Contract Analysis: From Concept to Market
How we transformed contract review from a weeks-long nightmare into a matter of minutes, launching an AI-powered tool that became a cornerstone of Thomson Reuters' legal tech suite.
Lawyers at Thomson Reuters struggled with a daily document analysis process, where manually identifying the repetitive and mundane tasks that characterize their work led to inefficiencies. This 0:1 initiative aimed to create a tool to streamline these processes and boost productivity in document evaluation.
My Role: Senior UX Designer
Team size: Two, a junior designer and myself.
As a Senior UX designer, I led the design of a tool that would benefit both users and machines. My responsibilities included analyzing definitions, identifying areas for improvement, and collaborating with cross-functional teams to bring the vision to life.
Hypothesis Formation and Early Prototyping
Based on our SME insights, we formed several key hypotheses when testing for real estate agreements (since they most consistently have the same format):
🔍 Hypothesis 1: Intelligent Clause Detection
Theory: AI can identify and categorize contract clauses more consistently than human reviewers, while learning from user corrections to improve accuracy.
Early Prototype: We built a simple clause detection engine trained on 10,000 contracts from our legal database. The prototype could identify 15 common clause types with >90% accuracy.
SME Testing Result: Lawyers were impressed by coverage but frustrated by false positives. They needed precision, not just recall.
🤯 Hypothesis 2: Risk Scoring with Context
Theory: Different organizations have different risk tolerances. The tool should learn each firm's preferences and provide contextual risk assessments.
Early Prototype: We created a risk scoring system that could be calibrated per organization, with explanatory text for each identified risk.
SME Testing Result: This was the breakthrough. Lawyers loved having their institutional knowledge codified and accessible to junior team members.
🤝 Hypothesis 3: Collaborative Review Workflow
Theory: Contract review is rarely a solo activity. The tool should support collaborative workflows with role-based permissions and audit trails.
Early Prototype: We built a simple commenting and approval system that tracked who reviewed what and when.
SME Testing Result: Workflow features were immediately adopted and became sticky - firms didn't want to go back to email chains and version confusion.
🎓 Hypothesis 4: Learning from Corrections
Theory: The AI should learn from user corrections and feedback, becoming more accurate over time for each organization's specific needs.
Early Prototype: We implemented a feedback loop where user corrections would retrain organization-specific models.
SME Testing Result: This created a sense of ownership and investment. Lawyers felt like they were training their own AI assistant.
Reflection: What We Learned
Quantified Impact
Matter Management
Crafted for optimal functionality, I took advantage of Thomson Reuters’ acclaimed matter management system and tailored it to center around documents in this specialized tool. The essence of the tool I designed revolves entirely around documents, making them the focal point of every action. I’ve strategically placed all the necessary actions right at your fingertips, ensuring a swift start to document analysis.
As soon as you upload a document, our tool kicks into action, analyzing it promptly to minimize any waiting time. Once the document is ready, you can effortlessly dive right in and begin your work. It’s all about a user-friendly and streamlined experience, prioritizing your efficiency in handling document-centric tasks.
Questions Panel
To ensure a comprehensive document analysis, our tool features a structured questions panel covering every category of queries a lawyer typically addresses. The system then systematically retrieves and compiles answers for each question. It’s crucial to note that, based on user research insights, lawyers consistently prefer to review and validate the generated answers, never relying solely on automated outputs.
Upon a lawyer’s meticulous review and approval, the tool marks the question as answered. This approach not only provides transparency in the review process but also enables a clear record of which lawyers addressed each question and the thoroughness of their assessments.
Split View
Through my user research, it became evident that the split view feature was essential for our tool. The true strength of Contract Analysis unfolds when we analyze two documents simultaneously. This functionality empowers lawyers to efficiently compare documents and cross-reference answers, facilitating the identification of necessary revisions.
Analyze Type / Layers
Contract Analysis does four different types of analysis in documents. They are:
Finding answers to questions for review
Showing definitions
Finding deviations
Discovering parties/ownership
Addressing a significant design challenge, I focused on enabling users to seamlessly navigate multiple layers concurrently. In collaboration with Thomson Reuters’ A11y metrics and an Accessibility (A11y) specialist, I successfully designed all four analyzing layers to be accessible simultaneously, ensuring compliance with A11y standards.
Mobile Optimized
Recognizing that the optimal user experience would be on a desktop, I also prioritized providing lawyers with enhanced flexibility. After conceptualizing and designing the ideal user experience, I took the initiative to fully optimize the entire app for mobile usage. Now, users can seamlessly execute all steps on their mobile devices, ensuring accessibility and convenience on the go.
Multi-Doc Analyzing
Once I honed in on the strategy of saving lawyers time through document analysis, the logical progression was to scale that efficiency. My goal was to instill confidence in lawyers by enabling them to apply the tool’s analytical prowess across a multitude of documents, exponentially saving them time. I crafted an intuitive interface allowing users to analyze multiple documents concurrently. To enhance comprehension, I integrated data visualization, providing lawyers with a deeper understanding of the analyzed information.
Knowledge Management
I crafted a dedicated section within the app to empower users in customizing the system according to their specific needs. In this section, users can educate the system by introducing new questions, providing deeper insights for existing queries, and even fashioning their own templates tailored to their preferred working style. This personalized approach ensures that the system aligns seamlessly with the unique requirements of each user, enhancing their overall experience.
Reporting
The last step I had to design after the lawyers did all this work to use the tool and analyze all their documents was a reporting function. Users can download a document that takes all the artifacts they made as a PDF or Excel document. I also designed a read-only view for users.
