Appen Limited Reports Growth in Generative AI Amid Cost-Cutting Measures

Earnings call: Appen Limited sees growth in generative AI amid cost cuts By Investing.com

Appen Limited (APX), a leading provider of high-quality training data for machine learning and artificial intelligence, reported a return to revenue growth in the First Half of Fiscal Year 2024, despite the significant setback of losing a major contract with Google (NASDAQ:GOOGL). The company's Q2 revenue increased by 16% compared to the same period last year, with H1 revenue up by 13% over H2 of the previous year. New generative AI projects from global and Chinese customers drove this growth. Despite a loss in Q1 due to the Google contract termination, Appen managed to control costs and achieve a positive underlying EBITDA in Q2, highlighting an improvement in profitability primarily through cost reduction efforts. In summary, Appen Limited is navigating a challenging period marked by losing a major contract but is showing resilience through strategic cost management and a pivot towards promising growth areas in generative AI. The company's focus on diversifying revenue and expanding its customer base, along with its commitment to innovation and efficiency in operations, positions it to capitalize on the burgeoning demand for AI and machine learning services. Operator: Thank you for standing by, and welcome to the Appen Limited First Half FY '24 Results. [Operator Instructions] I would now like to hand the conference over to Mr. Ryan Kolln, CEO and Managing Director. Please go ahead. Ryan Kolln: Thank you very much, Rachelle, and good morning, everyone. Welcome to Appen's H1 FY '24 results presentation. Today, I'm joined by our CFO, Justin Miles. There are four sections to the presentation as per the agenda on Page 3. Firstly, I'll share an overview of our H1 performance. Second, Justin will provide greater detail in the financial performance for the half. Third, I'll share an update on how we are tracking against our strategy. And finally, we will provide an update on July trading and discuss our FY '24 outlook. Moving to Page 5 in the presentation. Including the impact of the Google contract loss, we're very pleased that Appen is returning to revenue growth. In Q2, revenue grew 16% compared to the same period in FY '23 and was up 22% on Q1 this year. For the half, H1 revenue was up 13% on H2 last year if we exclude Google. H2 revenue has historically been higher than H1, so we are particularly pleased with this result. Much of the growth has been supported by new generative AI related projects, especially from our global and China customers. We are pleased with the revenue growth experience so far. Now onto Page 6. We achieved underlying EBITDA positivity in Q4 last year. However, the Google related revenue reduction impacted our profitability in Q1, resulting in a loss of $2.9 million. We acted swiftly to control costs and now have completed the previously announced $13.5 million cost reduction. While we were reducing costs, we were also winning new business which resulted in an underlying EBITDA of positive $0.6 million in Q2. This was a $7.8 million improvement compared to the same period last year. For the half, group underlying EBITDA before FX improved $13.4 million. The significant improvement is due to the cost out programs executed, with our operating expenses decreasing 33% compared to H1 2023. Profitability remains a key focus and something we will continue to manage towards. Turning now to Page 7. Generative AI continues to evolve rapidly. A major component of generative AI is high quality human data annotations, both for the training of the models and to evaluate performance. Our customers are investing heavily in generative AI and this in turn has been a major driver of revenue growth for Appen. In H1 15% of our revenue was from generative AI related projects. This was up from 6% in H2 last year. Throughout H1, the proportion of our revenue from generative AI projects has grown. In June, 28% of group revenue was from generative AI projects. This is up from 8.3% in January 2024. Finally, the number of customers we are working with continues to grow. We have 42 generative AI customers in H1, up from 28 in H2 last year. We remain very bullish on the impact of generative AI and have strong foundational capabilities to deliver high quality data to our customers. I'll now hand over to Justin who will take us through the H1 financial performance. Justin Miles: Thank you, Ryan. Good morning, everyone. A reminder that we report in U.S. dollars and that all comparisons are to the half year ended 30 June, 2023 unless stated otherwise. Starting with the H1 snapshot on Slide 9. Total revenue decreased 18% to $113.4 million, reflecting the termination of the Google contract. Excluding the impact of Google, revenue decreased by a modest 2%. Looking at our operating segments, global services revenue decreased 36% to $63.6 million. This also reflects the termination of the Google contract. New markets revenue increased 28% to $49.8 million due to strong growth in China and global product. This growth is pleasing as it reflects significant traction in multiple generative AI projects. Our gross margin percentage, which is revenue less crowd expenses, increased 0.4 percentage points to 37.7%. The increase was mainly due to a change in project and customer mix during the period. Underlying EBITDA before the impact of FX improved $13.4 million to a $2.3 million loss. The significant improvement is due to cost out programs executed during FY '23 and H1 '24. I won't talk to Slide 10 as we cover revenue in further details at later slides. Over to underlying EBITDA on Slide 11. As I just mentioned, group underlying EBITDA before FX improved $13.4 million. The significant improvement is due to cost out programs executed with our operating expenses decreasing 33% compared to H1 '23. The Global Services division reported EBITDA of $6.7 million, down 23% on the prior corresponding period. The decrease reflects lower revenue and gross margin, partially offset by the benefit of the cost out. New Markets EBITDA improved by $13.8 million to a loss of $7.9 million. The improvement is driven by growth in revenue and gross margin for China and Global Product, as well as the benefit of the cost out. Slide 12 shows monthly group revenue, underlying EBITDA and underlying cash EBITDA, both before FX. Revenue for Q2 shows positive momentum, which is pleasing as the early positive indicators of LLM-related growth has started to develop into significant opportunities. As you can see, EBITDA improved month-on-month in Q2. Turning to Slide 13. This slide shows Global revenue with Google excluded. Reduction in spend from a large customer experience during FY '23 stabilized in H2 '23, with growth returning in Q2 '24. Q1 '24 was pleasing given Q4 '23 includes some expected seasonality. Global product growth is driven by multiple generative AI projects. It is important to call out, given the early stage for some projects, volumes may be inconsistent during H2 '24 as customers adapt to their evolving needs and budgets. Global Services growth is driven by an increase in projects and volumes across multiple customers, including some early-stage LLM projects. Over to Slide 14. Our China business achieved consecutive quarterly revenue records during the half, with revenue for the half of $25.4 million, up 66% on H1 '23. The growth is driven by expansion in existing large tech customers, as well as new customer wins. Growth includes significant traction in LLM projects, and we continue to support