Appen Limited

Good morning everyone and welcome to Abbott's first half FY26 results webinar. I'm Sam Wells from NWR and joining me from the company today is CEO and Managing Director Ryan Cole and Chief Financial Officer Justin Miles. Following a brief summary of the results released to the ASX this morning, we will have some time for Q&A with the management team. There will be a choice of two options. First, research analysts are able to raise their hand should you wish to ask a verbal question of the management team, or will also take written questions via the Q&A function at the bottom of your screen throughout today's presentation. We'll endeavour to get to the majority of questions asked, in some cases combining questions on the same or similar topic. And with that, I'll pass it over to you, Ryan.

Thanks, Sam, and good morning, everyone. Thanks for joining us today for our H1 FY26 results presentation. My name is Jaron Collin. I'm the CR Managing Director at Appen, and with me is Justin Miles, our CFO. Today's presentation covers four sections. I'll start with the results overview. Justin will then walk through the detailed H1 FY26 financial performance. Then I'll return to cover our strategy and operational update, and we will close with an FY26 outlook and guidance before opening up the questions. Let me start now with an overview of our H1 results on page 5 of the presentation. The first half delivered on the themes that we've been building towards. At the group level, we delivered $119.9 million in revenue. This is 17% growth on the prior corresponding period. Appen China was a standout for the half. Revenue grew 80% to $76.2 million and achieved an annualized revenue run rate exceeding $175 million in June. That is up from $135 million at the end of 2025. The sustained growth reflects the strength of our relationships with the Chinese model builders and the ongoing demand for our data services in the region. Appen Global continues to make progress. Outside of our largest client, Q2 revenue grew 65% on Q1. We're progressing well in our ambition to expand across frontier AI labs in the USA. Capturing cost efficiencies via AI-enabled operations remains a focus for Appen Global. We have identified an incremental $12 million in operational efficiencies. 70% of the $12 million will be executed before the end of the year and the remainder in Q1 FY27. It's important to note that the cost gaps don't impact the ability to grow within Appen Global. On profitability, we delivered an underlying EBITDA of 4FX of $5.3 million for the half, a $7.5 million improvement on the first half of last year. EBITDA margins for the half were 4.5%. Our cash balance at 30 June was $44.7 million, equivalent to $64.8 million Australian dollars. I'll now hand over to Justin, and he'll take us through the financials.

