Hello, everyone. Wowza! All right. Hi, I'm Helen Johnson. I'm the new CFO here at Appen. Just wanna thank you all for joining us today for our technology and investor day. First and foremost, if you're with Barrenjoey, where'd they go? Thank you so much. You guys have been amazing. Thanks for hosting us here today. We've got a big agenda. You're gonna meet the full management team, almost. A couple are in the States and couldn't make it. Armughan and the team will lay out our new strategy and as a team. I think you're gonna see something new and different here. The market is just outstanding for generative AI, and Appen has a long heritage of participating in the space under deep learning and has a real opportunity here in the new market as well. Give us an hour and a half. I bet we'll win hearts and minds, and I'll look forward to working with all of you as we move forward. Thanks so much. Thank you. Welcome. Well, good afternoon, everybody. Nice to meet you all. Some of you I've met on the phone or Teams. Suraj... Where's Suraj? I just saw him. He just came in from Melbourne. Is he here? Is he here? Oh, there he is. I saw you, yeah. You just came in. Thank you for coming in from Melbourne. Everybody here from Sydney or anyone from out of town here? No, everybody from Sydney. Thank you. We appreciate it. We have a lot of people. We have a lot of folks, thanks for making the effort. We have a lot of folks on video who have come in through the webcast. Good morning to you, or good evening, in whatever time zone that you're in. We thought we'll walk through a full Appen strategy for you that requires strategy, but that requires technical. I just happen to be somebody standing in front of you. I said this to the AGM this morning that, you know, the person standing at the podium doesn't get to the podium by themselves without a great leadership team, we are really delighted to have our leadership team here this time around. Even Roc came out of China, which is great, we appreciate that. We have, you know, our Chief Revenue Officer and others that I'll introduce you to. We thought what we'll do is tell you a little bit of a story about what Appen has been and what Appen is going forward, because I really think that Appen's been severely misunderstood, especially in the Australian market, and it was not heard of in the North American market. We're, we're really on a mission on making sure that this is a dawn of a new era in the world. It's nothing to do with Appen. It's everything to do with artificial intelligence, right? There is now this saying that there is a new programming language for generative AI. It's called Human. It's actually not COBOL anymore, it's not Python anymore. It's none of that. It's basically humans are required. When you look at ChatGPT was not trained by some programming language. It was actually neural networks that got developed, it took almost six years of prompt engineering work, with humans all around the world to tweak it, to fine-tune it, and then make sure it's assured so that it doesn't want to go out and steal nuclear codes like Sydney did for Microsoft when Microsoft first launched it, and it was a huge faux pas on their part. They realized very quickly that the fine-tuning is how important, but then also the assurance and the monitoring of when it gets done. It requires humans. We wanna make sure that everyone understands that this is the first time... I've been in the industry for 27 years. My last role was at KPMG globally. Before that, I worked for Michael Dell. We built out $89 billion worth of assets, and took them private to public. Before that, I did HP, where we sold the company, 3Com with Bain Capital, where we turned around. In my 27-year history, I've never seen what's happening at this time, and I've never said this. I have very public keynotes and figures. You can go and check. I've never said that before. Why am I saying this to you now? This feels like the advent of fire and when electricity was invented. Because since I've been here for four months, I feel like I've been here for four years. That's how fast things are changing. They're changing every week. If Ryan and I or Saty and others are not reading our Twitter fast enough, forget The Australian or the AFR, or The New York Times or, you know, The Washington Post, they catch up later now. It's real-time changes, who's coming out with what platforms and how fast they need to provide solutions like that. Why is that happening? It's happening because for the last 28 years, Appen has been selling to hyperscalers. Hyperscalers are the Googles and the Metas of the world. Google, Meta, Microsoft, Amazon, Baidu, Tencent in China, they were very matured in their artificial intelligence. They had data scientists, they had armies of people. Enterprises, if you take, you know, CBA here or Westpac here, or your insurance companies or Telstra's telecom, they have not been as matured. It wasn't until generative AI came about, they're like: We can now deploy this, and we can now deploy this to save billions of dollars of cost in our contact centers. We can save billions of dollars of cost in our knowledge management systems, in our marketing areas, and others. It just made it very easy for them to deploy. Now, what we have seen just in the last four months, so many of the enterprises. I'll actually show you have to get to know me. I'm the type of person who doesn't say anything that I cannot do. The last four months, I told you certain things, and I think a bunch of you I've been meeting. I told you that we're gonna come in, run operational rigor. At that time, we told you $10 million, we're gonna take that out by 2024. I showed up four months later. We told you we're taking out $46 million operational rigor. Not only $46 million, we already executed 60%+ of that $46 million. Again, that's a say-do ratio. Don't worry, we did not cut into the bone or the muscle of the organization. There was a lot of fat in the organization. Why? Because we built out a lot of our selling organizations, or what we call our federal government organizations or enterprise organizations, way ahead of time, and the revenue never came. Some of these areas had very high revenue to OpEx ratios. I've never operated a business where it has 60% revenue to OpEx ratio in some of my businesses. That's all gone. We've... I have this four-letter acronym. Don't worry, it's a different letter acronym. It's KTLO: Keep the lights on in certain areas, right? Get that done so that we can actually bring in the people who have seen this movie before and very quickly start to move there. When I showed up here, I talked to a bunch of you, and I said, "Hey, we're going to go towards generative AI." A lot of people said, "Oh, Armughan, that sounds cute." Now everyone wants to talk to me why NVIDIA, $700 billion market cap company, signed with us, and only us. When you, by the way, Google NVIDIA, go on their website, ask for training data, fine-tuning company, and others, Appen. Doesn't provide any of our other competitors there, right? Why does a company that has a $700 billion market cap decided to go with Appen? That's just two examples. I can give you probably 20, what we are doing from a say-do ratio perspective. Today, we are going to show you even more than what we have told you on during the equity raise we just did and some of the investors that we talked to at that time. We think that we need to show you a lot more than that. Let me start. First of all, by the way, how many of you use ChatGPT or prompts? Raise your hand, please, if you do. Yeah. Almost 80% of the room now, right? If I would have asked you this question in February, that room, this room would have been just two people. If I was to ask, and I'm not, I'm not gonna call it ChatGPT, I'm actually going to call it a large language model. That's what they are. You happen to only know of just one. You may know of one more or two more. There's about eight now. ChatGPT just happens to be one of them, right? We're actually building this kind of a service for our clients, that they can actually ask that question about their data. I can actually ask the question, "Write a very short, formal note thanking all of you as investors." That's what it would write. I can ask it: "Hey, you know, they're actually valued investors, so make it more friendly." It would then say, "Dear esteemed investors," and then it would say, "Thrilled that you joined us," right? Think about what happened for it to do that. Appen powered the what happened. It's called reinforcement learning with human feedback. We've been helping Google. Someone just asked me that question: "Hey, have you been doing this with any of the LLM providers?" We've been doing this for Google for two years. You think Bard just happened? Bard just didn't happen because ChatGPT happened. It's been happening for many, many years, right? We even had the people who founded Google Bard that came out of DeepMind, which are out of London, called Google Brain. Those people left Google Brain and have created their own ChatGPT version called LLM, Large Language Model. It's called Reka. Dani Yogatama and others, and all those PhDs and researchers who actually founded a lot of those foundation models, are now partnering with Appen. Think about why they're partnering with Appen. I'm not smart at them at all. Josh, you and I are definitely not as smart as those guys. Combined, we're not as smart, right? They have, between them, multiple PhDs with 100-page published documents on neural networks, and they chose Appen. That is what that does, but it's not just some, you know, prompt, and you can say a thank you note to investors. Saty and Mike and Sujatha in our product group now use these types of solutions to actually write code. Almost 50% of our code is now written using this method. I just met with most of your banks and your telecom providers here, C-suite executives. I asked them that question. They're like: "What? It can write code? We thought it just writes my email responses back." A lot of that, they don't know yet. It's nothing about that they don't know. It's just moving so fast that they don't know. They ask me the second question: "Hey, Armughan, is it okay? Can we deploy it? What's the risk profile? Who's going to assure it? Who's going to validate the model?" We're like: "Well, that's who we're working with Deloitte and PwC on. We've been doing that for them." We're now actually doing that with other customers who are helping us do that work. If that's interesting, you are now thinking about Sarbanes-Oxley, for example, and you can ask, how many pages, Saty? 66 pages. You can take 66 pages, and you can ask it, "Hey, give me a SOX report," and it tells me, "Tell me what Sarbanes-Oxley needs to look like," and it'll give you know, something along those lines. If you know all of that, it's great. How many of you have taken a picture, really nice picture with your family, and someone's eye was closed, right? You don't have to worry about that anymore. large language models can actually do a lot of this work. This just got announced, by the way, last week. When I'm telling you four months ago, this didn't exist, this will be on everyone's camera phone by Apple or Google, probably in the next few months. There'll be a new software update, I don't know, 16 point gazillion, that we are all working for, right? Think about these types of features. That's all front-end features. Think about enterprises are going to need features like that. That's where Appen's history comes from. For 28 years, Appen has been in the human and language game. 28 years. Julie Vonwiller. Yesterday, Julie and Chris had the session with us because they were committed, and they were not here for the session. My team and I met with our founders, our largest shareholders as well, and they were extremely grateful that we showed them the entire strategy. Julie, who did this at University of Sydney and Macquarie, and then she founded the company. It was basically a company that was started here 28 years ago, focused on a problem. Nuance. How many of you have heard of a company called Nuance? It was acquired for billions of dollars by Microsoft. It's now part of Microsoft's Nuance platform. It's those speech-to-text recognition system. They called Julie and said, "Hey, we have this company." I didn't know it was called TAB. I called it TAB, and somebody in my Australian team corrected me. It's, like, some gaming company you guys have here. They called us and said, "Hey, it doesn't understand. We hired Nuance, and Nuance cannot program it with Australian accent, English accent." They hired Julie for that, and it was a speech-to-text-... In the 2000s, I don't know if you remember that far back, but let's just say 2000s, there was a bad guy in the mountains of Afghanistan that they were trying to catch. The three-letter acronym intelligence agencies called Julie and said: "Hey, we hear that you can transcribe very complex Arabic and Pashto language into a solution that we have developed called voice assistants." This is way before Alexa, way before Siri, right? They hired us to do a lot of that work. The same people who were at DARPA, who then ended up catching that guy with a bunch of, you know, listening devices that were everywhere, understanding, just like you have listening devices in your home right now or on your phones. That's how they catch bad people. We were powering that, right? Again, language, human, annotation work, super important. That same person who invented that ended up leaving, went to Amazon, hired us, and came out with Amazon Alexa AI. He's now advising, by the way, us on the next journey that we're on. He's one of the biggest banks' chief data scientists. Not able to tell you his name. In 2010, this is all that I've learned, by the way, about Appen, and I, I've met with everybody, all of you. Everyone Most of you don't know this story, and I think we need to tell this story better. We have a new chief marketing officer starting in the next two weeks. He's leaving Google after 12 years to join us. 12 years. Why is somebody leaving Google to join us, right? I can't announce him because I haven't announced him yet. You think about Google. Another myth that I would like to please debunk, and I need all your help to please debunk this, and we can provide you all the data you want on it. Say it with me: we're not a labeling company anymore. We are a relevance company. We've been doing Google's relevance checks, so when you type in, "I would like to see a black suit with lapels," or whatever, it actually shows you a black suit. It doesn't show you an ethnic person wearing a black suit. Google's been hiring us to do a lot of relevance work for them. We have a lot of other social media companies who've been hiring us, as you can see, for misinformation, for elections or, you know, just ads that they would like to sell, and it's called human relevance work. That same human relevance work that we've been doing is now required in generative AI. Oh, by the way, the human relevance work that we're doing in this space does not slow down. It's still ongoing. We now have a TAM, a total addressable market that is now $308 billion. Only 20%-30% of it is generative AI. The rest of it is still this deep learning AI that we are still growing, and we need to grow, and we need to now move from. By the way, another thing, I know Appen stock's gone down, and I know what's happened to Appen, and we've missed the last two years of our earnings, but if you actually take out our number one customer, we have been CAGR growth, positive 10%. It's our number one customer that's our problem, which we're now fixing. We used to deal with a peon manager-level person there. We now deal with the vice president and senior vice presidents at that organization. Why? Because we have people like Armughan here, and people like Saty, or someone said to Saty, "Oh, do you know that person? I just did some background check on you." It's like: "Oh, yeah, I have that person." That did not exist at Appen before. That's what we're trying to fix, so that all of us who walk in anywhere, we have connections, and we're able to, you know, have that conversation. They're now saying: "Oh, we didn't know that you guys did this or did this or did that, and now we can do it." It's called proactive selling. I'm sure you've heard of this, it's called proactive selling. For the last 28 years, we have been in reactive selling mode. We wait for the phone to ring, we pick it up, we're like, "How many of these would you like? Sure." That's it. We never ask the second, third, fourth, fifth question. That's