Thank you. Hello, everybody. Welcome to the conference call for Appen's first half results for 2021. Before I start, I hope you're all safe and well in the pandemic and for those in Australia, enduring the lockdown as best as you can. My name is Mark Brayan, chief executive. I'm joined today by our CFO, Kevin Levine, and our head of IR, Linda Carroll. Our presentation was loaded to the ASX website this morning, and I'll be referring to it throughout. We'll take questions after the presentation. Just as a reminder, all of our financials are in US dollars. To page three to start. We operate in a dynamic, growing, and exciting market. We provide training data for artificial intelligence and counts the world's largest tech companies amongst our customers. Training data is essential for the development of AI products.
There is no other way to build AI than with training data. Just as we learn from books, conversations, and classes, AI learns from data and improves with the amount and the quality of that data. Our market sits within a broader AI ecosystem that's forecast to grow to $110 billion by 2024. The market's growing due to an explosion in AI use cases, including areas where we have deep experience, such as speech recognition, chatbots, and search engines, to new applications, including augmented and virtual reality and geolocation data. AI mimics many things that humans do, such as speech, vision, decision-making, et cetera. The need for human curated data is very strong. It offers the highest quality outcomes. We're able to provide large, diverse sets of human curated data via our global crowd. Our customers want large volumes of quality data very quickly.
We blend our crowd and our own AI to increase the volume and speed of data provision with techniques such as AI-based pre-labeling, for example. Beyond data, there's growing need for more AI development tools and services such as data audit, data management, model testing, and lifecycle management. Our growth rate reflects the market. We've grown revenue at 38% annually over the last five years, and our new markets are up 31.5% half- on- half. This is driven by sales into our enterprise, China, and government markets, as well as product-based revenue from our global customers. We continue to invest to make the most of the opportunity before us. We invested 10.8% of revenue in product development in the first half of this year. Welcomed Sujatha Sagiraju as our head of product development, and today announced the acquisition of Quadrant to expand our location-specific data capabilities. To slide four.
Our transformation to a product-led business announced in May this year is well underway. We now cover more data types in our language-based origins, including text, speech, image, video, three-dimensional lidar data, augmented and virtual reality data, geolocation data, which range from satellite imagery to granular on-the-ground points of interest. Our ACV is growing up 16% in the half, and we're adding to our customer base with 74 new customers this half, and that improves the resilience of our business. To page five and the result highlights. As we advised in February, first half earnings were impacted by the skew in project delivery to the second half. Group revenue of $196.6 million was down 2% on the first half of last year due to lower global services revenue as a result of our major customers prioritizing new non-advertising product developments.
Global services revenue pardon me includes the work we do for our global customers on their platforms and includes many of our large relevance programs related to digital advertising and search. Group revenue growth was also dampened by the strong result in the first half last year that had negligible COVID-19 impact. In contrast, new markets revenue of $47.8 million was up 31.5%, driven by China, new enterprise customer wins, and product-led growth, including many new projects for our global customers on our platform, reflecting their investments in non-advertising products. Underlying EBITDA for the half was impacted by the higher first half cost base, and our restructure in May resulted in changes to the cost base that will yield benefits in 2022, and those benefits will be largely reinvested. Our balance sheet remains strong with $68 million in cash and no debt as of the 30th of June 2021.
We're also pleased to provide an interim dividend to shareholders of $4.50 per share. Over the page to slide 6, Our key focus areas provide a roadmap for the rest of the presentation. We're focused on maintaining and strengthening our core as the global leader for AI training data. We continue our expansion into new markets and our ongoing development towards a product-led future. Page seven emphasizes the strength of our core business. We've grown at a compound rate of 38% over the last five years to become the world's largest provider of training data. Our technology and crowd underpin the strength and scalability of our business. We're able to respond to all data types and use cases because of these core assets. Our customer relations are also key to the strength of our business.
We are trusted and relied upon to deliver high-quality data fast and cost effectively and securely. Our focus on Responsible AI is essential for our customers. Page eight, you can see that our global services revenue of $148.8 million was down 9.2% on last year. This revenue, derived from the work we do on our customers' platforms and typically includes our large relevance projects, was down as a result of our customers prioritizing new product developments as they diversify beyond ad-related products in response to regulatory scrutiny and ongoing privacy concerns. Their investments in new product developments is visible in our global product revenue on the right-hand side of the page. This revenue is derived from global customer projects on our platform and includes many new use cases. This revenue's growing at 32% annualized compound growth and is an increasing proportion of our global revenue.