Thank you, Ryan, and good morning, everybody. A reminder that we report in US dollars. and that all comparisons are to the half year ended 30 June 2025 unless stated otherwise. Starting with the H1 FY26 profit and loss on page 7. As Ryan already mentioned, revenue increased 17.5% to $119.9 million. Within our operating segments, Appen China revenue grew by 80.4% to $76.2 million, with Appen Global down 26.9% to $43.7 million. Appen Global made solid progress during the half. However, growth in new areas has not yet offset a reduction in traditional work. Gross margin reduced slightly, down 30 basis points to 36.7%. The decrease reflects a change in customer and project mix, noting China margins are traditionally lower compared to Appen Global. Underlying EBITDA before FX improved $7.5 million to $5.3 million. The increase reflects revenue and gross margin growth, prudent cost management and operational leverage within Appen China. I won't talk to slide 8 as we have just covered this data, so over to Appen Global Revenue and EBITDA on slide 9. The chart on the left shows quarterly revenue and on the right hand side it is underlying EBITDA. The charts demonstrate the progress made during the half. Q2 FY26 revenue reflects growth from expanding projects with leading AI labs. Pleasingly, outside the largest customer, Q2 FY26 revenue grew 65% compared to Q1 FY26. However, as just mentioned, growth in new areas has not yet offset a reduction in traditional work, resulting in revenue of $43.7 million for the half, which is down compared to H1 FY25. An important point to note is that traditional work has currently stabilised. EBITDA reflects some investment in winning new customer projects and does show improvement during the period. Ryan has already mentioned this, but again noting that approximately $12 million in operational efficiencies have been identified in the app and global segment, with 70% to be executed by the end of this year and the remainder in Q1 next year. Over to slide 10, which shows quarterly revenue and underlying EBITDA for Appen China, and reflects the strong market position Appen China continues to hold. Revenue grew each quarter, with Appen China achieving 76.2 million revenue for H1FY26, which was 80% growth on H1FY25. Growth continues to be driven by new and expanding LLM-related projects. Appen China exited the half with annualized revenue exceeding $175 million. Pleasingly, in addition to revenue growth, profitability has improved, with increased gross margins due to a greater mix of Gen AI projects and increased revenue from high-margin pre-built datasets. Appen China is also capturing scaling efficiencies due to tight OPEX controls as revenue expands. Turning to slide 11 for the profit and loss summary. I won't talk to all line items, however there are a few additional points to highlight. There was a decrease in employee and other expenses in Appenglobal, and this is highlighted later in the presentation. Employee expenses for Appenglobal were down 19% on PCP, and other expenses down 29% on PCP. The decrease was achieved through technology, innovation and automation. The decrease in Appen Global was offset by additional expense from the Appen China segment to enable the delivery of strong revenue growth. The $14.9 million NPAT improvement and $8 million improvement to underlying NPAT reflects the improved performance for the half, as well as a decrease in amortization. I'll finish up with the cash flow summary on slide 12. The cash balance at the end of the period was $44.7 million. The Australian dollar equivalent of the cash balance is $64.8 million. Despite the decrease in balance compared to the prior period, a strong balance remains. Cash flow used in operations was $2.7 million. In comparing to the prior period, it is important to note that H1F 125 was positively impacted by the receipt of a payment from a major customer in the first week of January 25 versus December 24 as scheduled. Cash flow used in operations for the period was impacted by the timing of customer receipts, annual payments during the period, and working capital required to support strong Appen China growth. Cash used in investing activities was $1.9 million higher compared to H1FY25 due to higher investment in product development and new facilities for the App and China division. Cash used in finance activities of $2.7 million reflects lease payments. Cash was used to fund operations and CapEx. That concludes the financial performance slides. I'll now hand back to Ryan.