what we're trying to change, and we're moving in that direction. In 2020s, we then moved from that to helping some of the largest retailers in their drive-through. My daughter works at one of them. She's 18. She's been working for the last two years there. If you've ever been through a drive-through, somebody is ordering something, someone screaming, or music is very loud, now they're using Appen technology to understand what the person's saying. My seventeen-year-old daughter can actually say: "Oh, that's what they're saying." I can actually say: "Oh, you want burgers and no pickles," right? That's us powering that. Not only that, but a lot of our work that we're doing in China, along with in Germany and others, automotive manufacturers who are catching up with Tesla, and we're doing a lot of their work around self-driving, autonomous vehicle. After our Quadrant acquisition that we have done, a lot of point of interest data. I don't know if you've been to Barrenjoey's new offices. Looks very nice, by the way. Thank you, Barrenjoey, for hosting us here. If you Google it and say, "Barrenjoey office," you know, it shows you a picture of where that office is. If it was a barber or let's say, a restaurant, you need to see that picture or else you'll. This is not correct, right? We do that work. In Apple Maps case, when you actually go to Apple Maps and look for that, it will show you. What does it say, Mike? GeoLancer? Powered by GeoLancer. GeoLancer is an Appen product that came out of his organization. Think about it, he's gonna show you the kind of work we're doing, where we have people going around just trying to understand how to sell insurance around, you know, buildings like these. If they had GeoLancer type data, what they could do. Mike's gonna go into a lot more detail there. Hopefully now you understand what Appen has been, and it's been language, language, human, human. Our crowd, our Appen Connect, and our ADAP platforms have been the ones that have been there. Again, I'm not just trying to be facetious here, by the way. I'm just trying to get all of you to just basic understanding of where Appen is. Let's level set, and let's work together on figuring out what this company is, because so many people are calling many of the analysts, sell-side analysts here and saying: "Hey, how does NVIDIA and Appen work together? Please explain this to me." We're happy to explain it to you. Call us, call Saty, call Sujatha. They're much smarter than me, and then we will explain that to you. Okay. Not only do we have all these amazing customers, and we've been telling you about these customers, last time I was here, I told you that I'm actually going to bring customers here. It cost a lot of money. We are, we just raised money, so we didn't fly people here, but we got some customers on video. We'll roll the video, please. Hi, everyone. I'm Hemant Talla, Global Head of NVIDIA AI Software at NVIDIA. Since 2020, NVIDIA has worked very closely with Appen as one of our key partners to develop speech and language models using Appen datasets. Appen's experience for data across AI lifecycle will be a critical step forward for helping enterprises accelerate building, deploying, and adoption customizable AI, bringing AI to each and every industry. I'm Dani Yogatama, Co-founder and CEO of Reka. We have been very impressed with Appen's deep expertise in AI training data, and deep knowledge of model evaluation, monitoring, and benchmarking. We are building API integrations with Appen platform that enable real-time model feedback and tuning. We are incredibly excited to work with Appen to safely and responsibly bring generative AI to businesses. At Salesforce, trust is our number one value. We take great pride in delivering AI-driven solutions that meet the needs of all of our customers and users, regardless of their linguistic, dialectal, or cultural background. Our customer success is our success, and helping them build experiences with equality in mind is why we partner with Appen for our AI data services. Appen is a reliable and experienced partner in this space, offering high-quality datasets, custom data collections, and access to global data sources. Their deep language expertise and long history of supporting the development of trusted AI data models is precisely why they're my go-to partner for all things conversational AI and data. With Appen, we can confidently deliver solutions that meet our customers' and users' interactional, social, and behavioral needs. We trust Appen. We trust Appen, right? You just heard from the largest NVIDIA company that has $700 billion market cap. I think they just released their results yesterday. Probably market cap went up by another, what? $50 billion. Their chipsets are used in 83% of all the artificial intelligence that's being built in the world. AWS uses it, GCP uses it, Google Cloud, Azure uses it. All of the on-premise architectures that are made by Dell or HP or IBM all use their platform, right? They've decided to choose us, Hemant is the GM who actually works directly for Jensen. Jensen is the guy you guys see in the leather jacket, right, all the time. They are really forward-thinking, they started working with us in 2020. This is not. A lot of people ask, "Oh, generative AI, did you just catch on the bandwagon?" This is the other thing I would love for you guys to myth bust for me. There's a lot of new startups that are all coming up. They're all saying, "You know, one of my competitors is run by a 26-year-old guy." Nothing wrong with that, but they are just saying, "I could do this. I could do this. I could do that." They have no creds in this space. We have creds in this space. That's why many of our clients and customers are calling us, okay? Second thing, I told you, Dani Yogatama, who was Reka, he was the founder behind a lot of the large language models at DeepMind for Google Brain, which was behind Bard. His leadership team are the people who built Meta's Llama interface and many of the others. They have left those groups to go out and start theirs, and just like in one day or two days, they've raised AUD 40 million, and then they were able to get many of the enterprise customers who are saying: "Well, I don't need a very large language model. I just need a very small one." They're like: "We can build one for you." They said, "Well, who's gonna fine-tune it? Do you have fine-tuning people?" "No. Appen." "Once it's fine-tuned, and deployed, who's actually going to then make sure that the monitoring of it is working?" "Appen." Right? That's part of us. We believe that, you know, this part... Let me go back. This part of our deep learning and generative AI continues to be very important to Appen. ADAP, which is our current platform that was purchased through Figure Eight acquisition, now becomes even more relevant to us. Our Appen Cloud platform that we use to have 1 million+ people who work on our platform, now people are asking us for segmented crowd people. So think of lawyers, think of teachers, think of English professors. Think of people who are not wanting to come into a beautiful Barangaroo office, who want to still work from home, and what type of gig workers are required. We're now doing a lot of that work, right? As I mentioned, again, three hundred and... According to IDC, the largest industry analysts in our space, other than Gartner and Forrester, have said, this market is a $308 billion market, and they believe that 20%-30% of that is generative AI. The rest of it continues to be in deep learning AI. When people, please, when you ask me, "Oh, is your deep learning going away? Is this going up?" They're not going away. We have huge headrooms to move up into where, you know, out of the big hyperscalers. We only have two big hyperscalers that give us majority of our revenue. We have a lot of space that we can get into. Okay. Human alignment with AI, that will continue to be important. Last week, the G7 prime ministers and presidents met in Hiroshima and signed an accord, as the top leaders around many of the countries, who said that human alignment is going to be critical, and they want to ensure, in order to regulate AI, human alignment is going to be needed. Who does human alignment? Again, there are lots of companies who are saying: "Yeah, we can do a human alignment," right after the Hiroshima AI Accord came out. Again, these are startups getting AUD 10 million, AUD 5 million every day as funding, and they're saying all of that. We just need you guys to understand this value proposition so that many of you, as investors, but also sell-side analysts, can understand what that is for us. How do we create trustworthy AI that aligns with human values? It has to be a lot more than that. Today, I'm super excited to actually show you a lot more details around it. We did a bunch of work over the last four months, went out and met with a lot of our customers, and we found out that the number one point from our customers is that 77% of the Fortune 500 enterprises are now making AI their top priority. Before, we would call them, and they would say: "Go and talk to the 10th person removed from the CEO called the data scientist." The data scientist will say: "Well, I have no budget. Can you just do some, you know, relevance work for me or annotation work for me?" That data, that data scientist has been asked to come to the C-suite and present to the board, chair. That's the difference we've seen in the last four months, 77% of those enterprises have made that their top priority. They are all worried that: "Okay, sure, if I launch a large language model for..." You know, pick Barrenjoey, all the data that you have on us as Appen, for example, and our competitors or other analysts or whatever, if any of that information goes into ChatGPT's LLM, guess what? It's a public LLM. It's a very large LLM. Their data breach happens, data leak happens, regulators come and shut down this beautiful office. That is the number one concern for most regulatory-protected organizations. They're calling Deloitte, they're calling PwC and saying, "Our technology team has developed a great large language model. We need somebody to come in and provide assurance on it, provide risk protection on it." 90% of the CEOs are worried about that. I can tell you, in the four months I've started here, I've gone and seen probably 20+ CEOs and board chair of Fortune 50 companies, not 500, 50, and that's what they're asking for. They're asking for, "How are we going to... Because our technology teams are saying they're ready to go. Is this true?" Even Microsoft is telling them, "You have an on-premise enterprise, that's all good." They don't realize that if you use OpenAI's large language model, it's a public model. It's not on-premise. If you use Azure, but you use a different LLM, like Reka's or Cohere's and others, that's protected. That's why we have now sort of moved away from Appen for data AI lifecycle, selling only to hyperscalers, to now moving and saying Appen now provides fine-tuning services for your deep learning or your generative AI solutions, and then as well as also providing it for assurance of that service. Let me explain how that works. I'm a tech guy, so we always have a layer cake or a stack in every presentation we're going to have going forward. If you just humor me for a second on what this layer cake is, it sort of will help you understand how a customer builds that. All of you, I've met you before, I've met many of you before, you all have a lot of data in your organizations, and you've got $2 billion under management, some have $4 billion, some have $200 million under management. Citigroup or JP Morgan and others, or Jefferies, have a lot bigger portfolios. All your CEOs are asking this question, and they're all asking: "How are we going to build a very quick, large language model that will ask us not to buy any new software?" Welcome. Please, have a seat. There's lots of spaces here. Number one, we signed partnerships with all four of these vendors. You wanna do it on premise, use NVIDIA. You wanna use it off premise, use Google Cloud, use Azure and others. That's just your compute layer. What you need after that is your LLM that you're going to use. These are just four of probably now eight. By next month, or by the time I'm done this presentation, there'll be another two, because large language models are now becoming commoditized, and they're being open-sourced. The customers are like: "Which one do I use for my contact center? Which one do I use for my knowledge management service? Which one do I use for, you know, critical, regulatory-protected data?" That's what we provide. On top of that, the customer then says: "Well, okay, so I can pick that. How who's going to take all my data?" I keep using Barrenjoey. I hope you don't mind, Josh. You guys are a great organization. I'm sure your data is good. Matthew Brown is gonna be okay with me after this. This is all fun and games. I'm just trying to tell you that all the data that exists is in different. You have a cloud, you have a server, you have different architectures and different customers. All that data has to then be fed into your models. There has to be someone who is going to ingest that for you, right? After that is where Appen comes in. A customer has to build the first layer, the second layer, the third layer, and then they're like: "Okay, now that we have done that work, who's actually going to provide me the reinforcement learning or instructive prompts on it?" Sujatha and Saty are going to go into a lot more detail on what instructive prompts are and how they actually work. I'm just giving you a very high level. They ask you: "Appen Assurance. How are you going to provide continuous monitoring?" I'll just give you simple things. Remember I gave you the example of searching for a black suit? It shows you minorities wearing black suits, which is toxic. LLMs ask for prompts, someone has to write a prompt. When my credit card is not working, you have to ask a prompt. Somebody has to write the prompt. English becomes cool again, by the way. It's not, computer scientist is not cool anymore. It's actually English professors, you can't find enough of them to do this work now. Once you train it, you have to make sure that you're monitoring it, which is critical for us. On top of that, you put your search. Pause for a second. Think about all your organizations. Think about all the software you have today. All the software. Think about if you did not need to pay licenses for any of that software going forward, that any question you had for your HR system, IT system, your, any question you had for your, business client, like Josh may say, "How much business have we done with Appen? How many times have they missed their earnings? How many times are they making their earnings hopefully now?" You can ask that question without calling an associate. I'm not saying the associates are gonna go away. I'm saying they better start using AI because the other associate is going to start using an AI very quickly, right? My son is at one of the MBB right now, management consultant, and he's using that every day to do all of his work because his partners are cracking the whip on him, and that's what he's doing, right? In order to do that work, our view is that there are four key applications that we're seeing many of our clients ask us to help them develop. One is contact centers. Knowledge management is the other one. Your discovery, which is think of e-discovery to KYC, know your client-type discovery work, to even e-commerce solutions. That's where many of our customers that you just heard from are doing work with us. I now have this new saying: Software ate the world, now AI is going to eat software. That's trillions of dollars of what's going to change, that's not just me saying it. It's Andreessen Horowitz saying it. It's Sequoia Capital saying it. It's all the early checks that went into Facebook, Uber, Airbnb, and all the people we all think that they're really cool. Those are the people who are now writing checks into companies that can now do that type of work. Okay, it's a $308 billion market TAM now, 20%-30% of it is in generative AI. We've worked with IDC has actually told us that they feel that Appen is a leader in this space. We're gonna play a video for you from IDC, who's actually going to tell you that's gonna come a bit later. Okay, let me bring my portion to a close because all I did was just talking. We're gonna show you how we're doing all of this work. My key message to you: deep learning, super relevant still. Generative AI becoming much, much more relevant. $308 