Subscribe
Share
Share
Embed
The Margin is a podcast from MGI Research that explores the evolving world of business monetization. Hosted by MGI Managing Directors Andrew Dailey and Igor Stenmark, the show features candid conversations with founders, CEOs, product leaders, and industry experts at the forefront of pricing, billing, and revenue operations. Each episode dives deep into the strategies, technologies, and trends shaping how companies generate, capture, and grow revenue—from subscription and usage-based models to AI-driven monetization. Whether you're in finance, product, or IT, The Margin offers practical insights to help you navigate complexity and drive growth in the digital economy.
Andrew Dailey: Hello and welcome to The Margin, a podcast exploring the forces shaping business monetization. I'm Andrew Dailey, Managing Director and Analyst at MGI Research. Today, we're exploring the future of contract management and the potential for artificial intelligence to disrupt both positively and negatively. The way companies construct, negotiate and manage contracts. Although AI and machine learning have been a part of select contract management solutions for years, the arrival of mainstream generative AI models has whipped the CLM industry and general counsels into a new, heightened level of hype, promises and fear. These are high-stakes decisions. Contracts define business relationships, financial obligations and corporate risk for legal teams and general counsels. Precision is paramount. 99% accuracy and one-percent hallucination just isn't good enough. So where does AI add value and where is it falling short? What, if any, are the practical use cases, financial rewards, and guardrails companies should consider as a part of their CLM strategy? To answer these questions, I'm joined by Grant Peterson, Chief Product Officer at Conga. Grant is a unique leader; he combines strategic vision with deep technical expertise and a prior life. He was the driving force behind DocuSign, market leading e-signature product. Today at Conga, he leads their technology and business innovation and works with Fortune 500 companies to responsibly incorporate AI into critical business capabilities like contracting and quote-to-cash. Grant, welcome to The Margin.
Grant Peterson: Appreciate the kind introduction, I hope I live up to it today.
AD: So what's been—you've spent a lot of time—obviously, this is critical to you, your business, to what you guys do. Conga is in a unique position dealing with both contract data as well as pricing data, which are probably two of the most sensitive areas of a business. So far, what's been your biggest surprise as it relates to Gen AI, both the positive and the negative?
GP: Yeah. So you know, it's a really fun question because there's a lot of terrain in there. We came to this with a lot of perspective. And so as a company, we've been working more on the ML side of the machine learning and extractive side of AI for probably the last, certainly for the last five years, pretty on a focus basis and have offered some products. And so when Gen AI sort of landed hard—I'd been tracking it for a bit, but sort of like kind of hit our world, hit everybody's world last December or so. And I think the biggest sort of surprise as, as having fully dug into it and really looked at what it can do, how well it can do it and where some of the maybe some of the edges are, the biggest positive surprise really was it can do so much more than I expected. Just the broad range of complex kinds of inquiries that it can attack and do so in a pretty effective way is quite a pleasant surprise. So I wasn't sure where the edges were going to be, what the limits were going to be. And a number of us sort of really tried to explore that. And it is super surprising what it can do in terms of kind of intuiting correlations between things and making reasonable brainstorming conclusions and frankly, even just identifying and pulling out relatively accurate answers most of the time to questions. The other side of it has been it's also been surprising how good the answers look. And in some cases, that it may be 80% accurate and 20% fully inaccurate. So it introduces a really interesting quandary, which is how to take the value without taking the risk for the things that—it wants to put three points in every answer, and it'll get two of them right for sure. The third one, who knows? And so, you know that that's really the other side of the issue in terms of surprises and things you need to work through.
Igor Stenmark: So it sounds like there's probably a lot of promise, but there's also a lot of hidden danger here. The scenario you described, the 80% accurate 20%, as Andrew calls it, hallucination. That's how propaganda works. We'll give you four correct answers and one that's like flaky. And you sort of begin to believe a flaky.
