Today, and welcome to the Constellation Software Inc. conference call and webcast. All participants will be in listen-only mode. Should you need assistance, please signal a conference specialist by pressing the star key followed by zero. After today's presentation, there will be an opportunity to ask questions. To ask a question, you may press star, then one on a touch-tone phone. To withdraw your question, please press star and then two. Please note this event is being recorded. I would now like to turn the conference over to Mark Leonard, President of Constellation Software Inc. Go ahead. Good morning. Thank you for joining us this morning. First of all, I'd like to introduce our panel. On the line, we have Cam, Chris, [Henk- Jan], and Paul. We're going to stick with the first names so these brave volunteers don't get overwhelmed by outreach from our shareholders or competitors. I'm going to address certain questions to them individually, particularly if there's an anecdote or data point that I think they can address well. I'll also throw open questions to any of them on the panel as well, and I may weigh in myself from time to time. Before we get going with the Q&A, I'd like to tell a story. It's a true story, and it illustrates a useful but unsatisfying lesson, which I think we should all understand. In 2016, Jeff Hinton made a long-term forecast. For those of you who don't know, Jeff is known as the godfather of AI and is a Nobel Prize winner for his work in the field. Long-term forecasting is very difficult. I talked about this before, and I'm happy to send you some source information if you'd like to delve into that further. Jeff's forecast in 2016 was that radiologists were going to be rapidly replaced by AI. Specifically, he said people should stop training radiologists now. In the intervening nine years since he made that forecast, the number of radiologists has increased from 26,000 in the U.S.—these are U.S. board-certified radiologists—to 30,500, or a 17% increase. That's outpaced the population growth in that period. The number of radiologists per capita is up from 7.9- 8.5. Jeff wasn't wrong about the applicability of AI to radiology. Where he was wrong was that the technology would replace people. Instead, it's augmented people. The quality of care delivered by radiologists has improved, and the number of practicing radiologists has increased. I told you the story to make two points. Firstly, you and I will never know a tiny fraction as much about AI as Jeff did. Secondly, despite his deep knowledge of AI, he was unable to predict how it would change the structure of the radiology profession. I think we're at a similar point today with the programming profession. It's difficult to say whether programming is facing a renaissance or a recession. Programmers could experience massive demand for their services if their efficiency improves tenfold. You can imagine not having to put up with software that does 80% of what you want. You'll be able to get software that does 100% of what you want that's customized to your needs. The cost of programming will drive that increased customization. What a wonderful outcome that would be. Equally, you can imagine a tenfold increase in programmer productivity driving massive oversupply of programmers and demand, particularly if demand for their services remains static. Similarly, if the 10x efficiency doesn't happen, if it's a 10% efficiency gain, you can imagine that there would be very modest changes to the current status quo. We don't know which way this is going to go. We're monitoring the situation closely. We're going to tell you stories from both Constellation and from third parties that we talk to that support many possible outcomes in the software development world. With that, I'm going to ask the operator to open the lines for questions from the listeners. I'll intersperse some questions that I've received by email from a number of our institutional investors in the course of the day. It's not just driven by the telephone lines. Dave, if you could introduce the first questioner. We will now begin the question and answer session. To ask a question, you may press star, then one on your touch-tone phone. If at any time your question has been addressed and you would like to withdraw your question, please press star and then two. Our first question comes from Thanos Moschopoulos with BMO Capital Markets. Please go ahead. Hi, good morning. Mark, maybe to start off with, can you just provide some context in terms of the four individuals? Are they like an internal AI SWAT team or just generally speaking what their roles are? Secondly, recognizing that you're a decentralized organization, but that AI is going to impact all of your businesses in some shape or form, have you thought about implementing some metrics and/or incentives to ensure that your businesses are actively leveraging AI and not falling behind the curve? Is your current framework sufficient to ensure that businesses will do what they need to do in that regard and not be laggers? Thanks. By way of background, the four panelists are people who are working at AI full-time, generally have been with us for considerable time, often strongly technical backgrounds, but they're not data scientists who are programming in CUDA on NVIDIA machines. They are application specialists for the most part, but you know obviously among the best and brightest. In terms of metrics that are standardized across the operating groups, some of the operating groups are very detailed in terms of the metrics that they're following. I'm going to report on one of the operating groups that sent me their data, who is not represented among the speakers. Others have taken a deeper, more nuanced approach to reporting progress on AI that is less, you know, 3% of business units have replaced people with AI. It's more sort of individual case studies and individual projects and looks at the maturity of the use of AI tools and things of that nature. Everyone's following AI, but there is no Constellation-level metric that you can look at right now. I don't doubt that we'll end up with one that is sort of a subset of the rolled-up metrics at some stage. Right now, you're going to hear about individual use cases and individual results from both business units and operating groups rather than a nice, easy answer at the Constellation level. You'll also hear... Sorry. You'll also hear some funny tales of what happens with radical decentralization, where you end up with duplication of effort. It's not like we're duplicating multimillions of effort. We're just duplicating modest amounts of effort, and sometimes there are lessons in that. Great. Just to... real innovation versus hype. Go ahead, Mark. Sorry, I was going to say, did you have a follow-up question? Yeah. Yeah, I did. Sorry. My follow-up is just a couple of the main use cases we're hearing from other businesses around programming efficiency and customer support efficiency. I presume that's some of what you're hearing. Just wondering if there's any other use cases you'd call out that seem to be broadly applicable across a lot of your businesses. I think your question was, you're assuming we're trying for programming efficiency. Is there anything else we're doing with AI? Is that the question, or did I miss it? Just whether programming efficiency and customer support efficiency are two of the key use cases that you're already seeing some of your businesses leverage and whether there are any other key use cases beyond those you would point to that are broadly applicable. Yeah, I mean, I think we'll get there, and those are sort of very 20,000-foot pan of questions. Maybe we'll move on, and I'll recapture this question at some later stage after we've talked about a bunch of individual triumphs and failures. Next question, please. Next question comes from Jérome Dubreuil with Desjardins. Please go ahead. Thanks for taking my questions. I have two, but I'll just start with one here. It seems like a lot of the conversation about AI and programming is all about what AI can do well. I'm wondering if there are things that AI isn't good at for coding. For instance, we've heard that debugging can be an issue with AI programs. I'm wondering about that as a way to determine whether this could be a major hindrance for the implementation of AI in DMS. Yeah, I think the challenge in answering a question like that, of course, is the state of AI is changing rapidly, and so one has no real sense of what it will be capable of a year from now. Why don't I throw it open to the panel if any of them have views on what, perhaps the other way around, where we're getting benefit from programming with AI and where we're not. I think he started. Why don't we do it alphabetically? Let's start with Cam. Anything you wanted to talk about there? Yeah, I think generally speaking, there's definitely some strength that any type of automated