Name: Legal contracting with AI: how DraftWise leverages Cohere’s models in the Microsoft Azure AI Foundry
Uploaded: 2025-07-17T16:30:04.957Z
Duration: 25 min 12 s
Description: Legal contracting with AI: how DraftWise leverages Cohere’s models in the Microsoft Azure AI Foundry

Transcript for "Legal contracting with AI: how DraftWise leverages Cohere’s models in the Microsoft Azure AI Foundry": Hi, everyone. Thank you so much for joining this webinar. We will get right into it. It is 11:30 on the dot. So I'm Heather, and I run our webinar program at Cohere. And before we jump into the great content that we have prepared for you today, I just want to share a couple of reminders. So first of all, if you are having any issues with hearing the speakers or if your video is lagging, there are some options at the bottom of the window to, unmute the sound or to turn on light mode, which may help if you're having any technical issues. And on the right hand side, you'll see that there is a chat panel, and we encourage you to drop in any messages and have some lively conversation there as the webinar takes place. And on the far right of the chat panel is a q and a section, and we ask that if you have any questions that you drop those in there so it's easier for our speakers to see them. And we're hoping that we will have some extra time at the end of the webinar where they can answer some of your questions live, and, otherwise, we can follow-up afterwards to answer your questions. And between the chat and q and a tabs, there is a docs tab which has a couple of links, that will also be helpful and provide some additional information. So we encourage you to take a look at that. And finally, we will be sending the on demand recording, to you afterwards via email, so please be on the lookout for that and share it with anyone who wasn't able to join today. And with that, I will pass it off to Nick to kick us off. Thank you, Heather. Welcome, everyone. Today, we're diving into a story of innovation transformation in the legal industry. So over the next thirty minutes, we're gonna explore how DraftWise, a pioneer in the AI legal space, partnered with Cohere and Microsoft to revolutionize the way that lawyers work. We're gonna learn about the their strategic decisions behind their AI model choice, the development of the RAG system, and really the impact that it's had on productivity and client service. So let let's get started. My name is Nick Morales. I'm head of customer experience at partnerships at Cohere where I have the privilege of working with customers every day on their AI journeys, and I'm thrilled to be the moderator for this session. I'm joined by Will Seaton Seaton from DraftWise and Trupti Parkar from Microsoft. So, Will, why don't you kick things off in telling us a bit about yourself and your role at DraftWise? Absolutely. Thank you so much, Nick, for having me, and thank you everyone for sharing your time with me today. So I am the chief customer officer here at DraftWise. My background is in product management and data science with, particular focus on natural language models. So, of course, I find the legal industry, how we can apply large language models and this latest technology to the law to be incredibly interesting. And we'll talk a little bit about an example of how we're using that today. And, Trupti. Hi, everyone. I'm Trupti Parkar. I'm a AI product manager at Microsoft. I'm so excited to be here. I drive Microsoft's partnership with Cohere, and we're excited to see how, we can use this in legal respects here today. Awesome. Thank you both for being here. I really appreciate the the industry and technical expertise that we have on this call. So, well, there's so much to get to, but there's actually no better way than start by seeing a demo of what DraftWise really, you know, cutting edge solution looks like, powered by the Cohere AI model. So I'm gonna turn it over to you first to show the audience what the intersection of AI and law looks like today. Yep. Absolutely. There's no better way to start than jumping right in. So let's show a demo of how, DraftWise is using Genentech AI and Generative AI, to augment the drafting review of a common real estate contract. So So here's the backdrop. We have here open a in Microsoft Word, not Versus Code, a purchase and sale agreement. That's basically an agreement you use when you wanna buy commercial real estate. And the background here is I'm a lawyer and my client said, hey, I really like what you did the last few times around. Just do the same thing for this deal too. So the first thing I'm gonna do is I'm gonna have Dropwise's agent perform some research to see exactly what the client meant. I'm gonna prompt it to look for similar deals to this one that we've done with that client that they were happy with, find opportunities to improve the one we're currently working on by comparing it to those other those other ones. So in short, we're trying to improve the position of the buyer in the, all the positions they're taking in this purchase. And in a deep research like way, what DraftWise is doing is it's actually doing that work, finding those deals, and it's able to do this because it's connected to the law firm's contract repository. And also, it can see all the negotiations that happened in the past. This next one speaks to the fact that we're just gonna make the the seller provide this material sooner. As we're in previously, we would have to tell the buyer or to tell the seller when we're good with the changes. But should we change our mind, we can't, reverse that. So this protects us against that. The next two talk about, making obligations optional or removing them entirely. This next one is kinda subtle. The words say one thing, but in reality, the implication is we can get a more robust insurance policy. A number of changes, about 10 changes, all of them, were found through analyzing other deals and comparing them to this one. And what's gonna happen is it's gonna stage the changes for me in a dispute so that I can easily review the changes it plans to make. And should I be happy with those changes, I can accept them into my document in a single click, or I can prompt it