative had been stalled in discovery for several months.
The existing third-party redirect created compounding problems: demo failures, cross-account document errors, and no native visibility into workflow status. Every new customer setup required manual handholding—and errors occurred with each of the first three instances set up. It was not scalable for beta expansion.
Users had no way to see what was happening with their lease submissions, no mechanism to verify AI-extracted data before it became the source of truth, and no discoverability for which properties were even eligible for abstraction. Embedding the pipeline natively was a hard requirement for General Availability launch.
Lease Administrators were the primary daily users—responsible for uploading lease documents, responding to abstractor questions, and managing the abstraction workflow. A key behavioral insight from research: their dominant interaction was passive status-checking. Not taking action—just trying to understand what was happening. The existing tool answered this poorly, if at all.
Asset Managers were secondary users focused on reviewing completed abstracts, using AI chat for portfolio-level queries, and approving or declining AI-abstracted lease data before it became the source of truth in VTS. Their core need: trust. They needed to verify AI output before committing it.
I was the sole UX decision-maker across all squads—the final call on interaction patterns, component architecture, status language, citation design, and approval workflows. I ran stakeholder reviews, defined the user research plan, co-authored success metrics, directed engineering handoffs, and managed third-party vendor relationships.
I also worked directly in the codebase, conducting design QA in live preview environments rather than isolated prototypes—and built AI-powered automation systems to accelerate delivery across the entire team. The scope was described internally as requiring a clone. I was the only designer in this role for the full delivery period.
Sole designer across 7–8 product teams simultaneously, spanning Lease Abstraction, Lease & Market, Activate (Visitor Management, Access, Work Orders), and Rise (Multifamily). The Lease Abstraction Pipeline was the highest-priority, most strategically critical project in the company’s AI roadmap.
01
User Research & Service Blueprint
Conducted user sessions with beta customers at RXR and MIT. Completed a company-wide Service Blueprint: 32 internal interviews across 13 roles and functions—the first formal blueprint ever produced for Lease Abstraction at VTS.
02
Workflow Mapping & Gap Analysis
Mapped the full abstraction workflow: upload through third-party processing, AI extraction, human review, approval, and final commitment to VTS as the source of truth. Revealed where the external tool broke down and where the native experience needed to be fundamentally rethought.
03
AI-Accelerated Prototyping
Built a production-fidelity prototype covering the full pipeline—upload wizard, dashboard, approval panel, citation system, and AI chat—at a pace that would previously have required multiple designers. Integrated Figma, GitHub, and Cursor into a continuous, partially automated delivery loop.
04
AI System Design
Worked directly with the LLM vendor on accuracy benchmarking, influencing how accuracy trade-offs between speed and depth surfaced in the UI. Designed the interface to set honest expectations—the AI was at 75% accuracy on structured data, 40–60% on unstructured. The design reflected the actual system, not an idealized version of it.
05
Structured Usability Testing
Developed UX playbooks for user interviews designed to be used by non-designers across the organization. Ran PURE scorecards via Lyssna targeting zero critical usability issues in the end-to-end flow across both user personas—with baselines defined and measured before engineering sign-off.
06
Live Design QA
Maintained a continuous QA loop by working directly in preview environments—catching interaction and visual issues in context rather than after the fact, and compressing the design-to-shipped cycle significantly.
A prototype shared internally on a Friday was being used in a live client demo by Monday—less than 72 hours later—when leadership shared it during a Piedmont client meeting without any prior setup or explanation needed. It performed in a live sales context before engineering had scoped the build.
The Abstraction Pipeline replaced a fragmented external redirect with a fully native VTS experience. Every design decision was grounded in the dominant behavioral insight from research: users’ most common interaction was passive status-checking. The architecture answered “what’s happening?” immediately, without navigation.
Abstraction Pipeline Dashboard — Task and document-set management with a clickable status chart, dual views, multi-level sorting, color-coded status chips, and an “Attention Required” section surfacing open questions and approval requests.
Document Upload Wizard — Multi-step flow guiding users through asset selection, tenant selection, document type categorization, and review—with auto-advance on selection to reduce interaction cost.
Approval Workflow — Right-side task detail panel with a structured approve/decline mechanism for AI-abstracted lease data. Mandatory comments required for declines. Full audit trail. Bridges the trust gap between AI extraction and the source of truth.
Citation System — A 3-case citation architecture linking AI-extracted lease fields to their source documents via bounding boxes and page-level navigation. Users can see exactly where the AI found each piece of data and validate it in context.
AI Chat Interface — Document-level and portfolio-level Q&A on abstracted lease data, with interface design informed directly by AI accuracy benchmarking—surfacing confidence levels honestly rather than presenting AI output as unquestionable.
Pendo Onboarding Guides — In-app onboarding at launch: homepage popup, 30-second video walkthrough, document vault guidance, and feature highlights—eliminating the per-customer manual support that had caused errors at every previous account setup.
72h
From first internal shareout to live client demo at a Piedmont meeting
1 mo
MVP delivered after the initiative had been stalled in discovery for months
0
Critical usability issues in end-to-end flow, per PURE scorecard targets met
The most important lesson from this project was about designing AI experiences honestly. It would have been easy to design an interface that made the AI feel more capable than it was. Instead, every interaction was designed around the principle that users needed to be able to verify, question, and override AI output. That trust-first approach is what made the approval workflow and citation system the most valued features of the pipeline.
The native pipeline directly unblocked VTS’s path to General Availability—resolving the operational errors and demo failures that had occurred with every previous customer setup. RXR went live with the first real document set through the new pipeline, providing real-world validation under actual operational conditions.