leading LLM model builders. Slide 15 has revenue for the balance of the New Markets segment being Enterprise and Government. The decrease in revenue was driven by lower volumes within some existing large Enterprise projects. The impacted projects remain active and the customers continue to be key Appen customers. Despite the disappointing H1 results, we have conviction in the revenue opportunity. However, timing is unclear around how enterprises will proceed with their generative AI investment. The enterprise market, the generative AI labeling software remains nascent, and we're yet to see material traction but market signals remain positive. We have a healthy government pipeline and remain optimistic about the federal market. Turning to Slide 16 for a summary of the profit and loss. We've already covered most line items. However, there is additional data on this slide worth noting. Employee expenses are down 39% and all other expenses are down 21% compared to H1 '23. This is due to the cost out programs. Statutory NPAT has improved by $25.5 million due to the cost out, lower restructure costs compared to the prior period and a reduction in depreciation and amortization. For the balance sheet on Slide 17, noting the comparison here is to December '23. The cash balance at 30 June '24 was $34.7 million, up $2.6 million from December '23. The cash balance at the end of July was $30.6 million. The decrease compared to June is due to the working capital cycle. Receivables decreased due to lower revenue in Q2 '24 compared to Q4 '23. Current liabilities were $2.6 million higher, reflecting the timing of trade payables. And the decrease in net assets to $79.5 million, primarily reflects trading during the period. Turning to the cash flow summary on Slide 18. As just mentioned, the cash balance at the end of the period was $34.7 million. Cash flow from operations increased to $10.6 million, which includes a positive impact from strong Q4 '23 trading. Cash flows from investing activities were down $5.9 million to $6.5 million compared to H1 '23. The decrease is due to a lower investment in product development and reflects the cost out program executed during FY '23. Cash has primarily been used to fund operations, some CapEx and one-off costs associated with the H1 cost reduction program. That concludes the financial performance slides. I'll now hand back to Ryan. Ryan Kolln: Thank you, Justin. I'll now talk to our strategy. Moving to Slide 20. Appen plays a crucial role in the AI ecosystem. We specialize in providing high-quality data that brings human expertise into AI model development. We support our customers across data sourcing, data preparation and model development. These are all critical steps in AI model development. Slide 21 provides a much more granular view of our capabilities. I won't step through all the elements on this page, but I did want to share more detail about the role that we play. The top section of this diagram outlines some of the AI solutions that we support. The common element here is that, most of these models are imitating human behaviors, and therefore, human-generated data is a critical element to ensure that the models act as close as possible to humans. The blue box in the middle outlines some of the core services we provide across data sourcing, data preparation and model evaluation. Each project that we work on is typically unique and involves one or more of these types of services. Technology is a critical component of everything we do at Appen. Our data annotation platform enables us to deliver high-quality data for our customers and include sophisticated annotation capabilities, quality and workflow management and specialized LLM features. A key part of our platform is the ability to include AI automation in the data development process. Finally, we support a diverse set of languages and workforce models that are tailored to suit the needs of our customers. This is particularly beneficial for customers that are operating on a global scale. There's a lot on this page, and it's reflective of the broad set of capabilities that we offer our customers. On to Slide 22. The breadth and scale of our global workforce is a key differentiator for Appen. We have over one million contributors that speak over 500 languages and have specialization in over 100 domains. The benefit of our crowd is that, we can create highly specialized and customized workforce to provide AI data but also deliver it very fast and at high quality. Speed and quality are the most important elements for our customers, and our workforce enables us to maintain our competitive advantage. On to Slide 23. We've spoken about the impact of generative AI on our business, and I'd like to provide more specificity into the type of work that we were doing. At a high level, there are four categories of work we do for LLM model builders. First, we provide data for supervised fine-tuning called SFT. In SFT, we are creating unique content that is used to train large language models. An example here is when we have our workforce create sample prompts and responses. Next, we perform preference ranking. This is where humans rank multiple generative AI responses across a specific set of criteria. This allows us to use human feedback to align generative AI models with human preferences. Next, we provide human evaluation of generative AI-created content. This is where our workforce evaluates a response for a given prompt. Similar to preference ranking, the criteria for how we assess models is often unique for each project. Finally, we help our model -- we help our customers evaluate the safety of their models. A common technique here is called red teaming. The common factor across all these work types is human subjectivity. Similar to the search relevance work we have been doing for many years, the subjective feedback of humans is a critical component for effective generative AI development. Now, on to Slide 24 where I'll provide an update on our growth strategy that we outlined earlier in the year. Firstly, we are replatforming our crowd and project management platform. This is the main interface for our crowd workers and how we deliver projects for our customers. The approach that we have taken is a complete rebuild of the platform that leverages best-of-breed capabilities in the market. This is an exciting evolution for Appen and will enable us to provide a better crowd experience, reduce project setup time, automated quality monitoring and advanced analytic in our -- all of our processes. This new platform will launch in September. Second, we are optimizing and automating the processes to create data for our customers. A big focus is the inclusion of AI in our data creation and quality management processes. We are increasingly using generative AI to validate responses and to automatically data quality check. This enables us to deliver the high-quality of data and improved unit economics. It is mostly applicable for projects that are being performed on our annotation platform and like most generative AI automation, we are early in the development with a lot of testing underway. The third element of our strategy is a SaaS platform for enterprise LLM customization. The thesis here is that, enterprises will need to connect their data science team with internal experts to train and monitor LLM performance. As an example, if a retailer is building an LLM chatbot for a specific product line, the best person to provide input data are the product experts spread throughout the business. We remain optimistic about this opportunity. However, the market timing is