Our client is working on a web-based AI-powered document automation product that aims to improve document handling productivity for non-tech background enterprise users.
For this version, the product team has been working closely with financial operation specialists at Google,to improve the contract processing workflows through AI document suggestion and validation functionalities.
January - April 2023
Save time for financial data extracting process including invoice processing, contract processing, etc.
Deliver MVP version to Google within a tight time frame to gain a long term big name customer, so that to expand business.
Flora Wang
Xijia Zhang
Xianli Zeng
Wen
As a Design Associate, I solved the key usability issues and redesign the experience of the review and validation process. The new design led to a 75% reduction in document processing time and a 85% increase in productivity.
Main Experience 1
Main Experience 2
Main Experience 3
We summarized what we found and built our main persona:
Before we started the partnership with stakeholders, engineers had done a demo version which prioritized technical feasibility but lacking of enough consideration of user experience.
How might we build users’ trust and reduce the frictions in AI suggestion and validation user flows?
How might we engage user during AI training?
How might we build user trust for Orby AI suggestion?
How might we optimize the key term suggestion and validation flows?
How to start the Orby experience?
Within the tight timeframe, we close worked with AI engineers and fast iterated. While we followed user centered design process, we offered flexibility to work on new use cases that were brought up by AI engineers during the process.
The process of AI training involves Orby learning from user actions like reviewing documents, identifying terms, and manually inputting them into sheets. As AI takes time to capture data and learn, it's important to consider engaging the user throughout the process. To address this, we came up with two solutions: Orby chatbot & Orby training mode editor.
After receiving feedback on the need for a more engaging experience, we made three significant modifications to improve user engagement:
2 Training Options
reviewing in PDF format or through the Orby interface.
Instruct users step by step.
Encourage users to complete the training process.
How might we optimize the key term suggestion and validation flows?
How might we build user trust for Orby AI suggestion?
To address these 2 questions, we analyzed user needs and incorporated the most useful functions and groupings that users most likely to use, and offered customized options such as sorting, search, and multiple filters. However, we received feedback stating that we had included too much information and hadn't considered the actual user flow, resulting in a disorganized hierarchy that made it difficult for users to find a starting point. These comments prompted us to reconsider :
How to create a Guided Experience?
Iteration
During the mid-fi stage, we prioritized the necessary features and developed two options and compared the pros and cons:
After evaluating the user experience and identifying the issues, we decided to merge the advantages of each into one. In the iteration, we separated the suggestion list and edit panel to allow users to avoid constantly looking left and right. Furthermore, by placing the most important information at the top, we can give users a quick overview of the contract.
Edge Cases
Though our clients are quite satisfied with the revised version, they also pointed out a few edge cases that we had not previously considered:
1.How to review long paragraphs?
2.How to review discrete information/cross-page terms?
3.How to add missing items to the suggestions?
1.How to review long paragraphs?
Our team has designed four floating cards that are of equal height but varying widths. This allows for easy customization to fit the length of different paragraphs while maintaining consistency.
2.How to review discrete information/cross-page terms?
Considering this use case, we add a plus button on the right corner, by clicking it, user can add another field and select content from the contract into the card, and by clicking “minus” button to remove.
3.How to add missing items to the suggestions?
When Orby cannot find the corresponding contents in the contract, this key term will move under the “not in doc” category.
By analyzing the issue of the existing design and gathering feedbacks from users, we found out that the suggestion panel needs to fulfill three fundamental requirements:
1.Users should be also able to Quickly identify different files
2.Users need to manually add/remove files in here
3.Provide access to the Orby web app interfaces
In the first concept, we explored 2 options - V1/V2
We received feedback that the card view is the best option for scalability. We applied it during the iteration and solved design questions such as:
How to quickly locate the files the users want to review?
How do we determine the priority of the information displayed?
Lessons Learned
1. Root in user needs and get design ideas from real use cases.
2. Keep the information hierarchy clear and concise, avoiding unnecessary design elements.
3. Be mindful of edge cases and different scenarios to ensure the system is adaptable to various use cases.
4. Keep stakeholders in the design discovery. This can help ensure that everyone is on the same page from the beginning and that everything is clear when it comes to the final product.