AI engine or AI coding assistant can bring to the fore. Most of the efficiencies that we've seen thus far are really around automating things like the aspects of unit testing, unit coding, test plan creation, store procedure optimization, vulnerability trapping, code commenting. Really niche type of specification. The reason why, I think, when we take a step back and really look at the code is most of the DMSs are written in tens of millions of lines of code. The context window size is just simply not large enough to be able to troubleshoot by taking all of it into account. That would be the weakness part. You know context window size is getting bigger every day. May I add something, Mark? Yes, [Henk-Jan], go ahead. Yes. For example, if you have just a clean sheet, you can just build your software from scratch, then you will be really productive. As Mark just said, the limit of the context window size, that's currently the limit for a broad support of software developers in all kinds of activities. Maintenance and support, bug solving, all those kinds of things are limited by the context window size. Currently, there's a rat race going on where the big vendors are setting up a multi-agent architecture where every agent can code with some scope, which is allocated to that agent. There can be an agent which makes, for example, a refactor plan, and the master agent can give assignments to client agents to do the job, all kinds of jobs which will fit in the context window size. It's going on right now. Within a couple of months, we see a lot, you can do a lot more jobs with a bigger context window size. That looks very promising from a productivity kind of view. Yeah, maybe following on from that, Henk. I'll fill it. Yeah, go ahead. We're doing sort of two approaches. I think one thing that is very, very important is the tools are fast evolving. To sort of make any statement about the tool today is inevitably going to be evolving and changing. We're actually seeing progress with a dedicated team that we're working with on almost every area of the software development lifecycle. Not everything is always as impressive as you hope, but you are certainly seeing sometimes substantial gains, things like multi-agent developing, sometimes looking at some of the areas you can push across. It doesn't matter whether it's testing, documentation, the coding itself, even interestingly, things that maybe take a bigger shift of people's mindset. It isn't so much about the tools themselves. It's sometimes about how we approach. One thing we started to see, and it is only early, and it is only through people who've become much more comfortable with the tool sets, is even in architecture and architecture design, being willing to maybe spend more time at the front of a project looking at how could you do this? How could you change the approach? Could this be done in five different ways that you would never really have been able to evaluate? That actually can lead to an intended or sometimes quite substantial benefit further down the cycle. I think, yeah, it's not in every business. It's not in every environment, but it's something we're pushing and trying to then broaden and get that thinking into all the business units, but allowing them to develop their thinking in that same time frame as well. I think today we can point to benefits across the lifecycle. I think it's also being very aware that the tools will continue to evolve and change. I'd certainly say we feel they're going to keep giving us more opportunities to keep driving that progress. Paul, I think you probably... May I just give a little bit of background, Paul? Paul gets to see what we're doing across a number of business units, works on projects with a number of our businesses on a for-profit basis, sort of think internal consulting for which you pay. Obviously, people really want him and his business units' input. Otherwise, they wouldn't be writing checks. He specializes in the AI sector. Did you want to answer the question, Paul? Yeah, just to add something on what was already said. Indeed, the models tend to perform better when it comes to writing new code and troubleshooting or kind of finding issues in existing code bases. That's mostly because of how these systems have been trained. Keep in mind that these large language models have been trained on publicly available data, which for coding means open source repositories. These systems have become quite good at generating new code, but there aren't a lot of data sets online when it comes to defects, like examples of data sets where the LLM can easily understand this is a defect or this is a bug in an existing system. Indeed, the context window is a limitation, but we're seeing practices around context engineering and best practices for developers that learn what to put into the context window in order to achieve good results. I agree with what Chris is saying that we're seeing velocity gains and improvements across the development lifecycle, obviously with different gains in different stages. Coming back to the original question, which was, should we consider rebuilding software than extending or maintaining what we already have, considering that these LLMs are better at writing code than maintaining code? Regarding this question, it's important to consider that even if we rebuild with AI, we're still going to have to maintain it. I think in some cases, AI will enable us to modernize and rebuild our solutions more effectively than we were able to do so before. At the same time, how we maintain, troubleshoot, and bug fix our solutions, we're still going to have to do that whether the code has been written by AI or by humans 10 years ago. To jump in and sort of drive home that point, it's really easy to get excited about 10x improvements in programmer productivity as you generate that new application. If that new application goes out into the field and generates scads of bugs reported by clients and is fundamentally difficult to change and improve, you may give up on the roundabouts what you made on the swings. You may end up with higher lifetime costs of the code base. Similarly, you have to take into account the efficiency of the code that the AI produces. I think we're very early days. We know there are some wonderful advances in programmer efficiency on the front end, and we just don't know the answers on the back end yet because we haven't lived with it for long enough. I think that's important. I'd like to add, Mark. Yes, Jérome, did you have a second question? I did. Thank you. Thank you so much. More on the financial aspect on the M&A side. You know, not knowing what's going to happen in the future sometimes comes with a higher discount rate, affecting terminal value in some instances. I'm wondering if the organization is now using a different discount rate and if there's an impact in terms of meeting your hurdle rates. The advantage of using a high discount rate already is that it minimizes the value of the terminal value in your overall assessment of the attractiveness of the investment. We've got that going for us inherently. We're already discounting the future a lot. However, I haven't yet seen, except in a few instances, acquisitors coming along and saying, "Yeah, AI is a real threat on this one. You know, we've got to increase the probability of a wipeout and a modest win in our distributions as we look at the probabilities of the outcomes." That doesn't change fundamentally our discount rate, by the way. It just changes the probabilities of the downside scenarios, which is important. It's really the essence of what you're asking. I'd say it's rare right now, but people are aware of it. In those sectors such as customer support software, they are factoring that into their thinking. In many of the verticals, it's not yet influenced the prices that are being paid. That's my sense. I don't think any of the panelists are really in a position to talk to M&A specifically. Let's move on to the next question. The next question comes from Stephanie Price with CIBC. Please go ahead. Good morning. Hi, Stephanie. Hi. Curious what Constellation would do if a competitor came out with an AI-embedded feature that was driving higher customer attrition at the vertical market software. Can you walk through kind of the thought process on what the competitive response would be from Constellation at a high level? I'd hope that we'd respond quickly and be a fast follower. Maybe there are individual use cases or instances that some of the panelists have encountered that talk to your question. Maybe let's start with Cam. Cam, have you seen anything where we're chasing a competitor with an AI solution? Yeah, not deeply yet. A lot of the industries that we tend to sort of focus on with the division that I'm with tend to be really slow followers. To use Mark's expression, they tend to be sort of in the utility space, telco space, and so on and