uncertain. We are utilizing our annotation platform as the underlying technology, but there is minimal technical investment were required. We continue to test the market, and we'll look to allocate more resources as the market shows sign of acceleration. Fourth, we spoke about a modernized sales and marketing functions. We have taken a far more technical focus in our marketing approach, including how we demonstrate our expertise and thought leadership. This includes an updated brand presence, but more importantly, providing our go-to-market teams with the content that resonates with our technical customers. Finally, we have continued to have tight controls around our costs. We have reduced our operating expenses by 33% compared to H1 FY '23, while also supporting new revenue growth. Justin and I continue to look for additional areas of cost optimization across the business. I'll now provide a July trading update and an outlook statement. Page 26 shows our July performance. Revenue for July was $17.6 million. While this was down from a strong result in June, we are very pleased that it's 26% up on July 2023, excluding Google. As Justin mentioned earlier, it's worth noting that we may see some lumpy revenue on a monthly basis due to the nature of some exploratory generative AI projects. Underlying EBITDA before FX was breakeven for July. June data included the write-back of share-based payment expenses, hence, why July is lower than June. Finally, underlying cash EBITDA before FX improved to a loss of only $0.3 million in July. We are pleased that we continue to trend in the right direction to profitability despite some lumpiness in revenue. On to Slide 27 where I'll provide an outlook statement. Excluding the impact of Google, revenue momentum is positive. We continue to see positive signals on LLM-related growth, including from our Global and China customers. Tight cost controls remain in place, in keeping with the company's focus on managing costs in line with the revenue opportunity. FY '24 will be the full year benefit of the $60 million FY '23 cost reduction program. And on the 12th of February this year, we announced a further $13.5 million of cost out initiatives that are now complete. We remain highly focused on ongoing cash EBITDA positivity, and our target is to become underlying cash EBITDA positive on a run rate basis in early H2 FY '24. Thank you. That concludes our presentation. I'll now hand back to Rachel, who will take questions. Operator: [Operator Instructions] The first question comes from Josh Kannourakis with Barrenjoey. Please go ahead. Josh Kannourakis: Just first question and well done on the growth in the gen AI space. Can you give us a bit more detail maybe just around, I guess, the margin structure of some of those projects when you are discussing things in those areas, just whether -- how the sort of visibility elements compare maybe with some of the previous projects you've had with the big global technology companies? And that's the first part. Ryan Kolln: Yes, I'd say you can sit in the margin quite similar on -- across the average of the projects we're doing with generative AI. We do see some work being done in domain specialization. So the dollar margin might be higher because we -- the people that we get that are domain-specific a charge higher than our generics crowd. But the margins on the whole are roughly similar. One thing to call out though is some of the work that we do is very fast and very rapid in generative AI. So there will be some projects that are higher margin, some projects that a lower margin, but on the net in the long term, we can consider the margin being very similar. Josh Kannourakis: Got it. And just in terms of project structure at the moment, like historically, I guess, you've had service agreements or SLA's with some of these players and they'll come in. What sort of -- what's the structure of the procurement pathway at the moment within some of these companies? And, I guess, how does it differ maybe from what you've sort of been used to previously? Or is it the same? Ryan Kolln: Again, at the whole, it's relatively similar. There will be quality metrics and time metrics that we need to hit. So not a huge deviation from what we're looking at. I think what we're seeing in the generative AI space, it's -- our customers are looking for data very quickly. So our ability to ramp up very fast deliver high-quality data is what's really important to them. And we've been doing this for a very long time, and we can ramp up very quickly, high-quality data is in our DNA. So what we're finding is that, the capabilities that we have are highly suited to the needs of generative AI customers. Josh Kannourakis: Great. And final one on the gen AI side of things. In terms of the competitive dynamics there, who are you seeing that you're coming up against more broadly? Because it does feel like the market there is segmented into some of the lower quality, high-volume work. And then also, as you mentioned, some of the domain-specific providers, which are also obviously getting a lot of work at various parts of the LLM model. So can you just get a feeling of who your sort of core competitors are in your view and how those competitive tensions or dynamics have changed at all since we caught up early this year? Ryan Kolln: Yes. So the focus of the model built is on very high-quality data. And in many customers, in many projects that we do, the quality bar of generative AI projects is significantly higher to the work that we've done in the past. I would say that the competitive set is largely the same as what we've experienced in the sort of traditional data annotation market. But as you call out, there has been some evolution. Domain-specific work has become a larger part of the generative AI market. So as an example, looking for people with coding background, mathematics, science, et cetera, which is a slight deviation, but I'd call it like a 15-degree shift from some of the work that we've been doing in the past. So many of the competitors that we're facing are largely the same. We do see some competitors who knew that have highly specialized workforce in a specific area, and they will do well in that specific domain, but we don't see them branching out into a broader set of capabilities. Josh Kannourakis: That's very helpful. And then just one on the numbers. In terms of the cost base, just -- so you obviously mentioned that you've done a good job getting the cost out as you've targeted. How should we be thinking about the sort of cost base into the second half and beyond? And, I guess, any potential for flex either whether you need to reinvest for growth further down the track or if there's further efficiencies that could be gained in the business? Justin Miles: Hi Josh, yes. So the cost that we took out during H1, 80% was done by March. And if you look at kind of the Google revenue coming out, you can kind of see the timing of that. So the majority of that would have been in March, with the balance in the second quarter. But what Ryan and I need to do very carefully is make sure we resource to deliver on any growth. So we're running this fine line at the moment to make sure we're efficiently managing our costs. But I think H1 is probably a good proxy or a good starting point, but there may be a bit of flex up or down as needed. Josh Kannourakis: Got it. No, that's really helpful. I'll pass it on to someone else. Thanks, guys. Appreciate it. Justin Miles: Thanks Josh. Operator: Your next question comes from Wei Sim with Jefferies. Please go ahead. Wei Sim: Hi Ryan. Hi Justin. In terms of