Lexion’s Gaurav Oberoi on Applying AI to Change an Industry


Founded & Funded IA40 Lexion with Guarav Oberoi

In this week’s IA40 Spotlight Episode of Founded & Funded, Investor Elisa La Cava talks to Lexion Co-founder and CEO Gaurav Oberoi. Lexion was one of the first spinouts from AI2, and Madrona invested in the first round the company raised in 2019. Like many companies that come from AI2 – Lexion is focused on applying AI to change an industry. Specifically, Lexion helps legal teams at enterprises manage contracts – and they are one of the companies that have added features using OpenAI’s GPT technology.

I’ll go ahead and hand it over to Elisa to dive in.

This transcript was automatically generated and edited for clarity.

Elisa: Hello everyone. My name is Elisa. I’m an investor with Madrona Venture Group, and I am excited to have the CEO of one of our portfolio companies on the podcast today. Please meet Gaurav. He is the CEO of Lexion, Gaurav, welcome to the Founded & Funded podcast.

Gaurav: Thank you Elisa. It’s great to be here. I’m excited to talk to you about Lexion.

Elisa: One of the things that we love to talk about is the founding journey. You’re not a first-time CEO, first of all, and you’re not a first-time founder. I would love to hear some background on what in your career led you to this point of founding Lexion.

Gaurav: Yeah, absolutely. I studied computer science when I was in college. I enjoyed writing code as a teenager, and I knew I wanted to get into technology when I grew up. I moved to Seattle a long time ago in 04′ to work for Amazon as a junior engineer. And it was a really amazing place to learn how to build software at scale, build amazing products, and work with incredible engineering talent. I’ve always had the entrepreneurial bug and soon after that, I left to bootstrap my first company. And in my career, I’ve bootstrapped three companies, exited two, and one still runs itself. This is my first venture back company. I started Lexion in 2019 is when we launched the business. And the year before that I had spent at the Allen Institute for Artificial Intelligence as their first entrepreneur and residence, working with their research scientists and team, looking at what are the new advancements that are coming out of AI and looking at ways to commercialize them. This was 2018 when I went there, and in the years leading up to it, we had these huge step-function improvements in a lot of machine learning applications, you know, ImageNet and our ability to detect objects and images and classify them, and then revolutions happening in NLP. And so for me, it was how do we take these innovations that are real and bring them into products that actually help people get their jobs done.

I met both of my co-founders, James and Emad, at the Allen Institute. And actually what has become Lexion today, really started with Emad coming in and pushing us to look at some of the new NLP breakthroughs that were happening and applying them to a problem that his wife felt at work. She worked in procurement at a very large telco.

Elisa: I’m sure you get this from other aspiring founders, aspiring CEOs, thinking about, “Hey, I have this inkling of an idea, I kind of want a co-founder to join me”. What advice do you give future founders on how do you find a great co-founder partner to build this big idea with?

Gaurav: So my advice to folks looking for a founder, find people who are complementary to you where they can augment your strengths and maybe not duplicate them. Find people that you can have a very honest relationship with right from the get-go. That is really central to building a business. And find people that you really like and admire. People you can learn from that you respect, that you’re like, wow I wish I had some of that person’s qualities. I think those things, the right strengths to help you build your business, a good trusting relationship, and someone you look up to, those things actually go a long way.

Elisa: Wow, such deep-seated respect there and it’s just incredible.

Gaurav: It doesn’t hurt that Emad is incredibly accomplished in the field of natural language processing and James is one of those hundred X engineers that you hear about. So those strengths definitely help, but I love hanging out with them. They’re wonderful people and that really helps us get through the difficult challenges of building a business.

Elisa: So I mean, you started in mid-2019. Now we’re in early 2023. You’ve been at this for a few years at this point. You’ve gone from, your three co-founders to now a team of, is it 50? 52?

Gaurav: That 50 number sounds great, except as I learned from my legal team, it comes with all sorts of new rules and regulations we have to adhere to. So there’s also that.