so forth. I think if anything, we're often finding ourselves driving the innovation sort of piece forward. Oftentimes, based on just the sheer volume of data that we have, running new models on it, we're able to all of a sudden look at optimizations from workflows or optimization of water consumption and so on and so forth that historically we may not have been able to. As we look around us in the competitive landscape and the market consultants that ultimately help these cities, municipalities, utilities, so on and so forth, to acquire new vendors, they are usually pretty good at giving us a bit of a sort of indication of how we, in a post-mortem fashion, fared and we always sort of score actually quite well in the AI space. Chris, any thoughts on how you respond to AI-enhanced competition? I think at a fundamental level, it doesn't change any of the business principles we all apply. We're fundamentally passionate about the customer journey being very, very close. It's one of the huge benefits of the decentralized model that we are very, very focused on our markets and our sectors. Frankly, it's something that's been around forever that, yeah, there could be a competitor bringing out a new feature, a new capability, and we'll address that. I think specific to AI, I think it is sector-dependent. There are some sectors, probably obviously the well-funded larger segments are always ones where probably there's going to be those opportunities. I think probably what I can talk to is we're seeing examples where our businesses, early stages, nothing yet that we could give you absolute definitions on, but are already starting to look through a different lens, looking at how they can actually provide new opportunity either in the existing space or even starting to maybe even look slightly wider. I think it does go back to an earlier point in a way is even with older technologies and legacy systems, there is a drag. I think AI across the lifecycle can actually help make that drag more efficient. We're still yet to see whether that's 5%, 20%, 100%, whatever. That does potentially allow people to then get back to that close to customer and start to innovate. We have one project running at the moment with a customer that essentially was, I suppose in a sense, getting frustrated. That's something they'd been hoping for for 12 months hadn't been delivered. We've changed completely the way it's being delivered, developed, tested, use cases being built, everything in terms of the lifecycle. It's seen a significant shift in velocity. Something that was already behind is actually now back on track and working through. It's more of a recovery, if you like, than a sort of a new area. It is providing opportunities that we do keep seeing. It's patchy. It's still evolving. I think, again, another to go back to our decentralized nature, the other thing we do is try and share the good and the bad. We're trying to make sure people understand the best practice, some things that don't work. That's giving real confidence as each business starts to see other people seeing those, let's say, social proofs that it's something that they can then take forward and move forward. I think there's lots of good examples moving forward. [Henk-Jan], are you deep enough in the weeds of the individual business units that you are aware of one where we are having to respond because of competitors' deployment of AI? There will become competitors who will come with initiatives based on AI, but we're not waiting for them. We really love the context richness of our VMSs, and it gives a lot of opportunity to enter fields of functionality in order to benefit our customers with, for example, functionality which can reduce a lot of hours. For example, in the field of education, we will release functionality which will enable mentors of a high school to save a lot of hours because we have all the data of students which we can convert to a holistic view of the students involved. You can give a better advice to a student where the mentor can give a better advice, and it saves them a lot of hours. We see a lot of opportunities. Our competitors will also develop kind of those functionalities, but we are very well positioned to profit from all the possibilities the AI technologies will give us. Thank you, [Henk-Jan]. Paul, are you actively involved in any projects where we're responding to competitors' AI initiatives? Yeah, there is one example that comes to my mind. The response approach was as follows. First, distinguish between actual value being brought by an AI feature and what's called AI-washing. This is a term which refers to companies adding the AI label on their products for sales and marketing purposes. A good approach is first to say, okay, is this competitor targeting some real value, actually making a meaningful difference in our market, or are they just adding the AI label on a feature to make it more attractive? After that, the response that we're recommending is make sure that you really understand the problem that you're aiming to solve with AI. We're seeing so many AI solutions which are actually solutions in search of a problem. It's starting with the technology rather than starting with the customer journey or with the customer problem that we're trying to achieve. In CSI, as all of you know, we're highly pragmatic in terms of our approach. Taking the same level-headed approach when it comes to responding to AI threats, I think, is the way to go. There is a risk that we're overreacting and overinvesting into certain things where, again, they're just solutions in search of problems. I know of an instance in one of our largest business units where we launched an AI-based business intelligence solution aimed at senior executives that would enable them to query the entire ERP and gather information from it. It's been successful. It works, but the utilization is flat. Basically, it was a solution in search of a problem as opposed to something that was driven by real customer need. As you say, you can use AI, but it doesn't mean you necessarily deliver value. Next question, please. The next question comes from Paul Treiber with RBC Capital Markets. Please go ahead. Good morning, and thanks for doing this call. One of the panelists mentioned data that was proprietary to Constellation or one of the business units. Can you speak to what you see as the characteristics of vertical market software that make it an attractive market structure? Specifically, things that are proprietary insights that your companies, Constellation's companies, have accrued over time that would make it difficult for new entrants to try to emulate the underlying attributes or functions of that software. Before we pass that on to the panelists to respond to, I'd like to distinguish when we talk about data between the data that is the customer's data and that lives in their system and the data that is our data about the customer and how we interact with the customer and lives in our system. The customer's data frequently is highly available to them. Even if it lives in a proprietary database, they frequently have some sort of data repository where the data gets dumped near real time, which they can access if they want to generate queries and reports and interfaces and things of that nature. Customer data, I don't think is a barrier to anything and anyone, whether it's AI or other vendors. Our data about our interactions with our customers is very proprietary and has the potential to be a pretty exciting tool. Why don't we do the old alphabetical approach? Cam, any thoughts on data as a barrier or asset when using AI? Feedback. Feedback. Yeah. It's a point one made. I think fundamentally, the customers have access to some of their data set and whatnot. I think the power that we're able to call that is really the aggregation of these various data sets as layered geographically, as layered based on the volume, seasonal changes, layering sort of weather patterns, so on and so forth. Taking that sort of history, recreating, running AI models to try to find novel correlations that may not have been there, all of a sudden makes path or makes way for net new insights and offerings where, sort of in a predictive fashion, you're able to detect certain events or leaks or collections events or what have you that would have been purely off the back sort of catalog of our data sets that have been there and sort of nuanced. We have a really, I think, reasonably good idea based on the specific set of circumstances that that given sort of city operates by and how we can more closely correlate it to a given customer that that might be sort of closely correlated. I think having a model reside on top of that data set and be able to provide us with the insights that could then be productized is sort of one key area that we're sort of focused on and making sort of reasonable sort of headway. We're sort of optimistic on sort