Qualitas Limited (QAL) Earnings Call Transcript & Summary

Good morning, everyone. Welcome to this investor briefing, where we're sharing a detailed update on Qualitas AI initiatives. I'm Andrew Schwartz, Group Managing Director and Co-Founder of Qualitas. Joining me today is Michael Kollo, Chief AI Transformational Officer; and Philip Dowman, Group Chief Financial Officer. Before we begin, I'd like to acknowledge the traditional custodians of the land from which I am presenting. I also acknowledge the traditional custodians of the lands from where you are participating today. We pay our respects to their elders, past and present. Today, I will open with an overview of the structural AI opportunity for Qualitas and why we believe we're uniquely positioned to capture it. Michael Kollo will then provide an update on our AI initiatives. Philip will close by quantifying the long-term shareholder value these AI initiatives are expected to generate. Finally, we'll open for questions. I'd like to open with why are we discussing AI right now, less than a week out from June 30. And the simple answer is that we've been spending a lot of time developing our systems that we believe will lead to efficiency gains at Qualitas. We didn't want this just to be the back page of a year-end financial presentation. The topic is way too important and it's deserving of a deep dive discussion. AI is fundamentally reshaping our industry. For Qualitas, it represents a transformational opportunity and it's worthy of its own webinar. At our full year '25 results, we announced the appointment of Michael Kollo as our Chief AI Transformational Officer. Over the past 10 months, Michael has led a number of AI initiatives across the firm. Let me start with the nature of our business. We undertake approximately 40 to 60 investments each year. We are a real estate-focused alternative asset manager with a deep network of repeat counterparties. While every investment is different, our assessment process is highly structured and intensive. We apply established frameworks across data collection, underwriting and credit assessment. The investment assessment process is highly labor-intensive. It requires significant time, significant resources to conduct rigorous due diligence of each and every opportunity. That combination of repeatable processes, large volumes of information and detailed credit assessment makes our business particularly well suited to AI augmentation. In response, we've developed a proprietary AI-enabled credit execution platform. It's designed to improve the efficiency and scalability across the investment process through generative AI. The platform delivers 2 core benefits. Firstly, faster and more efficient investment assessment; and secondly, a deeper analytical insight while still preserving the rigorous underwriting standards for which Qualitas is known. What sets us apart is our ability to implement this technology at pace. With approximately 140 professionals, we do not expect employee adoption to be a material challenge for us. Our team has already demonstrated strong AI engagement with more than 90% adoption of Claude and ChatGPT across the business. But of course, what we're discussing today goes well, well beyond Claude and ChatGPT. We believe our staff are flexible, keen to learn new ways to achieve their objectives. To be clear, we're not handing over decision-making to machines. That's not what we are saying. As ultimately, this is a business built on wisdom and careful human judgment. What we are discussing is how AI can make us a more efficient, increased throughput by assisting us with vast data collection and triangulation of information we undertake. Applying our human wisdom and our careful judgment is ultimately based on data analysis, and it's that data analysis is where we have focused our AI efforts. Interestingly, the conversation internally has shifted. It's no longer just about hiring more and more people to meet the growing demand of the business. It's about using AI to expand our capacity, enhance our productivity and scale more efficiently. We continue to be nimble and we continue to be adaptive. We have deep conviction in AI-augmented analysis with decisions firmly human-led as I've previously said. Now let me walk you through the specific objectives that we're targeting with AI. And this slide captures 3 areas where AI will deliver material benefits for Qualitas. As I said, our strategy goes beyond the adoption of tools like ChatGPT and Claude. We've built a proprietary AI platform purpose-built for our Funds Management business. It combines large language models, machine learning and 18 years of Qualitas institutional knowledge, including more than 400 proprietary investment committee papers. That gives us a capability generalist tools just cannot replicate. And the first advantage is the immediate efficiency it delivers. It allows us to decouple fund growth from headcount by increasing the number of investments per employee without a proportional increase in cost. AI allows us to decouple fund growth from headcount. The platform accelerates data extraction, our underwriting, our credit paper preparation and fund management workflows. Credit papers that once took days can now be generated in hours. Key documents, including contracts, development approvals, presale reports can now be extracted, structured and analyzed far more efficiently. This frees our investment professionals to focus on origination, critical analysis, conviction building and risk judgment. Over time, this supports operating leverage while raising the depth and consistency of our due diligence. The second benefit is better underwriting, faster decisions without compromising our quality. We're embedding 18 years of Qualitas underwriting expertise into each new investment. Every investment can draw on pattern recognition from hundreds of prior transactions. That improves the consistency, it reduces subjectivity and supports faster, higher conviction decisions. The platform pre-flags risks and validates assumptions. This allows the investment committee to focus on judgment, not data reconciliation, and it certainly reduces the iteration between investment committee and the investment team. The third benefit is that it provides AI-enabled intelligence. Every transaction, borrower interaction, market data point, and investment outcome feeds back into the platform. The more investments we screen the stronger the platform becomes. And that creates a compounding data advantage across origination, underwriting and risk assessment. The asset is unique to Qualitas. It reflects our transaction history, our repeat counterparty network and our institutional expertise. We see it as a defensible competitive advantage. So in summary, we're building a proprietary intelligent asset across transaction data, underwriting insights, borrower intelligence, and market information. Now I'll hand over to Michael, who will introduce the platform we've developed. By way of background, Michael holds a PhD in Finance from the London School of Economics, brings 2 decades of experience across BlackRock Fidelity and AXA. He founded AI ventures, advised boards and super funds on machine learning and now leads the AI agenda at Qualitas, Chief AI Transformational Officer. He's here to show you how we're putting AI to work. Dr. Michael Kollo, over to you.