Thanks, Justin. I'll now cover a strategy and operational progress that we've made in the half. So turning to page 14, to understand app and services, it helps to start where the value sits in AI development. There are three fundamental building blocks for AI development, compute, algorithms, and data. So compute is abundant and commoditizing. Algorithms are increasingly open and largely commoditized. While unique data is becoming scarce, and is a major source of differentiation for AI model performance. But not all data is equal. There are many facets of data used to train models, all with different uses. Public data is largely exhausted, and it's already been captured in existing models and offers little ability to differentiate. Synthetic data is reliant on other models to produce, does not solve new or novel situations, and can result in model collapse if overused. Real-world bespoke data, the kind that Appen creates, enables new AI approaches. It brings human expertise and interactions that improve and evaluate models in ways that alternatives can't replicate. This is the market that Atom serves. The primary way that we serve our clients is through a managed services approach, where we build custom and high-value data sets that are specific to the AI model needs. The usual first step is that a researcher comes to us with their data needs. It can be a description of the task, expertise requirements, quality rubrics, data volumes, and timelines. Our delivery experts work very closely with clients to deeply understand their intentions and translate that into a data workflow that typically coordinates a set of complex and iterative handoffs between humans and AI models to generate the data. We deliver through a combination of our research and delivery experts, our proprietary software stack, and our expert workforce marketplace. The output we provide is high-quality data for leading AI organizations. including the global tier one AI labs. What differentiates Appen is the combination of the platform people and our global reach. Our workforce is a core competitive asset. More than 1 million contributors across 200 plus countries and over 500 dialects and languages covered. We focus on building out our domain expertise in our workforce We now have contributors covering more than 100 specialist fields from computer science and mathematics to law, medicine and the creative disciplines. Generative AI demands a different type of contributor. It requires people who can reason, evaluate and provide expert level feedback. Our workforce has rapidly evolved to support the new expert level requirements of our customers. And our service offering continues to expand. It covers nine categories, and it's evolving rapidly with the needs of our customers. Some of the areas that we're working in include LLN training data, covering supervised fine-tuning, RLHF and preference annotation, multimodal data across text-to-image annotation, aesthetic scoring, video labeling, and body intelligence, speech and audio across a broad set of languages, domain expert annotation and medical, scientific, legal, and financial fields, amongst many others, model evaluation covering LLM and vision benchmarking and search quality performance, computer vision and physical AI for autonomous driving, robotics, smart home, AR, VR, and embodied AI, and we also offer reinforcement learning environments, off-the-shelf data sets, and platform and tooling solutions. The breadth of our offering allows us to evolve with the needs of the leading AI labs. We've recently expanded our data set offering significantly. These are the existing data sets that we either own or resell with evaluated services on top. We're seeing an increase in the demand for these data sets, and we're building out a catalog to meet the specific needs of model builders. Some of the areas where we have recently added data sets include reinforcement learning tasks, code repositories, book porpoises, enterprise data for agentic AI, and many other standalone datasets covering unique areas like medical dictation, STEM Q&A, and infographics. These products accelerate time to value for our customers and often also come alongside managed service projects to add value to the datasets. We anticipate this to be a solid growth driver in the near future. A unique proposition of Appen is our coverage of the two AI epicenters. namely China and the USA. We operate two dedicated businesses to cover these markets, each purpose-built for the specific customer requirements. Appen Global serves the USA and Europe, targeting hyperscalers, foundational AI companies, and vertical AI builders. Demand drives include AI capabilities, new customer expansion, and new data modalities. Appen China serves China, Japan, and Korea. targeting Chinese big tech foundational AI and vertical AI builders. In addition to the demand drivers, international expansion is an increasing driver of Chinese model builders look to complete in the global markets. Each business has its own dedicated operations and technology stack. That separation allows us to optimize for the distinct requirements of each market without compromise. As Justin mentioned earlier, Appen Global's technology roadmap has continued to deliver operational efficiencies. Employee expenses in H1 FY26 came in at $14.4 million, down $17.7 million in H1 FY25. Other expenses were $8.1 million that came in to $11.4 million in H1 FY25. We continue to be highly focused on driving technology-led efficiencies across our operations, particularly through the use of AI. We have identified approximately $12 million in incremental annualized cost efficiencies. Around 70% will be executed over the remainder of FY26, with the balance by the end of Q1 FY27. Importantly, there's been no operational impact from the cost-out work executed to date. We're capturing efficiencies through AI-enabled operations, not by reducing our ability to deliver high-quality data at speed. Let me now turn to our outlook and guidance statement for the full year. We remain confident in the AI data market and Appen's ability to contribute meaningfully to the development of leading foundation models. We continue to see positive signals on LLM-related growth from both Appen Global and Appen China customers. We're winning new work with leading AI labs and expanding existing programs. We remain focused on driving technology-led efficiencies across our operations. And as in previous years, Appen Global revenue is predominantly project-based and seasonality continues to skew revenue towards stage two. Considering all of this, we reaffirm our FY26 guidance of group revenue of $270 to $300 million and underlying EBITDA before FX margin of 5% to 10%. That concludes our presentation for today. Thank you for your time and your continued interest in AVEN. Justin and I are now happy to take questions.

Great. Thank you very much, Ryan and Justin. As a reminder, research analysts can ask questions by raising your hand on Zoom, and I'll aim to get to you shortly, while the remainder of the audience can submit written questions via the Q&A function at the bottom of your screen. We'll kick off with some pre-submitted questions before getting to any analyst questions today. First on profit sustainability, can you clarify how much of the recent EBITDA improvement is driven by a permanent structural cost reduction versus temporary project-based revenue?

Yeah, thanks, Dan. So we're definitely focused on a sustainable cost base and profitable growth across the business. The good thing is that particularly for China, we're seeing improved margins that's coming through the gross margin of China improving, but also we're getting the leverage of scaling efficiencies in the China business. And as we've called out, we've continued to drive OPEX improvement in Apple Global. So a multitude of factors, but we see this as a sustainable and ongoing trend in the business.

Okay, great. Thank you. And just to follow up there, what baseline quarterly revenue is currently required to maintain positive EBITDA through the second half of FY20?