just the revenue outlook, I appreciate we've called out some volatility and July was down a bit. But just in terms of the outlook into the second half of the year, should we expect something similar to what we had historically seen in terms of quarter-on-quarter build in terms of revenues into the back end of the calendar year? Or do you think with the different mix of customers and stuff that this profile looks different now? Ryan Kolln: Look, we're moving into -- there has been a bit of volatility in the past. So I think some of the trends that we've seen may or may not play out. We've had seasonality, as we've discussed before. But again, we're looking -- we're not providing specific details on the shape of the revenue. We're very focused on the customer. So we'll need to see how the second half of the year plays out. Wei Sim: Yes. Okay. Great. And then just in terms of the contract structure that we have now, I think one of the things that we had before was an issue with maybe some of the larger tech companies where they kind of like gave indications at the start of the year. And then we didn't see them necessarily hit those volumes and there was no kind of like recourse. Have we changed the contract structure with the customers that we've signed either existing or new customers? Or are we still looking at a similar kind of contract structure where there could be risk to the volumes which customers are committing to or indicating? Ryan Kolln: Likely the same contract structure. But I think what you can see is the stabilization in the spend. So we're not seeing that ongoing decline that we had in FY '23 in particular. So I think as a lot of our customers are stabilizing on the back of their generative AI focus, we're starting to see stabilization and growth in their spend. Wei Sim: All right. And there's no kind of like thoughts on, I guess, bolstering the contract structure such that we get a bit more visibility going forward or... Ryan Kolln: Well, right now, it's -- the market is moving very quickly. So our focus is on staying as close as we can to our customers so that we can adjust with them and support them as their needs change. So it's not a huge focus for us right now. Our main focus is staying super close to the customers and being as responsive as possible to get them the high-quality data that they need. Wei Sim: Right. Okay. That makes sense. And then just in terms of Google, losing them as a customer, you've pointed out that your services are quintessential to AI. Do you know how they're going about training their models now and where they're sourcing, I guess, their services from? Ryan Kolln: We don't have details there. I think in general, from a market standpoint, human annotated data is very important for AI any model development that's trying to replicate human behaviors. So I would assume that they are continuing with sourcing human annotated data. Wei Sim: Okay. And then when we talk about just like the Global Product and the volumes may be inconsistent in H2 '24. Can you elaborate on that comment a bit, please? Justin Miles: Yes. So Ryan alluded to it, and I did as well, that because this is new pieces of work for the customer, there might be a huge need for a certain amount of data and then there may be a pause or rethink while they evaluate what the ongoing needs are. So we're seeing a little bit of inconsistency there as these projects ramp up before they may go into full production or into a kind of a stability mode. So we are seeing that in consistency, and that's what we're calling out. Ryan Kolln: Just to add to that point, I think what's pleased with seeing some of the large global customers come to us for projects to be delivered on our ADAP platform, where we have a lot more control around the processes and the quality, and hence, why you see that big uptick in Global Product revenue. Wei Sim: Okay. Maybe the last one for me. Just in terms of the China revenues right now, looking very strong. Are you able to give any color as to how much is coming from Mainland China versus Japan and Korea? Ryan Kolln: Yes, we don't break it down. The bulk of it's coming from China-based customers. Operator: [Operator Instructions] Your next question comes from [Karen Goddard], a Private Investor. Please go ahead. Unidentified Analyst: Hi Ryan and Justin, it's very good to meet you guys. So it's good to see this deconcentration away from big tech customers, 68% revenue versus 77% revenue in FY '23. Will this trend continue to accelerate? Ryan Kolln: Yes. We're certainly focused on revenue diversification and looking for to expand our customer base and expand our revenue. But right now, we're focused on revenue growth in the business and getting back to profitability. So I'd say that's the key focus. But I think in the long run, as we see more companies investing, particularly in generative AI, it creates a really great opportunity for us to have a much more diversified revenue base. Unidentified Analyst: Okay. So second half of the question is, is the big tech revenue still growing while the deconcentration is happening? Ryan Kolln: Well, I think you can see from the Global revenue, there's an improvement when you exclude Google. So I think that the short answer is, what we've seen is yes. Unidentified Analyst: Okay. So I noticed that there was a lot more R&D in FY '23. But now there's considerably lower R&D. Is this because we are heading into the capital distribution phase where all the investing has done and now we are waiting for companies to start adopting generative AI? Justin Miles: Yes, it's a good call out. So the spend has dropped considerably. So the quantum has. And that's largely driven by the cost out program from last year. And what's important to note here is, while the quantum decrease, the quality of the output didn't necessarily and the reason for that is, one of the main strategies was to open up an engineering hub in India, which has proven to be really successful for us, whereas historically, we had engineering resources on the West Coast of U.S.A., just how we've organically grown up. So this is a much more cost-efficient structure for us, and it's yielding really good results for us. We're really pleased with it. Unidentified Analyst: Yes. That was a very good development in Hyderabad, what you guys did. Justin Miles: Yes. Unidentified Analyst: So my next question is, so I was looking at one of the interviews by Vinay from Figure Eight Federal, which is a subsidiary of Appen. So he mentioned that 80% of the top 100 Fortune 500 companies are clients of Figure Eight Federal. Can you tell us more about this? Ryan Kolln: Yes. So a Figure Eight Federal focused on the Federal division. So there may have been -- I'm not sure if that was the intent of Vinay's comment there. Unidentified Analyst: Okay. So my last question is, do you have any idea about the Hydra platform made by Figure Eight Federal? I mean, I was just looking online on their website. Ryan Kolln: Yes, sure. That's a location data product that came through the Quadrant acquisition, and that the Figure Eight Federal team has found opportunities in the Federal market. So we're pursuing that there's a lot of go-to-market activity around promoting our location data capabilities. Unidentified Analyst: Okay. And is that still growing in the automated vehicle space or... Ryan Kolln: So data is mobile location analytics and yet, we see continued growth and opportunity in that space. Operator: There are no further questions at this time. That does conclude our conference for today. Thank you for participating. You may now disconnect.