to revise or give it alternative feedback much as you would with a coding agent. So you can see here, it's making changes related to, that insurance provision, the thing that's gonna help us get a more robust insurance policy. It's making certain obligations optional as opposed to mandatory. Basically, all of the improvements that has surfaced before, it's found a way to fix them, to remedy them by using the language that we've used before. And what I'm gonna do is, you know, many law firms have a standard document of what they consider structural quality. I'm gonna cross reference it against that. I'm also gonna generally cross reference it and make sure that I haven't introduced inconsistencies. And that's what's happening here. And voila. No inconsistencies introduced, no structural problems introduced, and then tediously cross reference against our quality checklist. Awesome. So, Will, that's looks great, and it's so so awesome just to see it in action first as we as we jump into, you know, the you know, all the considerations that went to building it. And maybe we start there, you know, as innovators in leveraging AI for legal professionals. You know, what inspired the creation for DraftWise? And, you know, what were you thinking through as you identified specific pain points to address for for the industry? Yeah. Absolutely. I think it's always cool seeing how lawyers use DraftWise and the technology that we're building. But the original motivation came because when you ask a lawyer, why did you enter the law, why did you want to practice, they will say that they wanna provide strategic advice. They wanna guide their client, in best ways of reducing risk and practicing as a business or an organization. They never cite the tedious tasks as motivators for their career. They don't wanna be doing the cut and paste of, different language from one place to another. They don't wanna be doing, you know, searching through a document management system just to try and find, the right results. And so the founders of, DraftWise, they came from day law, practicing a decade of some of the best firms in the industry. They came from top engineering companies, where they were had a background in privacy and data security and advanced search. And they realized that by marrying all of this technology together, you could, help to shift how a lawyer spends their day from some of these mechanical tasks, which take too much time, into more of the tasks around understanding your client, advising your client, that they actually want to work on. And so as part of this journey, building technology for the legal industry, DraftWise is now used by over half of the ball 10. We're used by top law firms around the world at multiple regions around the globe. And all of that, I think, is underpinned by our ability to bring some of the best technology in the world, in particular, things like text generation, information retrieval, and search, directly to lawyers where they're working. Right. And and so and as the solution was developed, were there, you know, specific features and capabilities within Cohere's model that, you know, made it the ideal choice for powering the smart draft solution? And, you know, how do those features and capabilities align to the vision of the product? Yeah. I think there's two things worth talking about here. One is, what functionality did we need? And then another one is, how do we come to discover that functionality was out there? So what we needed, when supporting our lawyers with our particular problem, we needed the ability to find specific information in client contracts, communications that they had, get access to exactly the right, data, within our system. And so we wanted to lean into retrieval augmented generation as a process there. We needed to lean into, retrieving relevant information from a large database, using that information to generate a specific response to a specific query, and really focused on how can we get, the highest accuracy, the best retrieval, and deploy that against the problem of recommending what a lawyer, should be doing in each individual situation. So that's what we're looking for and then how we came to discover, that Cohere had a lot of solutions for this, that we needed to use Cohere models for this, is that we were going through, our deep relationship with Microsoft. We're integrated into the Microsoft Office suite. We're have an add in for Microsoft Word that allows lawyers directly to access this. So we're researching a lot of large language models on the Azure AI foundry and came across a couple models we'll go into more today, command for text generation, embed, and re rank for search and retrieval. And I think this combination of the ability to easily compare models side by side with the capabilities of Cohere's models specific to the problem that we wanted to go after, you know, it meant that we were very confident in the technology we could bring to lawyers. And as James says, once once we knew that we could run coherent models on Azure, that combination of technology for us, we were sure we have the best of both worlds and help lawyers solve their particular problems here. Yeah. Well, thanks. And, you know, it sounds like, you know, Azure AI founder was a game changer for for DraftWise and availability of the models there. So Trupti Trupti a great way to, you know, kinda segue into, you know, hearing more from your end around, you know, these advantages that, Azure AI Foundry brought to DraftWise and you bring to your clients, or for this particular solution. Yeah. So firstly, our Jour AI Foundry is highly secure and robust. So it really helps to scale as per customer demands. We have plethora of options when it comes to models, almost 1,900 plus models. And then users can test and evaluate whichever is best fit for them and ultimately find the best solution. And regardless of the model that they choose, our SDK is universally used. And you can simply change, the models by changing the configuration, making it easy and flexible for the users. We also ensure that we align with the responsible AI practices, which is very crucial when it comes to legal aspects. It has ensured that the