Led the design, research, and implementation of a solution that replaced inefficient manual document extraction with a faster, a more reliable process. It achieved over 90% accuracy, processed documents in under 3 seconds, scaled to handle 100,000+ documents daily, and resolved 95% of issues within 24 hours—delivering speed, reliability, and exceptional user satisfaction.

Pain Points
The manual data entry approach makes the initial document processing methods time-consuming. This leads to backlogs, an increased risk of human error, and delays in tax audits. Too many back-and-forth emails lead to omissions and loss of communication in an ongoing project.

Business Goal
A design solution that:

• Automates data extraction from various formats (PDFs, Word files, scanned images) within the XTM platform.
• Streamline workflows
• Improve translation accuracy and enhance user experience for tax auditing
• Enable seamless collaboration and communication between engagement teams
• Ensure compliance with each client’s data residency requirements,  enable users to collaborate seamlessly, and extract, review, and export documents effectively.

Key Outcomes:
- A seamless collaborating application
- With a strong conversion rate
- Measure success: Analytics/Research/Design/Testing

Due to the sensitivity and proprietary nature of this project, some information will be withheld and modified. This project will be referred to as DocXtract and the Client as XTM

At the kickoff meeting, we set the foundation for this project's success by bringing the stakeholders together, including the product owners, business analysts, frontend and backend engineers, and design teams, to align on the project's objectives, timelines, and key deliverables. During the meeting, we reviewed the project scope, defined the user stories and use cases, and agreed on communication channels and project milestones.

We collected functional and non-functional requirements for the XTM Document Extraction solution. This includes understanding the user needs, business objectives, and technical constraints to ensure a comprehensive solution, and deciding on MPV features and nice-to-haves vs must-haves.

I interviewed users in various roles within DocXtract to understand the challenges faced by engagement teams, the current state of data extraction, and processes across different access levels. I gathered their feedback and synthesized my findings -  created personas to map each pain point to a user scenario.

Additionally, I reviewed competitive solutions for document extraction to understand best practices, industry standards, and user preferences.

I led a design sprint/session with the product manager, product owner, engineers, and stakeholders to review requirements, analyze business logic, and understand relevant legal guidelines to build domain knowledge. Brainstormed and mapped out entry points, access, logic, user stories, user flow, and journey maps. This foundational understanding enabled me to engage effectively with stakeholders, leverage available analytics to understand the problem we are solving for, and structure my thought process.

How might we come up with a solution that:

1. Reduce the time it takes for engagement teams to collaborate, design for automation to suppress the manual approach

2. Suppress email communication, design for in-app chat/message functionality

3. Suppress manual Excel document tagging/mapping techniques that lead to human error/omission

I did some wireframing layout exploration to map out and tackle some of the product's complexity.  With time, I began to iterate on the wireframes to streamline the workflow and create a smoother experience that tackles the above pain points. 

Solving Manual Entry Friction

As part of the Document Extraction solution, I designed a flow where Engagement Members can upload documents directly to begin preparation, mapping, and model execution—all within the platform. This replaced the previous manual process of entering data into Excel, which was time-consuming, error-prone, and hard to standardize. By streamlining these steps into an intuitive workflow, the design reduces manual effort, improves data accuracy, and speeds up the overall extraction process.

Solving to Eliminate Manual Tagging Bottlenecks

To streamline the extraction process, I designed a workflow that allows Engagement Members to train a document model, map data directly from the uploaded document, and add contextual notes—all within the same interface. This replaced the previously tedious manual tagging approach, which relied on static tools and inconsistent methods. Once the extraction is complete, the document can be seamlessly routed to a manager for verification and approval, creating a more structured, efficient, and collaborative review process.

After rounds of reviews and iterations, I handed the design to the dev team. There's a full-fletched design system in place with a detailed storybook for developers to leverage. I worked closely with the dev team to review each feature as soon as it was developed, review component usage, and be sure there were no inconsistencies on the approved design to development. 

I leveraged Figma plugins and Web Content Accessibility Guidelines (WCAG) to ensure my designs were accessible to all user types. I used plugins like Stark and Contrast to test color contrast ratios, ensuring text and UI elements met or exceeded WCAG AA/AAA standards. Additionally, plugins such as Able helped simulate various visual impairments, allowing me to assess usability from diverse perspectives. I also conducted keyboard navigation checks and screen reader compatibility reviews to enhance accessibility for users with motor or visual impairments. This approach ensured an inclusive design that prioritizes usability for everyone.

The budget doesn't allow for prototype usability testing, but I worked closely with the QA and Analytics team to test each feature as soon as it was developed. We ran some moderated/unmoderated, and performance testing with the Engagement Partners, then documented our observations and how they interacted with the application.

The Document Extraction application aims to streamline the process of automatically extracting relevant data from various document formats, enhancing efficiency, accuracy, and scalability. Note: the above HI-Fi mockup is not a representation of the entire application, I narrowed down what is shown in this case study to a few key features.

Challenges
While the collaboration between design and development was strong, there were recurring challenges with implementing features to match the intended aesthetics and patterns, despite being functionally complete. This often led to time-consuming back-and-forth iterations to align the development output with the design.

To address this, we introduced review sessions where engineers demoed their work, and I provided feedback to refine it until it aligned closely with the design before the code was finalized. A key issue was some engineers' unfamiliarity with the Design System Storybook. To resolve this, I organized a knowledge transfer session between my scrum team and UX engineers from the Design System team to enhance their understanding and usage of the code base.

Additionally, I created a dedicated Confluence page to track design gaps for each feature, tagging the responsible engineers. These gaps were closed out once the refinements were complete, ensuring better alignment between design and development.

Success Metrics
Below are some metrics gathered from the analytics team some months into launching the application

• Extraction accuracy of at least 90%
• Processing speed of under 3 seconds per document (in comparison to hours of manual extraction)
• Now handles up to 100,000 documents per day without performance degradation.
• User satisfaction was tracked through feedback, with a target of resolving 95% of reported issues within 24 hours.
• This project was successful because it consistently meets or exceeds these metrics, providing reliable, fast, and accurate document extraction for the business.