NWR, and joining me from the company today is CEO and Managing Director Ryan Kolln, as well as Chief Financial Officer Justin Miles. Following a brief summary of their results released to the ASX this morning, we will have some time for Q&A from the management team. There will be a choice of two options. First, research analysts will be able to raise your hand via Zoom should you wish to ask a verbal question, or we'll take written submitted questions via the Q&A function at the bottom of your screen. We will endeavor to get to the majority of questions asked, in some cases combining questions on the same or similar topic. With that, I'll pass it over to you, Ryan.
Thank you, Sam. Hello everyone, and welcome to Appen's half-year results. I'm joined today by our CFO, Justin Miles. I'll now start with the presentation. Turning to page three, I'll start with an overview of the agenda for today. There'll be four sections to the presentation. First, I'll share some highlights of our H1 results. Second, Justin will provide greater detail on our financial performance. Third, I'll provide an update on our progress against our strategic initiatives. Finally, I'll provide an FY 2025 outlook and guidance statement. Turning now to slide five, I'm pleased to report that Appen delivered continued momentum in the half. We recorded revenue of $102.1 million USD, representing 2% year-on-year growth when excluding the impact of Google. 24% of our revenue for the half was from generative AI-related projects, demonstrating our continued success in supporting foundation model builders. Our performance in China was a standout.
Revenue grew 67% year-on-year, and we exited the half with an annualized run-rate revenue exceeding $100 million. This highlights the consistency and strength of our momentum in China. The turnaround in our core business is progressing well. We're deepening customer relationships, successfully entering new target accounts, and improving our operations. We're also seeing green shoots and recently secured a new generative AI engagement with potential annual revenue exceeding $10 million. Our technology transformation is delivering tangible operational benefits. In our non-China business, we have identified approximately $10 million in annualized cost efficiencies. Around 70% of these will be realized by the end of Q3 FY 2025, with the balance to be executed by the end of the year. In terms of the market, we're seeing customers' data requirements shift to areas where Appen has clear market differentiation.
Demand is increasing for large multilingual speech training data, a domain where Appen has deep expertise. We're also seeing positive market signals for LLM model evaluation, which is very similar to the search relevance projects we have been performing at scale for many years. China remains a key sector of the market, supported by growing demand for domestic and international operations. Finally, our balance sheet remains strong. We ended the half with a cash position of $60.9 million or A$92.9 million . We remain focused on delivering EBITDA positivity and are well-capitalized to support our growth objectives. In summary, the first half of FY 2025 shows encouraging momentum across the business. We're executing against our strategic priorities, diversifying revenue sources, and leveraging our technology and expertise to capture opportunities in the evolving AI ecosystem. We enter the second half with confidence and a strong foundation to build on.
I'll now hand over to Justin, who will walk through the H1 financials in greater detail.
Thank you, Ryan. Good morning, everyone. A reminder that we report in U.S. dollars and that all comparisons are to the half we ended 30 June 2024, unless stated otherwise. Starting with the H1 snapshot on slide seven, revenue increased 2% to $102.1 million. This excludes the impact of Google. Within our operating segments, new markets revenue grew by 20% to $59.6 million due to strong growth in China, which was 67% up on the prior corresponding period. The remainder of the new markets segment was impacted by short-term volatility due to the dynamic nature of the U.S. AI market. Global services was also impacted by the short-term volatility, as well as the termination of the Google contract in H1 FY 2024. The decrease in gross margin percentage reflects a change in customer project mix compared to H1 FY 2024.
It is noted that China margins are traditionally lower versus the rest of the group. Due to prudent cost management, there was an improvement in underlying EBITDA before FX, despite lower gross margin. I won't talk to slide eight as we cover revenue in further detail at later slides. Over to underlying EBITDA at slide nine. Group underlying EBITDA before FX improved $0.1 million to a loss of $2.2 million. As just mentioned, there was an improvement despite lower gross margin due to prudent cost management. New markets EBITDA improved by $4 million to a loss of $4 million. The improvement reflects a positive EBITDA contribution from the China business. The global services division reported EBITDA of $2.1 million, down $4.7 million on the prior corresponding period. The decrease reflects lower revenue and gross margin, driven by short-term volatility due to the dynamic nature of the U.S. AI market.