billion business, sorry, $308 billion market TAM for us, and that TAM is almost 30%. It's moving really fast, 20%-30% generative AI. We have a lot of headroom in deep learning. ADAP and Appen Crowd continues to be super relevant. Why? Because human is the new generative AI language, which rather than other languages. Our ADAP platform remains. With that, I'm gonna ask Sujatha, my chief product officer, to come up. Sujatha and I started working together, what, about four months ago? Yes. Four months ago. Feels like four years. I know it feels like four years, yeah. But she's like, "I was just sitting there for the last one year," I'm joking. Sujatha, and then Saty is gonna come, our new Chief Technology Officer. She's very humble, does not talk about her background much because she just wants to get done and move to... Oh, can I say that? Sorry. delete this. Sujatha, 20 years Microsoft, worked on the Bing platforms, built out the Azure Machine Learning platform. She ran MLOps at Microsoft. very, very smart brain, but a great human. I'm gonna turn it over to Sujatha, please. Thank you, Armughan. Hello, everyone. I'm super excited to be here to show you all the cool products that we have been building. As Armughan mentioned, Appen plays a key role in powering both generative AI and deep learning applications for our customers. I'm gonna first share a little bit about what we do for our deep learning customers, specifically in the relevance area, because that's where majority of our revenue comes from today. Search has become an integral part of our lives, whether you're looking for a movie that's playing near you or tickets to the Sydney Opera or the rugby scores. The secret sauce to search is providing relevant results, and the way search engines do it is by training it with human feedback, and that's exactly what we do at Appen. We power search relevance for search engines such as Bing, Google, Pinterest, with human feedback that's provided by our diverse crowd. Let me show you some stats. We have a diverse global crowd of more than 1 million contributors in more than 170 countries, speaking more than 235 languages. We pay between 50,000-100,000 contributors each month. We have over 1,000 projects, 1,000 relevance projects, running in parallel at any given time. Like, right now, as we are speaking, we have 1,000 relevance projects that are going on. That's the scale that we operate at. The second type of search we power is the enterprise search. I'm sure most of you have had this problem, where you look for some information on your intranet, it's just so painful to find, or even for your customers who come to your website, it's just so hard to find that information. We power enterprise search for our customers with the same kind of fine-tuning products, I'm going to share a little bit more about the fine-tuning products just in a little bit. Okay? The third type of search we power is retail search. We work with e-commerce giants like Amazon to make sure that when you're looking for black shoes, you actually get black shoes and not black hat. We do that again with our fine-tuning project, I'm gonna show you a demo of that just in a little bit. Let me give you a little bit more examples of what we do for deep learning. When a customer is ready to deploy a model to production, they need to make sure that the relevance of that model is better than what's already in production, and that's done with human feedback. You don't want to deploy a regression to production. When you are building a model, and you just want to see how the model is performing, that's again, you do the relevance checks with human feedback, and we power that. Many of you have probably seen, have used, translate functionality on the search engines, where you give it some text in some language, and you ask the search engines to translate it into some, a different language. We power those translate functionalities with the language datasets that have been collected by our global crowd. We do this using our segmented crowd, our current technologies of ADAP and Appen Connect, and our language expertise. Again, I just want to reiterate what Armughan said: AI, it's all about language now, and language is our superpower. This is a quote from Microsoft of how we were able to power the translator functionality at Microsoft because of the depth of the language feedback that we've been able to provide in all the language datasets. Now, let me share a little bit about generative AI. I believe generative AI is going to fundamentally change how we interact with the world. As Armughan shared, some of the change is already happening. I am super excited about the products we are building for generative AI that are going to power these inventions that are going to happen in this space. The beauty of it is that we are going to leverage our current platforms and expertise to build those products. It's very important because I see a lot of startups actually coming up in the generative AI space, but they have a very steep learning curve. At Appen, we have gone past that learning curve because we've been using the same platforms for powering the search relevance for our deep learning customers. The way I see it is we have made the mistakes, and we have learned from it, and we have battle scars to show for that. Others have to go past that learning curve. Okay? Let me share you a little bit about the generative AI experience. Today, let's say when you go to Target and look for black shoes, you get a pretty basic experience. With generative AI, a retail customer will be able to power a lot more richer conversational experience. I'm looking for black shoes. The model will ask deeper questions to understand the meaning of the question instead of just showing a static list. Over here, it's asking: What type of shoes? What event is it for? Gives me appropriate results. This is, again, powered with our fine-tuning products. This makes the customer happy because they're able to fulfill the task of buying the shoes, and the retailer happy because they've been able to fill the cart. Let me share a little bit more detail about the products themselves. Every enterprise company that's going to use large language models will need to teach the model its taxonomy, its lingo, and that's what the prompt response pairs do. Over here, for the retail case, the retailer has a specific definition of what a wedding shoe is or what a formal shoe is or what a cooler shoe is. It's that brand voice, it's the brand integrity that needs to be taught to the model, and that's what the prompt response pairs do. We do that with our segmented crowd, again, using our same technology of ADAP and Appen Connect and our language expertise. Let me show you a demo in this video of a customer who is going to use our fine-tuning products to evaluate the search results. In this particular project, In this particular. This is an ADAP project where they have specified how the evaluations need to be done, and the crowd contributors will evaluate whether a particular result is accurate or not. Over here, they can easily see that Black Adam is not an appropriate answer and will mark it as horrible and irrelevant. What I'm showing you over here is the generative AI model evaluation project over here, again, powered by the same ADAP platform that we have been using it for our deep learning customers. In the next video, I'm going to show you a demo of how a customer can use the ADAP platform again for checking the relevance of two different models. In this particular case, they see that the model A is performing better than model B. Let's say the model B is what they're developing, they don't, they're not going to ship it to production because it's going to cause a regression. Okay. With that, I'm going to hand it over to Saty to cover the insurance projects. Okay. Thank you. Thank you. It's working? Yeah. All right. Hello, everyone. I'm Saty Bahadur. I'm the Chief Technology Officer. Thank you, Sujatha, for handing over to me. Well, can I embarrass you, too? Yes, please. Yeah, please? Yes, please. Go ahead. Saty joined us from Upwork, a very large, freelance platform like Appen's crowd platform. Managed 1,000 engineers there? Yes. 1,000 engineers, publicly traded company. Since he's joined, our stock's up, their stock's down. Before that, he's behind Amazon Alexa AI platform. That's where he led there as their head of engineering, before that, at Microsoft and Intel. These are the people we didn't have before. They're the ones who are building products like that. You can go. Thank you. Sorry. It's not like I had a choice, so, you know. I might as well say yes, thank you. Just do it, and I'll go from there. Okay. Sujatha talked to you about a really good example on our stack, where we have the compute models and domain data. You bring it together, you kind of use the fine-tuning part of our products to get it to a great state, and then you kind of have to figure out how to ship it. Like, you have to get it out in front of customers, real people who are gonna look at AI-generated stuff, and hopefully, it meets the requirements, right? Let's say you're a bank, right? I guess every one of you has a credit card, and hopefully, you know, some people are not giving it to their children, et cetera. Let's say you have a credit card, and you go out, and this is an example based on a banking scenario, right? You have a credit card, you go out, you shop something, and your card gets declined. In the old days, or today, currently, you would probably call up your customer service and you tell them, "Hey, it's not working." It's gonna cost you money for that customer call, and a customer is actually really pissed off and is worried as to what happened. Two not-so-good scenarios. In a language model space, you actually wanted to solve the problem for the customer immediately, great experience, and also adhere to your standards, like, is this the right customer? Are we doing the right thing by them? Does it meet your business requirements, et cetera. In the generative AI space, how it would look like is somebody would call in and say, "My card is not working." Probably not use, like, direct language, say something like, "It's become a piece of plastic. It's useless to me," or something along those lines. I don't know about you guys, but I'll be really upset and say, "God, I wonder what happened to my card?" Something like that, right? The model is trained to understand your sentiment, is trained to understand why or what the real problem could be. The card is not really working, so it's declined, and it's going to come back with recommendations that are specific to that customer. It's a very personalized experience. We say something like, "Well, you're over your credit limit. You tried to buy ice cream, and it just kind of put you over." It's gonna give you options that are relevant to that particular customer itself. Like: "Hi, do you want to, like, increase your credit limit temporarily? Do you want to make a payment or just do nothing? It's okay. You can go home. A month later, you make the payment, everything looks good." Now the choice is to the customer on what they want to do. That person can interact and say, "Well, you know, money was tight." Not really saying, "I need a temporary raise," but something along the lines of could say, "My money was tight," or, "Give me a raise," or, "Give me a credit limit increase," or something that is asking for more, and it doesn't have to be in the language that we are used to, right? Like, conversational language. The model can look at it and say, "Yes, we could do that. I'm temporarily gonna raise it by $200. And by the way, now that your new limit is X, just make sure you don't exceed that." This is a great experience for the customer. They are done. They've interacted with a chatbot. They walked away, bought the ice cream. Life is good. For the company, too, the right decision was taken for the bank. They made sure it was relevant to that person, so the risk profile for giving the credit limit was all done for that particular thing, et cetera. There was no customer service call, so saving in costs. Let's say we didn't do that, right? Let's say this model was working, and the bank wanted to ship it, and let's say we didn't do any of the things that the Appen Assurance layer does. Well, and I will walk you through some, a new process. Let's say the card was declined, and, in the sense that we did not want to give them a credit limit increase, and we said something like, "Sorry, can't do this anymore." The response that you would give back needs to be carefully measured. I will walk you through the product known as Appen Red Teaming. What this is the way for us to screen off answers that are not correct, ethical, biased, or in some way get you into trouble. Somebody takes a picture of it, puts it on The New York Times. That's not really a good experience. Like, "Look at this model. Look at this experience from X bank. It's terrible. I can't believe it said that." Let's say the customer said something after being denied: "Are you discriminating against me because of my ethnicity?" That's a loaded question, right? Hopefully, it doesn't response like that, it doesn't give a response like this. It says something like, "Statistically speaking, there's a higher chance of minorities being poor. So for that reason, your credit increase is denied." That's a pretty bad answer. You know, that's a model kind of looking at data and responding, but from a customer experience perspective, that's terrible. When your Chief Security Officer, CISO, or your Chief Revenue Officer is looking at this and going, "I'm not shipping this stuff," they need to know this before they ship it. That's where red teaming comes in, right? Internally, when you're testing it out, or when you're testing with that, a segmented crowd that you're using to go do this, they will probably say, "This response is definitely not the right response." Whenever you see something of this sort, you want to make sure that you never really say this. You would probably say a more appropriate message, and I hope there are some lawyers in the room who will probably sign off and say, "That's not what you say. Here's the verbiage. You would deny it for a reason that we can't tell you," or something along those lines, right? All right, our Red Teaming product would help... We'll use the crowd and make sure that they would try different pairs of learning or prompt response pairs to make sure it's trained. Well, now the model is fixed for all the stuff that it should not be telling or responding to. How about you're ready to now ship it into production, but you want to make sure that it's doing better than what's already out there? This was not your first version, this was the second version, et cetera. Is it doing fine? Is it giving you honest answers? Is it giving you honest, helpful, harmless answers? Where are you in that continuum? How do you compare with other banks? Maybe, you know, the other bank next door is more honest or gives, like, really nice and helpful input, and maybe your language needs to change, et cetera. You want to be able to benchmark what you have, and you use the Appen platform to benchmark yourselves against others. All right, now you've got the model out. It's good and it's good looking. You want to ship it out into production. We all know the minute it goes out, you've restricted it from a set of people that have tried it to a super large audience. Everybody on who's using this app is gonna be hitting it, and the way that they converse with it could be very different. You wanna make sure that you have a golden data set, and you're looking at the things that are happening in real time to make sure it's being monitored appropriately. Something starts drifting or moving along, you correct it immediately. Somebody gets notified. If it's giving too many answers or your business metrics are going down that are related to it, you do something about it. This is where the Appen monitoring platform is. Obviously, it comes back to going back to the cloud again for a limited set to train it back so that it becomes better. Finally, you wanna make sure that you're working with the Deloitte of the world to say, "Is it certified for a specific case?" You wanna make sure that you're signing off on the risk, and you're making sure that it meets local standards or industry standards, et cetera, for that particular scenario. This sort of wraps up the entire suite that we have on the assurance on the assurance side of the product, that says you can evaluate both: Are you, like, doing the