of what's ahead there. There are also some dead ends that we've also hit in doing some of these sort of investigations. Complexity of the vertical market software space in the utility slash sort of government sector is that every town, every village, every county, every sort of thing may completely be a bit of a unique snowflake. There is no real rhyme or reason why they've adopted this billing methodology and such. That certainly adds some complexity in the correlation of data. Despite that, there's been some good ones identified so far. I think the main point there was the fact that across customers, you can extract insight that you might not be able to extract within a particular client's data. I think that for sure is powerful with most of our businesses because, of course, we have multiple customers in a particular vertical. Chris, any thoughts? Feel free to pass if it isn't your cuppa. I agree with everything Cam said. I also feel that aggregation of understanding, which I think is more than just data, it's processes. Understanding how they drive value in their customer markets actually just layers on top of all the opportunities that you can actually drive forward. I also feel like there's a whole range of opportunities coming with things like predictive that could drive capability there. Clearly, still to be proven. I do think there's capability in all of those areas. [Henk-Jan], any thoughts? Yes, because, of course, the data of the customer is from the customer. No discussion about that. I think it's all about the dynamics of the data. A lot of functionality only works if it's almost real time, the real-time status of the data. Of course, customers have invested a lot of time in their way of working. I think because we can influence the dynamics of the data and use it with the possibilities given by the AI tooling, we can maximize the value for our customers in terms of all the context richness we have and we can serve through the AI tooling in order to make great functionality, great value for our customers. It also makes things—the scope we can act in is also more broader and deeper. I think a lot of new value can be unlocked to our customers. I'm looking very positive in that way to the future. We have a lot of examples which prove that AI tooling can give a lot of value. For example, in healthcare, people can just ask their questions in natural language. Because AI understands the semantics of the question, it can give answers they really want in terms of their information needs. Just because of natural language, it's understood by our systems. That's what I also mean by the dynamics of the data. If you have a question about some patient, it can get a very precise and correct answer because it understands really the dynamics of the semantics of the question. A lot of possibilities in this perspective, I should say. Cam, I wouldn't mind looping back. [Henk-Jan]'s comment made me think of the demo that I circulated to the board last quarter of a customer service interaction between an AI-powered support agent and a customer inside of one of your utilities. It was really remarkable, the fact that it wasn't a chatbot, it was a telephone call, and that the voice recognition was working in real time. The interaction was very sophisticated, lengthy, nuanced, and you didn't at all feel like you were being held hostage by a low-intelligence chatbot. Can you talk a little bit about what you were developing there and how close we are to being able to roll it out? Are we still limited by the quality of the AIs that are out there or the expense of deploying these kind of solutions? No, I think very clear to the question. Sorry. Go ahead, Cam. Yeah. Okay. Thanks. Yeah. I think for us, the catalyst was because we have all of the workflows already sort of established, all of the various click-throughs that an agent would have to do. Instead of using just a regular agent having to do it by hand and such, we essentially looked to tie those processes, those actual workflows to our natural language structure and LLM that's specifically that as its asset class of specialty. The key component to decipher there, then the complexity that we have to overcome is oftentimes the right or the same LLM isn't the correct LLM to tackle the various challenges at hand. Mark alluded to it based on the weaknesses of the actual LLMs and such. Having to embed a router structure through that architecture as an example where it's able to decipher which ones to pick in order to tackle mathematical computation versus those that can just be purely word comments as an example and multilingual, and so on and so forth. The capability of doing this in real time was really a critical component of making that successful. The neat thing for us is that it's the reliance on our own systems and not a third party that's capable of doing that, which would be far more difficult to retrofit. I'm not sure if that answers your question, Mark, or if there's anything else you want me to kind of zero in on. Fundamentally, I think the key thing is that proprietary applications tied into our actual workflows, natural language interaction with the customers, and able to automate flows in a much, much more intuitive way than the historical way of doing things. I think to Paul's point, which is the access to proprietary data about the workflows of that particular client, I think you've answered that nicely. I guess what I'm asking is, is it ready for prime time? Are we going to install this on every utility next month, or is it keep a human in the loop 95% of the time just in case? Okay. Gotcha. I think right now, it is out in pilot with customers as we speak. Should they greenlight it and be comfortable, then we're off to the races. The biggest complexity, especially about the understandable byproduct of the industries we tend to operate in, where getting things in any type sort of erroneous can have catastrophic impacts, the cities, the utilities, the health districts, or what have you, they really want to ensure that their CSRs ultimately can rubber stamp the thought process and how the AI came up with its reasoning prior to pushing it through. That user-in-the-loop component will, over time, as we get enough data and they build enough comfort with the statistical proof that it was not overwritten and this has now circulated 5,000 times and so on and so forth, will we start decoupling things from being user-in-the-loop to fully automated end-to-end? I think that type of partnership is going to be key for us to push it forward. Quite frankly, it's something we encourage our customers to do as opposed to let it be as is from right off the rip. It also helps us gain greater and greater comfort, certainly. Thanks, Cam. Paul, you always get stuck at the end of the list. Do you recall the original question about data and whether it was a barrier or an asset? Any thoughts? Yep. Just a quick thought. From my perspective, there is a lot of focus on data as being valuable and the fact that it will act as a moat. In my view, what might be more important than data will actually be processes and workflows. Our businesses have incredible knowledge of the end users' processes and workflows, often better than the end users themselves. I believe this will be the big opportunity for us in looking at those processes, trying to reimagine some of them by embedding AI in certain areas and going from systems of record, systems that capture data and allow you to edit and retrieve data, to systems of action, which in some cases, the AI agent or the AI solution will do certain steps that will automate more of the human work. One of the things that we're recommending our businesses is, again, looking at data is an important first step. As has been discussed, the customers already have access to that data. In some cases, the end customers are experimenting with other AI solutions on top of that data. The processes, the business rules, the workflows, that's something that we build into the system. I think we're going to be able to leverage quite nicely there. I agree with Paul entirely. I believe that vertical market software is the distillation of a conversation between the vendor and the customer that has gone on frequently for a couple of decades. You distill those work practices down into algorithms and software and data and reports. It captures so much about the business. Being able to examine that in a new way because of AI creates a new opportunity to modify and change and suggest new approaches. I'm hopeful that that unique and proprietary information will be of value. Exactly. That was a call from likely spam. Why don't we do one more call from the lines? I'm going to ask some of the questions that I was sent beforehand. The neat thing about emailing in questions is you can be less pleasant in emails and you can ask tougher questions. Let's take another one of these nice questions from the line, and