Over to slide ten. Ryan has already talked about China's strong revenue growth. Revenue grew 67% compared to H1 FY 2024. Pleasingly, the China business exited the half with an annualized run-rate for June exceeding $100 million. This is a great milestone for the business. China customers include leading LLM model builders, along with leading technology and auto customers, and there are revenue opportunities in both China domestically, as well as China customers expanding internationally. It is important to note that the China business has a more predictable revenue profile compared to the remainder of the business due to the engagement and delivery model. Turning to slide 11. Pleasingly, there has been a positive EBITDA contribution from the China business for the last five quarters. China contributed $2.9 million underlying EBITDA to H1 FY 2025, a $2.8 million improvement compared to H1 FY 2024. Over to slide 12.
This slide shows quarterly global revenue with Google excluded. As already mentioned, global revenue was impacted by short-term volatility due to the dynamic nature of the U.S. AI market. Despite the short-term volatility, our customer relationships and conviction in the revenue opportunity remain strong. However, it is important to note that timing is uncertain in relation to the resumption of large volume LLM projects. The uncertainty comes from customer LLM teams planning and going through multiple reorganizations, as well as the nature of the work, being large volume short-term projects. As well as the strong customer relationships, our quality metrics with our largest customer are at an all-time high. We continue to win new projects, albeit at a low volume of data during the half compared to PCP.
A recent win with a $10 million+ annual revenue potential are all signals that support the progress we are making and conviction in the revenue opportunity. Slide 13 has revenue for the balance of the new market segment being enterprise and government. The decrease in revenue was driven by lower volumes within some existing large enterprise projects, including some projects coming to an end. There is a healthy enterprise pipeline, and conviction remains in the revenue opportunity, although timing of revenue conversion remains unclear. U.S. policy uncertainty has meant generating meaningful short-term revenue opportunities within the U.S. government division is challenging. Given this challenge for the U.S. government division, a decision has been made to wind back investment. This will result in approximately $4 million in annualized OPEX savings. The majority will be executed by the end of Q3 FY 2025 and the balance by the end of the year.
This decision does not have a material impact on existing revenue. It is important to note the $4 million cost savings are in addition to the $10 million cost efficiency gains from the technology strategy that Ryan spoke to at the start of the presentation. Turning to slide 14. This is a summary of the profit and loss, and we've already covered most of the line items. However, it is worth noting the expense lines. Employee expenses are down 9%, and all other expenses are down 11% compared to H1 FY 2024. This is due to the cost-out programs that were executed in H1 FY 2024, as well as continued prudent cost management. The cash flow summary is on slide 15. The cash balance at the end of the period was $60.9 million, up $6.1 million from December 2024. The Australian dollar equivalent of the cash balance is A$92.9 million.
Cash flow from operations improved by $2.3 million to $12.9 million. The period was positively impacted by the receipt of a payment from a major customer in the first week of January 2025 versus December 2024, as scheduled. Cash flow used in investing activities was $0.8 million higher compared to H1 FY 2024 due to a higher investment in product development and new facilities for the China business. The new facilities also impacted financing activities, with cash used $0.3 million higher due to lease payments. Cash was used to fund operations and CapEx. Now, before I hand back to Ryan, I would like to call out a potential change to the reporting segments. With the strong growth in China and the wind-down of investment in the U.S. government division, we are exploring whether to simplify the reporting segments as we head into the end-of-the-year reporting. That concludes the financial performance section.
Back to you, Ryan.
Thanks, Justin. I'll now provide a strategy and operational update. Moving to slide 17, we are focused on three critical areas for long-term success. First is an expert-led approach to sales and marketing, where our technical teams build deep, trusted relationships with AI researchers and product teams. Our customers are highly technical, and we are retooling our organization to bring our expertise to the forefront of conversations with our customers. Second, we're evolving our data offering to address the rapidly changing needs of our clients, supported by a flexible technology infrastructure. Third is the need for technology-enabled efficiencies. We utilize generative AI heavily throughout our operations, including how we validate the quality of our data and provide real-time feedback to our contributors. These priorities ensure that we stay responsive to market trends and deliver consistent value to our clients and shareholders.