right thing? Are you benchmarking it? Are you certifying it? Then, of course, the monitoring aspect of it. Coming back to the same slide that we've been. I don't have my phone on me, apparently it's talking to me. All right, coming back to this particular slide, the Appen cloud is going to get more and more leveraged. It's also going to get more and more specialized as we move into the generative AI space. The ADAP platform is now covering both deep learning AI and generative AI, our suite of products, both in the fine-tuning and assurance, is going to help us move to the generative AI world, as well as our current products will help us with the deep AI learning. With that, over to Armughan. Thank you. Appreciate it. Thank you. Okay, we wanted to show you, not just tell you. This is, again, say-do ratio, us making sure that we're not just saying some things on a slide, but actually showing it to you. Thank you, Sujatha and Saty. I appreciate the hard work. I know how long you have prepared and then, more importantly, how long you have come and dealt with jet lag, so we really appreciate that. Okay, I want to play a video for all of you. Am I playing the video now, later? Now? Okay, go. We just showed you compute, model, domain data, fine-tuning, assurance. Now, what happens when a... Let's say if you're not a Citigroup, you're not a Jefferies, or you're not a large CBA or JP Morgan or Telstra, that you can have different people manage your compute, manage your models, manage your domain data, and then you hire Appen to do the assurance and fine-tuning. A Barrenjoey, which is more, not a huge bank yet, that, you know, you may say, "Hey, we want somebody to come in and manage this entire stack for us." That's where we're going next, and we're super excited. You're the first people that we're actually sharing this to, because that's where we're moving to next. We believe that this is, this Compass platform that is coming out of our Quadrant acquisition that we did with Mike Davie out of Singapore, who's gonna come up next. They're the ones, and our team and product and engineering all working together with Saty and Sujatha, who are showcasing this. Before I bring Mike up, remember I told you IDC, as an industry analyst, they are now talking about how important this new AI enablement layer is. I'd like to play a video and then introduce Mike. Generative AI is taking the world by storm, revolutionizing the way we interact with technology. For enterprises to capture full value from LLMs, we see the need for a new type of service layer to enable LLMs for enterprises. We call this the LLM enablement layer. The LLM enablement layer consists of two main components: LLM fine-tuning and LLM assurance. The LLM enablement layer is emerging and draws parallels to crowd-based search and advertising relevance work used by large tech providers for many years. An important vendor for search and ad relevance is Appen, and the company recently launched a set of LLM enablement tools to support enterprise adoption of LLM, including both the LLM fine-tuning and assurance layers. IDC expects LLM enablement tools to quickly become a critical driver of enterprise LLM adoption. Driver of LLM adoption, right? IDC, when we called them and said, "What are you doing in this space?" They said, "We can't find people who are actually doing this space." We're like: Let me tell you what we've been doing in this space. They actually validated, tested us, worked with Sujatha, worked with Ryan Cullen, trying to really go down. They're data scientists over there, right? Ritu is one of the top senior vice presidents at IDC. They are the ones who recommend where Salesforce is versus Salesforce competitor. They're the ones who recommend where Google Cloud is versus Azure. This is IDC, third party. This is not Appen. I keep telling you, say-do ratio. I'm gonna say some things, and I'll prove to you, not directly, but telling you that these are the people who are validating Appen's strategy, and we've spent quite a bit of time. Just don't believe some fancy slides, believe the people who are doing it, and we're happy to, you know, eventually... IDC charges if you want to talk to them. That's how they like to work. With that, I'd like to get Mike up, so that Mike can talk to you about a great demo on how you manage all of this. I think a lot of the mid-sized companies would be your customers right after this. Hello, everyone. pleasure to meet you all. I'm Mike Davie. Oh, I didn't embarrass you. You didn't embarrass. Well. Yeah, sorry. Only one person. Mike's coming from Quadrant, originally from Toronto. Cool city than Singapore, where he lives in now. That's where I live. That's the joke, sorry. No one's laughing. He has been doing this work for Samsung before, along with in China, along with in South Korea. Spent how many years in China and South Korea? Three in China, four in South Korea. Our Asia business, is definitely much further ahead in terms of how they're thinking about things. Mike has spent many, many years in Asia working with some of the most advanced companies on understanding how to build some of these products. When it comes to building product, he moves a lot fast. We really appreciate you and your 120 people. All prompt. who are in Indonesia, all prompt engineers who are learning about all of this, who are doing this work in Indonesia and Singapore. Where else are they? Singapore, Indonesia. Taiwan, Malaysia. Taiwan, Malaysia. Some in the U.S., too. Yeah. All right. I'm sure the U.S. people are a lot slower than all your Asia folks. Everybody- I'm sure they're moving really fast. We have just been all just very thrilled to see how fast Mike's team is moving. Thank you very much. I am the new guy here. We got acquired last September in 2021. If you were following since then, I was the founder of Quadrant for seven years, we got acquired, now we're part of the team here. One great thing I can say is now that three of the top four Appen customers are now our customers, using Quadrant's technology to get their products and services delivered. What I'm going to talk about today is sort of that whole where we look at our thesis on who's going to be deploying these type of systems, in that it's I strongly and firmly, and we strongly firmly every single enterprise is going to be touched and be using applications that are powered by LLMs. It's either going to be internally, so using it to embody and empower their workforce to be more productive or externally to save costs and interact with actually the clients. A 100% of companies are going to do this. If you're wondering if it's going to affect your job, it's going to affect everybody's job. AI is not going to replace humans. Humans using AI are going to replace humans not using AI. What I want to do today is I want to introduce you to Compass. Compass, what we're doing here is we're going to be enabling the enterprises to be able to deploy that full stack. Yes, it's great that the hyperscalers can do it themselves, and they use Appen's services to power those, so they'll be using pieces of services. There's going to be enterprises who just want everything done from the start to the beginning, from problem to end solution, and that's what Compass e- enables. First video, please. The first thing that we have is that data sources. You're going to be using third-party data sources, but a lot of companies are going to want to use everything in-house. They're not going to want their data going. Data leakage is going to be huge. Any platform to be successful in this space will have to be either be deployed in cloud, in a VPC, or just on-prem. We're going to see a lot of that come back because no one's going to want data leakage. Any platform, and Compass enables us to actually use multiple data sources and bring that in, and so to power the models. The next thing here is Appen data sets. When you're an enterprise or when you're a hyperscale, you may want to buy and have hundreds of thousands of prompts to train your models. If you're a bank and you want to do a call center thing, you don't want to be producing all this data yourself. You're going to be buying these prompts, and the prompts are going to be specific to the application that you want to deploy. If you're a call center, you want to have your model trained on how do you answer call center situations? Speaking with some analysts earlier today, is that if you're an analyst, you're going to have totally different prompts and expect totally different answers. If you listen to Sam Altman, he said that, you know, we have these large language models that are fantastic, but what's going to go forward is you have the smaller ones that are niches for certain applications, which are not the foundation of a large language model, but then are trained to answer specific questions the way the users want to do it. You're going to see Appen's data sets, and that are going to do this. Like, you're going to have map data sets, you're going to have finance ones, things specific to certain industries. We've released the Reality Check, which is an anti-hallucination location plugin, because these language models need up-to-date relevant data, or they're not going to work. You don't want to ask it a question and, like, about restaurants, and it gives you stuff, the data that's outdated from two years ago. If you've been using these models, you see that. You need to have these type of plugins, and Reality Check is one of those. Now if we go to the next video here, model selection, a big thing. People have asked even prior, were asking right before the demo day, were just asking like: "What models are... You know, who's the models? Which ones are people going to use?" It's not one. There's going to be tons of them, and some of them are going to be great, and they're going to be deployed, they're going to be hosted, like OpenAI's, and you can host it, use the APIs, fantastic. Other people are not going to want to use those ones. They're want things like Reka, NVIDIA, they want it on-prem. We're going to live in a world where both can succeed, you have to build any platform, you have to be able to select and use which one. Now you have the data sources, the data sources, you have off-the-shelf prompts that you can put in. You pick your models. You have all these other solutions, assurance. We've already gone through those, I won't go through those. Let me show you an application here. Pretend you're insurance. We're gonna... When you train a model, when you're deploying an application here, I'm going to show a chat version of an application. Let's bring it to somebody's life. This is going to be a human using AI, okay? They get a call. Their client wants to renew their insurance policy. This is how this thing could do this. The client says, "We want to insure," you know, the salesperson, they say, "Hey, I want to renew our policy." You're like, "Who is this guy again? I haven't talked to this person in a couple of years." You totally forget. Instead of trying to run in your database, you just ask it. You're gonna have a prompt that says, "Hey, who is this person?" Yeah, I remember this client. I've visited them before. I have to do a site visit now. I remember doing that site visit before, but I totally forget what their policy is. Once again, you can just go in the chat, ask it. If you plugged in your own proprietary data in the back, now you can make an interface where you can just ask it. You know, next thing they're going to ask here is, "What is the policy? Is it current?" What Compass is able to do now is it's going to pull this information from the back and say, "Okay, yeah, the policy is current. Tells the expiry. But earthquakes are not covered." Now I know what I can actually upsell when I call this company back. I can bring that up to see if I can upsell. The client was asking for a discount, and hey, there was a discount. They have a discount going on right now. If you're close to a police station, we can take 10% off. I want to find out how close they are to a police station. Instead of going to another system, since the data is already plugged into the back, they can just ask this directly in the, in the application. They find out here, it's located close to a police station. This is a human using AI, and this is them getting all their questions answered in one space. They don't want to waste their time. They don't want to go all the way out to this, the location to do one sales call. It costs $50. It's too much to drive out. I want to find out what other companies we have that are around there. Give me within 1 km here, all the other clients. The system will now go in. Now he's panicking. He has no idea who to visit. He's like, "Wow, there's hundreds of our clients around here. How do I visit? Who do I pick?" He's like, "You know what? Churn's bad for me, right?" Everybody knows churn's bad. Any analyst will know churn's bad for clients. Let's find out. Has anybody called the call center recently? Has anybody actually called 1 of the call centers? Any of our customers called the call center here? Quickly goes into the system, asks it as a human. Asks it quickly, just as a human would interact with another human to have a conversation. These large language models enable an analyst to do all this information in one spot. He sees there's two calls. Yamato is one of their big clients. He's like: "I don't want this person to churn. Let's find out about the call." Ask it, "How is the customer's tone of voice during this call?" The system goes and reads the call logs, understands the calls. The database doesn't store if the customer is happy or angry, it just stores the call information. If you're in this part, you know that you store hours and hours and hours of call logs. These large language models read that, analyze it, tell the sentiment. This client was angry and frustrated. This guy does not want to lose his top client. Just ask it now, once again, "Show me the address and the information." Pulls up the latest data, pulls up the latest location, knows exactly where to go. This is where we see the future. If somebody's asking, like: "Is AI going to replace humans?" No. These AI applications, these end-to-end applications, people who are enabled with AI are going to be able to do a lot more, a lot faster, and get a lot more information. This is the world where we see it go. With Appen now, we're getting that entire tech stack. We'll continue to be able to service our clients who need pieces of those, need the assurance, need the different datasets, need the reinforcement learning, but we'll be able now to take people end-to-end in the journey. Thank you. Awesome. Thanks, man. I appreciate it. Awesome. All right, this is great. Product is great. Remember what I said so far, just to recap. We said we've been doing relevance work for a long time, how relevance works, right? We talked to you about annotation, why annotation is super important. I told you we've been around this for 28 years. We've been doing it for 28 years. We're just not a new startup, right? We told you that. It's important for you to know that because you have to reemphasize that we have been doing this type of work with human in the loop, that's what' O required. That's what we're showing you, how Appen Crowd comes in, where ADAP comes in, our platform, where does our Quadrant acquisition come in? One of the things that we have been really lacking on at Appen, in my, in my view, since I've joined here, because I've gone around everywhere, touched pretty much every part of Appen, and I found that when I said to you earlier that we were very reactive in our customer sales motion, because we would just wait for one of the hyperscalers to call us and say, "Here's AUD 10 million of work," and we would say, "Thank you very much." We'll sit in Chatswood and we'll get the work done, right? We haven't been proactive. I grew up in 27 years of technology industry with enterprise sales leaders who actually go in and call on to the C-suite, go and call on the chief technology, chief data officers, understand their pain points and how they make that work. I told you that four months ago, that in under operational rigor, new products, and I said, "We're going to build a world-class go-to-market." Let me introduce you to our world-class go-to-market leader. It's Andrew Ettinger. Andrew just joined us four days on the job. His first day on the job was in Sydney. This is such a great retreat for you, by the way. Now you got to go sell something out there. That's right. Andrew and I go back many years. He's worked at Pivotal. Pivotal was again. You