Elisa: Well, maybe you put some of those contracts through Lexion to help you understand everything.

Gaurav: Oh, yes. Absolutely.

Elisa: Well, let’s, use that as a perfect segue into talking about how you and Emad and James and now your expanded leadership team, including Jessica, have thoughtfully considered how do we grow Lexion into this incredible outsized business over time. What was that philosophy going in together?

You mentioned earlier, Emad is the real deal when it comes to natural language processing and his understanding of ML. James Baird, an incredible engineer. How did you think about that formation at the very early stages when they sat down, you all began to code, what did you do?

Gaurav: So for us, the early days were really about recognizing that there is a huge opportunity in automatically understanding the contents of long-form documents. At the time you have to recognize all of the big cloud providers Microsoft, Amazon, and Google did have APIs for text parsing, but they were all really aimed at short-form text. So think a customer review or a support ticket or a Tweet. There was really nothing out there that would help you understand the contents of even a page, let alone a 50-page contract. And none of them also dealt with the complexities of the very nasty format that many contracts are in, which is PDF, which is a very poorly standardized format. So that was the opportunity. And we knew from the early days that there are myriad applications and we’ve seen companies over the years expand into many of these, contracts being won commercial contracts that we focus on, but there are also companies that focus on looking at case law and what’s happening in the courts. There’s companies that focus on patents. There’s also companies that look at big banking contracts and the big LIBOR scandal and moving away from it, there’s a whole business around doing that.

So we knew that opportunities abound because back then, while those companies didn’t exist in their current forms, there were businesses that were doing this with lots of human labor. And for us we’re like, just fast forward five years, if you do the thought experiment and this NLP technology matures, the world isn’t going to look like that. There’s going to be technology either assisting these humans or removing some of the really low-value tasks that they do. So right from the get-go, our hypothesis was if we can build a system where you can come to us with a pile of documents like Wilson Sonsini did in the early days for proof of concept. They came to us with a pile of venture financing documents and said, “Hey guys, if you can quickly design models that will go through term sheets, certificate of incorporations, voting agreements, stock purchase agreements, and spit out for us answers like what are the liquidation preferences, what are the pari-passu rights, what’s the pre-post money?” That could be really interesting because we spend considerable time extracting this information when a deal is done.

Elisa: And let’s pause for a second. We’re talking about thousands of pages of documents that normally a lawyer would need to review, get an eye on each page going into different clauses, finding out what different terms are, tracking them down, and then organizing it all in one place, and then getting onto the part that is their job, which is then what do you do about it and what are the next steps?

Gaurav: That’s right. And with our prototype back then, our goal was how can we quickly go from, here’s a pile of documents, to here are very high-quality AI models that can extract this information reliably. And also a UI that helps the user then gain confidence and identify any issues. Because I think these things go hand in hand. So that is where we started, was the crux of our technology, and we knew that we have something. Because Wilson Sonsini from those early days based on that proof of concept, entered into a commercial relationship with us to help do this processing for their knowledge management teams. But then also became an investor. And David Wong, their chief innovation officer, still sits on our board and has been an enormous help. So that was the early days. You know, and the other cool thing to recognize is in 2018, there was a groundbreaking paper that came out of AI2 called the Elmo Paper. And soon after that, Google built on top of it and cheekily called it Bert. Bert was one of the early large language models that distilled language in a way that made it richer and made representations of language or embeddings, richer so you could get more value out of them when you work with them and fed them into your neural nets.

And so that was one of the innovations. The other innovation that we were highly inspired by was looking at the deep dive project, which eventually became the Snorkel project. And looking at their approach of weak supervision. And so we combined a variety of these ideas and built the initial technology prototype that got us to launch the company and go raise a seed round. So those were the early days.

Elisa: And this is exciting because what you’re talking about is the intelligent repository product, which was your very first product, and you were testing it and ingesting and training with information from Wilson Sonsini, the law firm. But then how did you then move into your sector of customers today, which looks a little bit different?