I think it's fairly similar to where we're at today. In Q2, as we said, we've made some efficiencies across the business. So there's not a material uplift required to deliver profitability through the remainder at a quarterly basis.

Great, thank you. And on cash runway and free cash flow, given the cash balance, what's the current projected timeline to achieve consistent positive free cash flow?

I'll throw that one to Justin. Thanks, Ryan. Thanks for the question, Sam. Obviously, part of the strategy, we're talking about the efficiencies and the performance of the business and the growth in Appen Global. We're well towards... heading in the right direction and well towards achieving that. That is the goal, sustained profitability and free cash flows. So, we've got enough cash. We've got a strong cash balance. There's enough working capital. We're confident that the runway is there. There's no additional funds required to manage the growth. So... We're definitely heading towards it. We're heading in the right direction and we're not too far off.

Okay, and another follow-up there. Should shareholders expect the current cash reserves to be sufficient to fund operations until self-sustainability is... Based on everything we know to date, yes. Thank you. And just one more pre-submitted question before we get to the analysts. On revenue diversification, outside Athens core hyperscaler clients, what specific momentum or contract wins are we seeing in the broader enterprise AI market? And how long are the typical sales cycles for these newer revenue strengths?

So we're very focused on... the large foundation model builders and the Neo Labs that are popping up, typically spun out of the research division of these large AI labs. So that remains the focus of Appen at the moment. That's where the bulk of the spend is in the market, and it's highly aligned to the capabilities that we're building. In terms of the deal cycle time, it can vary significantly. quite extreme. Some of the deals are very, very, you know, short lead time. Some, you know, if researchers are, you know, contemplating specific areas they want to work through and there's a good amount back and forth, that can introduce, you know, a longer sales cycle. But it's certainly trending towards much, much shorter sales cycles.

Okay, great. Thank you. Next question comes from Nicola Wilmette at Barenjoie. Nicola, please unmute your line and go ahead. Nicola at Barenjoie, would you like to ask a verbal question? Please unmute your line.

Hi, guys. It's Josh Kanerakis here. Can you hear me?

Yeah.

Oh, cool. Sorry. Nicola's just looking at me a bit funny here. All good. No worries. So just the first couple of questions to get kicked off. First one, just around in the global business. So I know, you know, obviously there was some reasonably chunky customers in terms of that you had contracts for that you ended last year on. You know, some of those contracts were potentially coming back. Like what have you seen in terms of the start? And I know visibility is obviously not high, but what are some of the, I guess, the conversations and scope for work that you see from both some of the traditionally big customers that you have, but also, as we said, some of the foundational other customers into the second half. And just to talk a little bit about, yeah, what confidence you have in that ability to deliver into the second half there.

Yeah, thanks, Josh. So there are a few things that are giving us some really good confidence at the moment. So first is confidence. we've been able to penetrate into some new areas within existing customers that are really focused on the foundation model build. And we've started in specific areas, starting as a first project in a specific domain, and now what we're seeing is a much broader set of conversations around growth opportunities, not just within the projects that we're working on, but across a broader set of domains. And a lot of these growth opportunities are in areas that we've been strategically investing our capabilities in. So some of the things that we've called out, like coding, finance, healthcare, really pushing into the more valuable part of the market at the moment to support AI development in specific data modalities.

Got it. That's helpful. And traditionally, the margin on those sorts of projects as well versus maybe where the historical gross margin is. Can you give a bit of context on that?

Yeah, it can vary, but you should think about them broadly similar to the traditional margins that we've seen in the business.

Got it. And just more on global, just the competitive... the competitive environment in terms of what you're seeing out there. Obviously, there's a number of players. What are you guys seeing in terms of when you are in those new markets? What the competition is? Is there price? Is price coming into it? Is it more around the deliverability or the quality? Maybe just to talk through some of those frameworks that you think customers are using to choose the vendors and other places.

