Legal Document Drafting

The Problem

• Time-Consuming and Error-Prone Process: Traditional legal document drafting involves extensive cutting, pasting, checking, and updating from existing documents. This process is not only time-consuming but also prone to errors.


• Tedious Proofreading: Lawyers often spend a significant amount of time proofreading documents to ensure accuracy. Despite careful checks, there's no guarantee that errors won't slip through, leading to potential legal and financial consequences.


• Risk of Errors: A Thomson Reuters study highlights that manual drafting is associated with a higher risk of errors, which can be costly and damaging to a lawyer's reputation and client trust.

The Goal

To introduce a document drafting capability that utilizes generative AI in the application. This will assist with:

• Time Efficiency: Generative AI significantly reduces the time spent on drafting documents. For instance, AI-powered tools can produce accurate first drafts quickly, allowing lawyers to focus more on strategic activities rather than on time-consuming drafting tasks.


• Reduced Risk of Errors: By automating the drafting process, generative AI minimizes the risk of human errors, such as incorrect details or formatting issues, thereby enhancing the overall accuracy and reliability of documents.


• Enhanced Productivity: Lawyers using generative AI report increased efficiency in reviewing documents and mitigating the risk of errors. This technology enables them to use their time more effectively on high-value activities, rather than on repetitive and mundane tasks.

Design Process

Research

We conducted a survey through WalkMe and collected responses from Jan 26 - Feb 20. The purpose of the research is to:

• Gauge how interested our customers would be in having a document creation tool within SimpleLegal
• Gain insight into what vendors use today and what about them is liked/disliked
• Learn what document creation feature are important to clients

Our results showed that the majority of respondants do not use a document creation tool as many are wary of the high cost of such a tool. We wanted to explore how we could offer a document creation tool inside of SimpleLegal, thus allowing our users to draft, brainstorm, strategize, and generate content for any type of legal document more efficiently.

Ideation

This was the first generative AI feature for the product, and I took on the sole design responsibility for this project. I researched several different applications that had document generation capabilities, and brainstormed on ideas to implement this feature into the application.

One of the biggest challenges for me was testing out the prompts and designing the prompt inputs so that we could get closer to accurate and relevant results. We wanted flexibility with the prompts so that users can create first drafts of various legal documents, including contracts, memos, and legal briefs, based on learned patterns and data from existing content, but also give enough context to receive accurate results. After numerous testing, my project manager and I decided to provide these fields and supply the final prompt to generative AI:

• Include user input from the Drafting Context field in the prompt.
• Include user input from the Practice Area field if provided, in the prompt.
• Include user input from the Type field, if provided, in the prompt.
• Include user input from the Writing Style field if provided, in the prompt.
• Include user input from the Tone field if provided, in the prompt.
• Set the temperature for GenAI based on the value selected for the Creativity field.
• Include user input from the Length field in the prompt.
• Include user input from the Language field in the prompt.

Design

I explored multiple design approaches for how users input prompts and preview generated documents, drawing inspiration from industry best practices and our own product ecosystem. My initial concept used a modal where users could enter prompts and select relevant fields. However, after testing and considering usability, I realized this approach introduced unnecessary scrolling and disrupted user context. To address this, I pivoted to a more integrated design that repurposed an existing, familiar component within the application. This solution not only preserved context and reduced cognitive load but also made better use of screen real estate, allowing users to understand their actions in the flow of their task.