Appen Limited (APPEF) Q2 2024 Earnings Call Transcript

Ryan Kolln - Chief Executive Officer and Managing Director Justin Miles - Chief Financial Officer Thank you for standing by, and welcome to the Appen Limited First Half FY '24 Results. [Operator Instructions] I would now like to hand the conference over to Mr. Ryan Kolln, CEO and Managing Director. Please go ahead. Ryan Kolln Thank you very much, Rachelle, and good morning, everyone. Welcome to Appen's H1 FY '24 results presentation. Today, I'm joined by our CFO, Justin Miles. There are four sections to the presentation as per the agenda on Page 3. Firstly, I'll share an overview of our H1 performance. Second, Justin will provide greater detail in the financial performance for the half. Third, I'll share an update on how we are tracking against our strategy. And finally, we will provide an update on July trading and discuss our FY '24 outlook. Moving to Page 5 in the presentation. Including the impact of the Google contract loss, we're very pleased that Appen is returning to revenue growth. In Q2, revenue grew 16% compared to the same period in FY '23 and was up 22% on Q1 this year. For the half, H1 revenue was up 13% on H2 last year if we exclude Google. H2 revenue has historically been higher than H1, so we are particularly pleased with this result. Much of the growth has been supported by new generative AI related projects, especially from our global and China customers. We are pleased with the revenue growth experience so far. Now onto Page 6. We achieved underlying EBITDA positivity in Q4 last year. However, the Google related revenue reduction impacted our profitability in Q1, resulting in a loss of $2.9 million. We acted swiftly to control costs and now have completed the previously announced $13.5 million cost reduction. While we were reducing costs, we were also winning new business which resulted in an underlying EBITDA of positive $0.6 million in Q2. This was a $7.8 million improvement compared to the same period last year. For the half, group underlying EBITDA before FX improved $13.4 million. The significant improvement is due to the cost out programs executed, with our operating expenses decreasing 33% compared to H1 2023. Profitability remains a key focus and something we will continue to manage towards. Turning now to Page 7. Generative AI continues to evolve rapidly. A major component of generative AI is high quality human data annotations, both for the training of the models and to evaluate performance. Our customers are investing heavily in generative AI and this in turn has been a major driver of revenue growth for Appen. In H1 15% of our revenue was from generative AI related projects. This was up from 6% in H2 last year. Throughout H1, the proportion of our revenue from generative AI projects has grown. In June, 28% of group revenue was from generative AI projects. This is up from 8.3% in January 2024. Finally, the number of customers we are working with continues to grow. We have 42 generative AI customers in H1, up from 28 in H2 last year. We remain very bullish on the impact of generative AI and have strong foundational capabilities to deliver high quality data to our customers. I'll now hand over to Justin who will take us through the H1 financial performance. A reminder that we report in U.S. dollars and that all comparisons are to the half year ended 30 June, 2023 unless stated otherwise. Starting with the H1 snapshot on Slide 9. Total revenue decreased 18% to $113.4 million, reflecting the termination of the Google contract. Excluding the impact of Google, revenue decreased by a modest 2%. Looking at our operating segments, global services revenue decreased 36% to $63.6 million. This also reflects the termination of the Google contract. New markets revenue increased 28% to $49.8 million due to strong growth in China and global product. This growth is pleasing as it reflects significant traction in multiple generative AI projects. Our gross margin percentage, which is revenue less crowd expenses, increased 0.4 percentage points to 37.7%. The increase was mainly due to a change in project and customer mix during the period. Underlying EBITDA before the impact of FX improved $13.4 million to a $2.3 million loss. The significant improvement is due to cost out programs executed during FY '23 and H1 '24. I won't talk to Slide 10 as we cover revenue in further details at later slides. Over to underlying EBITDA on Slide 11. As I just mentioned, group underlying EBITDA before FX improved $13.4 million. The significant improvement is due to cost out programs executed with our operating expenses decreasing 33% compared to H1 '23. The Global Services division reported EBITDA of $6.7 million, down 23% on the prior corresponding period. The decrease reflects lower revenue and gross margin, partially offset by the benefit of the cost out. New Markets EBITDA improved by $13.8 million to a loss of $7.9 million. The improvement is driven by growth in revenue and gross margin for China and Global Product, as well as the benefit of the cost out. Slide 12 shows monthly group revenue, underlying EBITDA and underlying cash EBITDA, both before FX. Revenue for Q2 shows positive momentum, which is pleasing as the early positive indicators of LLM-related growth has started to develop into significant opportunities. As you can see, EBITDA improved month-on-month in Q2. Turning to Slide 13. This slide shows Global revenue with Google excluded. Reduction in spend from a large customer experience during FY '23 stabilized in H2 '23, with growth returning in Q2 '24. Q1 '24 was pleasing given Q4 '23 includes some expected seasonality. Global product growth is driven by multiple generative AI projects. It is important to call out, given the early stage for some projects, volumes may be inconsistent during H2 '24 as customers adapt to their evolving needs and budgets. Global Services growth is driven by an increase in projects and volumes across multiple customers, including some early-stage LLM projects. Over to Slide 14. Our China business achieved consecutive quarterly revenue records during the half, with revenue for the half of $25.4 million, up 66% on H1 '23. The growth is driven by expansion in existing large tech customers, as well as new customer wins. Growth includes significant traction in LLM projects, and we continue to support leading LLM model builders. Slide 15 has revenue for the balance of the New Markets segment being Enterprise and Government. The decrease in revenue was driven by lower volumes within some existing large Enterprise projects. The impacted projects remain active and the customers continue to be key Appen customers. Despite the disappointing H1 results, we have conviction in the revenue opportunity. However, timing is unclear around how enterprises will proceed with their generative AI investment. The enterprise market, the generative AI labeling software remains nascent, and we're yet to see material traction but market signals remain positive. We have a healthy government pipeline and remain optimistic about the federal market. Turning to Slide 16 for a summary of the profit and loss. We've already covered most line items. However, there is additional data on this slide worth