Gaurav: Yeah they’re very different. We did not keep growing into big law. We actually moved to selling to corporates. There were a few reasons. The biggest one is just the market size. There’s a much larger market serving corporates.

The second was the magnitude and urgency of the problem. We just found as we started talking to in-house council that there are all sorts of challenges they run into with managing their agreements, specifically with managing executed agreements. Things like, “Hey, what are all the vendors that are coming up for renewal next month? Oh, we don’t know”. Or, “Hey, there’s been a data privacy change in California. Which contracts do we need to go and amend to address these issues?” We don’t know. So all sorts of analytic questions that one, help with the day-to-day running of the business. Finance, trying to do revenue recognition, or somebody trying to pay a vendor, or, “Hey, is this NDA active or not?” Two more complex queries that would happen periodically, like the ones I described, laws changing. So urgency of the problem was there. We also had insight from Emad’s wife, working in procurement, and other friends looking deeper into the bowels of large enterprises and seeing how those manifest there, like “Hey, until we structure these contracts and type them into large systems for procurement, for example, we can’t really operate our procurement systems correctly.

So a lot of energy and effort goes into people reading contracts and then typing in the handful of things that are in there. So it was clear from those that the market opportunity and the need, the urgency both exist in corporates. And so we started to make our way towards corporates. Unlike a lot of other CLM companies, we took a very different route in that we didn’t really build the end application, the repository with all its features first. We instead focused heavily on building our machine learning stack, and it meant that it took us longer to get to market. It was a higher initial capital investment. But it was a very conscious decision on our part because we recognized that, again, if you look five years into the future there, there will be players that are going after this opportunity, but the ones that will differentiate themselves are the ones that can dramatically alter the amount of effort that human users are having to put in.

Elisa: I love what you’re touching on because one of the ‘aha’ moments over the past few years is falling into that realization of these in-house councils, there’s often only one person supporting and lobbying back and forth these contracts. There’s the whole sales team that’s trying to get their deals done. There’s an ops team that is supporting, maybe there’s a customer success team or account management teams that are working with existing customers trying to get them renewed or upsold and in a certain way there’s this critical bottleneck that happens with the in-house council in working with all of these contracts. And so the ‘aha’ is what tools do these people have at their disposal? Not many …

Gaurav: Email. Spreadsheets.

Elisa: Email and spreadsheets which is a very natural workflow and, you know, you make it work. But they’re a fairly under-resourced and super critical part of what we’re kind of now calling the quote ‘deal ops process’. So thinking through that intentional investment in embedding AI and ML deep into the platform at the very beginning so that it could create that seamless magic for the in-house councils. And what you see now at your current customers is paying off in dividends.

Gaurav: It really is. And you’re absolutely right. We have since evolved the platform to help all of these teams get deals done faster. If you go back to the beginning, we started with let’s invest heavily in a deep NLP platform and we continue to invest in that. On top of it, we built a repository application, which will do things like, as soon as your contract is signed in DocuSign, your general counsel doesn’t have to download the PDF, open it to find out who it was with and when it was signed, rename it at their file convention, and then drag it into the right folder in Google Drive. No. Instead, we’ll just automatically do all that. We’ll pull it from DocuSign, identify it was with, and index it away. We’ll give you great reporting. We’ll automatically give you a list of alerts and reminders. Boom. So that was our repository. And then as we continued to grow the business, we started to hear customers say, your repository is amazing. It helps us with post-signature contract management after it’s signed. But what about pre-signature? What about all of the stuff that goes up to signing it? Generating it from a draft or a template sometimes. Just the negotiation and managing all the various red lines and drafts that happen. Getting approvals is a huge part of the process. And then things like task and status reporting. And so we started to see a lot of demand from our prospects. And this led to the next evolution of the company, which is where we are now, helping operations teams accelerate deals.