then I'll pose some of the tough ones from the emails. Okay. Our next question comes from David Kwan with TD Cowen. Please go ahead. It's Paul. I was wondering, Mark, if you're seeing many of your probably larger customers using GenAI to internally build solutions that could displace or maybe are displacing some of your PMS solutions. Maybe a response that helps put that question into context because I think it's a great question, but it's one that we have been confronting forever. I see vertical market software as sitting somewhere between horizontal applications that are cheap and cheerful and do 50% of what you want and highly customized systems that do exactly what you want. Obviously, you have to hit a certain price point to live in the middle where most vertical market software companies live. You frequently have professional services to provide some customization, but those professional services, whether it be custom programming or otherwise, are expensive, and only a certain class of client can afford them. Those who graduate from horizontal point solutions laced together with Excel to vertical market software at the low end are going to have very little in the way of professional services and customization and custom interfaces and custom reports. At the high end, they're going to have highly customized systems where we are willing to do whatever they want as long as they have budget. Our people aren't cheap, and they're going to have to pay for that. That has always been the case, and the very largest clients frequently see their software as strategic. It isn't just the tools to do business. It's a way that they differentiate themselves from their competitors. If you're dealing with a highly differentiated large client, they're going to want to have that be proprietary to them. They're going to try and capture as much of that information technology advantage within their own realm as they can and control it. We frequently do lose large clients to an SAP implementation or a proprietary implementation, and that's always the case. We capture the small companies as they graduate from horizontals. We take them, and some of them grow enormously and become very successful large companies. They graduate to no longer using our systems, but to using a much more proprietary system that they have a much stronger hand in driving. AI has the potential to allow us to do way more work on making the client happy and customizing our solutions, but it also allows the client to potentially do that. There's a natural tension there. We obviously would love to capture that. Our clients, if they don't have a list of five years' worth of IT projects to get to, would obviously love to capture that as well. I think to some extent, whenever we go see a large client's IT Director, we're in a negotiation regarding what we'll do and what they'll do. It's not going to be an easy answer. It's going to be somewhere in between. AI makes it potentially way more exciting for us to provide customization, but it also makes it much more likely that the client will do it themselves. Why don't we do the reverse alphabetical? Paul, any thoughts on this? I know you and I have talked about it a little bit. Yeah, nothing to add to what you've mentioned there, Mark. I think that natural tension will still remain. I think there will be big customers, maybe with the new CTO, being tempted to say something like, "Hey, let's give this a go. Let's try to write it ourselves or build it ourselves with the help of AI." I think some will be successful. I think most will minimize the effort and the complexity of replicating some of our offerings. To me, what this really means is that this should be a trigger for us to spend more time with the customers, to refocus on customer intimacy, to make sure that we're understanding whenever they're considering this type of approaches, and to make sure that we remain as competitive as possible. [Henk-Jan], a ny views? Yes. What we're seeing is that AI add-ons are proposed by clients. That's okay with us. Of course, we want to be proactive in those kinds of new functionalities. We see both working close together with clients in order to connect to AI tooling and, of course, provide it to our customers. Mostly, those kinds of tooling are more stretching on the surface, the real functionality which needs integration and deep integration with our solutions. Those are the things we are working on, in close interaction with the customers. Those are not the areas which will be claimed by customers because then you have to have deep knowledge of the workflow and of all the functionalities deep within our software themselves. Both is happening. Yep. Chris, any views? I think also, the original question in terms of customers building their own effectively DMS, don't underestimate the complexity. Even with AI, it still requires high skill, high capability. There are easy headlines to sort of say systems are able to be rebuilt. Yes, we're seeing efficiency and huge change in what can be achieved. I think reiterating Paul's point and what we've been saying all along, if you maintain the customer intimacy, we see ourselves as a company needing to move forward, be able to offer more capability through AI, probably actually help a lot of those customers. Again, many of our customers are in mature, less dynamic marketplaces and actually just interested in their own business and making money. Sure, some large customers may try, but I think overall, if we've got good relationships, good connectivity to what they're trying to achieve, it may shift in terms of exactly what we're doing and how we're doing it. I feel like we will still be there to drive new opportunity and trade the capability. Only if they've got a particular dogma, which as Mark Leonard said, has happened many, many times over the last 30-odd years of software where they make a decision to shift to a different product set or build themselves. I don't think that will shift its dynamic to anything we've seen in history or going forwards. It's tempting to hang you by your own petard. You've got a couple of use cases where you've seen like 50x productivity improvement and 10 x productivity improvement. If our large clients have those same sort of improvements in their development, are they going to chew into our value-added? Yeah, I feel like this was more so a trend that sort of I've been in this space for, give or take, 20 years. I would have seen that well in the past. The key reasons why the customers have largely been increasingly dissuaded from wanting to do this is often driven by regulatory changes that all of a sudden hit their mandate. They're now all of a sudden on a time crunch and having to withstand those governmental pressures independently. There is often a sizable benefit of them hitching their wagon to a best-of-breed solution like ours where, if there is, let's say, the concept of water conservation that they're all of a sudden having to abide by governmentally, there's a really good chance that we would have equally done such a functionality in different zones or for different cities, in different states, and so forth. We're able to re-leverage much of the architecture, much of the structure, what had been built there to greatly accelerate their current piece. We don't anticipate AI changing much of that. I think it's ultimately the customization that's required governmentally that comes with a penalty or a time crunch. It's not something that they have a great deal of appetite for. Yep. Just to sort of attack the underlying thesis of David's question, we have a large business unit where we've had AI programming tools in place for a year. We've certainly seen significant increases in the number of lines of suggested code over that period of time by the AI, and the percent of lines adopted has stayed pretty stable. If we look at the actual programming efficiency, it's almost entirely flat. You know it isn't a panacea. We aren't always going to get enormous increases in programmer productivity in this large business unit as an example. We haven't used programming agents in this particular instance. We've been using fairly simple tools, and we are moving to the next stage and trying out agents that are much more sophisticated. We'll see. There are a couple of other instances where we're seeing overall efficiencies in the 10%-12% region. As you may have heard, Google has reported something similar. There are also instances where we do see some very significant improvements. Whether those will be through the full life cycle of maintenance and support, whether those will be maintained is yet to be seen. Here's a tough question that came in, which was that AI will eat software budgets. If clients have a budget and they view software and AI as one particular pocket that they're going to spend in the coming year, and they are being approached by a host of horizontal AI vendors