noting. Employee expenses are down 39% and all other expenses are down 21% compared to H1 '23. This is due to the cost out programs. Statutory NPAT has improved by $25.5 million due to the cost out, lower restructure costs compared to the prior period and a reduction in depreciation and amortization. For the balance sheet on Slide 17, noting the comparison here is to December '23. The cash balance at 30 June '24 was $34.7 million, up $2.6 million from December '23. The cash balance at the end of July was $30.6 million. The decrease compared to June is due to the working capital cycle. Receivables decreased due to lower revenue in Q2 '24 compared to Q4 '23. Current liabilities were $2.6 million higher, reflecting the timing of trade payables. And the decrease in net assets to $79.5 million, primarily reflects trading during the period. Turning to the cash flow summary on Slide 18. As just mentioned, the cash balance at the end of the period was $34.7 million. Cash flow from operations increased to $10.6 million, which includes a positive impact from strong Q4 '23 trading. Cash flows from investing activities were down $5.9 million to $6.5 million compared to H1 '23. The decrease is due to a lower investment in product development and reflects the cost out program executed during FY '23. Cash has primarily been used to fund operations, some CapEx and one-off costs associated with the H1 cost reduction program. That concludes the financial performance slides. I'll now hand back to Ryan. I'll now talk to our strategy. Moving to Slide 20. Appen plays a crucial role in the AI ecosystem. We specialize in providing high-quality data that brings human expertise into AI model development. We support our customers across data sourcing, data preparation and model development. These are all critical steps in AI model development. Slide 21 provides a much more granular view of our capabilities. I won't step through all the elements on this page, but I did want to share more detail about the role that we play. The top section of this diagram outlines some of the AI solutions that we support. The common element here is that, most of these models are imitating human behaviors, and therefore, human-generated data is a critical element to ensure that the models act as close as possible to humans. The blue box in the middle outlines some of the core services we provide across data sourcing, data preparation and model evaluation. Each project that we work on is typically unique and involves one or more of these types of services. Technology is a critical component of everything we do at Appen. Our data annotation platform enables us to deliver high-quality data for our customers and include sophisticated annotation capabilities, quality and workflow management and specialized LLM features. A key part of our platform is the ability to include AI automation in the data development process. Finally, we support a diverse set of languages and workforce models that are tailored to suit the needs of our customers. This is particularly beneficial for customers that are operating on a global scale. There's a lot on this page, and it's reflective of the broad set of capabilities that we offer our customers. On to Slide 22. The breadth and scale of our global workforce is a key differentiator for Appen. We have over one million contributors that speak over 500 languages and have specialization in over 100 domains. The benefit of our crowd is that, we can create highly specialized and customized workforce to provide AI data but also deliver it very fast and at high quality. Speed and quality are the most important elements for our customers, and our workforce enables us to maintain our competitive advantage. On to Slide 23. We've spoken about the impact of generative AI on our business, and I'd like to provide more specificity into the type of work that we were doing. At a high level, there are four categories of work we do for LLM model builders. First, we provide data for supervised fine-tuning called SFT. In SFT, we are creating unique content that is used to train large language models. An example here is when we have our workforce create sample prompts and responses. Next, we perform preference ranking. This is where humans rank multiple generative AI responses across a specific set of criteria. This allows us to use human feedback to align generative AI models with human preferences. Next, we provide human evaluation of generative AI-created content. This is where our workforce evaluates a response for a given prompt. Similar to preference ranking, the criteria for how we assess models is often unique for each project. Finally, we help our model -- we help our customers evaluate the safety of their models. A common technique here is called red teaming. The common factor across all these work types is human subjectivity. Similar to the search relevance work we have been doing for many years, the subjective feedback of humans is a critical component for effective generative AI development. Now, on to Slide 24 where I'll provide an update on our growth strategy that we outlined earlier in the year. Firstly, we are replatforming our crowd and project management platform. This is the main interface for our crowd workers and how we deliver projects for our customers. The approach that we have taken is a complete rebuild of the platform that leverages best-of-breed capabilities in the market. This is an exciting evolution for Appen and will enable us to provide a better crowd experience, reduce project setup time, automated quality monitoring and advanced analytic in our -- all of our processes. This new platform will launch in September. Second, we are optimizing and automating the processes to create data for our customers. A big focus is the inclusion of AI in our data creation and quality management processes. We are increasingly using generative AI to validate responses and to automatically data quality check. This enables us to deliver the high-quality of data and improved unit economics. It is mostly applicable for projects that are being performed on our annotation platform and like most generative AI automation, we are early in the development with a lot of testing underway. The third element of our strategy is a SaaS platform for enterprise LLM customization. The thesis here is that, enterprises will need to connect their data science team with internal experts to train and monitor LLM performance. As an example, if a retailer is building an LLM chatbot for a specific product line, the best person to provide input data are the product experts spread throughout the business. We remain optimistic about this opportunity. However, the market timing is uncertain. We are utilizing our annotation platform as the underlying technology, but there is minimal technical investment were required. We continue to test the market, and we'll look to allocate more resources as the market shows sign of acceleration. Fourth, we spoke about a modernized sales and marketing functions. We have taken a far more technical focus in our marketing approach, including how we demonstrate our expertise and thought leadership. This includes an updated brand presence, but more importantly, providing our go-to-market teams with