At the time, I was both excited to go after this opportunity and a little hesitant. The excitement came from natural market poll. When you hear something from your customers, listen to them, and we kept hearing it again and again. So they’re people ready to open their wallets for this. On the other hand, it’s a very competitive space. There are other contract management and other workflow automation companies that try to help with a variety of these areas. And so before we jumped into it, we spent a lot of time speaking with over 50 in-house councils, current customers, users of competitor products, and prospects to understand what is working and what is not working. And we heard one thing very consistently. It’s that a lot of these CLM projects end up failing just in the implementation phase. When you look at the effort companies put into this, it’s not just the license fee you pay to the vendor, it’s the training time, it’s the change management of it all. And if you don’t realize the benefits, it’s very taxing on the organization because everyone still has to get their daily jobs done. And that’s when it clicked. We said, “You know what? The way that we can really make a product that addresses change management and adoption is if we really focus on making something that the rest of the business can use out of the box.”

Elisa: And that’s an incredible learning. How do you iterate and hear feedback from an early set of customers or design partners that are helping you understand what features are most valuable? How much pull are you hearing and repetition about certain things that they want to be able to do moving forward?

Just hearing you talk about that now and what we can help share with other founders is listening to your customers. Come in with a point of view on an understanding of the workflows, not just of the team you’re helping, but the surrounding teams that they touch and how do they get their job done, and then what are different ways that you can remove friction in that process makes any software solution something that the core users are going to say, ” over my dead body, you’re going to take this away because it’s so critical to helping me get my job done.”

Gaurav: That’s right. And it is common advice. Listen to your users, and iterate based on them. There are a few things we did that helped us get there. One very early on, we recognize that this is a very specialized group of people — in-house council. And so we were very lucky to bring Jessica Nguyen on board. She has had a career, spanning lots of different scales of business, working in-house from Avalara in the early days, to Microsoft and working in a large tech company where she got to meet Emad, and then to being general counsel at PayScale. She kind of saw it all, and she was really the ideal person because she understood the problem. She has a lot of respect in the community, and she’s just an incredibly charismatic person. She helped us in so many ways. One, really understanding what are the problems customers face, what is useful to them, and what is fluff. How do you message so that people listen to you and aren’t just turned off, with, oh, you said AI, and I don’t really know what that means? And it sounds buzzy versus, oh, actually, you know, if we just tell them, we’ll automatically file your contracts for you and give you a reminders report. Okay, now you’re talking of value. From those kinds of things to testing product features and then all the way to helping us bring in customers, and run webinars — she really helped us understand that.

The other thing I think that has really helped us is, right from the beginning, our culture has been very customer obsessed. One of the core traits we have in our culture is to always ask why. Think it really shows when we talk to customers and they tell us I really need help with pre-signature. Why do you need help with pre-signature? And then we’d start to hear stories like I’m a team of three and me and two other council, and we’re supporting 60 salespeople. And our company is growing, next year we’re going to grow to a hundred salespeople, and I’m not getting any headcount, so I need help.

Why? Why do you need help? What do you do? Oh, all these contracts come in. Some of them are bespoke. We have to negotiate them. Others, I feel like we don’t need to touch, I want to automate them. How do we do that? And then we start to get the picture and we start to hear other stories. Oh, at the end of the quarter when we’re doing a deal rip, we’re going through our 40 deals, and these 11, just say legal. What the hell does that mean? And I tell my AEs to email them a week in advance to get an update. But it takes days to get an update. So I’ll ask the council, why does it take days? Because there are three of us and it’s all in our mailboxes and we have to update a, a spreadsheet. And you begin to understand that these teams are trying to do the best they can with the tools they have. And so that has been our natural evolution being closely embedded with these teams — listening to them, we’ve gone from a core NLP technology prototype, which is still at the core of our machine learning infrastructure, then to an intelligent repository for your contracts to an end-to-end workflow automation platform for these operations teams.