that are doing voice recognition and OCR and a variety of other sexy things, are they going to siphon off a bunch of money that they would otherwise spend with us to those other AI vendors who are incredibly good at promoting their solutions? Why don't we start with Cam and go alphabetical? Yeah. Is it going to eat our budget? I think if we are in front of the customer, proactively speaking with them, talking with them, partnering up with them, the way that we tend to really sort of engage with our customers, then part of our job is to sort of not give them a tremendous amount of reasons to want to sink their teeth into other sort of AI one-offs, if you will. Ultimately, this is not a new phenomenon where sort of third parties of any kind at any conference that we sort of may share ultimately sell their own sets. There is sort of complexity of Frankensteining it all together and making sure that it all works adequately well, and hidden fees that they may not have factored in and such. I think there's all of these components to really think of. As it relates to the AI piece itself, there's this perception, depending on the viewpoint that a person has and the use cases, that it could be relatively inexpensive, the actual token cost of an LLM as an example. If you factor in, and as we have where we're crunching an actual invoice that evokes the LLM numerous times to do analysis or anything computational-wise, or if I have an example set, that could become expensive very, very quickly. One of the pieces that we've gone about combating, and that speaks to the pricing thoughtfulness, if you will, and making sure that we're conscientious of not eroding margins and whatnot, is to use some of the pre-existing assets that we have proprietarily sitting, so hardware and servers and so on and so forth, and then run some of our own models, run some of our own LLMs that we end up morphing into one that's trained with our data sets, with our specialty, with our additional intelligence and whatnot, where in most cases, those LLMs are independently able to address the given AI needs and cost, therefore no incremental token costs, because it's really all within our own four walls, if you will. I think that's one of the approaches that we've taken to minimize the cost. I'll pass it on to the next speaker. Chris, any thoughts? Yeah, very, very similar to really what Cam's saying, but I just generally think that if you continue to talk to customers, you talk about relevant pain points. I think one of the responsibilities we have is to understand the potential, not to get carried away with the hype, but to understand, be sort of healthily skeptical, but driving where we can see potential opportunity. Just continue as we've always done to work with customers in their specific markets and show that evidence. We have one example in the education sector where they've been talking to their customers and always had an ambition to try and be something very different and very focused on that particular segment. It's really now allowed them to explore that. They actually feel that's going to give them revenue growth and real opportunity, still to be proven. It will be over the next couple of years that that will come through. It has allowed them to change the conversation. I think that's probably the most relevant part. If you keep going and talking to people about things that were relevant a few years ago or just in the traditional environment, yeah, sure, in some circumstances that's relevant. I think it's really also being able to talk in a sort of modified language with the sort of cautious skeptic in your mind as well to actually ensure that you're still driving those new opportunities. I think if we do that, then I think the budget worst will stay the same and could even increase. [Henk-Jan]? Yes, I think the thesis assumes that the world will be steady. The opposite is, of course, true. There's a lot of dynamics around answering this question, I should say. We saw it with low code. Low code was the promise of automating all the processes, and there wouldn't be very sophisticated software developers needed anymore. Anyone could contribute to low code systems, and reality was completely different. I think because the scope will broaden and deepen, there will evolve very interesting business cases. If you talk about software budget, it's only costs compared to the very interesting business cases which will evolve. I should say it could be that software budget will become much bigger because of the broadening and deepened scope. I think that will be the case because there are a lot more possibilities to optimize businesses. Of course, all the businesses using our software also will have stronger competition. They have to respond to their competitors. Of course, AI and IT will be the means to compete with competitors. It will become even more important to have a unique selling proposition from the businesses using our CMS. I would say software budgets will certainly not be eaten by AI. It will be leveraged by AI. And. They can't see in the future, of course. That's my personal opinion. Yeah, yeah. My personal belief is that we all can't see in the future. Paul, I'm going to pose the next tough question to you. Specifically, it's AI introduces a new COGS element, not historically present in software. Does this change the economics of the business model? I think specifically what we're seeing here is the adoption of AI is being massively subsidized by the AI companies, the AI model companies. At some stage, they will want to recapture that investment. Are they going to be in a position where we face very large switching costs, and they are going to be able to capture a large chunk of our value added through what they charge us for their systems, whether it be on a per token or per whatever basis. Paul, any speculation? Yeah, maybe it's worth briefly outlining what we know right now before we look into what might happen in the future. As of right now, these model providers are charging anything between $1 to $3 for 1 million tokens. You can think of tokens roughly as words. There are many studies that show that AI platforms users consume on average per month between 50,000 tokens, which are the light users, and 1 million token users, these are the heavy users. Based on the data that we have right now, we can infer that if a CSI customer would start to embed AI features into their product, there's going to be an estimated cost per user of anything in between $1 and $8. I think we can easily cover this and maintain our margins by having premium add-ons where our end customers, if they want to leverage these features, can buy those premium add-ons. Indeed, it's hard to predict how this will play out in the future. The good news is that currently, these large language models do not have a very large moat around them. Some of you might have read that when GPT-5 came out, there were some doubts regarding its performance. They had some issues with their deployment. Overnight, there was a huge switch from OpenAI models to different providers, with virtually one line of code being changed into the consumer of these LLMs. This tells me that there will be high competition between these model providers, and this will keep the pressure on the cost. I think in the long run, the price per token will go down. Again, we can speculate about how these companies will start to build up their moat so that it's harder to switch from one LLM to another. So far, we do not have clear data or indicators to work that. I believe that as of right now, if CSI companies are adopting AI features, we will be able to maintain our margins. If one of the big providers is starting to hike up their prices, we always have the option to use things like model routing, using smaller models for different tasks, or even on-premise LLM inference by leveraging open weight LLMs. Cam, you have thought about this probably more than anyone else inside of Constellation. You've architected much of your efforts around AI, around the threat of third-party LLM providers preying upon their customers. Do you want to talk a little bit about what you've done? Sure, yeah. We've essentially created our own centralized sort of platform that removes the various factions that are currently going on, where to a certain extent, you have to largely be within this cloud provider to have access natively to this LLM and so on and so forth. There are these turf wars being created across the various cloud providers and whatnot. Our strategy has been to really play a very neutral sort of Switzerland type role where by centralizing things both through strategic relationships, either directly with the model providers or with the platform providers and so on and so forth, we've managed to negotiate, I think, some really aggressive deals and remove the element of these factions. They're all willing to play nice with us in the sandbox. That