the content that resonates with our technical customers. Finally, we have continued to have tight controls around our costs. We have reduced our operating expenses by 33% compared to H1 FY '23, while also supporting new revenue growth. Justin and I continue to look for additional areas of cost optimization across the business. I'll now provide a July trading update and an outlook statement. Page 26 shows our July performance. Revenue for July was $17.6 million. While this was down from a strong result in June, we are very pleased that it's 26% up on July 2023, excluding Google. As Justin mentioned earlier, it's worth noting that we may see some lumpy revenue on a monthly basis due to the nature of some exploratory generative AI projects. Underlying EBITDA before FX was breakeven for July. June data included the write-back of share-based payment expenses, hence, why July is lower than June. Finally, underlying cash EBITDA before FX improved to a loss of only $0.3 million in July. We are pleased that we continue to trend in the right direction to profitability despite some lumpiness in revenue. On to Slide 27 where I'll provide an outlook statement. Excluding the impact of Google, revenue momentum is positive. We continue to see positive signals on LLM-related growth, including from our Global and China customers. Tight cost controls remain in place, in keeping with the company's focus on managing costs in line with the revenue opportunity. FY '24 will be the full year benefit of the $60 million FY '23 cost reduction program. And on the 12th of February this year, we announced a further $13.5 million of cost out initiatives that are now complete. We remain highly focused on ongoing cash EBITDA positivity, and our target is to become underlying cash EBITDA positive on a run rate basis in early H2 FY '24. Thank you. That concludes our presentation. I'll now hand back to Rachel, who will take questions. [Operator Instructions] The first question comes from Josh Kannourakis with Barrenjoey. Please go ahead. Just first question and well done on the growth in the gen AI space. Can you give us a bit more detail maybe just around, I guess, the margin structure of some of those projects when you are discussing things in those areas, just whether -- how the sort of visibility elements compare maybe with some of the previous projects you've had with the big global technology companies? And that's the first part. Yes, I'd say you can sit in the margin quite similar on -- across the average of the projects we're doing with generative AI. We do see some work being done in domain specialization. So the dollar margin might be higher because we -- the people that we get that are domain-specific a charge higher than our generics crowd. But the margins on the whole are roughly similar. One thing to call out though is some of the work that we do is very fast and very rapid in generative AI. So there will be some projects that are higher margin, some projects that a lower margin, but on the net in the long term, we can consider the margin being very similar. Josh Kannourakis Got it. And just in terms of project structure at the moment, like historically, I guess, you've had service agreements or SLA's with some of these players and they'll come in. What sort of -- what's the structure of the procurement pathway at the moment within some of these companies? And, I guess, how does it differ maybe from what you've sort of been used to previously? Or is it the same? Ryan Kolln Again, at the whole, it's relatively similar. There will be quality metrics and time metrics that we need to hit. So not a huge deviation from what we're looking at. I think what we're seeing in the generative AI space, it's -- our customers are looking for data very quickly. So our ability to ramp up very fast deliver high-quality data is what's really important to them. And we've been doing this for a very long time, and we can ramp up very quickly, high-quality data is in our DNA. So what we're finding is that, the capabilities that we have are highly suited to the needs of generative AI customers. Josh Kannourakis Great. And final one on the gen AI side of things. In terms of the competitive dynamics there, who are you seeing that you're coming up against more broadly? Because it does feel like the market there is segmented into some of the lower quality, high-volume work. And then also, as you mentioned, some of the domain-specific providers, which are also obviously getting a lot of work at various parts of the LLM model. So can you just get a feeling of who your sort of core competitors are in your view and how those competitive tensions or dynamics have changed at all since we caught up early this year? Ryan Kolln Yes. So the focus of the model built is on very high-quality data. And in many customers, in many projects that we do, the quality bar of generative AI projects is significantly higher to the work that we've done in the past. I would say that the competitive set is largely the same as what we've experienced in the sort of traditional data annotation market. But as you call out, there has been some evolution. Domain-specific work has become a larger part of the generative AI market. So as an example, looking for people with coding background, mathematics, science, et cetera, which is a slight deviation, but I'd call it like a 15-degree shift from some of the work that we've been doing in the past. So many of the competitors that we're facing are largely the same. We do see some competitors who knew that have highly specialized workforce in a specific area, and they will do well in that specific domain, but we don't see them branching out into a broader set of capabilities. Josh Kannourakis That's very helpful. And then just one on the numbers. In terms of the cost base, just -- so you obviously mentioned that you've done a good job getting the cost out as you've targeted. How should we be thinking about the sort of cost base into the second half and beyond? And, I guess, any potential for flex either whether you need to reinvest for growth further down the track or if there's further efficiencies that could be gained in the business? Justin Miles Hi Josh, yes. So the cost that we took out during H1, 80% was done by March. And if you look at kind of the Google revenue coming out, you can kind of see the timing of that. So the majority of that would have been in March, with the balance in the second quarter. But what Ryan and I need to do very carefully is make sure we resource to deliver on any growth. So we're running this fine line at the moment to make sure we're efficiently managing our costs. But I think H1 is probably a good proxy or a good starting point, but there may be a bit of flex up or down as needed. Josh Kannourakis Got it. No, that's really helpful. I'll pass it on to someone else. Thanks, guys. Appreciate it. Your next question comes from Wei Sim with Jefferies. Please go ahead. Wei Sim Hi Ryan. Hi Justin. In terms of just the revenue outlook, I appreciate we've called out some volatility and July was down a bit. But just in terms of the outlook into the second half of the year, should we expect something similar to what we had historically seen in terms of quarter-on-quarter build in terms