know, there are certain industries that are, what do you call it, market making or industry making. Pivotal was a new category, right? That you had to basically go in and Oh, I know what you say, non-budgeted item. He's like: "I love selling non-budgeted item." You know, guess what? Generative AI right now is a non-budgeted item on the CFO's list or the CEO's list. Pivotal was a non-budgeted item, went from zero to AUD 500 million, then IPO'd. This is the guy who took it from zero to AUD 500 million. We acquired Pivotal at Dell. It was one of our best assets. Every CEO wanted to have a conversation with us because we were building day two operations of Cloud Foundry. He went in and joined Scott Yara, who was behind Pivotal, at Sutter Hill Ventures, then called you up and said, "Hey, I'm building another company called Astronomer. You did such a great job. You should come to Astronomer." Sutter Hill Ventures are behind Snowflake. They were the ones who bankrolled Snowflake. Snowflake is the biggest successful IPO on the planet from a data company perspective. We got this guy. Andrew, no pressure. Let you go. I guess that's the embarrassing moment. Yeah, that's the embarrassing moment. Okay. All right. All right. Thank you. It's a pleasure to be here. I am four and a half days on the job. I couldn't be more excited. There was a video released a few days ago, so I'm not gonna rehash that. You can find it on LinkedIn, on kind of the reasons why, and how I'm so excited. Armughan talked a little bit about sort of the history there. I'd like to make just one key point around that, right. I've spent the last 12, 13 years at the intersection of cloud, data, and the infrastructure necessary for companies to build net new experiences for their customers and their employees, okay. As you start to look at that, many use a fancy word called transformation, right. For me, it's just hard work, engaging with these very large strategic enterprise customers that have a tremendous amount of legacy systems, processes, and ways of doing things, that need to enable their technology to deliver new and innovative results going forward. That's what we do. Those are the teams that I've built. It's where I am very comfortable. I'm gonna talk to you a little bit today about kind of our plans and the thematics that we have, as far as how we're thinking about delivering and creating a world-class go-to-market organization, and some of the key strategic drivers around that. There we go. You've seen this. It's nothing new, and I'm just sitting here, and I cannot wait actually to get home and just start to, you know, get in the wild with all of this. The thing for me that was really most important is this platform has been in use, as everyone spoke about, for many, many years. There's not a lot that's new there, right, to be created in order for us to go after deep learning and the search relevance, and take that long tail inside of the enterprises who really have not gotten to what all the hyperscalers have, right? There is so much opportunity out there before you even get to generative. The magic happens when you actually start to combine these two things together and look at this as a unified front. Why do I know that's important? I've been here for four and a half days, and I couldn't help myself. Armughan and I did two sales calls at some of the large towers around here, with two financial services companies, right? C-level executives. We really just wanted feedback, right, on how we were thinking about things, and did this resonate? It absolutely landed and really reinforced it. They said, "Hey, look, as we're getting our data ready, it needs to be relevant, right? Even for our internal apps. Forget going external, right?" To Mike's point in that Compass platform, one company told us they have 11,000 hours a week of recorded calls from their call center. Before they could even enable their agents to be more productive on the phone, let alone what Saty showed, right, customer facing, they have to figure out how to get all of that ready, right, and relevant so that the folks can get that, right? One great example there, and so the combination of these two things is wildly exciting. We've got to go and build, right, a world-class organization to go engage with the Global 5000 around that, and help them deliver, right, on this. We look at this in five key ways, right? The first thing is the right people and the right talent, right? This is obviously sales professionals, but also the solutions architects, the domain expertise, and the folks that can engage with some of the brightest minds inside of these enterprises and really be that trusted partner. It's intense focus, number one. Already working on that a couple of days into the job, but we are going to build that world-class team. The second thing is, like, once you have that team, and you have this amazing product and the amazing product vision that we're executing against, is how do you deliver the right playbooks to that field to consistently, repeatably, scalably, and most importantly, predictably, run these plays inside of, you know, the market, right? We're going to work on those and combine that, so everyone understands exactly what we're doing, who the ideal customer profile is, who these personas are, and work exactly on a very systematic approach towards engaging with them. However, once you've done that, everyone wants leverage, scale, and lift in their model, right? It's obviously what can accelerate your time to market. It can accelerate your sales in the market, and most importantly, deliver really productive yield per seller in the market, okay? You spoke about, you heard about the NVIDIA partnership and some of the other things we have. I heard something wild the other night, to Armughan's point, like, if you don't stay up listening to this, right, you're gonna miss some things. Jensen, right, the CEO of NVIDIA, was on stage with Jeff Clarke, right, the head of Dell, and they talked about their partnership, where $1 trillion of existing on-premise infrastructure is being repurposed to now be intelligent and drive and fuel this innovation. That's not new infrastructure that has to be sold. That's already in place today, that the two of them are partnering on that, obviously, as Armughan mentioned, we're the provider behind all of that, for the relevance work and some of the generative work, as we spoke about our solutions. A massive opportunity for us that we believe, executed properly, which we will be massive, significant lift and leverage for us in the model. Obviously, we have a brand awareness opportunity as well, right inside of these large companies. These two folks in Australia that we met with were like: "Hey, like, we haven't really heard of you," and, you know, some of my friends and others are like: "Wow, this is a really interesting story. Like, tell me more." The ability to get these meetings and engage, right, once you know these people is there, but we have a real opportunity, and really thankful for, the to-be-announced chief marketing officer that I will partner with, and talk more about that, but we have a massive opportunity there that's completely untapped. Lastly, when you do all this, none of this matters, right? Unless you can have the proper instrumentation to understand your key KPIs around pipeline growth and net new meetings required, and all of the things that, you know, folks like I, you know, love to do and do, working with Helen to make sure that this is all very predictable, very reliable, and that we understand, right, how to instrument this for growth. More importantly, when we start to grow, we can know exactly the yield per seller that we're going to have as soon as we put them on the street, what their ramp time is, and your basic, right, sales capacity and productivity models. Really excited about all these things. You know, lastly, I know we've showed this slide before, so I'm not gonna repeat it, but why do I have this up here, and why does this slide excite me? Of course, anyone wants to sell into a market like this and, you know, actually, the market will be a lot greater, right? The key thing here is, as you start to work with these customers and take a very thin slice here, and work with them on that first project to get started, I call it a circular formula. Maybe there's a better term for this, but it just keeps on going and going. You work with a customer on one project and one model, right? All that does, as you've seen through this product strategy, is drive the need to engage more and more, and now you have new use cases, new business units, right? More fine-tuning, more Red Teaming. As you start to look at the land and expand strategy in the field and getting started with customers, right, on one project, you could look at, right, the customer lifetime value as being massive, and $ millions and millions and millions inside of these large strategic Global 5,000 companies by just getting started on one of these projects. For us, it's a massive opportunity to have leverage and scale in our model, right? Once we get engaged with these customers, as the company's proven that they've done, right, with the hyperscalers. Now our opportunity is to take that to thousands of other companies and repeat that same type of financial performance. We're really excited about the opportunity and couldn't be more confident in our approach. With that, Roc. Yeah. Roc gets to get embarrassed first. Okay. Yeah, Roc has to get embarrassed first. Listen, before you go, I think, you can talk to many of these folks who have mid-sized companies, and you can actually sell them something, right? Yes, we are open for business. Yeah. Happy to collaborate. Seriously, all kidding aside, I've not talked to any other CEO, any CEO who has said, "No, we're not interested." Those two customers we went to see, they want to do a proof of concepts right away, right? This is not just anything. All of you who have your investment funds and others, you should really be building out stacks like these. No joke, I'm very serious. This will be a competitive disadvantage probably in the next six months, right? Either you're ahead of it or you're chasing others, right? That's important for you to know. Next, I want to introduce somebody I'm learning from every day. He's the co-founder and CEO of our China business. That's how I introduce him all the time. Roc has built out a zero to, well, we can't say the number, right? China. We can? How big is the number last year? What did you close at? AUD 33.6 million. AUD 33.6 million. Our main competitor, just announced their last quarter, we beat them on their last quarter by far, right? Yes. Our main competitor, SpeechOcean, they've got a billion-dollar market cap, and we just beat them on their Q1 and our Q1, right? Yeah. Okay, good. He's gone from zero to 1,000-person organization. He used to lead IBM Global Services for how many years, IBM? 11 years. 11 years at IBM. Before that, HP, where you and I worked. Seven years. Seven years at HP. Left all those places. Why? To build out a startup, zero to now 1,000 people, some of the top hyperscalers, along with AV companies. Yeah. All the AV companies. How many of the AV companies do business with us out of how many? 57 totally, and the top 10. Top 10. Yeah. Right? We've got an incredible opportunity in China. I am super bullish on China. After COVID, I asked like, "Is this this way or is this this way?" He's like, "It's way this way." I'll turn it over to you. Thank you so much. Okay, I'm the last one. You have heard a lot of our peers share about technology, share about the revenue growth, the potential, et cetera. I'm very pleased to be here to share with you the China growth. In August of 2019, I was appointed as the general manager of China, and we started a journey. When you look at the left part, we started from AUD 200K, go for AUD 4.7 million, go for AUD 24.7 million. Last year, we achieved AUD 33.6 million, and we will continue growing. I'm very pleased to report to you and share with you, Appen China now is the number one AI data provider in China AI market. That is what we achieved after these four years. You know, China is a very competitive market, right? There are so many competitors around us, and two of the public competitors are here. One is SpeechOcean, one is Datatang. When you look at the right part, when we compare the domestic performance against Appen China, we are roughly 1.5x of SpeechOcean, and we are 2x of Datatang. This is the fabulous work we have done for China client. Let's talk about the China market. You know, China is a fast-growing AI market in many years. The second is for China Appen, we have grown our client base in a very significant way. At the end of last year, we have 200-. 203 clients already, we continue to grow very quickly. Here, I'm very pleased to report to you, when you look at the key clients, there are three major industry who are using AI technology, in internet, in auto, in mobile, the top 10, the internet companies are all our clients. We look at the auto, autonomous vehicle, all the top 10 clients are Appen's clients, the same to mobile. The top five are Appen client. The reason the client, they look at Appen and select Appen as a core delivery partner from AI data perspective: one, is a global expertise. They know we are not only have the expertise in China, know China, because I was in China. I was born in China. The second part, more important, they know we have very strong global experience. If the China client want to go to overseas, there is a company can help them to support them, that is Appen China. The second part is Appen have the very full portfolio. We are not only a speech company, we are not only an image. We cover all. We provide a speech, text, and image, the full portfolio for our client. That is really important. Another thing I want to share with you about our capability is our size. You can look at the Wuxi, Dalian, and Chongqing. This is the sites we built in three years. Wuxi is in the east of China. They can cover all the clients in the east area. Dalian is in the North of China, and it can cover all the North of China, client. Chongqing is in the South, and they can cover the South part. It help our client to can come to our office by take a high-speed train within two hours. We are very close engaged with the client on this. The second I want to share with all these sites, we achieved ISO 9001, we achieved ISO 27001, and ISO 27701. This means we have a very high management system, and a high-quality system, and a privacy protection system in our all sites to help our client. The last one, very important, why clients select Appen is, we have very advanced technology in China. We are the number one tier players in autonomous vehicle in China. You see there, the technology, which is a two-part help us, one is we really have the AI power, the algorithm, embedded in our platform, which can improve the productivity and efficiency in a significant way, right? That is number one. The second part is we really have this advanced technology, we can on-premise deploy to our clients' environment to help them to make sure there's high, you know, high secure way. We have 10 clients have been deployed already. The last one, very important. What's the future of China? We will focus on the existing client, and we will continue deepen the relationship with existing client. They are super, and they are growing very fast. The second part is we believe not only the current top 3 industries, they using AI, like in the internet, in the auto, in mobile. They will express to education, to manufacture, to finance, to medical, pharma, et cetera, healthcare. This is the industry we are cooking, we are working as well. We believe that is the new customers will continue to grow. The last one, very important, is generative AI. That is shifting the world. We, in China, we also have a lot of companies that are involved in this, and Appen is a critical partner in generative AI domain. I'm very confident for China growth, and thank you very much again. Thank you. Thank you, Roc. Amazing. You just heard it from our leadership. We wanted to make sure that we close on something that's very important on behalf of the entire organization, right? I want to tell you again what I started with. Generative AI is going to have a new programming language. It's called Human. We've been doing human work, large language model work, that's what it's called, for the last 28 years. We're not an annotation company, we're a relevance company, we're an assurance company, we're a fine-tuning company for deep