Because here’s what we also learned when we went back to council and said what about these 11 deals that just say legal? They would say, “Dude, no three of them are in legal.” The others. There are a few that are in IT security review. Finance is looking at some because of commercial terms, and like two are sitting with the CEO, who’s the one complaining about it. It’s in his mailbox. We need approval. When you really dig in, we began to realize, wow. This isn’t just about the legal team to get deals done in the company, whether you’re closing a customer deal, onboarding a vendor, even getting an employee onboarded, or doing an RFP or a security review, you’re going to touch a lot of the same teams.

You’re going to touch sales and revenue ops or procurement. You’re going to touch IT. You’re going to touch finance, you’re going to touch legal and maybe HR if it’s an employment matter. And in all these cases, the activities are also similar. You might be generating something from a template or you might be negotiating something back and forth with red lines. And then you’re going to, at some point need to get a bunch of approvals. As soon as you close a customer order, finance is going to say, “Hey, where is it? Because I need to send an invoice.” And that invoice data is in the contract. It doesn’t go into your CRM. So…

Elisa: Right. And by the way, you’re going to be renewing said customer a year down the road, and they might get a new contract. This process starts over. To put a finer point on one of the things that you just said around the evolution of the product. One of the beautiful mixes of art and science in early-stage company building is how do you decide what features to build and in what order. You mentioned, deciding one day as a team we’re going to add pre-signature work into Lexion. Based on what I remember, pre-sig, it was at least a few more quarters out on the roadmap at that time. It wasn’t an imminent build scenario. You had a lot of other things going on and a lot of other high-priority items the team was working on. But I would love to hear how did you decide which capabilities in this case, pre-sig need to be so prioritized that they have to be pulled forward. How does that decision-making process work for you?

Gaurav: As you know, these decisions don’t happen overnight. There’s some undercurrent of, we’re hearing this from customers and we discuss it in a meeting and it’s ” Hey, we should definitely do it in the second half of next year. That’s how it gets into a later part of the roadmap. So you start to build a thesis on why this is important. What can cause it to tip over and say, actually we, we need to accelerate the priority on this is, you know, one thing I found really useful are board meetings in bootstrap companies, you don’t do board meetings because you’re just going to hang out and chat with your founders or you don’t always. But one thing I found very helpful about our board meetings is it’s a time to pause and reflect on the business and look at some of the data because day-to-day you are, you’re very involved in getting it done. Shipping product, closing customers, and helping existing customers succeed. And these board meetings allow us to reflect and some of the reflections we would look at is our wins and losses in our pipeline. And as we started to look at the reasons, we increasingly heard comments like, we love your repository, but we also wish you had pre-signature workflows. And when we talked to some of our customers that had purchased our repository but had another product for workflow, they expressed the same. They said, yeah, we would love to have one platform. We only have two because neither platform has everything we need. So it was a combination of having a huge amount of signal in our active customer pipeline that told us that if we made this change now, we would actually see different results in a quarter from now because our salespeople would actually be able to close these customers that we’re having to turn away. So in this case, the acceleration was a combination of customer demand and having built up a thesis as we’re doing this of customer research and building a point of view on, if we were to execute this tomorrow, what would it look like? How are we going to do something that’s really going to stand out and give us a moat as opposed to just copying, maybe what other products are doing.

Elisa: What you’re describing is what is classically referred to as reinforcement of strong product market fit. You are hearing pull from your existing customers saying, we love what we have and we want more. And so this is an amazing moment where as a founder you probably said to yourself like, “Hey, we’ve got something. We are like rocking. We have momentum. Let’s keep going.” We need to pull forward, these really critical…

Gaurav: This is what you wait for. This is what you hope for in Startupland — when a bunch of customers with a real problem tell you ” this is a problem. If you solve it, it’s really worth it to both of us.”

