DraftWise helps lawyers reduce tedium, focus on strategy with Azure AI Foundry Models

Throwing an AI lifeline to lawyers mired in data

If you ask a lawyer what the most rewarding aspect of contract creation is, they’ll tell you it’s the chance to be creative and offer the client novel strategic advice. Ask them where they spend most of their time, and the script flips. 

“Lawyers I talk to often refer to themselves as ‘cut-and-paste drones,’” says James Ding, founder and CEO of DraftWise. “Our research across the profession confirms this.” As a software engineer, he has led teams across multiple industries where privacy, data security, and advanced search capability are crucial. So Ding knew the legal profession could create a sea change by implementing state-of-the-art generative and search retrieval AI models to draft higher-quality contracts faster. 

Collaborating with customers, creating purpose-built solutions

The DraftWise developer team developed the strategy for bringing their AI-powered contract drafting (Smart Draft) and review (Markup) products to market. Because of their deep trust in Microsoft based on extensive experience with Microsoft Azure, the company founders turned to Azure AI Foundry Models. In the model catalog, they discovered Cohere’s set of models: Command for text generation, with Embed and Rerank used for search and retrieval. “Once we knew that we could run Cohere models on Azure, we were sure we’d have the best of both worlds,” explains Ding.

The DraftWise team went directly to prospective customers to fully grasp the needs and preferences of lawyers. They found a customer base committed to the ideals of their profession. “We knew we could provide the best experience for lawyers by aligning with the tools they use every day and deliver DraftWise convenience from within their current workflow,” says Will Seaton, Chief Customer Officer at DraftWise.

The screenshot above shows how DraftWise serves contract clause suggestions.

Microsoft also offered a perfectly aligned cloud platform and AI tool set. Azure fulfilled the DraftWise commitment to security and data privacy, coupled with reliability at scale. While leading engineering teams at his previous position at Palantir, Ding was impressed by the platform’s unified capabilities. “Microsoft serves a customer base in highly regulated spaces,” says Ding. “Azure offers reliability, availability, and control over data that our engineers need. And Microsoft offered more model choices than its competitors.”

Azure AI Foundry is highly secure and robust; it easily scales to customer demand. Aligned with Microsoft’s responsible AI principles, it helps to ensure the customer’s data repository stays safe. DraftWise thus retains precise control over which information is collected, stored, and used for training purposes.​

Developers could search across more than 1,900 models within Azure AI Foundry Models, then test and evaluate from a preferred pool for the optimal choice. Regardless of the model one chooses, the SDK is universally used and therefore publicly documented. That means developers can change models simply by changing configurations.

Fusing Microsoft and Cohere ingenuity

DraftWise selected the Command model for its high-quality text-generation capabilities, which provide a natural, approachable interface to its clause-generation tool within Smart Draft.  

The company powers its DraftWise retrieval-augmented generation (RAG) system with the Cohere Command, Embed, and Rerank models. The Cohere Command model generates citations that provide the sources for the information it presents and offers strong text-generation capabilities. But lawyers often aren’t looking for general knowledge about a legal case. They're looking for specific answers found in their clients’ contracts, communications, and other documents.

The DraftWise platform incorporates Cohere Embed, a multimodal embedding model that enables advanced search through these documents using semantic meaning rather than simple keywords. Cohere Rerank then ensures that the most relevant results show up at the top. These three models provide accurate information to support complex legal cases with verifiable citations so the user can easily confirm material in source documents. Cohere Embed can process all data, including images, diagrams, tables, and graphs, so the DraftWise solutions aren’t limited to text searches.

“We’ve built our models for real-world business use cases,” says Nicolás Morales, Head of Customer Experience and Partnerships at Cohere. “Cohere’s secure, enterprise-focused generative and retrieval AI models seamlessly integrate into your workflows, delivering powerful results with broad multilingual support, high accuracy with citations, and cost-efficiency.” 

“Azure AI Foundry Models is an absolute game-changer. It’s ignited development for us, improving developer efficiency by 60% over traditional methods,” adds Ding. “In the early days, we were trying out various model providers to see which one offered the best out-of-the-box performance. Had we chosen a single model provider with a proprietary SDK, we wouldn’t have the flexibility we’re now getting from the Microsoft single SDK that powers everything.” 

Building an AI solution with trusted Microsoft tools 

The DraftWise development team also turned to Azure AI Foundry Models to access the o-series reasoning models it uses to augment search results returned by Cohere. The team applies reinforcement fine-tuning to improve model performance further, using DraftWise’s proprietary data to tailor the results to the legal profession. Ding describes a typical interaction between the various models that underlie DraftWise solutions: a lawyer asks DraftWise to find an example of a termination clause in a license agreement. They may add a statement like, “Include all the extra features and make this friendly to the side I represent.” The Cohere model retrieves several examples. Then the fine-tuned o-series reasoning model responds to the lawyer’s request—“Tell me what I’m missing and generate the change for me”—to adjust the suggested draft. 

DraftWise internal benchmarks show a 30% improvement in search result quality since incorporating the fine-tuned models into their solution. Reinforcement fine-tuning shows significant improvement in these results, too. “We have the models illustrate their reasoning before executing on the answer, but the thing that shines is the ability to sharpen the response with reinforcement fine-tuning,” adds Ding.

Lifting legal customers up with an AI helper

The threefold increase in application API calls made by DraftWise solutions to models over just the first quarter of 2025 indicates spectacular growth in usage. That comes as no surprise to Seaton. “We’re seeing tremendous excitement about AI in the legal profession. In addition to the productivity boosts and faster new lawyer training our customers are experiencing, they’re advising their clients on how they can also use AI to improve their own operations,” he says.  

Ding and his team are gratified by the changes that DraftWise tools make for lawyers. “It’s incredible how capable our best users have become with DraftWise since we’ve introduced Azure AI Foundry Models into the product,” Ding concludes. “Several of our most elite users have been promoted to partner at their firms. The time they save on mundane tasks goes into value-added work that supercharges their performance. Building our solutions with the technology we trust is the key to dramatically impacting the industry.”

Discover more about DraftWise on LinkedIn, X/Twitter, and YouTube.