Thank you, Andrew, and thanks, everyone, for joining. Andrew just walked us through what an AI augmented underwriting means for us, enhanced quality and speed together a capability that keeps learning and compounds our data advantage over time. What I want to do now is to take that one step more concrete and to show you why we chose to apply AI inside our due diligence and credit execution process specifically and what we've actually built as well. So let me start with the workflow. When a new investment comes to us, we run an initial evaluation, draft a heads up paper for the investment committee and then move into a detailed due diligence process, including an ESG assessment across many different angles. And depending on the complexity of the investment, that can take months because each investment will be different and have many different unique angles as well. It's enormously data-heavy, first of all. A single investment might carry 160 documents and sometimes significantly more, covering everything from the borrower to the builder to the transaction summary to cash flow reports, to legal contracts and elements and details of the actual building designs and so on. And so there's a whole mountain of data there that our analysts have to understand and analyze. The investment team has to pull the right numbers out of the valuation reports, contracts and financials, know which ones matter, confirm that they are correct, especially when different accounts disagree or more importantly, there's nuance and context around different kind of numbers that could be used for analysis. This is not just a data problem. It's a real domain expertise and judgment problem. There's actually quite common in financial services and certainly in financial statement analysis as well. Then the second part is once we have the data and we've extracted the context and the right kind of numbers, from all the heap of different kinds of documents and forms that the data comes to us then comes the second mountain, which is the analysis. Risk has to be assessed from every single angle. The borrower on the same principles as any credit assessment and the loan itself, perhaps, whether it's a building loan or residual loan or land loan, analyze the way our policy dictates and shape by what 18 years in this market have told us what to watch out for and what to think about. There's no template that you can simply roll out and tick boxes against here. This is a core investment intelligence, and it's really at the heart of Qualitas and how these deals are analyzed. And so then, of course, the question becomes why start here with AI? Well, there's 2 main reasons. So the first one is really around data and data extraction and it's probably one that's kind of most commonly understood in the market, which is what generative AI and language models will specifically do very well is that they read unstructured documents. So they extract data from PDFs, images, contracts and extract not only the information, but again, the context around that piece of information surrounding it as well. And in finance, context is often where the meaning lives. So if you lose that, then you lose much of what makes that number matter or indeed how that number comes into analysis as well. And this is probably one of the main reasons why this form of AI is differentiated from more traditional technology or automation processes that we would have had in finance over decades, which is that it's really able to go into that unstructured database and pull out all the relevant information. The second is reasoning. So these systems don't just extract data. And increasingly, what they are able to do very well is that they help and think and reason through the analysis itself, and they do it very transparently in a very traceable way as well. We can get a very clear audit trail running from a document to a data point to the analysis that we built upon it. And that keeps us accountable. And that keeps a human in the loop, and that keeps us a lot more comfortable in terms of understanding why certain analysis is happening. And I think it also helps us refine and improve the analysis as new information comes to light as well. For people working in the space, they will know that information is constantly moving and updating, analysis needs to be revised as well. And so the systems that are handling that analysis need to be adaptive to that. So both of these have improved dramatically over the last couple of years as AI systems have become ever more capable as reasoning models, for example, have been rolled out more widely, and had been matured and more and more capable models have arrived, not only in the commonly understood world of things like ChatGPT and Claude, but also in much of the open source models as well that have become a lot more capable over time. So we expect that trend to continue over the number of years. And I think our opportunity is to integrate these systems into our investment process now but also knowing that they're already highly capable, but they will only get more and more capable over time and likely cheaper as well to execute as open source models and alternatives come to market that are just as capable as some of the closed form models that we have today as well. So I'm just going to jump now to -- let me show you what we've actually built. So let me give you some high-level numbers around this, and then we're going to drop into a little bit of an architecture here. So let's start with the investment committee paper. It's a substantial piece of documentation. It has to capture analysis done over months by a team of experts. So it's no surprise that a system augmenting that work, producing much of the base analysis is very substantial, too. Here's the scale of it represented in these 8 numbers. So first of all, we start with -- there are 9 assessment chapters. Again, this mirrors are the structure of a typical investment committee paper for a certain type of loan. There are different kinds of loans, again, different levels of complexity that may all require more or less chapters, but I've taken that as the average. Behind them, we have about 33 different agents. And so the way to think about agents at Qualitas is a very deliberate analyst that is doing one type of task, so very cost specific. They don't have a general purpose agents roaming around our systems, moving data around. Each one is responsible for a single well-defined piece of risk assessment. It's highly traceable, it's highly monitored, and it's really understood in terms of why it's doing and how it's doing it. And I'll show you that in a moment in a more graphic kind of way. On the documents, as mentioned before, at least 160 documents per investment. We run more than 370 verifications and checks across those documents, which are primarily checking that they are correctly signed, that the right individuals and entities are named. A lot of the kind of standard but very important groundwork of truth that we have to do in order to have trust and faith in the data that's coming in. It might be mundane, but it's exactly the kind of detail that has to be right and across a huge variety of documents. Once we've confirmed we're working for the right documents, we extract about 400 data points and metrics, many of which we calculate ourselves. So this is going to be the