So the ability to deliver high-quality data is always number one. We don't see price as a major factor, particularly for the AI labs. We see quality and speed being, you know, the two primary considerations. And, you know, we, you know, used to call out there are some competitors that are, you know, growing very rapidly. And that's built on a lot of... They're established within the companies. They've got the trust of the researchers, and the researchers go directly to them because they trust their ability to deliver quality, and it speeds up the cycle rather than running an RFD process. And that's what's giving me a lot of confidence in the Atom Global momentum that we're seeing is because of the conversations that we're having in specific labs and it's across multiple labs. They're really satisfied with our work. The quality is really great. That's leading to bigger opportunities. But it is on the back of their confidence that we are delivering high quality and that we can turn around the data really quickly for them. So the thematic that we're seeing with some of our competitors is really starting to play out within Appen Global.

Right. And just moving on to Appen China, another stellar result, obviously. But when we look at that business, you know, now it looks, and we've seen, I think, well, probably one thing that's changed a little bit since we last talked is just the rapid release. including, I think, even yesterday, one of those sort of GLM3 coming into one of the Chinese models, sort of record things. And then Quinn and the like as well coming out with some fantastic models. I mean, there's been a lot of talk around the geopolitical aspects here, but, you know, it does feel like the Chinese models are definitely trying to accelerate into the US and other markets as fast as possible. And I think some of the open router stuff is saying it's over 60% of the tokens that are coming through from those models. So, like, I mean, I'm just interested in maybe under the hood what the trends you're seeing and how should we sort of think about the breakout for that work, the continuation of that trend. Obviously, before there was a bit of an acceleration and then margins were also tracking. So, maybe just a bit of a context around The type of work you're seeing, the type of customers, your confidence in revenue, my mind, and then confidence around, you know, they're the margin upside there.

Yeah, and look, it's some incredibly exciting and impactful work coming out from the open source model builders in China, which I think we all get a lot of visibility into and we can all kind of predict that that's going to be a continued focus. I think some of the things that we see on the ground in China, which don't get as much exposure outside, is the real focus on, I'll call it consumer-based AI, where through the super apps, the companies in China, the AI labs, the focus on things like healthcare advice, financial advice, We're starting to see a lot of video generation, AI video generation, particularly in short form videos. And I think they're a fair way ahead of the US on the video generation side, or at least getting the applications out that are supported by these models. And then the third really big driver in what we're seeing in China is the international support required to support Chinese technology companies that are heavily reliant on AI. So you can think about social media companies, e-commerce companies. There's a really big driver for supporting their international ambitions.

Okay, great. And just final comment there just on margins. You know, obviously that fourth, sorry, the second quarter margin was very strong. Do you think that can continue and how should we think about, I guess, you know, the cost base versus margin perspective there in terms of what further expansion we could see across this year from China.

Yeah, we certainly expect that trend to continue and we're seeing good operating leverage come out of the China business and that's a trend we expect to continue also.

Okay, great. Thanks very much. Appreciate it.

Thanks, Justin.

Thanks, Josh. Next question, sticking with China. Is your Chinese lab revenue recurring in nature, evaluation and data that's refreshed every model cycle, or one-off data set bills? And a follow-up to that, roughly how much of the $175 million run rate would repeat if customers shift to new models next year?

Yeah, thanks, Sam. So it is a mix of what we do. There is work that is directly related to the development of new models. There's a lot of work that we do which is related to the evaluation of existing models, making sure that they're working in applications, et cetera. There's also a lot of work that we do that is very experimental with the researchers that actually may never make it into a model. So it's a difficult one to dissect because the needs at a project level change and vary. on an ongoing basis. But I think what we're seeing in both the Chinese and the US market, there's certainly no slowdown in the model advances and the model release cycles. If anything, it's speeding up. So I think the notion of if China stopped building models, what would happen is not one that we're too worried about. We're very focused on supporting them as they evolve into new models. But as I called out also, different applications, particularly on, you know, things like short-form video, robotics, speech. There's a coding development. There's a very rich ecosystem of applications that sit on top of the existing models.

Okay, great. Thank you. And just a follow-up there. Does anything built in China, like data sets, tooling, capacity, get sold back to global customers, or does sovereignty and customer requirements keep the two segments commercially?

Yeah. It is largely separate and the data export controls in China kind of mean that there's no data that's sold back into the global customers.