puts us in a very unique position where technically we have access to 15,000 unique models. That's because we're essentially coalescing anything that otherwise couldn't be or reside within other platforms. The other piece that I had touched on very briefly and Paul alluded to as well is using AI on-prem-based assets where and when possible. To the extent that the LLM needs to be or the AI model needs to be hyper-specific or a specific trained one that resides with a pre-existing best-of-breed provider, then sure, that may make sense to tap into that one. For basic, let's say, translation service, summarization service, and a myriad of other hosts of functionality and whatnot, the on-prem one is plenty sufficient and capable of doing its own thing. I think there ought to be some thoughtfulness of the whole, do we need to go with this? There's a website, I think it's called, sort of like there's an AI for anything. At the end of the day, going through these repositories, do we build or do we partner? In many cases, it's been fairly painless to augment the functionality to be able to natively create, let's say, slides or Excels or presentations and so on and so forth. The functionality is becoming richer and richer, and the adoption rates internally from our business units, that's the closest thing that we've, or our narrative internally is that it's the closest thing to a viral application that we've ever had. The adoption has grown, I think, month over month, something sort of 450%, give or take, continuously. The lion's share of our staff are using it for a myriad of things, and we're excited by that. I think it's just being smart about the way we architect stuff and not reinventing the wheel 10x over if we don't have to. Dave, why don't we take another call from the lines? The next question comes from Samad Samana with Jefferies. Please go ahead. Hi, good morning, and thank you for taking my questions. I'll echo the sentiment of others. I appreciate you guys doing this. Maybe just, I know we've talked about the implications for Constellation 's portfolio, but how should we think about how this changes the nature of M&A you may pursue, whether that's changing the size of the company you may look at, whether that's targeting different verticals, or maybe where they sit in the software stack? How does it, maybe, how does AI change your M&A strategy? How does it actually change your appetite for M&A volume? Meaning, is it better to be in more of a holding pattern right now, or is now the time to really lean in? We've seen some large M&A deals announced by private equities, so I'm just curious what your philosophy there is. Thank you again for taking the questions. I think the underlying assumption is that we're not opportunity-constrained, and we are opportunity-constrained. Narrowing the aperture is a bad idea. We end up sitting on a whole pile of cash. When you're striving to generate very high rates of return on your investments, sitting on cash is not a good plan. We already are working hard to look at things outside of strictly vertical market software. We've done some horizontal stuff. We've done some hybrid hardware software. We've done some hybrid data software. I would say that AI is not reducing what we're looking at. I'd say it may influence the pricing on certain things where we see it having current impact. It isn't changing much in the M&A world. Understood. Yeah. Oh, sorry. I'm going to pose a question to myself, which is, tell me more about the operating group where you have stats on their business units, etc. I'll ask the other folks on the line, who none of them is from this particular operating group, if they have similar stats or if they have an impression on how they stack up versus this particular operating group. First off, 27% of the business units in this operating group are developing an AI product for their customers. I wondered, why don't you start, Cam? In the BUs in your operating group, what percentage of the BUs do you think are developing an AI product for their customers? I think for us, we mandated for all of them to experiment with the creation of a solution. Technically, the answer is really all, I think, based on the footprints that I'm also able to capture using our centralized tool set where whatever LLM it is that they would have wanted to use, irrespective of vendor or whatnot, would have been captured here. They're all in various degrees of progress of having a solution offered. I think there's some low-hanging fruit that most of our business units ought to really, as a table stake, starting point to have. I don't think that they're striving for ideas that they don't already have. Some already have some products out there and in full sales mode, whereas a lot of them are in varying degrees of the build, if you will. I would say the nuance of the question, I think that I don't want to misrepresent this piece, is there's a consideration of where AI is internal focused versus external focused. The customers will see benefits oftentimes by way of better quality of support and so on and so forth as an example, but not necessarily in a new front-end screen or widget right off the bat. These projects vary between internal focus versus external focus. To the tune of roughly 50/50, it's sort of a 55% is facing, so a 45% is process-related. Was what you were saying there that the products being developed for the customer, 50% of them are aimed at the customer's customer and 50% are aimed at making the customer, our customer, more efficient? Perfect. Yeah, you bet. Okay. The 100% mandate, I mean, one can order people to do all kinds of things and you can get lip service as opposed to strict and enthusiastic compliance. Sometimes it works and sometimes it doesn't. I suspect the 27% that was reported by this operating group is people who are seriously implementing a development for an AI product for their customers. I would agree. As opposed to. yeah. That makes sense. Chris, any comments? Yeah. I'd probably say we're trying to look at it at three levels. If you said, are people using AI in some way to drive something either internally or even just engaging with a customer, I would say that's very high adoption. For us, what we're really then trying to focus on is, I know it's very heavily used, but the sort of 10X principle. Where are things being done that are genuinely shifting either thinking or approach or even product for the customer? I would say that drops back to probably in the 20s as a game. What we're also trying to do is, I guess, differently to the sort of 100% mandate route. The joy of decentralization, we've been taking a much more educational proof point, trying to show best practice. I think we now start to see that that really comes through with a lot more people now trying to outpace that mentality. Today probably in the 20% region for something that is meaningful, but I think we'll see that accelerate through the next 18 months. Yeah, and I think that meaningfulness criteria is, you know, clearly CEO fodder. Managers love that stuff. I also love AI tools, and I think so do some of our clients that are aimed at relatively small demonstration type activities. We have, for instance, inside of our operating groups, three separate initiatives all designed to inhale contracts of businesses that we're looking to acquire and analyze those contracts. All of them have been relatively modest efforts and quite successful and easy to benchmark against not using AI and easy to benchmark against commercially available AI. I think we're all feeling tough with ourselves about those developments. That's wonderful. I think from a morale and proof point of view, it's terrific, but you know, maybe we're optimizing 100 people here out of our 75,000 with those three solutions. Good on you. I'm glad people are doing the experiments and are learning from them and are applying them. Not everything needs to be order of magnitude meaningful for it to be useful from both a morale and a customer enthusiasm point of view. Let's move on to the next category on the reporting that I got here. It says this BU is currently using for customer service, currently using AI for customer service. The result was 29%. I guess, [Henk-Jan], any sense of whether inside of the operating group that you're associated with, AI is being used for customer service? I think it's more than 50%. It's a very logical area to have AI support within because you can have all the questions with all the data, all the questions which were asked. Of course, it's very easy to let AI support you and to have a very correct and good answer to end customers. I think there's still space to gain, but more than 50% is supported by AI in my group. Yeah. I had expected this to be higher, and I'd also expected it to have better results. I've a couple of instances where I've gotten the data about our AI customer service agents, and I'm seeing core diversion of 10% - 20%, no, 50% - 60%, which I guess I was disappointed because we already had comprehensive knowledge bases for our support personnel from which the AI could be trained. I had sort of hoped that it would be a lot more effective. Obviously, with time, it will get better. At least I'm hoping it will. I had thought this would be the most prevalent area for us to apply AI. Let me move on to the next category. This BU is currently using AI for sales and marketing. 