foundational bedrock of a lot of the analysis that follows from here. That's a really important point. So these 400 metrics are somewhat regarding financial statement analysis, cash flow analysis. Some could be details of the buildings. Some could be other types of sales contracts involved. There are a whole bunch of different kind of metrics and data points that we calculate. And not only do we then use AI to extract and make those calculations, but then we second and third check those numbers often with another set of AI auditors on top of them. So AI checking AI, rechecking these numbers again and again and again. Now for a person collating our 400 metrics and rechecking them relentlessly is painstaking and time consuming. But for AI systems, it's actually quite straightforward, and it can be done within the process itself. So we get a lot more faith and trust in the fact that those numbers are going to be correct and have the right form and structure. So from there, we moved to the 260 analysis steps and that number is really a measure of how sophisticated the Qualitas credit system is. This number is really derived from the internal workings of the company, spanning everything from sensitivity analysis, the financial modeling through the trust and corporate structure sitting across the transactions. And beneath these steps sits a library of, in this case, more than 1,900 different risk questions or angles of risk that we can be thinking about for any given investments. Now not all of those questions are going to be relevant to every single investment. The more sophisticated the transaction clearly the more potential angles to investigate it on, but the systems come pre-armed with a battery of these 1,900. And I expect this number to increase over time because as Andrew mentioned already, a lot of these systems are adaptive and they will essentially be learning from us as we go through and do more deals and learning how to refine and as well as include new angles of risk as well. So when these 33 agents go to work, they're carrying out enormous research efforts. They're gathering the data, they're building the analysis, they're answering those questions. They're synthesizing all the way back up into something that a human can actually see and more importantly, make judgments upon and understand. So added up, there's more than 200 pages of analysis in the bottom right corner, sometimes even more than that for any single investment. That's a huge library of information that's collected and the system also has the ability to compress it down into the most relevant points, for example, for investment committee to make decisions, to ask questions to interact with that information a lot more in 2 ways. And I want to be clear about what this is and what it isn't as a result. This is an investment analysis engine that leads up to a human decision maker, an expert who knows how to weigh the different kinds of risks through experience, through expertise, through time, through judgment. And what the AI is doing is getting better and better at producing exactly the analysis that, that person needs to make that call or to understand it, but also to engage and interact with that person to help their understanding. It's not designed and will not replace human judgment. So designing these systems isn't just a matter of working through a large checklist. I mean, what I've given you here is 8 numbers that make the system feel very comprehensive and large. But what I haven't really shown you here is how that interlinks with each other. And asking a set number, a large set of number of questions or doing a fixed number of analysis in automated ways, it resembles a template, and this is not a template. It's a deeply integrated system that draws information from right across the process. And we wanted to show you visually of what that could look like, and we're going to do that now. Right. So give me a moment to explain what you're looking at here because it can be a little bit overwhelming at the beginning. So this is a simulation that helps us visualize the entirety of this investment platform. We have 9 different pillars there that you see at the top there. Each 1 represents 1 element of risk in these chapters of the investment committee paper and what you're seeing here is the interrelationships of those and how they build up into those chapters and all the way up into the IC paper at the very, very top. And so essentially, what you're seeing here is the -- and I'll talk through a minute an example how this structure all comes together to go through documents to metrics, to analysis to chapters and then to the investment committee paper as well. So we're going to take 1 pillar, which is the builder pillar in the middle there. And so that's basically everything in red. And so what you'll see at the very bottom of that pillar will be the documents in this case, that are being extracted. The next level up is going to be the metrics that are being pulled out and then the analysis that is happening from there and then all the way up into the little kind of red square, which is the summary of that analysis there. And so what's happening here is that a little bit of animation that you see is the data flowing up the chain, I suppose, as we get more documents in a analysis, as the analysis is running in the background. And each one of these things is being executed by, in this case, one of the number of the 33 agents looking at the builder specifically. So we've got the animation back, which is lovely. So we've got the analysis steps going up to that level. And then from that level, we've got into the builder chapter as well. But that would, again, would kind of position these 9 pillars that's quite separate entities. But what I wanted to show you here is that not only are they separate entities, they are very interlinked. And in fact, a lot of the interlinking that you see here in terms of the path between them is really how the system is designed to use context and data from different elements as well and to bring it together into a single coherent picture. So in the case of a builder, for example, the information we gather about a particular builder that's undertaking a project may also be relevant to, for example, the way we think about stress testing the cash flow models. And so what you see there is the analysis chapter being linked to the builder chapter as well. So it's quite sort of, I suppose, a complex web of interrelationships. But the whole point for us to show you this is to give you a sense that these systems and this new age of AI that allows us to do analysis for these more complex systems is not simply a templated to-do list of many, many different things, but it's really an interlinking type of cognitive system almost and it's idea that every system is connected and is listening to the other parts as well. So that's the platform and the real takeaway, I suppose, from all of this is that building it takes genuine care in the detail of how every single one of these points works with each other and within itself and the real discipline in how all of that work builds up to a single point of decision and the investment committee paper that represents faithfully our investment process. So that brings my section to a close. I'll now hand over to the Chief Financial Group Officer, Philip Dowman.