GP: Yeah. I mean, I do think that's—and I don't want to—I'm not afraid of it. You know, I think there's a totally a way to get the value from it. But you need to be aware of and sort of do some good adulting around some of the risks. And I do think the idea that it's that that's what good product propaganda looks like is, is, I think, a great way to think about it because if somebody is making a fully false statement, it doesn't have much impact. If they're making an almost entirely true statement with one very dangerous falsehood in it, that's some pretty important issue because you know, it's tempting. I saw it over the holidays, actually, and had it write poetry. I had it review contracts. I had it write essays, I had it—a spouse on topics I felt that I was reasonably up to speed on. And one of the things I learned is it’s believable, very believable. But you go through with it and pick out the points that are false or just inaccurate. Therein lies the danger. So I think what you have to do is you have to figure out what's the right role for it to play right now.
IS: Yeah. Looking specifically at CLM, where some people would believe that AI generative AI can potentially help completely redefine CLM, but you could abandon the current platforms almost entirely and create CLM from scratch using generative AI. What do you think about that idea?
GP: Well, I don't think that that's very likely. I will say I do think it's going to completely redefine CLM. I mean, I don't think there's any question. We ourselves are redefining how we look at CLM using what's possible with generative. I think when you really look at CLM, there's lots of facets to it, right? One facet to it is this whole notion of supporting the authoring of a contract. Another facet of it is managing the process of negotiating and ratifying a contract. Another facet of it is dealing with things that come into that lifecycle on the fly that we tend to call third party paper addendum security requirements, these kinds of things. And another facet of it is, frankly, the ongoing management, which could be from sort of the mundane to managing renewals all the way through to the more sophisticated, which is sort of diving into risks or managing sort of specific activities that you want to do, like migrate a contract sort of terms forward or so there's all these aspects of it. I don't think generative necessarily changes any of the structural value of a CLM the sort of keeping managing sequencing segmenting contracts, keeping the guardrails around your company policies. These are sort of still really the fundamental value propositions of a CRM product. It can help you in other ways, though. And so I think what I'm kind of running into is there's sort of three major buckets that the way at least the way I'm looking at it, one major bucket is really the classic one we've been attacking for some time, and that is sort of sort of a extracting key clauses or key terms in order to categorize your contracts in a way that you can manage them. That value proposition has been critical to every company that has contracts that still live in boxes or in multiple systems and remains critical. Now, it turns out we'll talk if you want. We can talk about how generative can help with that. But I think that value proposition remains the fastest growing value proposition around CLM we have right now. The other two I think are new and they're interesting. There's things that that really were not possible to do before, and I put those in the categories of convenience. And I think convenience meaning it can make you do your job faster, make it make it easier for you to do your job, and make you feel more supported in doing your job. And those are things like characterizing a red line or identifying maybe a clause that has implications that aren't obvious. So just sort of kind of drawing your attention to things and sort of making the entire experience a little bit faster for you. That is good. And I think pretty safe. And I think the third category is probably the most interesting one, which I would call advice. I don't personally, and I would caution anybody to think about this carefully, I don't think we're ready to let software practice law, and I'm pretty sure the legal system is going to agree. I understand the circumstances of the planet. Again you go to law school, you have to be barred in every state in which you're going to practice law, and I don't think anybody's quite ready to let AI do it alone. I'm pretty sure that wouldn't be considered appropriate. And so I don't think the right role for, for for generative, really at this point is to practice law. I think the right role is perhaps to ideate and give the the attorney that might be sitting at the helm of these things some perspective that like basically having a great colleague next door that could take you could chat through some of the risks or terms or opportunities. It can probably do some base authoring for you. But I do worry about people. Which is you let it author a contract. Maybe nine out of ten people aren't going to actually just push that forward. But maybe one will, and it'll be pretty interesting to see what happens when that starts to happen. I think we all saw a court case on that recently where a brief was filed by Gen AI and the attorney was sanctioned for submitting it, so there's a line in there.
AD: One thing we haven't talked about, but I think it's important to bring up, which is what happens to machine learning in this conversation? It kind of gets washed over. And yet there's still arguably a lot of value to be extracted there. And the users even care if it's machine learning, AI, Gen AI, or trained monkeys?
