Patty for Uber Intranet is an end-to-end AI-powered contract document recognizing and processing app. The Dev team brings AI power to everyday work and automates mundane and repetitive tasks. Our team build AI-powered contract reviewing and processing tool and fine-toned the core for Google Financial's process outsourcing team.
Project Type
Mentor project, AI-powered intranet platform, UX research report
Team
Yitong Du, Starkey Shi, Eva Wang, Jinhan Wu
My Role
Project manager, UX research, UX design, Prototyping, Usability testing, Design System
Duration
Mar 2023 - Aug 2023
OVERVIEW
About Patty for Uber
The design process is not linear
As impact investing and climate change were unfamiliar and complex areas for our design team, we devoted considerable time to researching, outlining, and redefining our objectives. We consistently scrutinized and updated our knowledge throughout the design journey. I discovered that alternating between the discovery and design phases was advantageous, as it allowed us to gain fresh perspectives and modify our design approach over time.
OVERVIEW
My Contributions
Conducted usability testing sessions with the education and engineering teams to gather feedback and iterate on the design.
Created high-fidelity prototypes and handed them off to the engineering team.
Built and maintained a design system to unify the user experience across mobile and web platforms.
REFLECTION
Takeaways
The design process is not linear
Conducted usability testing sessions with the education and engineering teams to gather feedback and iterate on the design.
Created high-fidelity prototypes and handed them off to the engineering team.
Built and maintained a design system to unify the user experience across mobile and web platforms.
REFLECTION
Next Steps
Conduct two rounds of iterations with pro reviews and usability testing results.
Due to the project schedule, the design team only ran one usability testing session. It helps to frame the high level user flows and key screen layout. We are looking for having more usability tests to get feedback and input on:
Cr.
Built and mai.
Align the measurable success metrics with stakeholders and track the data.
I appreciate the significance of evaluating the impacts of design solutions to confirm the hypotheses and guide ongoing refinement.
Given the opportunity, I'd work diligently to define quantifiable indicators for the design results and monitor their influence to grasp their efficacy thoroughly.
OVERVIEW
Project Brief
Google financial specialists spend much time reviewing contracts and extracting data manually. It prevent them from doing more strategic work.
Hardly few solutions currently exist in the market to make the document process easier. Automation tools available would not save their time and effort.
OPPORTUNITY
Design Opportunity
The product team proposed a AI-powered document processing and managing platform for users, where they can streamline the document review process, reducing the time spent on manual tasks and enhancing the accuracy of entity extraction.
STRATEGY
Design Process
We aligned the client team to follow an iterative design process in which designers focus on the users and their needs in each phase. Our design teams involved the stakeholders throughout the process via a variety of research and design techniques to create highly usable products for the target users.
USER RESEARCH
Competitive Analysis
Based on the design scope from the client, our team conducted a competitive analysis and a couple of stakeholder interviews to learn about initial insights:
USER RESEARCH
Persona
USER RESEARCH
User Journey Map
USER RESEARCH
Insights from Research
Insight 01
Priority
Provide the users with more valid info and help them establish work priorities to increase efficiency.
Insight 02
Navigation
Need a clearer navigation pattern between a key term checklist, original document and extracted document
Insight 03
Trust-building
Prioritize building user trust to help the user gain usage of the Orby system.
DESIGN
User Flow
DESIGN
User Flow Findings
Training workflow Iteration summary:
We provide two training path - Original Traing Workflow V.S. Orby Training Editor. Training in orby allows users to become familiar with the system earlier and helps to pass time-consuming training steps. It also allows users to select the initial training workflow if desired.
DESIGN
Ideation
How might we encourage users to finish the training process?
How might we help users build trust in the AI process?
Chatbot UI - AI Training Phase
A chatbot has been designed as an entry point for users to start interacting with Orby during the daily workflow.
How might we streamline the validation flows?
Page Layout - Validation Process
The two-column layout provides more space for contract display and follows the Z-readable pattern for easy entity review.
Navigation - Validation Process
Our aim is to direct the user toward examining the terms attentively using a tabbed filter, thereby enhancing precision.
How to assist users in identifying different tasks when there are lots of files with abundant information?
How can we display as much information as the users need within a regulated area?
How to add tasks that are not recognized by Orby AI?
Suggestion Panel Layout
A chatbot has been designed as an entry point for users to start interacting with Orby during the daily workflow.
How might we help users find undetected entitles?
How might we help users review different types of entitles?
Additional Contents Not in Doc - Corner Case 01
"The review every month will be too much for me. I don't think I will do it every month." - Participant 1
"Filling such a long review sheet is time-consuming." - Participant 2
Final Round
Long Piece Contents - Corner Case 02
"The review every month will be too much for me. I don't think I will do it every month." - Participant 1
"Filling such a long review sheet is time-consuming." - Participant 2
1st Iteration
Final Round
Discrete Contents - Corner Case 03
"The review every month will be too much for me. I don't think I will do it every month." - Participant 1
"Filling such a long review sheet is time-consuming." - Participant 2
1st Iteration
Final Round
Identifying Table Contents - Corner Case 04
"The review every month will be too much for me. I don't think I will do it every month." - Participant 1
"Filling such a long review sheet is time-consuming." - Participant 2
1st Iteration
Final Round
DESIGN
Hi-Fi Prototype
01 Tutorial for first time users
Minimize effort to learn how to use the product by providing quick and easy tips on key concepts and features of Orby AI.
02 Get started and check work orders
Orby will notify user when there are new tasks, users can open it by clicking the icon. Suggestion panel will help user prioritize their tasks productively.
03 Add new tasks
Orby can auto-sync files and allows users can add a new task manually if Orby has not detected certain refreshed tasks.
04 Increase the success rate
Streamlined classifications into All, Low confidence, and Not in doc minimize unnecessary data and error frequency—tabbed filtering supports users in precisely verifying terms.
05 Optimize streamline the process
The Orby score increases as the user corrects the terms that establish confidence between users and the Orby. After all terms being reviewed and confirmed, Orby’s confidence score goes to 100 with firework animation. And Orby will help user export the data to desired platform.
REFLECTION
Takeaways
The design process is not linear
As impact investing and climate change were unfamiliar and complex areas for our design team, we devoted considerable time to researching, outlining, and redefining our objectives. We consistently scrutinized and updated our knowledge throughout the design journey. I discovered that alternating between the discovery and design phases was advantageous, as it allowed us to gain fresh perspectives and modify our design approach over time.
Work with stakeholders
We developed this product for a professional client, the Google Financial team. Collaborating intimately with our client, we held numerous interviews and discussions. We offered a complete hand-off package and a well-communicated cooperation experience upon delivering our product.
REFLECTION
Impact
The project resulted in an 80% increase in reviewing efficiency, decreasing processing time from 60 min to 5 min per task.
Google confirmed the cooperation for the following AI-assisted invoice processing and data comparison phases.