Thank you, Michael, and good morning. Let me now translate what you have just seen into financial terms and explain what these AI capabilities mean for long-term shareholder value creation. We believe Qualitas is one of the first Australian private credit managers to develop a proprietary data trained AI underwriting support platform with broader implementation planned through FY '27. As you've seen, we are systematically building AI-driven operating leverage into the Qualitas platform. This is powered by two reinforcing drivers. Firstly, faster AI accelerated investment decision-making and secondly, the scalable application of intelligent automation across the broader Qualitas platform. Together, these translate directly into structurally improved margins as we scale FUM faster relative to increases in our cost base. By building a proprietary AI platform with a dedicated in-house AI team rather than licensing generic off-the-shelf solutions, we are achieving a faster time to benefit realization. Over the long term, we expect these AI initiatives to enhance our funds management margins and support sustained compounding earnings growth. At our 2023 Investor Day, we set then a long-term funds management EBITDA margin target for the Australian business of over 50%. We have now achieved that target in FY '24 and FY '25. Supported by our AI transformation and broader operational efficiency initiatives, we are today upgrading that long-term target. We are now targeting a funds management EBITDA margin for the Australian business of over 60%. Our business is not only growing, it is also becoming more scalable and as a result, more profitable. That concludes the formal part of our presentation. Please feel free to submit any questions you have in the Q&A portal. Thank you.