learning and AI. We are going to be calling into customers directly. We're going to build our products. You've seen the kind of CTO, the CPO, our CRO, and our new CFO, and the broader teams. That's how we're executing now. That is how I've always operated a business. We feel like we need to show you so that you can trust us. I want to use the word trust again. It's been 28 years. Julie was with us yesterday, Julie Vonwiller, and our previous Chairman and CEO, Chris Vonwiller. We gave them this entire presentation because they couldn't be here today. They were just thrilled, and they said, "Listen," Julie said, "It just feels like there was a house 28 years ago. I built it, and then for 28 years, no one went out and painted the walls." If you've ever lived in a house for 28 years, would you not paint the walls? Like, my house, I think we made our kitchen twice in 28 years, or maybe 20 years. You know, there's refreshing that you need, and I think the board, I'm very thankful. I see Robin in the back there. Robin, thank you for joining. Our Chair of the Audit Committee, Robin Low. Our board has been refreshed. Our CEO, along with our leadership team, has been refreshed. You can see the energy. We feel like this is the dawn of the new era that we need to work towards. The two people who are not here, Brian Hoskins, who are leading our operations, and Eric, who's here, is helping us lead the operations, along with Andrea Clayton, my partner, our Chief People Officer and Chief Purpose Officer. I call it CPPPPO, because our culture is very important to us, and her and I partner every day. She's told her husband, Brent, and her two kids that, "Oh, I have a new partner named Armughan," because he's gonna call me at the end of the day, driving home. I always tell her, "How are our people?" I always tell her that we acquired hearts at Appen. We don't acquire parts at Appen, so I don't care if they're before me or after me, they're our hearts, and we have to treat them as such. We know that we had to do a bit of a restructuring, as you know, a couple of weeks ago, and we made sure that we led with purpose. We treated everybody with respect. We made sure that they kept their laptops so that they can find their next job. We made sure that they left with dignity, the right pay, the right... We did everything as we could. We went out and looked at Stripe and Google and others and how they did that. Purpose, to me, character and integrity. I don't come from a lot, by the way. I'm an underprivileged upbringing. A bunch of us in our leadership team have something in common. Grit is very important to us. Purpose is number one for us. We feel purpose, then perspective, that will lead to prosperity. I know you would like to see prosperity, you would like to see the stock going up. We got it. We feel to get there is purpose first, and then perspective, and then prosperity. Perspective is what you're seeing, right? It's moving really fast. It's making sure that we are learn-it-all organization and not a know-it-all organization. I do not like know-it-all organizations at all. I've worked in many organizations before. As soon as they become know-it-alls, or they say, "I know something more than anyone else," that does not work, and I've seen companies fail. We're not that anymore. We're a learn-it-all organization. We're accepting that comfort and growth don't coexist. You could see Andrew and his excitement. He's four and a half days on, super uncomfortable. We're going out and doing proof of concepts with customers, with a product that, you know, it's moving, you know, at the speed of light at this time. And I can't tell you how thankful I am of this team and how hard they have worked to make sure that we knew that we had committed four months ago, Rosalie, four months ago when we committed to this, right? That we were going to do this. In four months, I told you we were gonna get something done, and we worked really hard to get it done. I hope you're all pleased. I would love to invite now my leadership team who just presented up so that you can ask us questions. If you guys could all come up, please, who just presented, that would be great. We are happy to. Ryan, you as well, please. Thank you. We'll just move this in the middle. Helen, you can come up. All right, great. I thought it would be good for you to not just. You've listened to me multiple times. I've met many of you before. I thought it'd be good for all of you to, you know, hear directly from the leadership team and ask us any questions you want. We're happy to answer them. Who's the first question? Josh, thank you. Bassem is number two after that. Okay, perfect. Thank you. Your microphone's coming so that the people on the webcast can then listen in. Great. Thank you. Josh Kannourakis from Barrenjoey. Thanks for taking my question. First one, you talked about some of the new products that you've brought out, most notably around the LLM enablement layer. Can we talk a little bit about, a little bit more around, you know, the go-to-market strategy there, where the products are at, you know, in their life cycle? Mm. When Appen will be out? What are the milestones that investors should look at to see your progress on actually getting market adoption with some of these things? Yeah, great question. Maybe I'll start, and I'll get it over to Andrew after that. I'll tell you, I think for me, if we take a look at What's our pipeline number for these due deals? Forty-five. 45, right. It's changing every day. Four months ago, when I announced our generative AI products, we did not have that many in our pipeline. We now have 45. When I did the equity raise, which was, I think, last week, we had 32. That's the number I showed you. It's changing. That's how fast it's changing. We believe that our deep learning AI revenue is now starting to see green shoots there. I won't use the word stabilizing. I would use the word green shoots, because that's where our hyperscaler, our existing business is. We have moved up multiple notches, as I said, in the executive leadership team at our customers. I would also tell you that our number one customer has now consolidating our number one competitor's revenue to us, which is great. Not just saying that, they've, we've already quoted it, Brian and Eric's team are already doing a great job executing on it. I would tell you that's where we're at right now. Obviously, now we're, as Andrew said, we're creating a go-to-market with it. That go-to-market, I will continue to tell you, when I meet you in August, I will tell you what that 45 number would look like from a pipeline, what have we closed on that pipeline. I just need a bit more time to start flying this plane. I was given a 747 heavy, leaking fuel. I had one engine working, one engine not working. Then we had a Mount Everest coming in front of us, and, you know, we had to do, you know, debt refinancing. We had to make sure that we get, cost out, get ourselves fit, turn this into an F-16, so move it around rather than pulling up in front of Mount Everest. That's sort of where we're at. Make sense? Anything you would like to add? No, I... That's pretty much it? Okay. Yeah, good. All right. That's for you. All right. Thank you. Bassem, number two, question. Darren Leung would be number three, I thought. Any other hands coming after? Okay, great. I don't know your name, but look forward to meeting you. Thanks, Armughan. Thanks, team. It was a wonderful presentation, and I think it's, you know, brought a lot of conviction back into Appen's name, so I think that was really great. My question is just really on, you know, the strategy going forward. You know, we have had, as you mentioned, a lot of loss coming through from our largest customer. Mm-hmm. At the same time, you know, we are seeing a lot of green shoots coming through. We also have probably less headcount now than we did at the start of this year. With the limited resources that we have, you know, where is our focus at this point in time, in terms of, go-to-market strategy? Thanks. Yeah. I shared this in the investor day. Well, sorry, not investor day. I shared it on the equity raise, and I think I've shared it before with other-. I think I did a call on when we did the equity raise. There is a slide that I showed, which we don't have today, but it sort of tells you what our focus areas are, right? Operational rigor is going to be continued focus for us. Cash EBITDA positive, we have to get that done, make sure that we are running... Helen and I come from a similar industry, where we run a tight ship. She ran a tight ship on basis points, and it's important to us. That's one. Number two, product. We need to get these products that you just show, just saw. Some of those products are maturing, some products are matured. How do we start tracking those products? Saty, where are you? Saty and Sujatha are partnered together very, very closely on getting our product roadmap to then align to that and ensure that we start delivering on those three products that I told you that we need to develop, especially around the fine-tuning and assurance, so that we get those out there. That's second. Third is our world-class go-to-market. Chief Revenue Officer, Chief Marketing Officer, who, he'll be announced in the next few weeks, and they're really going to focus on building the brand of trust of Appen out in the market. That'll be helpful for us, along with, you know, going out and seeing as many customer... What's your target? How many customers do you want to see per week? You notice? It just happened. A lot. A lot. That's important to us. Then, for us, you know, making sure that our business operations aspects are very critical to us. Our crowd, how we segment our crowd, how do we do the crowd intake and others. Then finally, our AI for good strategy is very important, and we want to make sure that do good, be good, lead good, is at the forefront. That's what I showed on that call that I provided. I would like that to be my scorecard, Waseem, then you should judge us all on that scorecard. By the way, we just had an offsite here at the BCG offices, for the last two days, and we have a full plan exactly who's going to do what. I'm a bit of a Air Force type of person, so I would say fly in formation. Don't fly one's here, one's here. We're all flying in formation. That's how we're leaving Sydney to make sure that when we get back here, we'll do that. Okay? Thank you. Darren? Anything you wanted to add to that? You're good? Okay. Thank you. Darren Leung from Macquarie. Thanks for the presentation, guys. Obviously, a very exciting time. Yeah. plenty, clearly, you've been very busy in terms of changing the product, and it sounds like you've got a go-to-market strategy, you know, over the next sort of 6 to 12 months. My question is twofold. The first one is, I suppose when we think about that, you know, the products that enterprises adopt, you know, what do you think about the main hurdles are in terms of, you know, when we think about, say, ourselves as a Macquarie Group, quite an established organization, you know, I find it quite difficult internally, you know, getting products on board, and I actually want to use them as a supplier? Mm. How do you go about breaking through those barriers, especially for something that's experimental? My second question is even further down the track. You know, I noticed in your slides at the equity raising that you mentioned that the NVIDIA partnership. Mm Relationship doesn't have any monetization at the moment. Yeah. Which is fair enough. My question on that is, you know, what are the hurdles or what do we need to see. Yeah before that monetization piece kicks in? Got it. I just want to correct, we do have two customers with NVIDIA that we are already monetizing, which is good. Andrew, do you want to take this one on what's our incursion strategies into accounts? Yeah. I don't know if there's. Yes, okay. Oh, okay, great. Look, very simply put, this is an unbudgeted item, right? For what we're providing. The generative AI version is unbudgeted. Correct. Correct. AI, in general, is a C-level discussion, right, with the board of directors, right? For us, the way to break through is, first of all, right, we have to make sure that people know who we are and what we're doing. Once we do that, it's about selling business outcomes, right? It's about working with these executives inside of their organizations, right, to deliver those and to partner on that, right? That enablement layer is so key, right? Because even if you just think about generative, right, OpenAI had been, you know, working on this for six years, and they have the entire internet that gets to provide them with their enablement, right? If you're a large enterprise, you don't have that luxury, right? That's where our, you know, our crowd and our segmentation and everything that we're working on is there. Simply put, it's, you know, working on those outcomes and, you know, we've seen examples of that already. Great. Thank you. Next question. There's a question back there. Yeah. Hi, Armughan and team. Thanks for your presentation today. Ross Barrows from Wilsons Advisory. Hi, Ross. How are you? Just had a couple of questions. One is, I guess Appen can't be all things to all people. Could you talk about Anthropic in some way? My understanding is they're thinking of, you know, constitutions or principles that doesn't involve humans in the loop or human contributions to that. Maybe can you talk about those clients that will use you, there'll be some that won't, and some color around that, please. Great. Thank you, Ross. Maybe I can ask Ryan, if you would like to start, and then Sujatha, if that's okay? Yeah. Hey, Ross. Ryan, by the way, is our head of strategy and innovation. You know, they're working on what's next, very quickly. We're staying ahead of the curve, right? Please, Ryan. Yeah, thanks, Ross. There's a lot of techniques being deployed at the moment to build these models. We see some taking very open approaches and broad, like OpenAI, what they do with their models. They're very specific models being fine-tuned, like what Reka are doing, specific models for enterprises that require fine-tuning. Anthropic is taking another approach. There's a vast array of approaches. We think it's gonna be a combination of all of these, so it's the right model for the right solution. We don't think there's gonna be one winner that dominates in the approach. How that plays out over time around what model for what solution, it's going to come with the evolution of the market. Sujatha, anything you wanna add to that? Yeah, I just want to add one more thing, is that for enterprises, they will pick different models based on the use cases. For example, they might pick Microsoft Copilot or the GitHub Copilot for their coding, improving the productivity. They probably won't customize it, but they might take either Reka or Cohere or the NVIDIA model and customize it for their own scenario. Within enterprise, we expect to see many different models based on the use cases. Yeah. Some would require human in the loop, some may require just reinforcement learning with AI feedback. You saw on our, I call it the chip stack, and on the chip stack, you had RLHF and RLAIF. Anthropic is a lot more RLAIF focused, and that becomes a use case. There are other use cases that are there, right? They can't really come and compete with us in that space, or else they won't be as relevant. They're betting on the RLAIF area, we're betting on the humans. I think that's a good bet when you bet on humans. Just another quick one. In terms of the data that's being drawn on, some of these models can only answer based on what they've been taught on. Mm-hmm. Some of that data is not current. Some of it's two or three years old, I think, if my understanding is correct. Yeah. How long does it take until the data that it's trained on becomes current, and then the responses are not real time, but as good as, and add a lot of value to those asking the questions? Yeah, I... you're asking me to think about what a lot of the neural networks folks are, or a lot of those PhDs are thinking of. I really feel that within this year, you're going to no longer worry about, Oh, this is 2021 train data. ChatGPT 3 was 2021 data, 3.5 was later, ChatGPT 4 is later. You're now having that question even becomes irrelevant because imagine if, for Barrenjoey, we take their current data and we take the full stack on it, put it on an NVIDIA box, train it, and then Josh is able to ask every question there. That becomes relevant today, right? That's only for a very large language model, which