Our proprietary platforms, Mercury, ADAP, MatrixGo, and CrowdGen, are foundational to delivering quality at scale. Each platform has been enhanced in H1 to improve efficiency, quality, and contributor experience. Mercury is our workforce and project management platform. We've continued to make improvements and automate many manual steps in how we deliver projects. This has enabled us to significantly enhance the efficiency of our operations. ADAP is our data annotation platform. It's a highly configurable interface for contributors to produce accurate annotations supported by built-in AI quality checks. MatrixGo is the data annotation platform that we purpose-built for the China market. This platform underpins much of our growth in China. Finally, CrowdGen serves as a dynamic interface for our crowd workers, enabling them to explore earning opportunities, enrich their profiles, and get paid seamlessly for their tasks.
Moving to slide 19, where I'll provide an update on progress against our 2025 strategy. First, we remain highly focused on our core market of human annotated data for AI, including supporting the development of large language models. In H1, we have strengthened our go-to-market team with key hires from competitors. We've also made significant changes to our marketing efforts by developing content that is highly focused on technical thought leadership. Second, we're driving operational efficiency and automation by utilizing LLMs across our operations. We have been successful in incorporating generative AI into our platforms for high-quality data. This includes both real-time AI quality checks and using AI to post-process data for further quality improvements. We've also recently launched an LLM-powered customer service agent that auto-resolves over 30% of contributor tickets, freeing up internal resources and improving ticket response time.
Third, we're accelerating technology innovation by utilizing LLMs across our software development teams. Our engineering team is now utilizing AI coding tools extensively, increasing the output of the team. We're also running a high-velocity innovation approach enabled by AI software development tools. This allows us to build software prototypes in hours that previously took weeks. Fourth, we're growing our people, particularly by expanding our technical expertise in our project delivery teams. In the past three months, we have made strategic hires from our competitors that have brought significant new capabilities to Appen. Fifth, following the launch of CrowdGen, we are evolving our data annotation workforce. We now have an LLM interviewer that is in beta testing that will streamline the onboarding process. We're also utilizing the latest LLM models to better match contributors to tasks, which is a major factor for improved data quality.
Finally, we're committed to prudent cost management. We're capturing the benefit of our automations and productivity improvements and ensuring that we can remain lean as we scale. By executing in these six areas, we're positioning Appen to deliver superior data quality, ensuring that we meet the evolving needs of our clients and drive profitable growth in the business. Moving to slide 20, where I'll share more details on the benefit of our operational efficiency initiatives. Investments in our platforms have been heavily focused on automating operations. This improves the speed and quality of how we produce data while also delivering material operational efficiencies. Due to the platform improvements, we have identified approximately $10 million in incremental annualized cost efficiencies that will be executed over the remainder of FY 2025.
It's important to note that the $10 million in cost savings are net of talent upgrades we are making to enhance our team's technical expertise. In addition to the above $10 million, as Justin mentioned, we are winding back investment in the U.S. government division. This will result in a further annualized OPEX saving of around $4 million. Now to slide 22 for our FY 2025 outlook and guidance statement. Appen remains confident in its long-term revenue opportunity, including previously provided longer-term targets. As it relates to the balance of FY 2025, there is limited visibility related to specific timing for the resumption of large LLM projects. In addition, recent U.S. policy uncertainty has reduced the likelihood of generating meaningful short-term government revenue, resulting in reduced divisional investment.
Considering this, Appen reaffirms its current FY 2025 guidance of revenue towards the low end of the $235 million- $260 million range and positive full-year underlying EBITDA. That concludes the presentation. Thank you, everyone, for joining. We will now turn to Q&A.
Great, thanks Ryan, thanks Justin. As a reminder, research analysts can ask questions via raising your hand on Zoom, so I can unmute your line, while the rest of the audience can submit written questions via the Q&A function at the bottom of your screen. Moving to a couple of pre-submitted questions, market tailwinds, you've identified in the slide deck at a high level some of these tailwinds. Can you just identify what specifically you're monitoring to underpin confidence around full-year guidance?
Thank you, Sam. We engage very closely with our clients, and we're seeing increased conversations around areas where Appen has long strengths, notably large-scale multilingual data related to speech interface systems. This is largely for large language models and generative AI. The second is broad-scale generative AI evaluations, which, as I mentioned, is very similar to the work that we've been doing to search and ad relevance for a very long time in the business. We're very confident that the capabilities that we have are very aligned to where we see the market demand heading.