50% of the BUs reported that they were. I've also heard some very nice individual anecdotes about AI being used for sales and marketing that have led to significant new business that we haven't seen previously. Anyone have any comments regarding that and within their business unit? Any sense of what the penetration of AI is in sales and marketing? I can add some thoughts on that. Paul, go ahead. Go on, please. Okay. We definitely see we've seen higher adoption in sales and marketing when it comes to AI tools compared to any other bucket. That's also in line with what other organizations outside CSI are also reporting. At the same time, I think what's important to remember is that sales and marketing is really about differentiation. I think that we will start to see gains initially while we're maybe some of the first using AI in sales and marketing. Once everyone uses AI in sales and marketing, that in itself will not be a differentiator anymore. If everyone is generating LinkedIn posts with AI and they all start to sound generic and kind of the same, I don't think that in itself will drive better sales and marketing results. One of the things that we're discussing with the businesses is how can you use AI to kind of enhance your edge and enhance the relationship that you already have. Otherwise, in a couple of months, when everyone is using AI, it's going to be quite hard to stand out just because you are an AI user. Yep. I don't disagree. Any other comments, or should we move on? From a sales and marketing perspective, oftentimes for us, where we sort of factor things in is we have a better success rate of selling assets when we're able to present a more comprehensive solution set. As we continue to acquire more and more assets, having them results in thousands of products and whatnot. I think a novel way that we've been able to make some, I think, reasonably good progress on the sales and marketing front is by way of actually putting what the customer is through an actual LLM-based assets knowledge base that we created that essentially has information about every single one of our collective [Harris], then the other partners and so on and so forth, assets in order to be able to coalesce, if you will, the right solution set that the customer could want. That has led to successes in where the standalone product we didn't feel had a really good chance. I think that's maybe a neat way of using it in a sales and marketing function. The other piece is if we factor in the RFP within the sales function, I think that's another neat area to trim based on the sector that we operate in. I think there's some really good opportunities for us to dwindle that down based on the hundreds that we've historically done answering these thousands of questions and subcases. There were two more data points in this survey. The next one was this BU is currently using AI tools in R&D, and this operating group reported that 61% of their business units were using AI tools in R&D. Cam, any sense of your operating group percentage of R&D teams? Yeah, it would be really close to that. I think there's a number of sort of tool sets that have been very largely embraced. The interesting piece is really what the output percentage of efficacy is, it quite ebbs and flows. There's some that are gaining abnormally high because they're using it a certain way or their tech stack. I think Paul touched on it where the AI natively is perhaps better trained at specific instances or circumstances and programming language in this case versus some that are more niche-based assets, if you will. The adoption, yeah, about the same, but the output of efficacy is the biggest delta. And Chris? Yeah, I'd say very similar in terms of, I think there's two metrics we would be looking at. One is very similar, certainly more than 50% in terms of adopting, using the tools, starting to see the progress. It's also almost the enablement, the mind shift to use it to get the maximum delivery. I think that's probably a lower percentage at this point, but we're seeing that drive up as well. Yeah, very similar to Cam. [Henk-Jan]? More than 70%, but the level of using AI is still that there's a long gap between because we see AI is preserved a bit like a toolbox full of tools, enabling them to choose the right tool for every job. The knowledge to know which kind of tool you use in which situation, that's still a learning curve for a lot of developers within our operating group. Everybody is touching the use on issues, but the deepness will hopefully rise the coming month. I enjoyed how [Henk-Jan] assesses the AI maturity of the business units because I think that's an important way of distinguishing between those who use it casually and those who have really understood how to use the tool. The last question, and I thought this was an interesting one because it relates to a lot of the underlying questions that I got by email, was, has this BU replaced any roles with AI tools? I think when they say roles, I think they mean people. Roughly 3% of our BUs reported that they had replaced people with AI tools, which is lower than I thought. Actually, I think it's probably a good thing, but I'm placing value assumptions on the question. Why don't we start with Paul? Have you encountered any instances where business units have been able to replace people with AI tools? Not yet, at least. No, because. Sorry, go on, [Henk-Jan]. No, because, of course, we have roadmaps for our tooling, and all the freed up capacity will be used to develop new value for our customers, not things to do. No, we didn't replace software developers. We could give them other tasks to develop. Of course, we also see that if you have young and experienced software developers, they have a much steeper learning curve. We want to grow, and we are growing. AI helps us to make our software developers more productive and develop functionalities with AI to make the value our customers are willing to accept. I think even in support, when we've seen 10% - 20% core diversion, you can, in our relatively small businesses, translate that to the removal of a person. I think in most instances, we just keep the people and try to respond faster to the calls that aren't diverted and try to do a better job with them. Once again, much like in R&D, you don't get to reduce the people cost. You redeploy it, hopefully in the case of support and net promoter score improvements. I'm going to stop the questions there. We've been on the line for over an hour and a half. I'd like to thank you all for attending. I know we have several hundred participants still on the line, so there's obviously real appetite for this kind of information. Let me encourage you not to listen without healthy skepticism to what you read. In the last few weeks alone, I've heard that a major soft drink company increased its sales by 7% to 8% because of AI. I had a look at its stock and it went down. I've heard from the founder of a major software investor that AI just increases TAM, and that's, you know, wonderful, but you got to consider the source. He's not about to say that software is threatened by AI. I've heard from a bank CEO that AI is revolutionizing their business and is going to lead them to a brave new world. It's really important to dig in and try to understand, to be an anthropologist, to observe and test the claims that you hear and try to understand the current state of the art. There are two ways to do it. One is obviously through sort of sleuthing claims that you hear. If you have trusted partners from whom you're getting evidence, that makes life a whole lot easier. The other thing you can do is be a scientist instead of an anthropologist and observing. Actually, run experiments. Try AI and ideally try it against the alternative and see if you get significant improvements in whatever it is that you're endeavoring to do. Predicting the future is really, really hard, particularly at times like these, but monitoring what's happening in real time is a whole lot easier. You just got to approach it with, as Chris said, a healthy skepticism. Thank you for joining the call. Really appreciate it. This is important to us to share with you where we're at on the pursuit of AI and hope that you learn something from today's session. Thank you, Dave, for teeing up the calls, and you can end the call now. Thank you. The conference is now concluded. Thank you for attending today's presentation. You may now disconnect.