We had another such moment earlier this year where one of the things we started to hear from customers was, “Hey, look, you do have ways to gather approvals in Lexion, but we need more sophisticated automated approvals.” And we, we again, talked to our customers and looked at what competitors were doing. And this was a, another case where it was clear we needed to do it. But Emad and Chris, who runs the product and came from Smartsheet and is an incredible product manager, both of them said there’s a missed opportunity here if we just go and wrote build a sort of approval chain like a lot of the other products on the market. The real opportunity here is to build a generic workflow automation tool. Because if you think about it, you could say, if this is greater than 50 grand, then seek approval from finance. But what if the action was to seek approval from finance? And then update the Slack channel saying, “Hey, finance, we’ve sent you a ticket, if someone can jump on this, it’ll be helpful.” Or once finance approves, why not then go file a Jira ticket to it to do the security review because they work in Jira and that’s where they want to be. And so the realization became if we do this right, we can build an incredibly flexible tool that will allow these backend operations teams to further stitch together the systems they use and express even more complex chains automatically. But using the same UI and the same infrastructure, it’s really triggers and actions. What’s the trigger? Hey, did you submit this ticket? And then what’s the action? Ask for approval, add a follower so they’re in the loop, set an owner so they work on it, update a Jira ticket, and so on.

And so this was a case where we actually knew that if we raced to build this feature, we would close more deals in our pipeline and have more upsell opportunities for our customers. But we actually made the conscious decision, much like our investments in ML, to say, okay, we have to look at what these teams are going to be doing on our platform. And if we pause and build a richer infrastructure, then we’ll be able to deliver a lot more value.

Elisa: Having that thoughtfulness to understand that customer journey, that user workflow, I mean, this goes for any company, is the most critical in understanding how do we start and build the infrastructure of our platform today so that we have option value for areas that we already have some kind of conviction will become critical for our capabilities in the future.

Gaurav: Yep. I have to say this is possible because of a few amazing things that happen at Lexion. One, Emad has just built an in incredible engineering org. We have a very small engineering organization for the surface area of product that we support. And Emad has been very methodical in how he’s built up the team. But then the other is making sure that everybody in the company, down to the engineer building the feature, is very aware of why we’re doing something so it doesn’t feel like there’s whiplash.

In fact, we follow a great road mapping process. Chris is again, is, he is just an amazing product manager. I love going to Chris and saying, “Chris, oh my God, we really need to build this thing. I really think it’s time.” And he is like, “Yeah, I agree with you. What are we going to cut? Let’s look at the list.”

Elisa: Ruthless prioritization!

Gaurav: Uh, And sometimes I walk away being like, oh no, you’re right. Actually, we’ve already done this exercise. So we really have an amazing team.

Elisa: Everyone says they have an amazing team. Everyone says, “Oh, my CTO is great. We have AI built at the core.” But I think in the case of Lexion, the power of the customer love that you guys receive, kind of echoes and reverberates across the investor community even where we perk up our ears and say, Hey, what’s going on with Lexion? And Lexion was a winner this year of the Intelligent Applications 40, which is an award that we run Madrona, where we ask 50 of the top VCs across the U.S. To vote for which companies did they believe are building the most intelligence into their products. And Lexion was a winner in the IA40 this year for the early-stage company category, which is incredible.

Gaurav: We were so grateful for the recognition.

Elisa: Yes. It’s amazing.

Gaurav: It’s been a really good year for us. It feels wonderful to get this recognition because of the amount of effort and investment we put into our AI. One of the things that we’ve learned over the years is if someone comes to you and says, I have 5,000, 10,000, 20,000 contracts that have been sitting in my SharePoint folder or Google Drive, and we really need to understand what’s in them because we want it to be loaded into Salesforce and Coupa and understand renewals, etc. We need your help. And from the beginning, when we would ingest these contracts, we know that the algorithms aren’t perfect. Depending on the models, you could have 95% accuracy or it could be a little lower, and so you’re going to have some errors. And right from the beginning, we knew that there should be some human in the loop before we hand this off to the customer.