GP: I'll take that second part of the question second, so I'll have to think about that for a second. But I'm going to start by saying I don't know what it’s going to look like three years from now. I don't know what it's going to look like two years from now. I will tell you as of now, because we've done some fairly extensive testing, you know that my conclusion to date is that Gen AI and traditional machine learning and trained models are actually very compatible in certainly the way we're going to proceed with it desirably better together. And the reason is essentially what we found is that if you're going to go with a trained model approach, which we currently have, then don't ask what Conga’s been doing. We've been sort of using bootstrapping, using some pre-trained models. But for the most part, when you get to any customer project, you'd spend a couple of weeks sort of working on the training. We spent the last year basically making it possible so customers could auto train so they could do through UI training. So we've been approaching it where you have to invest in that training to a degree in order to get a great result. What I've found is by introducing AI in front of the trained ML sort of strategy. So you run it through Gen AI and then you run it through the trained pipeline. What you get is Gen AI does a surprisingly good job of extracting relatively common things, and for obvious reasons, I mean, if you've got a couple hundred billion provisions that have trained the model that the LLM that's behind, say, ChatGPT 4, it's going to know a fair bit about sort of relatively standard language and it can pick out clauses fairly well. It does well. And so it's sort of pre-trained to an extent that's impressive. But what happens when you start running contracts that are specific to a given company that may have sort of structural aspects that that really aren't in the public domain, or maybe terms of art, or maybe they had doesn't circumstance that were documented in a particular language. You're looking for those things. It just can't find them. Its accuracy for things that would be considered confidential or proprietary, that aren't in the large document sets that were used to train Gen AI those things can't be found. And so what we're doing, how we're approaching it, is run it through the Gen AI pipeline, find the stuff that it can find, and then and then focus your training on the things that are more more local, more specific to a given customer's needs, and that that allows us to get accuracy fast, but then to get very high accuracy on things that wouldn't otherwise be in the model. And as a side benefit, because I do worry about this a bit, it also allows you to do that without introducing some of that material into the larger LLM data set. You can kind of keep it in the private trained model. And I do think that one of the things we're all kind of have to really grapple with and learn is really how to keep our companies and our customers and our and our partners proprietary and confidential material confidential. I realize that the private instance approach is sort of promoted as safe and hopefully it is, but I feel good about keeping some of that stuff sort of closer in an ML model and what we're able to do is, frankly, we were able to get into the very high accuracy rates through training in the past. We can get there substantially faster, like weeks faster by doing both. So I mean ultimately better experience and higher accuracy in the end. And there's really no reason not to do more than one technique.
IS: What do you think, Grant, users should do while looking at CLM or currently using CLM if we want to be well-positioned to take advantage of not only machine learning, but also Gen AI in terms of technology, skill sets, budget, finding money to pay for all this stuff, expectations that we have from a business side? What are the top things in your mind would be important to do?
GP: Yeah. So I think that's a great question. You know, honestly, I think the most important thing for every company to be doing is considering introducing this kind of technology into their workflows is think about that for yourself first. Right? So think about what what risk does that impose for you? What benefits would you potentially like to get? To what extent are you willing to let your material be part of a broader trained a training set that other people might participate in? Do you actually believe that just because something's been shredded into an LLM, that it won't necessarily expose confidential material? Because I don't. And I think you have to think about what are you trying to gain. Is it a productivity gain you're looking for? Is it an accuracy gain? Is it an intelligence gain? And it's probably some of all those things. I think you should think about what you're trying to get, what you're trying to accomplish, what those risks might be to you, and what the value might be to you. That's that's I think that's fundamentally what I would consider to be the responsible place to start. And then I think you can go out and look at your tool set suppliers and CLM vendors and sort of take their perspective. But I'd encourage you to develop your own first, possibly avoid a little bit of the I think we're all seeing the word “generative” stuck in front of everything and it's not entirely clear what exactly the value proposition is the that that's being offered by a lot of these sort of people—and people are doing this, I'm not going to make any statement about whether I think there is or isn't value in everybody's case. But the question is, is what are you trying to get out of it? Because at the moment, I think there's sort of two kinds of buyers that I'm running into. There's the folks that are just super excited about the concept, and frankly, they want their CLM vendor to kind of tell them what the state of the art is going to be. And then there's the folks that really are thinking about, what am I trying to do in my trying to save labor? Am I trying to accelerate contract execution? Am I trying to find insights? Am I trying to make complex strategic evaluations? Given your goals, you should be able to match up with the right functional set. That would be my advice. But I you know, I guess I would add one more thing—to some degree or another, every single vendor you have in every single industry you're in, including CLM, is going to have the exact same set of generative capabilities because none of us are building them, folks. We're buying them from the folks that are building these large LLMs and some vendors like ourselves have been pipelining multiple tool sets in order to get complex results. And really it comes down to the experience is going to be as good as the prompt engineer in the company and how they integrated into creating value in your workflow. So if you're real clear about what you need, and your vendors are real clear about what they're offering, you should get a good match. At the moment, I don't understand how every company is trying to differentiate by something that's actually just the new normal.