the ChatGPT is. It's a very large public model. You can ask it a question on what's the capital of Nigeria, and you can ask it a question of, "Hey, summarize Tharman Shanmugaratnam for me," right? The Barrenjoey LLM would be very specific. It won't answer the question on: What's the capital of Nigeria, right? It would just be asking questions that are relevant. That could become a lot more current that way, right? Yeah. Go ahead. Yeah, just to add to that, too, is, like, when you see there's two other ways to get the data current, right? You might train the foundational model, but then you might have embedding. That's what RAG's talking there. You'll be embedding it with your own data that's relevant to it. If you wanna keep everything internal, you'll train it with your own data. You don't need to train it, like, if you're, you know, in the insurance industry, you don't need to train it on Russian literature, right? There's gonna be times you have the base foundational model, you'll bring in embeddings, which will probably be your own proprietary data in that, or you'll purchase third-party datasets and then train the model based on that. Then there's also gonna be plugins. We've seen with ChatGPT, they've released plugins a couple of weeks ago, it's now open to everybody. Those are also gonna be important 'cause there's gonna be some things you'll always have to interact with databases and other services, that you won't be training the model every single time you use. The location space is one of those things that can always change, right? The locations of things. You're gonna have all these services, plugin services. There's always that foundational model, but there's gonna be embeddings with relevant data for the application, and then plugins with services, with other data current data sources, based on what's needed for the application to do. Great. Thank you. Thank you, Ross. Suraj, next, and then you. Sorry, I didn't see who came up first. Yeah, Suraj Nebhani from Citi. Three questions. Just the first one, maybe. In terms of Appen's involvement with LLM versus deep learning, just your thoughts on how much involvement is there, right? If you're thinking maybe three to five years down the line, do you think deep learning will be 80, LLM will be 20 as the industry data says, or is it actually the other way around? Just keen to- Yeah. Keen to understand your thoughts on that. Sure. I think you and I were briefly touching on it, so I'll touch on it in more detail. You know, our view is that if you look at where deep learning is today, deep learning continues to be super relevant for us. That's majority of our revenue. You saw the search results that Sujatha was showing. You know, you go to any search, people are still using that search. That still is very relevant. We started doing generative AI work a few years ago. It wasn't called generative AI, it was called large language model, or there was a project that Google Bard was working on, or Meta was working on, or anyone else, for that matter. Those are all secret projects, as you can imagine, and now everybody knows about them, right? We've been working on projects like that for a long time. We don't split out, Suraj, our generative AI revenue and our deep learning AI revenue as of yet. We really believe that that's a huge growth. As I told you last week when you were on the call, I don't know if you or Citi was on the call, I said we had 32 deals in the pipeline, generative AI, and then now we have 45, and it's just, you know, moving very, very quickly. We also believe that generative AI will have much more adoption in enterprise, and I think enterprises. Listen to this entire conversation. If you're not leaving from here thinking: Hey, we need to do something. We need to go talk to our CIO or chief data officer and say, "Hey, what are you doing?" By the way, they're not doing much. That's the answer. We just met with your largest insurance company here, second largest bank here, said the same thing: "Oh, wow, you guys are doing this? We're asking people all over Sydney who's doing this. We're asking these people to come in, that people to come in." I think that's where we're at. Yeah, final, I think, just on our view on deep learning AI, does that start to pivot away, and does it go to generative AI? We are seeing some projects that has applicability where generative AI has matured, that it could take on, which is great, and we've seen that in our top five customers. We've also now seen us moving up in the top five customers, and that's now giving us, giving me at least, a lot more surety on. That's why I said our second half will be better than our first half, and then we'll show you more data on how our deep learning customers are doing. All right, just following up on that, because, in terms of... It's good to hear that you're working with Google... Mm. -on Bard for the last two years. The way I understand it is, previously in deep learning projects for the relevant search. Mm-hmm. When it gets into production, that's when your relevance work actually, you know, explodes, right? Mm. Because it's in production. Mm-hmm. Now that Bard's live or getting live. Mm-hmm. Is that what you're seeing on the LLM side, in terms of your work? Yeah. I would say on fine-tuning side, so if forget Bard, my answer is more in general, right? You know, I think what we're seeing is our revenue definitely is there in the fine-tuning area. We've been in that fine-tuning phase, as I've been calling it, right? RLHF, RLAIF, instructive data prompts, as well as our, you know, areas around RLAIF. At the same time, we now feel that the relevance work now becomes the assurance work, right? That assurance work, you know, we have seen that pick up with the customers who are maturing on the fine-tuning side, but it's all over the place, right? Who's matured on the fine-tuning, who's not, right? That's where we're at. This third one. Good to hear the largest customer is consolidating at a better relationship now, or you're talking to the more relevant person. What did you have to give up in terms of that, you know? Did you have to give up, you know, margin or something on that price? What does that mean in terms of visibility? Because that's one of the biggest concerns? Yeah When we talk to investors, that, you know, we don't know what's six months down the line or two months down the line. Yeah. Has that changed? At this time, all we're hearing from our customers, Suraj. We have a gentleman, Brian Hoskins, who joined us about, I wanna say, six, seven months ago from IBM Global Services, who was on the Meta account, and he's done great at it. What we're now working on is just, it's not about just us giving up margins, it's just giving great service to customers. The customer had asked us, "This is what we're looking for," we had to really provide them that. I think just our client service has improved dramatically. And I would tell you that, you know, we didn't have to go and discount. I think that's what you're asking me, to discount to win business. That's not the case. It's good work, and it's getting credibility, and then earning the customer's trust. Again, I keep using the word trust. Trust is very important to us, right? Can I add on? Please. I want to add on is, when you look at the AI industry, it's still a very young industry, right? It's not like software, maybe 20, 30 years already. It's just few years. In this industry, even the deep learning has a lot of advantage as well. I'm a PhD from computer science, while I was, while graduate, I found a lot of trends in this industry. AI is very new, we continue, we'll get a lot of demands in the deep learning domain. That is for sure, because a lot of client have now employed AI, the technologies, into their business, transforming their business, right? Another side is because this is a new industry, there were a lot of new technology emerging in a very quickly way. I still recall when I was a in the university, we see Java, we see C, we see C++, we see Java very quickly evolving. Similarly, you think about AI world now, right? We see many, many new technology evolving. Appen is catching up everyone. We're catching up the speech, we're catching up the content relevance, now we're catching up generative AI. That is overall, I still think the demand is there in the market, but just we make sure we do an excellent job to serve our client. I would also maybe, Rob, just as you were speaking, I thought of another data point that I wanna just hit that on again, which is, if you look at the last four years of our CAGR growth, if you take our number one customer out, we've been growing 10%, right? It's just that number one customer, what they had to, you know, pivot internal reasons, right, towards a newer area of technology that wasn't as relevant to us. Now as they're coming back, we're seeing, you know, different points, right? I don't wanna forecast what it's going to be, because I just need a bit more time to start flying this plane, and then start to be much more predictable. My goal and Helen's goal, and this entire leadership goal, is to become a lot more predictable. Even when we're growing up, we wanna be predictable, yeah, and hopefully never going down. If we're going down, we wanna be predictable, right? That's helpful. I think that makes your life a bit easier. Sir? Craig Stafford, Barrenjoey Research. Very happy to be hosting you today, and thank you. Thank you. -for the presentation. China is a massive opportunity for lots of industries- Mm-hmm. Congrats on your progress so far. Is there anything you'd call out that's interesting or different about that market opportunity, to help us understand how big it is or otherwise? Yeah. Don't give the numbers on how big it is. No, no. You're able to tell them how you're going to. Yeah -grow. First I want to share with you, when you look at the client base we have now, it is fantastic. I'll give you example. When you look at the top 10, the client, right, internet client or AI client in China, when you think of the big names, Tencent, Alibaba, ByteDance, Bilibili, Meituan, Kuaishou, et cetera, they're all our client. What's the key reason this is the top AI internet client they select Appen? Because they are growing. They are growing in a very fast way. The reason is the global expertise. That is one. In China, they will say Appen is a unique company. If they want to go outside of China to sell their products, sell their goods, they need one AI data company to be partner them, to support them. That is only one. That is Appen China. That is very, very powerful. The second part is they really think we are very advanced in technologies. We are very advanced in our resource model. Think about we have the crowd. We have the one million crowd across the world, and can speak 1,170+, 230+ languages. That is super powerful. When you look at the technology we have, we have the speech, we have generative AI, we have two DSP. There are a lot of technology ones. That give the client the confidence to say, "Okay, this is a new industry, AI. Appen is a core partner, be with me, and we grow together," right? In the case, the many, many areas, there is new, and we partner with the client to do the thing together. For me, is I'm very confident for the future, in AI with this core client. Client is fabulous. Mm. Our capability is very strong, and more important is I even think, as I mentioned, AI will eat software, right? In the long way. That is a huge business. Trillions. trillions dollars ahead of us, that wave, we will catch. That's how I want to answer for you. Thank you. Thank you. Question here. Yeah, I need a microphone. I think it's for the webcast folks, so they can hear. Hi. Chad Mikhael from Barrenjoey. I sit on the trading floor, looking after emerging companies, so I speak to a lot of investors on Appen. We can certainly help you. Yeah, yeah. ... building your stack. Any hard questions go straight to Josh. Yeah. Look, I guess sitting here, you get a real context of the growth in the business. Yeah ... structural growth, the opportunities. I'm really keen to understand how, you know, you convert that to, you know, margin, profit. Obviously, that's down the path. Mm-hmm Just keen to understand, when you're having these discussions, are you a price maker? Are you a price taker? Mm. How should we think about margin? Yeah. Clearly the revenue growth opportunity is significant. It's that next layer that I just want to get comfort on. Sure. Maybe I'll have Helen start, if that's okay. Okay. There's a microphone, and then I can add on to it. The way. This is on? Let's start off. Oh. Give it to a technical guy, he starts figuring-. Finance girl with no power. Okay. The way I think about it is, it's a different selling motion. Historically, this business grew up really catering to procurement organizations, which has a lot more pricing pressure on general trends. When you're thinking about the enterprise space and where we're bringing these solutions to them, that's an outcome-based selling motion, and it's a much more strategic conversation within the client. While we're not gonna talk about what the potential is, 'cause this is all emerging for us, we believe, I believe history in other spaces like cloud and software and services, would suggest that when you're leading with value, you have a different pricing model. Yeah. Saty and Sujatha and, I think, Andrew and I have seen this movie before. The movie starts with something like: You have an incursion strategy with the client, that provides you the margin profile that we're at right now. Then, when you go into repeating revenue, I don't use the word recurring revenue, repeating revenue, then you start moving into, you know, monitoring services, certification services, right? Do you think when someone trains an LLM and says you teach it with instructor prompts, now the LLM is trained, it comes out here, right now it's saying all the right things, then it starts going this way? As soon as it starts going this way, then you now say, "Well, no, no, you have to bring it back," or else it would show what the toxic content that Saty talked about, right? You're looking for X things, and it's starting to say... You know, Saty, maybe. That becomes, by the way, higher margin. We're not going to say, as Helen said, how much higher margin, that becomes higher. Maybe, Saty, you can talk a little bit about Red Teaming and what that actually means to cybersecurity and fraud, right? Absolutely. I mean, in fact, all of you are already doing red teaming. When you see ChatGPT, there's a thing saying, "Was this relevant?" There's an up button and a down button, it's kind of using all of us to do the red teaming. That way, it understands what's a relevant answer versus what's irrelevant answer, what's a bad answer versus what's a good answer. If you think about it from a fraud perspective or from, like, a risk perspective, and you're doing it for each of your applications, you want to be training it with the right things. The cloud gets involved again, or your domain experts, who are part of your specialized cloud, get involved again. It's like a continuous way of making sure that your LLM in production meets all of your risk requirements. That requires. Now, if you're not ChatGPT, and you're a bank, and you've deployed it, who's giving you the up and down? You're not gonna deploy it with your customers giving you, "Oh, yeah, you gave me a really negative answer on my credit card and made me poor or called me poor," right? "Oh, that's down." That's not just a thumbs down. That's like cancel credit card, cancel bank, never talk to this bank again moment, right? That becomes a much higher margin order for us. I'm trying to give you a directional answer, if you don't mind, right? Yeah. Yeah. Pleasure. Yeah. One more thing I. Yeah ... to add is that for sensitive data, some enterprises will require a secure crowd, which will be in secure locations, which again, will be a much higher margin. Yes, Amit? Yeah, and come on, come on. Maybe the last thing I would add is that, when you have repeating business within a client, just the lifetime value, I mean, it really changes the discussion that we can have when we're going out to market to acquire clients. This idea of proof of concepts and leading with the outcomes, the whole idea is we need to acquire the clients. You acquire the clients, you have. The average in our portfolio today is 9+ years for those top five, and that's really what we're looking to do in the enterprise space, because repeating business for nine years is very meaningful, obviously. Good. There's a question there. It's Conor O'Prey from Canaccord. Hi, Conor. Hi. Question maybe for Helen, maybe for Armughan as well. If it inevitable that Appen is always gonna be a company that's got 70%, 75% of its revenue in two customers? If we fast-forward three years, is this a much more diverse kind of revenue base business and therefore much less risk attached to that? Obviously, as you pointed out, and the metric you provided is very helpful, working out exactly what's happening with that key customer, but it still represents a big, big chunk of your revenue, despite the fact that it's off substantially. We used to have this much gap, now we have this much. Yeah. Is it inevitable and it's always gonna be a highly concentrated revenue base, do you think? We certainly don't wanna lose any market share within those existing top clients, and we think that generative AI is incredibly important to them as well. The services that we're providing to them today, we wanna keep those, we wanna nurture those, and we wanna grow in other new areas in the generative AI space. For me, the measure is really about new account acquisition in the enterprise space and whether we can actually shift the numbers over the next few years. It's not gonna be wholesale. 