Some of our large customers have in-house platforms for certain use cases such as relevance. Our platform covers all data types and many use cases and provides utility unavailable with in-house platforms. Page nine illustrates the impact of changes to the online landscape. The revenue we derived from ad-related projects was down this half as customers prioritized their resources towards new product development. Non-ad related revenue was up half- on- half, reflecting our customers' ongoing commitment to AI product development and maintenance. We mentioned in prior releases that our results were impacted by the deferral of a few large projects. This chart shows that those large projects were predominantly ad related and reflect market forces, including increased regulatory scrutiny and privacy concerns. Our large customers rely on ad base revenue.
We expect revenue from ad-related projects to grow in the second half of this year, but at a lower rate than non-ad projects. Slide 10 further dissects the direction of our large customers. In the first half of 2021, we derived $20.3 million from 100 projects that commenced this half. The balance of our global revenue, $150.8 million, came from 185 existing projects. This shows the extent of development in new projects and that projects can start small and grow over time. Also 97 of the 100 new projects cover a range of non-ad applications, showing that our ad-based customers are investing in new areas. We do see some new ad projects, though.
We expect that the future AI solutions that solve regulatory and privacy issues will be data and machine learning based, and our capabilities and strong relationships set us up nicely to support our customers on these projects. Slide 11 shows some of the use cases we're working on with our global customers and illustrates the range of applications from augmented and virtual reality to geolocation and e-commerce. To slide 12 now, and our push into new markets is going well. New markets capture the success, pardon me, of our product-led and customer expansion strategies and includes product-based revenue from our global customers, along with revenue from our enterprise China and government divisions. New markets grew strongly this half, as illustrated on slide 13. Up 31.5% on the first half of last year.
Growth was driven by a combination of factors, but mainly China, enterprise, and global projects on our technology platforms. New markets EBITDA reflects the ongoing investments to grow product-based revenue and further diversify our customer base. We won 74 new customers in the half, giving us over 320 active customers. The chart on the right shows a solid start to the year with respect to new customer wins. Slide 14 breaks out our global product in enterprise China and government divisions. Both are growing well at 32% and 27% annually, respectively. Product revenue is more retentive than services revenue, and the fluctuations in the global product chart are due to variations in data volumes that occur as part of the natural life cycle of an AI product.
To slide 15 now and two views of our committed revenue, annual contract value or ACV on the left, and how that translates as a percentage of our total revenue on the right. Progress this half hasn't been as rapid as prior halves, but we are nonetheless very pleased that we've improved revenue quality and visibility. China, on slide 16, is progressing extremely well. Maintaining its very high growth rate of 60% quarter-on-quarter. Growth is coming from China's tech giants, new customer wins, and we are building a good position in the autonomous vehicle market. Our team in China have their own technology stack to give them the agility they need to respond to local requirements and to protect customer data.
We use a crowd in China for many tasks, but are also leveraging facilities to protect customer data and respond efficiently to some use cases such as computer vision. To slide 17 now and our enterprise division that is growing at double-digit rates and continues to add new customers across a wide variety of use cases. The breadth of use cases shows how many ways AI is being deployed. The opportunity before us and the value of adding more data types and capabilities such as the acquisition of Quadrant that we announced today, and we'll discuss in a few slides. Our government business on slide 18 is ramping, but not as fast as planned due to the slower nature of government purchasing and a cautious approach to doing business in the pandemic.
We're actively engaged in projects directly with agencies, through contractors, and with government research labs. They have been the font of many impactful technologies over time. Overall, we're well positioned and optimistic about the government market. To page 19. The opportunity before us is simple. Our customers need large volumes of many different data types of high quality training data to build an increasing variety of AI products that benefit their customers and their businesses. The more we can support them in this journey, the better. Sujatha Sagiraju will soon join us to lead product management. She has 20 years experience in engineering and product management in areas including search and AI. She told me the other day that AI will change the way we live. We're still in the very early stages of its adoption.
She's thrilled to be joining us because she sees training data as the key challenge for anyone developing AI. Our product-led strategy aims to achieve two things. We're using our products to digitize and improve the productivity of our internal and crowd operations, and we're developing products to add more value to our customers. These combine to enable us to deliver more high-quality training data to our customers to help them achieve their AI objectives. We're investing in this strategy with 10.8% of revenue invested in product development this half, up from prior periods. Our product suite on page 20 includes Appen Connect, which is our crowd management platform that enables us to operate at scale. Our annotation platform that provides tools our workers use and our customer interface, and a number of new products. Appen Intelligence, a suite of machine learning models that pre-label data and automates internal functions.