of revenues into the back end of the calendar year? Or do you think with the different mix of customers and stuff that this profile looks different now? Ryan Kolln Look, we're moving into -- there has been a bit of volatility in the past. So I think some of the trends that we've seen may or may not play out. We've had seasonality, as we've discussed before. But again, we're looking -- we're not providing specific details on the shape of the revenue. We're very focused on the customer. So we'll need to see how the second half of the year plays out. Wei Sim Yes. Okay. Great. And then just in terms of the contract structure that we have now, I think one of the things that we had before was an issue with maybe some of the larger tech companies where they kind of like gave indications at the start of the year. And then we didn't see them necessarily hit those volumes and there was no kind of like recourse. Have we changed the contract structure with the customers that we've signed either existing or new customers? Or are we still looking at a similar kind of contract structure where there could be risk to the volumes which customers are committing to or indicating? Ryan Kolln Likely the same contract structure. But I think what you can see is the stabilization in the spend. So we're not seeing that ongoing decline that we had in FY '23 in particular. So I think as a lot of our customers are stabilizing on the back of their generative AI focus, we're starting to see stabilization and growth in their spend. Wei Sim All right. And there's no kind of like thoughts on, I guess, bolstering the contract structure such that we get a bit more visibility going forward or... Ryan Kolln Well, right now, it's -- the market is moving very quickly. So our focus is on staying as close as we can to our customers so that we can adjust with them and support them as their needs change. So it's not a huge focus for us right now. Our main focus is staying super close to the customers and being as responsive as possible to get them the high-quality data that they need. Wei Sim Right. Okay. That makes sense. And then just in terms of Google, losing them as a customer, you've pointed out that your services are quintessential to AI. Do you know how they're going about training their models now and where they're sourcing, I guess, their services from? Ryan Kolln We don't have details there. I think in general, from a market standpoint, human annotated data is very important for AI any model development that's trying to replicate human behaviors. So I would assume that they are continuing with sourcing human annotated data. Wei Sim Okay. And then when we talk about just like the Global Product and the volumes may be inconsistent in H2 '24. Can you elaborate on that comment a bit, please? Justin Miles Yes. So Ryan alluded to it, and I did as well, that because this is new pieces of work for the customer, there might be a huge need for a certain amount of data and then there may be a pause or rethink while they evaluate what the ongoing needs are. So we're seeing a little bit of inconsistency there as these projects ramp up before they may go into full production or into a kind of a stability mode. So we are seeing that in consistency, and that's what we're calling out. Ryan Kolln Just to add to that point, I think what's pleased with seeing some of the large global customers come to us for projects to be delivered on our ADAP platform, where we have a lot more control around the processes and the quality, and hence, why you see that big uptick in Global Product revenue. Wei Sim Okay. Maybe the last one for me. Just in terms of the China revenues right now, looking very strong. Are you able to give any color as to how much is coming from Mainland China versus Japan and Korea? Ryan Kolln Yes, we don't break it down. The bulk of it's coming from China-based customers. [Operator Instructions] Your next question comes from [Karen Goddard], a Private Investor. Please go ahead. Unidentified Analyst Hi Ryan and Justin, it's very good to meet you guys. So it's good to see this deconcentration away from big tech customers, 68% revenue versus 77% revenue in FY '23. Will this trend continue to accelerate? Ryan Kolln Yes. We're certainly focused on revenue diversification and looking for to expand our customer base and expand our revenue. But right now, we're focused on revenue growth in the business and getting back to profitability. So I'd say that's the key focus. But I think in the long run, as we see more companies investing, particularly in generative AI, it creates a really great opportunity for us to have a much more diversified revenue base. Unidentified Analyst Okay. So second half of the question is, is the big tech revenue still growing while the deconcentration is happening? Ryan Kolln Well, I think you can see from the Global revenue, there's an improvement when you exclude Google. So I think that the short answer is, what we've seen is yes. Unidentified Analyst Okay. So I noticed that there was a lot more R&D in FY '23. But now there's considerably lower R&D. Is this because we are heading into the capital distribution phase where all the investing has done and now we are waiting for companies to start adopting generative AI? Justin Miles Yes, it's a good call out. So the spend has dropped considerably. So the quantum has. And that's largely driven by the cost out program from last year. And what's important to note here is, while the quantum decrease, the quality of the output didn't necessarily and the reason for that is, one of the main strategies was to open up an engineering hub in India, which has proven to be really successful for us, whereas historically, we had engineering resources on the West Coast of U.S.A., just how we've organically grown up. So this is a much more cost-efficient structure for us, and it's yielding really good results for us. We're really pleased with it. Unidentified Analyst Yes. That was a very good development in Hyderabad, what you guys did. So my next question is, so I was looking at one of the interviews by Vinay from Figure Eight Federal, which is a subsidiary of Appen. So he mentioned that 80% of the top 100 Fortune 500 companies are clients of Figure Eight Federal. Can you tell us more about this? Ryan Kolln Yes. So a Figure Eight Federal focused on the Federal division. So there may have been -- I'm not sure if that was the intent of Vinay's comment there. Unidentified Analyst Okay. So my last question is, do you have any idea about the Hydra platform made by Figure Eight Federal? I mean, I was just looking online on their website. Ryan Kolln Yes, sure. That's a location data product that came through the Quadrant acquisition, and that the Figure Eight Federal team has found opportunities in the Federal market. So we're pursuing that there's a lot of go-to-market activity around promoting our location data capabilities. Unidentified Analyst Okay. And is that still growing in the automated vehicle space or... Ryan Kolln So data is mobile location analytics and yet, we see continued growth and opportunity in that space. There are no further questions at this time. That does conclude our conference for today. Thank you for participating. You may now disconnect.