Now when we look at the competition out there, we’ve seen companies that have themselves gone out and said, oh, we have lower gross margins because we have a large team that helps do the human in the loop part. Which to me really meant that their algorithms are not doing the amount of work that you’d want them to do. They’re not working as well as they should. For us, I remember early board meetings whereas our volume started to grow and customers started to come in, even we discussed, “Hey, should we look at augmenting our labor by partnering with larger offshore teams and things like that.” We made a very conscious decision and said we don’t want to create a tech-enabled services business. On the contrary, we should keep higher skilled people on our team full-time, but far fewer of them. And we’ve reduced that team size over the years as volume has gone up. Instead, we’ve said our whole process should be, if somebody comes in with a bunch of contracts and their errors, we’re not going to fix any of them manually. There should be a close collaboration with the ML team, and we should build the right tooling so you can click, retrain the model, rerun it on our validation set to make sure it hasn’t regressed, and then roll it out to the customer and improve the customer’s results. So, you know, there are significant investments there. We’ve built our own annotation tool, we’ve built our own model training, model versioning, stage deployment, and it has really paid off because it’s allowed us to accelerate how we improve these models, accelerate delivering new models, and serve a higher volume of customers with more contracts at an incredible price because we’re using technology to solve the problem, not people.

The other thing we do is we stay abreast of the latest research. We’ve been lucky to hire an incredible machine learning team. And with them, they’ve brought a culture of paper reading, of wanting to contribute to academia. And along with this, when GPT-3’s API was first released in 2021, we started to experiment and said well, first we need to understand how do these models perform at the task we already do. Can they extract parties and dates and summarize clauses better than we are and faster at a better price? And, you know, we were quickly able to identify not really, there’s still a lot of gaps to get there. But what new things can they do and how do they match to customer value? Even back then we knew that eventually, we would want to get into helping teams negotiate the actual contract itself or complete a security questionnaire itself.

Elisa: Everyone’s talking about ChatGPT and foundation models. What does this mean? And you’re saying, Hey, we were testing this a year ago. But the great news is what that evolution meant in terms of new magic in the product.

Gaurav: Yep. Absolutely. What it’s meant for us is, that earlier this year we released a Word plugin. This was in preparation for releasing AI features alongside your contract editing experience. And even there we talked a lot about whether we should we build an in-browser editing experience or should we do it in Word, which is where a lot of our customers live. And that’s where we ended up going. The initial release was really about just getting the canvas out there and providing some value. So, you can just click a button, edit in Word on Lexion, the document opens in word, you can edit it there and hit save, and we’ll do version management. It’ll go back into Lexion without you having to upload files and rename files and all that. We have a roadmap on what we want to release, starting with being able to identify clauses and say, “Hey, here are clauses from your playbook or from your prior contracts that you may want.” And all the way to automatically looking at a contract and saying, “here are the high-risk and low-risk areas.”

When Da Vinci 3 was released in late ’22 we continued to experiment and said, “Hey, let’s go see how this performs against some of the tasks that we’re giving it. And we saw remarkable improvements in generating contract clauses, and in proposing edits and because we have several in-house lawyers — Jessica is just one of them. But Krysta, who runs BizOps has lots of experience with negotiating deals and contracts. Staci on sales enablement used to be a lawyer. We have a lot of people with experience that we can test these ideas with. And when we started showing it to them, the ‘wows’ were immediate and we’re like, okay, this is great. Let’s start showing it to some customers in this prototype form. And again, we started to see, oh wow, this is very helpful. And that’s when we knew we should really move around a bit of our roadmap and accelerate getting this into the hands of customers.

And so we rolled out AI Contract Assist in December. It is a Word plugin that helps you generate new clauses or edit existing ones. You can do things like highlight a big old payment clause and in colloquial language say, “Hey, can we modify this to be net 90 and quarterly billing, and I don’t want any payment fees.”