customer's data repository stays safe. And DraftWise does retains the precise control over which information is collected, stored, and then used for training purposes. This this really helps DraftWise a lot by training. Yeah. No. So security and the optionality, the, of of AI foundry, you know, massive for this use case and for the industry. So now if we go one layer deeper, well, you mentioned the models, and we think about the large language models available through Azure AI Foundry that you partnered with Cohere for. Could could we go a layer deeper and talk about, you know, the innovation from these particular models and how they did support the, the smart draft tool? Yeah. 100%. So, DraftWise has developed, many tools, and we're able to deploy many models for specific tasks that a lawyer wants to complete. In particular, our our retrieval augmented generation approach allows, information to be brought back and provide accurate revisions, but that has multiple steps as part of it. And so we're able to use three different models from Krakir, to produce the highest accuracy and the best results for each of those steps. So command, embed, and re rank are models that we went with. The re rank model ensures that only the most relevant search results show up at the top of a lawyer's result page. We're able to put the best results in front of them first. Embed, which is a multimodal embedding model, actually enables advanced search through any source document. It allows us to pull out context from the file, by semantic meaning, not just keywords. And it can operate at scales. We can process a lot of data including images, diagrams, tables, graphs, so that regardless of how the contract is formatted or structured across jurisdictions and across regions, we can use the embed model to get back the information that we need. And then lastly, the command model, this offers strong text generation capabilities, and so this allows us to provide a recommendation for how to mark up a contract to make a revision. They can help us to source back to source documents. And it's a natural, interface for what we want to do as part of our workflow of find the best information within the contract data that a law firm or a legal team have, rank it in the best way so that a lawyer can can see the best results first, but then choose from a couple different options, and ultimately generate a revision based on their selection, based on their judgment for what they want to use. And you combine these three models together, and it produces really good high accuracy and verifiable results based on the data that DraftWise can make available. Yeah. And, you know, when we saw the demo, we I'm instantly, I'm thinking through several use cases around purchase agreements and order forms, contracts where you just see instantly how quickly, you know, DraftWise can identify how to help, how to make changes, do the red lines instantly within the the same experience. So the the value just just pops out. But, you know, talking about the results, and this is always a a theme when it comes to implementing AI in production. Can you talk about some of the the results and measures for success? It's gonna be some of the improvements you saw in performance or the enhancements that directly impacted your clients. Yeah. I think there's results we've seen both on our offering for what we're providing to lawyers, but also on our own internal use of these tools. So we've seen, a 30% improvement in the search result quality as a result of using the different coherent models here. We've also seen internal productivity boosts being able to, develop, review, and release code much faster. We can put new features and ideas from lawyers back in front of them much more quickly. And so I think we see both these outcomes we can produce for our users and our customers, but we also see ways that we can operate better internally by using models made, in Azure AI Foundry and models made by Cohere. So we've seen results on both sides of the house, which is great. Yeah. Yeah. Yeah. And that's awesome when the again, the technology and the ROI are are there, and it just opens up so many more use cases that that I think we're gonna be seeing from DraftWise. We're we're very excited about. Now you we we've talked about Microsoft and we heard a little bit from you, but I'd love to go deeper with you. Can you share more around, you know, how your team ensures that secure deployment of models. Right? Security. So important to being able to move from, a pilot to to production, and how this contributes to the overall success of of DraftWise solution. Definitely. So yeah. So, basically, Azure AI Foundry helps infuse AI in any of your applications. And let's start by understanding the ecosystem that we provide. At the center, we have the model services, an agentic AI platform, which is surrounded by a layer of observability, search capabilities, and also content safety best practices. This really helps you provide an end to end ecosystems. And when you're starting your journey, developers can start with using their familiar tools like Visual Studio Code or whether it's GitHub. And then when they are ready, they go through the life cycle of, this Azure AI kind of foundry, and then they finally ready for their deployment, they can start leveraging, serverless options or if they need full control, they can also leverage AKS. This really helps as per their price and performance needs. And all of this is really brought together with our developer friendly portals, APIs, and which really helps them, start with, like, the quick start templates and get their application to production. Next slide. With the Jira Foundation, you get unmatched breadth and enterprise grade reliability. We offer models from leading partners and open source community. This really helps from whether you're using Cohere or different other models. We help choose whatever is the best fit for you. And, next slide. And lastly, you can also choose multiple options to leverage these models. If you wanna deploy on your own Azure compute or leverage our on demand compute, we can help you in your journey and scale as far as your needs. So this really our goal is