Okay, great. Thank you. And switching to App and Global, can you just elaborate on the board's long-term strategy for the global business? Will growth be primarily organic or is there a And if organic, which higher-value services and initiatives will drive growth to improve gross margin?

Yeah, we're certainly focused on growth in Appenglobal, profitable growth. We think there is a significant pathway and runway through organic growth in the business. So inorganic growth right now isn't a high priority for us. And in terms of the services that we provide, it's continuing to work very closely with the AI labs to meet their needs. And we called out in the presentation some of the ways that we're evolving. That is in a really close response and feedback to what we're hearing and getting requests for from the AI labs. So we will continue to evolve to meet the needs of the data for AI training, for AI evaluation across LLMs and the future variants of AI models.

Great, thank you. And sticking with Appen Global, how deep and durable are Appen's relationships with core clients? Is Appen pursuing frontier AI companies such as Anthropic? And how does it convert pilots into recurring larger scale work?

Mm-hmm. Yeah, so our target customers are all of the leading AI labs. We work with a real majority of them today. There are a few that we're targeting and making really good progress on to break into. In terms of how we convert a pilot into a larger project, it is really, and this is a very typical sales cycle where we come in and do a small piece of work and based off the ability to deliver really high quality at speed, for that piece of work, building across with the researchers, and then showing them our capabilities and the work we're doing across different clients, that confidence that we build based off the project, based off what we're seeing elsewhere, and the conversation that we have and the way that we bring value to the AI labs is the best way to get up that, we'll call it the revenue curve as quickly as possible.

Got it. Thank you. Just a follow-up question from Josh at Baron J. Josh, please unmute your line and go ahead.

Great. Thanks, Sam. Just to follow up, guys, on the robotics side of things, there's been some interesting news this week. Figure one of the big humanoid robot companies have come out with a platform for building real-world training data. And, I mean, I'm just interested. Obviously, they're big. They're very well-funded. But what's the opportunity you guys see in robotics? And If you think about the capability and the muscle you've had to build, are there any other options to create a specific sort of more white label product or a specific product for robotics that could be used by some of these humanoid companies both in the US but also in China, which is obviously a huge focus out there? Yes.

Yeah, thanks, Josh. Good question. So you can think about three sources of data that are needed to train, particularly the humanoid robotics, which are getting a lot of focus at the moment. So one is what's called egocentric data collection, where data is being collected by humans who would typically have cameras strapped to their body, sometimes on the forehead, sometimes in different positions on the body. And this is capturing what humans are doing in the real world around things like anything that requires some type of physical manipulation, particularly with the hands and arms. That's a really big focus, and that's some of the data that Figura AI released was around these egocentric data sets. This part of the market is very interesting. It's a little bit of a... commoditizing really quickly because there are many companies that are out there going collecting the data, et cetera. The second part of data needs are related to the annotation of that data. And it's less about the annotation of, you know, what we saw with autonomous driving, which is putting a bounding box around all of the images. It's more assessing and filtering for quality standards, making sure that the instruction set matches the image. because there is really vast quantities of data. There are LLMs being run across it, but what we're finding is that there is demand for human involvement in the quality assessment of that data that's being captured. The third data source is more simulation-based, where it's getting humans to describe in a simulated environment the tasks that the robot should be completing as a way to provide the training data. There is one fourth bucket, which is related to teleoperations, which we don't really play in too much, but that's another interesting evolution of the market. So it's, you know, like the LLM front, it's evolving very quickly and the needs of the robotics builders are changing really quickly. And we're trying to find, you know, the best place, as you say, Joss, that's got a, you know, a durable and longstanding value add. but is also going to generate good margins for the business.

Right. And just a final one from me. Just in terms of, you know, obviously being quite a bit of movement there, changing strategy from one of your, you know, potentially big social media customers out there. I know you had a fair bit of work from them towards the end of last year. Have you seen any of those large projects start to resume or get any signalling on when you expect some of those projects to resume?

Yes, I mean... communication with all of our clients on their needs and what we find is that these shifts can come very, very quickly. So it's, you know, there's a lot going on across all our clients. They're changing strategy really quickly and we stay super close to meet their needs.