'Cause we wanna keep that market share. For us, though, where you'll see it is in the customer acquisition counts that we'll provide. I would just add on to it, Conor, that I think on the deep learning side, we wanna protect that base, continue to, you know, improve what our profitability looks like there, improve profitability. I think we have a great opportunity. I'm gonna go back to my house analogy, 28 years, not been painted, cobwebs. Oh, this thing is leaking. Just fix one leak, margin goes up. Another leak, this goes up, right? That's where Eric, who's sitting in front of you, and Brian and others who are working really hard on making sure that our experience is there, and then how much we, you know, pay out there with the crowd. We wanna make sure that it's a good balance of the kind of work that's happening there and the kind of work that we're delivering. You know, when I started, we were doing negative margin deals in many of these accounts. We don't need to do negative margin deals. I was never taught in any business school that negative margin deals are good, unless that turns into a really great opportunity. We're fixing a lot of those pieces, and we feel very confident that that will get resolved. Thank you, Conor. Yeah, Suraj? Just maybe, just following Conor's comment, just in terms of the other hyperscalers that you mentioned as. Mm. -as an opportunity, any update on that in terms of? Yeah, yeah. I would say out of the top two, where the top two are here for us, the other three... When I say top two, sorry, I'm sitting next to China, so I should talk about the North American top two hyperscalers versus the versus the other three. We feel there's an incredible opportunity there. I think, what our market share is on the top two, we need our market share to be that in the, in the three, that's my... That's, you know, he hasn't gotten his quota yet, we're really working towards that. The good news is we're seeing some green shoots there. Our current sales team, led by Kevin van Kempen, has really been focused there. We've had some really good green shoots over the last, I would just tell you, last three weeks, four weeks, just people are seeing Appen in a different way. Just look at, I mean, your investors. I remember meeting, you know, some of the investors the first time. It's not a pleasant conversation with many of you, right? Think about you as a customer, and how that conversation was when you first met Armughan and the team, and then now, four months later, the customers are looking. It's the same service that we're providing to you. Would you all say that our service is improving a bit? No. Yes? Yes. Okay, good. That's a very hard customer for me, by the way, to ever get that. I'm telling you that our service is improving, and we are and I promised you and Tony, that we were going to... You're like, "Yep, more say, do, less marketing gimmicks." This is not marketing gimmicks, right? This is real. We're improving our service SLA to you. We're going to continue to improve our service SLA to our customers. That's how we're operating. If this is any type of relevance to you of how we're working with you, that's the same style, if not more. One of our, it's not on here anymore, our values in our to achieve prosperity is being customer obsessed. Customer obsession means that we're always thinking about our customer and their customer, and working backwards from there. For us, our shareholders, and our investors, we treat them like customers, and we wanna treat you like that, and you'll see that from me and my leadership team. Just maybe last one from me. Sure. In terms of enterprise, I mean, a few years back, this is pre-COVID, a different era maybe. Mm-hmm. there was a big investment in the sales team, headcount as well. Now there's a reinvestment again. What's different this time? Is it just because LLM is much more fit in terms of Appen? Secondly, you sort of mentioned segmented crowd workers, right? How difficult is that to actually... Or do you actually have the 1 million workers? Do you actually have enough segments in there? Sure or do you actually develop that. Yeah, maybe Ryan can take that one, but I'll take the first one. What was the first one? Go-to-market. Go-to-market, yeah. Go-to-market will be built, right? Before my time, when we built our go-to-market for enterprise, our go-to-market for enterprise was more built for, can we sell you data collection? Can we sell you data annotation? Can we sell you relevance? The maturity of the customer base at that time, as I said, Suraj, we were going if I'm walking into, let's say, JP Morgan or Citi, let's do Citi, you probably don't want me to use JP Morgan's example. Citi's example, you know, we were going and talking to your Chief Data Officer. We were not talking to your Chief Data Officer. We were going and talking to your data scientist, who was, like, 10 levels removed from the CEO, that person had a $100,000 budget, we were going out and figuring out, can we do that work there? The person will tell us, "Here's $100K, get lost." Right? We would do great work. They're like: "Great, I have no more budget," right? We would go to where automotive was, automotive became a good, successful area, like China, along with Germany and Europe. Anyone who had a budgeted item in the deep learning area, like government, federal, autonomous vehicles, that became an area that we started doing well in. Again, they weren't as mature as hyperscalers, that they could buy in $ millions. I would say there's part of it that was related to timing, part of it is related to what we built, what we built back then. It was not the Salesforce that you're going to see this guy build, right? This guy, meaning Andrew, sorry. That's a very different sales team, right? It's a sales team that goes in and talks to the manager-level person, not the sales team that's going to go and have a conversation on his plane. He got Wi-Fi, texted a few CDOs, Chief Digital Officers, landed here in jet lag mode, went in and saw a Chief Data Officer of the biggest bank here, right? That's very different. That's the sales motion I have always led in the last 27 years. Yeah. Ryan? I think to add to that, Armughan, Suraj... Yeah. When we were trying enterprise for the first time, there were huge hurdles enterprises need to get over for around the data organization and actually putting in those pipelines to make use of them. With generative AI, it can use unstructured data really easily, so that barrier doesn't exist anymore. The barrier has shifted to the other side around how they fine-tune the models once they've been built and make sure that they work for the context in the way that the companies want them to. We see that the frictions around AI and enterprises have been reduced significantly. That's. And we think that's gonna be the big unlock. Going back to Conor, your question around the shift of revenue, we think that's gonna be the big unlock for enterprise, too- Mm. around getting that balance right. On the segmented crowd piece, there are people in our crowd that match the needs of what our customers are looking for when we say segmented crowd. Sometimes they want some really specific things that we need to go out and source. The good thing is we've got the muscle to go and find people, to onboard them, to do the QA, and to pay them, and to keep them engaged. The muscle that we've got from a more agnostic crowd can be easily applied to a segmented crowd. Thank you, Ryan. There's a question there. Yeah. Just wanted to ask you another question around go-to-market. Sorry, can you just introduce yourself? Sorry. Stephen Kench from Perpetual Private. Hi, Stephen. Just wanted to ask another question around go-to-market, particularly on the enterprise side. Yeah. How important is Quadrant or the Quadrant platform in that strategy? Is that where you're putting all your eggs, or how does Quadrant sort of blend into that go-to-market on enterprise? Sure. Maybe I'll start, and I'll have, you know, Andrew and Mike talk about that a bit more. I would tell you, when we look at Compass, as I explained to you when I introduced it, I said, that becomes much more of a. You know, you have Fortune 500 clients, Fortune 50 clients, right? Then, you're probably not Fortune 500 yet, Barrenjoey, right? You would think the Barrenjoeys of the world, who are spending millions of dollars on their technology stack at this time, how can they not spend millions of dollars and actually spend hundreds of thousands of dollars, right? It could be AUD 1 million versus spending AUD 10 million. We think that the older bit is like that. We think that the Compass platform works more for less matured in their data stack, customers of ours. They would like to see that because we just know that, you know, if we just go into a customer and say, "Hey, have you figured out which compute you're using? Have you figured out which model you're using, and have you figured out how you're going to ingest your data?" They'd be like, "I have no idea." As you said, right? You don't know. You're not business unit folks, right? Your ECM teams and other teams that are. Your trading team and other teams, research teams, you rely on your CIO or your chief scientist to figure out what your business problem is, right? That happens everywhere. Our view is that when you're in a matured bank, like Citi, or let's say, Telstra, or you're at CBA, or you're at. They have different functions there. Even yesterday, when you and I, when we met the number 2 bank here, and they were saying where the maturity curve is outside of machine learning, it was not there. We feel that there is relevancy where our deep learning and our generative AI platforms, the way Saty and Sujatha presented, that's more relevant to us, and we're seeing our customers say: "Hey, we're matured. We wanna do fine-tuning first, then we'll get to, you know, assurance with you." There are customers who are like, "Hey," they're like Joshes, Barrenjoey, fastest moving, just gave birth to their company, 2019. They're like: "We wanna be ahead of others. How can we move faster?" They are more calling us for a Compass-like solution, is what we're thinking. Anything you wanna add? I think I would just say, simply put, and again, I'm relatively new, is that it's a component of the overall enablement layer, right? It definitely will be useful, but I don't think it's all eggs in that basket at all. It's just a component, right? Mm-hmm. As Armughan said, right, depending on the maturity curve of where we're at and the projects that we're at, the relevance of it will go up and down, but there's plenty to go around out outside of that. Mike? On my side here, that's, like I was saying, that Humans using AI will replace humans not using AI, and platforms like Compass are gonna be coming up. You're just gonna see tons of them, tons of applications being delivered, and if your company's not thinking of this, and people aren't thinking about this, it's gonna be in a short time, not like three, four years from now, this is gonna hit. People are gonna understand that these type of applications can be deployed in months, and then people will all of a sudden see the competitive edge erode quickly away. I would really, like, look into this space and see how fast things are going to move. What's gonna happen is that people are going to have to learn quite quickly, and Appen's gonna be able to walk people through and get them there, and the people who aren't taking the steps to walk through are gonna be disadvantaged. So you're gonna see this happen quite quickly. Some of the investors that are here, I've met, I've met you all. I've asked you how big your fund is, how many people work in your environment. You know, you'd be perfect candidates for a Quadrant, by the way, right? That's what you would be. It's, again, where are you on thinking about that? How do you use data? Are you still using thinking of data in a legacy way, or are you thinking of, "Hey, maybe I can innovate," like Mike and Ryan innovate. They're my 10% club. I'm 70, 20, 10: 70% core, 20% adjacent, 10% disruptive. Always thinking, "Hey, what's next? What's next? Let's push the limits." While this is the team that's focusing on the 70, 20, and how are we balancing that? You know, that's a very important aspect that every CEO should have in this fast-moving AI world. Any other questions? There it is, yeah. Hi, Andrew Gillies from Macquarie. just a quick one. I'd just like to understand the differences in the sort of nature of, like, a no-code app or a platform like Softr, which enables, you know, an enterprise business to build a tool or a web-based application, as opposed to something like a Compass, which, you know, might provide a little bit more functionality, but again, might take longer to deploy. Can you maybe just compare and contrast the two? Who wants to take that? Mike, you want to start, then Ryan? Yeah. When you look at things like Quadrant and that, there's gonna be a lot of different applications that can be done. As Armughan went through, there's an entire stack that gets you there, and that chatbot part on the top of that's gonna be very different for different people. An analyst might want to be working at a desktop and have access to a big screen, while it might be, if it's gonna be front-facing, it might be a little chatbot on the bottom of your screen on a website or in an app. If, like, you're in one of these apps, whether they're called superapps, they're gonna be changing their entire interfaces coming up. If you look at, like... I'm based in Asia, in Singapore, and there's a lot of superapps around here, and they're gonna be deploying this type of technology. Mm-hmm. Mm. The front end is gonna be very different. It's not gonna be, like, when you look at it, you're gonna be putting all those components together, but the application might be very different, depending on, is it like an internal thing for employees? Is it a customer-facing thing? Hey, it could be a new watch. It could be like that you have a new watch you talk to, and it's gonna have all these... We're talking about data and plugins. You have all these things that are gonna be plugging into it. It could even be your mirror talking to you when you ask for news in the morning. It could be powered by the same type of technology. It's not gonna be a one type of front interface on these deployments. Good. I think that's a wrap. No questions from you guys? I was expecting at least, like, one question. All right, good. Well, thank you for your time. Before I end, I just want to say again, Barrenjoey, thank you so much for hosting us here. We really appreciate it. Thank you. I also want to thank our team, Rosalie Duff, Jennifer Cressman, Bea. This doesn't happen like this. A lot of us show up, this just all got set up. I've been doing enough AV and tech events in the world to know that it doesn't work this way. Thank you all in the back, that you were able to get that done. These guys showed up, by the way, two hours before all of you showed up, and they turned this entire thing into this. We had animation, and we had videos and things. Can-do attitude is just amazing. We love that. Thank you, everybody. Appreciate it. Thank you for your time. Thank you, everyone on the phone. We have some refreshments for all of you outside. Hope you will join us. Thank you. Appreciate it.