AD: That's a good question, is if everybody is claiming Gen AI capabilities, at the end of the day, who has anything that's unique and different? And in this space in particular, there's been, what, half a dozen, a dozen startups heavily funded on the premise of having some unique Gen AI magic.
GP: I'll be as interested as everyone else to find out if there's any reality to that. But from where I sit, I think introducing Gen AI to do something like characterize a red line, I mean, I can take a contract with red lines in it, and I can hand it to the beta and get an answer to that question. So I think to a degree, Gen AI is a huge windfall for all of us because we can bring that new extra capability and sort of bring it into our experience, but it doesn't alleviate any of the rest of the whole operational, procedural, and sort of functional, and approvals, and putting contract guidelines in place, that stuff all remains hugely the core of what we're doing with CLM. It just makes CLM a little more approachable and a little bit more intelligent with regards to how you might sort of look for a result. I do think that on the generative front that for the most part, it's going to be pretty much the same experience across the board because we're all basically benefiting from these massive LLMs. And it really at that point comes down to how good a job you do putting in your user experience and how good a job you do doing the prompt engineering and, frankly, how many different ways you can fine to expose it as a value to your customer.
AD: Who picks up the cost?
GP: Well, that's a really great question.
IS: Where is the money going to come from, too?
AD: Yeah.
GP: You know, like maybe that was just a bit of the engineering behind the answer I gave second, and what you should think about before you bring before you target bringing AI and a vendor into your workflow is what's the value to you? Because I think we're all going to have to figure out—I will tell you this, it's not inexpensive, right?
IS: To run it as well.
GP: If you're going to put everything through it, it's actually got quite a heavy cost to it. And so the question is how does that just create adequate value that the end user communities are up for paying for that directly? Does it just go into the license cost, or do you use it in a somewhat more targeted and selective fashion so that it becomes a bit of a transactional choice? I mean, this is what we're all working through right now, but it's not inexpensive and if the industry is ready for a pretty significant hike in the floor of price, then it can be table stakes. But I doubt it. I think this is part this part's left to be seen.
IS: What do you think is the best-case scenario for the suppliers? Is Gen AI going to generate revenue in 2027? Meaningful revenue?
GP: I don't think so. Not this year. And I don't think it's going to in the next year I think what it what it probably the I do think it will generate new revenue. And the reason I think it'll generate new revenue is because it's captured people's attention, I think it's causing people to do something. Maybe they were looking at their contract lifecycle as just something that just was sort of at a point where it had created all the value it could create. Well, now we have this new set of capabilities. And as companies think about to what extent they want to allow some of this new capability to transform and transform how they're thinking about their contract lifecycle, that should create overall some great new value. It'll also create just more eyes on and more interest in what's possible with contract lifecycle management so that there will be some growth that will, I think, help with the funding the underlying infrastructure, all sort of additions that go with this but I think it really comes down to things that that have huge value generally will have some pretty significant costs associated with them. And, and ultimately it's on us as vendors and customers, as customers to decide how much of that value is worth paying for. Essentially, I think the three key issues that I've been trying to think about since the beginning in this is how to create great value with this rather than great marketing. The second has been how to do it in a way that protects customers confidentiality, and so that we can sort of give incredibly high confidence in terms of how this affects their risk. And I think the last one really just comes down to what new does it make possible? So I think balancing those out is there's definitely a risk side of this, there’s definitely a cost side of this, and there's definitely a value side of this. And I will be super excited to participate in and watch kind of what happens in the next sort of probably 12 to 24 months as people's sort of approach to this, their public approach to this, moves from basically just saying “we have generative product” to what are the customer use cases? How have they experienced new value creation? Have there been any issues that had to be managed? I think getting to that sort of network vendors are approaching it more less from an excitement and marketing perspective and more from an actual expert guidance in terms of how to design and get value and manage risk through the experience. And so it will absolutely, positively change a lot. But I do think there's a lot ahead of all of us to learn.