Thanks, Philip. The first question has come in. How does Qualitas ensure the accuracy and reliability of AI-generated assessments?

Michael, over to you on that.

I love that question. That's a great question. Thank you. And probably one of the questions that we thought about first when we really thought about the system. And the answer is through a lot of detail, a lot of detail, but I probably don't have time to go into this particular call. So when data is recalled by systems, we have a series of steps that ensures that we understand what document is coming from, where in that document is being brought in and why. And we all have a lot of context around that single data point. We then make that transparent to the investment team. And then we have a number of different auditor agents that run over that to double check and often triple check whether the number is not only correct, but also contextually the right data point that we want to use for that analysis because oftentimes, it's not necessarily about absolute correct, it's about context awareness. And so those agents that repeatedly get us to a point that we feel extremely comfortable in terms of the accuracy. So the system runs slower, and it certainly runs in a more iterative way than otherwise it would be. But we've really optimized primarily the system for accuracy and reliability over speed or other things as well. So we take a lot of care with that. Ultimately, all of these analyses and steps we make extremely transparent for the investment team because to Andrew's point previously, they build conviction in the deal and part of building that conviction is understanding the numbers and why the numbers are what they are. So that transparency in the calculation is afforded to that team so that they can build that conviction and their reliability and also the transparency underneath it. So it's a topic that we spent a lot of time thinking about, and it's a great question.

Thanks, Michael. Let me just bring up the Q&A function. Second question, how are you protecting proprietary information with these implementations?

Michael?

Fantastic. Another great question. Thank you. So all of the data that we use within models are constrained within our Claude environment. So everything is locally hosted in the same way that Microsoft's OneDrive or something else would host your files. All of the models that we call upon and that we utilize in this particular tool is contained within our Claude environment. So nothing leaves a Claude environment. Nothing is used to train models. Nothing is in any way kind of leaving, especially for our investment process or any kind of private information that we have as a course of the deal gathering as well. Again, that was probably #2 on our most important parts of our list when we first started thinking and building the system.

Thanks, Michael. Next question, the last long-term margin target was met within 1 year. How do you think about the timing of your new long-term margin target?

Look, we have said long term and for us long term is between 3 and 5 years. This is a slower burn. And therefore, long term is within that 5-year time horizon.

Next question. Which underlying models are you using for your AI system? And will you be buying or renting GPUs to run your model?

Michael.

More and more specific questions. It's really good question. So at the moment, we are using the frontier models. I won't disclose here what they are, but I mean there's only a few of them really to choose from. And what we are certainly developing at the moment is a lot of testing around new open source models, which I kind of hinted at during the presentation. That open source models are probably 1 or 2 generations behind the frontier models in terms of capability. But we're already seeing that the frontier models without even the big names of this world can do a lot of the analysis that we have created in part in the way that we have created analysis tiers as well. And what that allows us to do is probably within a year or 18 months or so, we'll be able to use a much wider assortment of models to do this kind of work, not only the premier frontier models, which are the more expensive but better models, but also a wide variety of other models as well. We don't require on-premise GPUs to run this. We rent them through the cloud. And so this is as-you-go type of arrangement at this point. And I think in terms of our calculations for the economics of that, that makes more sense than hosting them on-premise. But again, technology moves very fast. So we are keeping an eye on it.

Next question to Philip. What are the expected FY '27 implementation costs? And how much will be expensed versus capitalized?

Yes, that's a great question. Look, the bulk of the implementation cost, well, the implementation cost of the new platform is relatively modest. And we have a tendency to want to expense as much of that as we can just through our operating P&L. There may be some opportunities to pick up some R&D credits along the way, which we're certainly investigating. But at this stage, our expectations for FY '27 is that the bulk of the cost will be expensed.

Next question to Philip. What's been the P&L cost and realized benefit of the AI platform in the current financial year? Has it given a net benefit yet?

I think the first thing to really point out is that this journey on AI was well, now Michael started early in the FY '26 year. So our initial investment in AI capabilities is within our current budget and within our current guidance. In terms of benefit realization, that is really something that is not a feature of FY '26. FY '26 is more about the build.

Next question to Philip. Of the uplift to 60% EBITDA margin. There are a number of drivers, including scale, AI, deal size, how much of the margin lift is directly attributed to AI?