to make sure that we can help you innovate wherever you are in the journey and develop successfully AI applications. Alright. Thank you, Dhruvdeep. And so as we as we wrap, you know, we have time for a few questions, Heather. Mhmm. We can we can take a few. Exactly. And our speakers want to look through and see if there are any questions that are jumping out to you to answer. I know we talked already a bit about the ROI, but possibly some of these questions that are around, the the data storage and and the security, which I think you've already spoken to, but if there's anything else that you wanna add around that. I'm happy to jump in with one answer. I see Kevin asked a little bit about ROI of the solution and how we compare this versus, current tools or current roles in the industry. And so we focus a lot on the value that we can provide to lawyers from, yes, producing, certain legal work faster. You can review a contract, faster. You can produce a markup faster. But I think the real value in, the approach that we're taking is in providing you more of a learning opportunity, more results that you can reference for any individual deal. You can go through 10 comparisons instead of a single one that you may have access to, and that ultimately produces a higher quality contract with lower risk involved. You can spend more time reviewing the legal risks you want to eliminate, instead of spending that time searching through your documents, for example, and choosing you have the option to choose what to include or exclude, as a way of exhibiting legal judgment. And I think we look at this versus how people are using tools today, how roles are operating today, and we see that roles across law firms and legal teams are starting to take advantage of generative AI and agenda k I, Whereas people might have searched, a old storage solution and spend their time doing that, now they're using DraftWise to actually do research and prepare analysis that they can, report up to their partners, for example. So we've got users who are paralegals. We've got users who are junior associates week one. We've got users who are senior partners, thirty years of the career. And so I think we see a lot of, exciting shifts in what people are working on, bringing all roles a little bit closer to that end goal of advising their client. Heather, I see a few questions around coherent models deployment I can I can kind of address together? So Cohere models are, developed, from the ground up by the Cohere, team. They're designed specifically for, enterprise enterprise use cases such as, the one we just reviewed with with DraftWise. Absolutely, they can be deployed privately in partnership with, with Microsoft as well as on prem, in data centers. So, in terms of region availability, yes. Absolutely. Please, you know, reach out to our team and we're happy to to engage with you ensuring that the models are available, privately, where you need them. Great. Thank yeah. Go ahead, Will. There's another one you wanna answer. And just one more that I think is, great to touch on. So how do you define the scope of search in a document management system? I think what's interesting here so DraftWise has developed functionality allowing you to define rules that will then automatically curate documents that match that rule. From a DMS, it will automatically reflect any kind of access or ethical walls that you've set up for stripping access. And then this is where, some of those scope rules, some of those, search universes that we've created pair really well with the capabilities of a coherent model. We can help to define, the scope of data made available, and then we can deploy, Cohere Search, Cohere ReRank to work over whatever documents that user can, in particular, see and then return really good results only from the set of documents that they have access to. So this is why I think you start to get, different companies developing technology with different lenses into the problem. You bring it all together and you get a much more, powerful output, well controlled. You can control exactly what is made available, exactly what is blocked entirely due to restrictions from being accessed here, and all of that can be returned very easily and quickly. So a lot that goes into defining the scope of search and ultimately returning good quality results from that. Great. Thank you. And, I know, Trupti, we haven't heard from you in a few minutes. Is there anything else that you would like to add, any questions that are jumping out to you, you know, any anything else that the audience might find valuable from your perspective. No. Yeah. I saw some questions around on prem and private cloud aspects. Right? Microsoft Azure AI platform does provide a flexibility to serve your deeds as per the capabilities you're looking for. We have a lot of options around, how you can deliver this compute. And we are here to support your needs and, you know, help you solve the problems, and we'll take care of the platform for you. So we are excited to kinda demonstrate today how this could be used in in DraftWise scenario. And hopefully, you know, it helps answer all the other questions that folks have as well. Wonderful. Well, I think we answered a lot of questions and and had a really interesting discussion. So, Nick, if you wanna say, a couple a few words and then we can, start to close out. Sure. Yeah. So first, wanna start by extending a huge thanks to both Will and and Shruti for sharing their incredible insights and expertise in in this domain. Your contributions, you know, really bring to life and, you know, seeing the demo and talking through the use case in our ROI. That transformative power of AI, specifically in the legal industry. When I thank the audience for joining us and the engagement for the questions, you know, the con conversation continues. So, you know, I think we've shared some links around the, DraftWise Microsoft story. You can find more information, and, of course, we'd love to engage with you offline, with, Microsoft and DraftWise to, help you on on your AI journey. So we wanna thank you all again for joining us, and we look forward to seeing you at our next event. Cheers. Thanks a lot.