AD: To that end, we haven't talked about some areas like say implementations. Is this something that could we see in the near future, say implementation times get cut in half?
GP: That that's a solid question. Probably. Probably is the answer that—I mean, that was one of the very, very first things. You know, it's funny because I go pretty far back in the CLM space. I mean, long before I came to Conga, I also been then been working on some of these things. And one of the things I said, jeez ten years ago, 12 years ago was one of the biggest opportunities around CLM would be to be able to build your clause library in an effective and almost immediate fashion by ingesting your legacy documents. You know, just like take a sample set of documents and sort of create sort of a hierarchy of the most probable and most valuable clauses. You know, that was always probably possible, but no one really attacked it. Gen AI can attack that, and it can do a pretty decent job. And beyond being able to sort of create a lot of that clause library sort of input fodder, it can also kind of make recommendations about what you might want to do to harden a clause if you train it adequately and you put the right prompts together to understand what you consider to be a risk in a clause. So I think one of the things nobody really is talking about yet, but I know we were working on, I imagine others, is how you can take contracts that are currently in your set, how you can take contracts that are not currently under management year, in your product like stuff that's being ingested from either paper or other digital sources and extract from that or recommended set of core key clauses. You know, I will say one of the very first things I've looked at is, is there's interesting domains out there like ISDA and ISLA where there's a bunch of variants of core clauses around a set of very specific set of transactions. You know, can we ingest that in order to create a pre-created sort of core clause library for people that are doing ISDA and ISLA agreements? And so the answer to that, I'm positive, is going to be yes I don't know that we've got it fully ironed out yet, but those are the kinds of things that I think everyone's talking about being able to do a red line or being able to do whatever, table stakes and clause recommendations. But I think the really huge value is going to be in terms of how do you set up your core material, how do you create what are your best practices? You know, we rely on humans to figure that out, Gen AI, I can actually help with that. So I do think it will change pretty significantly the sort of set up and implementation time.
AD: Is there a day when a CLM vendor becomes no longer a provider of a productivity app, but a marketplace for for agreements and clauses and contract language?
GP: There's unquestionably an opportunity there. I don't know how that will affect the distribution of where people spend their money, I guess. I don't have a crystal ball, but I do think it opens up a possibility. Before, if you had wanted to do that, you would have had a probably had a deep, deep expertise in that particular area that there are already boutiques that that have that practice, right? You know, like, do these people do financial instruments and these people, do you know, M&A instruments. And that kind of already exists in the realm of consultancy. Does it make it possible for somebody to sort of go into the business of doing that? I think it probably does. And the question is will the existing vendors leave that opportunity open? We'll see. It could be that just everybody gains the ability to ingest a set of contracts and suggest a clause library, sort of maybe making it less valuable to have somebody do it for you. But we'll just see. I would hope that if you have a practice area that has a particular set of contract types, that I can pull a clause library out of it using Gen AI and some engineering on our side. So you might not need to go buy it from somebody, you probably already own it. You just need to extract it. So, I don't know. Yeah, I don't know. But clearly having more of that material available easily in some way or another will change the industry for the better.