Enter. Literally, you can type that and we will show you red lines in line that you can then accept or reject or ask for another suggestion. And this is just the beginning. The fact that we have your entire team’s workflow. We have the email that came in, “Hey, the customer wants quarterly billing and net 90, are we cool with this?” And we have all your historical red lines because we manage drafts and we have the final executed contract. We have a wealth of data to train these models. This is a very exciting area of research, but also exciting because we think this is going to turn into products very quickly and start adding value.

Elisa: The magic of what you’re talking about is incredible. I know this has been a hot item since you’ve shown it to some of your customers and, we’re very excited for the year ahead. We’ll pause there. One of the things we like to end on is a couple of lightning round questions. So to wrap things up, the first lightning round question is what do you think will be the greatest source of technological disruption?

Gaurav: It’s probably already passe to say GPT X or N. I don’t know. There are so many exciting and interesting things happening in the world right now. It’s hard to say, which is going to be the most disruptive. Some of the things that I look at, we’re going to see AI become pervasive in applications. I’d say this is a foregone conclusion and not something anybody is surprised by. But I think we will also see changes to how we consume and produce energy over the next 10 to 20 years and how we think about long supply chains versus shorter ones. And I think there’s going to be a lot of opportunity in how we think about tackling climate change using energy-consuming products.

Elisa: Next question. Aside from your own, what startup or company are you most excited about in the broader intelligent application space and why?

Gaurav: I’m inspired by GitHub Co-pilot. It’s a remarkable addition to a developer’s tool set. I think it’s very far blown that Co-pilot will start writing entire blocks of code. Having been a developer and enjoying writing code for fun, I see the benefit of removing a lot of boilerplates, making it easier to work with new libraries, of being able to jump in and produce something a little bit faster. And think it’s the early innings for co-pilot. I think it’s going to be quite a transformative product for GitHub, but time will tell.

Elisa: All right. Next one is, what is the most important lesson likely something you may wish you did better, that you’ve learned over your startup journey?

Gaurav: I think energy and emotion management is closely tied to sleep and diet and exercise. I find that if I’m having a bad day or, you know, it’s, it’s a tough day, I’m able to deal with it a lot better if I’ve just taken care of myself and done those things. And it’s a lot harder if I haven’t. I’ve generally not been very good at listening to my body and taking care of those things, but making a conscious effort really pays huge dividends in my ability to have quality output in a given day. I know it, it’s such obvious advice, but I think a lot of people don’t really take care of themselves in the way they need to. And when you’re in a high-performance dare I say sport, like running a startup, you need to be in good physical and mental shape. And it is just as important to take care of these things as it is any aspect of your business or your team, or your family.

Elisa: So important because it never really slows down so then you just have to make space for yourself. Okay, last question. What’s something you’re watching or reading right now?

Gaurav: I am a science fiction junkie, so I just finished reading “Termination Shock” by Neil Stevenson, and now I’m reading, “Dodo,” his other book. I’m a big fan of Neil Stevenson. He’s also a Seattleite, which I love.

Elisa: All right. Well, Gaurav, thank you so much for this amazing conversation. It’s been such a pleasure to work together the last few years, and so excited for the direction Lexion is headed and this big opportunity, and it’s just such a pleasure to chat today. So, thank you.

Gaurav: Thank you for having me. This has been really fun.

Coral: Thank you for listening to this week’s episode of Founded & Funded. To learn more about Lexion, please visit Lexion.AI – that’s L-E-X-I-O-N.AI. If you’re interested in learning more about the IA40, please visit IA40.com. Thanks again for joining us and come back in a couple of weeks for the next episode of Founded & Funded.


Related Insights

    Our View on the Foundation Model Stack
    Generative AI — Overhyped or Underhyped?
    A-Alpha Bio’s David Younger on machine learning in biotech, building cross-functional teams

Related Insights

    Our View on the Foundation Model Stack
    Generative AI — Overhyped or Underhyped?
    A-Alpha Bio’s David Younger on machine learning in biotech, building cross-functional teams