Got it. But yeah, hasn't sort of resumed some of those larger ones as of yet. Just more so a second-hand story.

Yeah, there's always a traditional skew to the second half, so we're confident that we'll see that growth come through.

Okay.

Thanks, guys.

Okay, thanks, Josh. Next question comes from Connor O'Prey at Canaccord. Connor, please unmute your line and go ahead.

Yep. Managed technology. And, Ryan, a couple of questions. Maybe if we go back to the... I guess if we go back to two years to the previous peak of the business, one of the defining characteristics was a heavy customer concentration, really around two customers, driving a lot of the revenue growth. I'm wondering, and I think you and I were both in the business, observing the business from the outside. We're both around that. I'm wondering what, at that time, I wonder what lessons you're taking from that, especially in the China business, which is going through a sort of similar analogous kind of growth path. Are you able to diversify the revenue across more customers to sort of lessen those risks?

Yeah, Conor, it's certainly a focus for us. And I think the difference between what, you know, that period that you're explaining where there are a couple of big customers that contribute a lot of their revenue The China AI ecosystem, there are some very dominant players, but there's a decent number of them that it's not market skewed towards too big of customers like it was traditionally. So the focus for us is we want to serve the big accounts to the best that we can. There's a huge amount of growth potential there. We also work with a really large number of customers in the China business. This covers the Tier 1 labs, Tier 2 labs, startups that are getting into the space. So we... I'm less worried about that diversification risk that we had previously vis-à-vis what's happening in China at the moment.

Thanks. And then just back on global, is it fair to... characterize the revenue trends there as kind of the legacy business deal? I guess we would call it the content relevance piece. Is that in a sort of a structural decline? Is that a decline? How do you characterize that? On one hand, I'm seeing that possibly swamp all the other kind of good stuff that's going on, or is it, I'm guessing it's more complicated than that, but maybe you can sort of break that apart a little bit for us.

So there's certainly an element of that, Connor, where we've seen some of our more traditional work decline as its LLMs are able to replicate some of that work. But what we have seen, particularly in the first half of this year, some good stabilization across the vast majority of that work that we're doing, and we are seeing the uptick in the newer areas related to LLM development. So it is certainly a factor, what you're describing there, and quite accurate. But we are, as we called out in the presentation, seeing the stabilisation of some of the revenue from more of our traditional work.

Thanks.

Great, thanks, Connor. Next question. Just on cost out, you've identified $12 million in annualized cost efficiencies in App and Global, with 70% to be executed by the end of Q4 this year. At what point does the incremental margin benefit of this program start to visibly flow through?

Yeah, Justin, I'll pass that one to you.

Yeah, thanks.

Thanks, Sam. So there will be some benefit towards the end of the year, but it's not going to be material the way the timing works. So I think there'll be incremental benefits in the second half and in Q1 with the full benefit from the start of Q2 next year.

Great, thank you. Just a couple of final questions here. Is the board considering acquisitions or other strategic transactions? If so, what capabilities or businesses would Avan specifically target?

Yeah, like I called out, M&A isn't a We're very happy with the organic capabilities that we're building. The market's changing very quickly and we need to be dynamic in response to the needs. So we're largely focused on organic growth in the business.

Okay, thank you. Final question. As profitability and cash flow recover, would the board consider share buybacks, dividends or other capital returns?

So the board will... closely look at their capital allocation and consider all options as we continue to improve the cash reserves in the business. So yeah, that's certainly for consideration.

Okay, thank you. That's all the time we have allocated for questions today. If you do have any follow-ups, please feel free to send them through to me via email and we'll endeavour to get back to you. And maybe with that, Ryan, I'll just pass it back to you for any closing comments.

Yeah, thank you, everyone, for your time today. I think you've heard your continued interest and support in Appen. We continue to play a major role in the AI ecosystem. It's a very fast and rapidly evolving space that I'm sure of you are keen interest in. But Appen's role for over 30 years now has been at the forefront of AI, and we look forward to continuing to supporting our customers and delivering great financial results for our shareholders.

Great, thank you. Thanks very much for joining. That concludes Athens First Half FI26 results call. Enjoy the rest of your day. Thank you and goodbye.