IS: So, couple of questions. One is more tactical and one and leading to something a bit more broader. If I really combine all the clauses I have on my electronic paper from various parties, various agreements, doesn't that expose me to a risk of having some amount of data leakage or IP leakage, even though with terms and conditions of a contract and pricing are supposed to be completely confidential? What's the risk? And then I want to talk a bit more about it.
GP: Well, unfortunately, my answer to that is I don't know how big a risk that is. I think that's exactly what we need to be careful about right now. You know, I guess the garden answer to that would be if it's done in a private instance, and that private instance is private to the one company, then then if everything we have been led to believe based on the way, the way Gen AI is set up and the way private instances work is true, then there probably isn't significant risk. So the question is, will you be able to have your own private instance? I guess the second question is will there be any mistakes in that? This is a brand-new technology that's come to market for the first time so is any how comfortable are you taking the risk when you run your material through it, through that cloud, that Microsoft and the other folks that are operating these almost got it right the first time that we'll have to learn that together. I mean, I’m always going to be a little bit risk cautious because I've been in the industry for decades and I've seen lots of claims, and I've seen lots of reality that follow those claims. So I think caution is appropriate. But yeah, there are solutions that sort of resolve for that. It's part of the reason why I'm not really all of that. I like my approach here is, is I think customers should run their stuff through Gen AI voluntarily, rather than me doing it every single time. And the reason is it gives you a little more control. And secondly, it allows us to mutually control the costs a bit. So I think the bottom line is we don't know. I think there's good theory on this. And I and by very smart people that will tell you there's no problems, no risk.
IS: We just don't know what we don't know. And we may require additional technology, additional development. I can imagine a scenario where you say you, Conga, pull contracts from 100 different suppliers you have and customers you have, and this data gets commingled and you get to put some additional guardrails around it as well. But I want to get into a broader topic of, IP protection. We talked about this issue, remember, but kind of describing it as the greatest—
GP: I like the provocative tagline here.
IS: —theft of intellectual intellectual property in history. Is it? I mean, we’ve had lots of court cases being prepped by some pretty smart people. What do you think?
GP: Well, I guess I'm probably not able to fully justify the accuracy of the comment, but let's talk about the intent of it. These LLMs, the reason they're amazing is because they're trained with astonishing and just outrageous amounts of material really, John, I is not born of a fully different set of AI technologies. It's born of the somebody having had the inspiration to run pretty much everything that could be acquired in digital form through to create hundreds of billions of connections in the model. And so, really, the reason it's so human is because it has had so much information that it could actually just do that. And so where did that information come from? Well I think it's pretty much the entire digital published set of literature that's ever been published, most of which is still under copyright. It is the contents of, public domain sites such as GitHub, which is publicly accessible by on purpose but pretty much 100% under some sort of license. Everything given even open source, and things that come out of publicly accessible get libraries. They're all under some kind of copyright or GPL or whatever they're having. Licensure is they generally require things like attribution and republication of any derivative works and who knows what else. Those are the things I know for sure. I don't know for sure what other source code repositories whether people where the privacy was respected or not. Nobody actually knows exactly the extent to which it went in. But we do know for sure that a lot of data went in was under some sort of restriction. So I think the argument, the core argument, is that because it's only being used to learn, the copyright is not actually being violated. And we'll have to see how that holds up. I bet you a nickel if I ask it to write a poem similar to something Keats would write, there will be some of Keats poetry in the response. So does that or doesn't that respect the intellectual property? So I think so. But the basic answer is I don't know. But it sure seems like a problem to me. That material that had restrictions has been consumed. And I guess the question is, is really what it's going to come down to is if I read a book and I happen to memorize a portion of the book, and then I use a portion of what I memorized in in a book I write, and somebody claims I violated their copyright, did I?
IS: Getting into boundaries of what's fair use and all this other stuff. And the question is, is it going to get resolved strictly through a course of some people like Barry Diller think right now
GP: I doubt it.
IS: Or is it going to be a regulation or some combination?
GP: I think it's going to be all of it.
IS: Right? Yeah, probably all of it. It's going to be chaos probably.
GP: Yeah. We've seen this movie before. I mean they called it Napster. Yeah. And that got resolved lots of different ways, Right?