Okay, great, thank you. Just on the China business, it delivered over $2 million of EBITDA during the second quarter of this year, which reflects about a 7% margin. How much incremental investment is required to keep growing that business?
We don't think there's any material incremental investment required. We think we're at a really great point now where we've delivered profitable growth, and we expect to continue to deliver profitable growth as we move forward in China.
Sorry, next question from Josh at Barrenjoey. Josh, please go ahead, thank you.
Great, thanks for taking my questions. First one, just on the market and then on some of the financials and cost out. Ryan, just firstly, with regard to some of the market trends, I know there's obviously a lot of disruption with the Scale and Meta deals. Has there been any follow-on from that in terms of any of the customers that were using Scale and a large capacity that have looked at Appen or evaluating any of Appen services currently?
Great question, Josh. There's certainly a lot of discussions that we're having with clients following the Scale news. It has created a big hole in the market that we are looking to fill actively. As I think we've shared previously, we see this as a real net positive for Appen and our services going forward.
I guess there's some particularly big ones out there that haven't been using your services for a while. Is there any re-engagement with any of those larger customers?
We continue to have active conversations. That's really a big focus for us for the tail end of this year and as we head into next year.
Okay, great. Next one, just on the cost out, just to clarify on that. The $10 million is across the year. How should we think about, and you did mention that was net of some of the new hires, how should we think about, I guess, the net in-year cost benefit and what that means for the run rates into next year as a result of that and the incremental $4 million?
Yeah, I'll take that one. Hi Josh. Of that $10 million, around 70% of it will be executed by the end of September. I think take the end of September as kind of the line in the sand for that calculation. That is just distilled on the net $10 million, which is about right. We're looking to get people into the business as soon as possible. In terms of next year, there are a few swing factors as we go into next year. Obviously, the full $10 million will be out. The $4 million OPEX from federal will be out from the U.S. government division. The swing factors are STI is likely to be higher than the half year. There were a few write-backs in H1, so that's going to be higher. The STI can be a swing factor as well.
If it's at 100% next year, it may be higher than what's reflected in the numbers this year. There may be some CPI around existing employee wages as well. They're the three swing factors. What is important to note, and I'm not giving you the definitive numbers here, is that Ryan and I continue to push super hard on getting more cost efficiencies out of this technology strategy. Some of that will be offset by these further gains as we head into the next year. That's the way I think about it and certainly the way we're planning as we go into next year.
Got it. I might just squeeze in one super quick one, just on China as well. Obviously, some great progress and profitable progress there, which is great to see. In terms of the customer spread there, has that changed at all to any of the actual, some of the LLM or other players that are sort of building out, or is it primarily still the bulk of it, the large technology players that are obviously building their own LLMs amongst other things?
Yeah, there's pretty big overlap between the customers that we've been working with for a long time and the customers that are building LLMs at the moment. We're working with many of the major companies in China that are building out LLMs. A good portion of our work in China is focused on LLMs. That's helping us out on a few factors. One is that the gross margin for the work that we do for LLMs in China is typically higher than our traditional work. That's a real positive factor that's helping us drive the profitability in the division.
Great. One final one, just CapEx in terms of how we should be thinking about across this year and next, you know, the sort of capitalized sort of spend in the business.
Yeah, there's a slide in the appendix, Josh, and we're not looking to spend a whole lot more. I think we may get some efficiency gains there, but I think using the H1 as kind of a proxy is a good starting point.
Great, thanks for taking all my questions, guys. Appreciate it.
Thanks, Josh.
Thanks, Josh. I think there were a couple other questions, but they seem to be answered on the OPEX. Just in terms of the recent project win for over $10 million annualized potential, any more detail you can share there? Is it U.S.-based? What's the expected ramp-up with that timeline?
It is U.S.-based. It's for a foundation model builder doing evals, and as I mentioned, that's a tailwind for us. We've done this work for a very long time. I think we'll see the ramp through the remainder of the year and into the first quarter of next year.
Okay, great, thank you. I think that's all the time we have for questions today, and that's the only ones we have coming through. Maybe with that, Justin and Ryan, I'll pass it back to you for any closing comments, please.
Thanks, Sam. Thank you, everyone, for your time today for joining the call, and we appreciate the continued support. We look forward to touching and getting in contact with you again in our Q3 results, which will be at the end of September. Thank you, everyone.
Thank you. Goodbye.