IS: Lots of different ways. Lots of things came out of it.
GP: Yeah. The whole digital music industry at a dollar per click came out of it like probably one of Steve Jobs least credited accomplishments. But at the end of the day, it'll be interesting to find out if it's 100% an issue, how it will be resolved, what way it will go and in what period of time and what kind of what that will mean to all of us, I guess we're all we'll find out together.
IS: Yeah. So it's something to keep in mind.
GP: I definitely do imagine there'll be some regulation. I mean, we've already seen it, right? Yeah. A lot of what was happening with the actor strike the writers’ strike was based on a lot of based on their rights and relative to AI. And then like whether their images can be used to create. And so this will not play out fast.
AD: if I'm a buyer or if I'm a prospective buyer. Vendors come in and they talk about ChatGPT 4, 5, Anthropic. Should I care about that? And if I do care, how should I think about that? Does it matter that you have a particular flavor of the day? How do I insulate myself? Do I need to? How do you, as a consumer of these things, think about it?
GP: We have we've developed an AI acumen, but we're largely an AI consumer. We buy it and package it to create an experience, which I think is kind of what everyone does. It's funny if someone's telling you that they're differentiated based on their sort of core AI expertise, I would dig hard on that, because honestly, in order to be a core AI expert right now, you have to be able to make multiple billion-dollar investments around data aggregation. And I just don't think anyone can do that outside of some obvious folks. So like I first and foremost, I would be very cautious of anyone that's telling you that, that that they have secret sauce. So that would be my first advice. I guess to the question, should you care? Maybe, maybe you should care. I'll tell you, I care. And so I first got to weigh into this in the CLM space three years ago when I, when I joined this company and then made some difficult decisions that were made a couple of years ago when we did a couple of acquisitions in the space, and I made the decision that I didn't think there was a right way to do it. That, long term, one model would never win. And so very first thing we did is build a multi pipeline AI model. An AI infrastructure. We were already using at least two vendors, we had our own sort of training set that we were building around a bunch of public LLMs. So we had three ways of doing it. And rather than getting into the sort of process of deciding which one way we were going to do it, I decided we were going to never do it any one way. So I guess I would ask your vendor if they're telling you, but if your vendors telling you that there's one way to do it, I don't think they're telling you good—I don't think they're giving you good advice. I think there's always going to be evolution. I mean, we we all were sort of stunned when we saw ChatGPT drop, right? And we were all able to access the data and get in there and really experience what was possible. Anybody want to bet we don't see something new like that in the next year or two? So anybody that tells you there's an answer is probably wrong. The answer is be ready for the new answer.
IS: It's only going to come and going come faster.
GP: Yeah. There's no right answer, you know. And I will tell everybody I think our company is this competent in AI is is most maybe significantly more. We're definitely focused on it. We're not experts. What we are is integrators. We're we're trying to consume and create a customer experience at a reasonable cost with great value that doesn’t open risks to them. And I think that's what a good vendor looks like. Somebody tells you that their generative product is going to transform the world it’s like, well, you didn't build that, folks. Everyone gets to buy that. It transformed the world, all right, but not just your world.
IS: Yeah. We we talk internally here, but Gen AI is essentially being with silicon layer.
GP: New level playing field, folks. By the way, you said something a little while ago, Andrew, about that and about startups that are popping up around this. Yeah, That's because some of the traditional barriers have been kicked down. Right now, it's a disruptor’s opportunity. And surely there will be some of that, and we'll learn from those those guys, those folks, too. So my guess is there'll be disruptors and consolidation. The industry is going to be pretty fun for the next year or two.
IS: Well, this has been terrific. Grant, thank you so much for being here. And thank you all for joining us today.
AD: Thanks, Grant.
GP: Thank you.
AD: Thank you for listening to The Margin. If you have questions about today's episode or if you'd like to schedule a call with an MGI analyst, reach out to us at insights@mgiresearch.com. You can also reach us on LinkedIn, Facebook, and X, and you can find more information about our research and advisory work at mgiresearch.com. Until next time.