In-Platform Audit that analyzes training data sets for quality and bias, Appen Mobile that provides greater utility to our crowd workers and enables a growing set of data collection projects. Our own use of machine learning is fundamental to our scalability and growth. Slide 21 shows several applications of our over 20 machine learning models. The first four are examples of the use of machine learning to improve productivity of our crowd workers and lower the unit economics of each data point. For example, we've automatically created speech databases to high levels of accuracy without human intervention. The last two examples use machine learning to improve our internal operations, automatically catching fraudulent workers and predicting which workers will be best at which task to improve project speed and quality. To Slide 22, in further support of our product-led direction and expansion of our training data capabilities.
Today, we announced the acquisition of Quadrant, a leading provider of high-quality location data. Location data has many applications, including in e-commerce, supply chain and delivery, mapping, and marketing and advertising. For example, the rapid rise of home delivery services means that delivery companies need more granular location data. Not just the address of the restaurant, for example, but exactly where the pickup point is in order to streamline the pickup and delivery process. Over to slide 23. The combination of their innovative technology and our global crowd enables us to collect and provide high volumes of granular location data in almost any country around the world. We're very excited by the combination of Quadrant and Appen and the opportunities in location data. As a global and responsible company, we have the opportunity to contribute beyond our core business.
Slide 24 highlights our work with the World Economic Forum on Responsible AI, an initiative that is setting standards to ensure that AI works for the benefit of all, it also shows our approaches to reduce emissions and the risk of modern slavery. Further, over the page to slide 25, we're supporting our crowd workers in a variety of ways, including providing work and income to many that don't have opportunities to earn. Our partnerships with diverse communities provide income and opportunities around the globe. I'll now hand it over to Kevin to take you through the financials in more detail.
Thank you, Mark. Good morning, everyone. On to slide 26. Different dynamics from the first half of last year make comparison with the prior period challenging with respect to revenue and costs. 1H 2020 revenue split was 49% as compared to the historical split of around 44%, which is closely aligned to how the split is forecast for this year. As Mark has explained, first half revenue has been impacted by a range of factors, resulting in reduced ad-related revenue from our global customers. Our cloud and ad-related projects, where we derive 75% of our global services revenue, we have had strong growth in new projects and expect that to continue. Ad-related projects are expected to grow in the second half of 2021, but at a lower rate than non-ad-related projects. Our new market's growth has also been strong.
Cost of sales, which is comprised of payments to our crowd workers, increased as a percentage of revenue from 59.4% in the comparative period to 61.6% in the current half. This was impacted by the mix of customers and projects comprising the revenue impacted by the decrease in global services revenue and the number of early-stage new projects. Operating expenses for the first half were higher due to the fully annualized impact of growth investments in FY 2020 and the investment in new markets in the first half of 2021. This is represented mainly in employee expenses. The other expenses line in the P&L reduced mainly due to lower recruitment job board expenses. The overall expense increase was partly offset by a true-up adjustment of share-based payment expenses. Similar to impairment testing, share-based payments need to be assessed.
Following an assessment of the probability of achieving hurdles for the 2020 long-term incentive plan, a true-up adjustment was processed in the first half of 2021. Underlying EBITDA margin of 14.1% was impacted by lower revenue and gross margin and a higher cost base. Further to that, underlying NPAT reduced 35%, additionally impacted by increased amortization of product development investments. In relation to the organization restructure announced in May, a restructure charge of $2.3 million has been taken in the half. This charge reflects costs incurred in the half, as well as the provision for costs to be incurred in the second half. The effective tax rate of 20.5% was down from the prior corresponding period. The effective tax rate is subject to fluctuations from the tax effect of movements from expensing, investing of employee performance shares, and differences in overseas tax rates.
Excluding these fluctuations, the normalized tax rate is around 28%. On to slide 27. We continue to focus on driving efficiency and productivity in our core expenses to facilitate continued growth investments. Core expenses as a percentage of revenue have been reducing since the second half of 2019. On slide 28. In the first half, we invested $21.2 million in product development, representing 10.8% of revenue. This focus is important to drive customer growth and repeatability, as well as quality improvement and margin expansion. Since FY 2019, we have strategically invested in engineering resources to develop new products and enhance existing ones. 53% of product spend was capitalized in the first half, consistent with the FY 2020 rate, reflecting our commitment to development of new products and tools. We expect product development spend and the associated capitalization rate in the second half to be in line with the first half.
On to slide 29 to talk about our amortization policy. We take a conservative approach to amortization in that we commence amortizing product development in the year in which the spend originates. The purpose of this slide is to show how we effectively apply our amortization policy. The table on the left shows how the annual amortization expense each year is comprised of the different layers of expense pertaining to the years in which the spend was capitalized. The chart on the right shows the amortization rate per annum, i.e., 33%, relative to the amounts capitalized in each year. On to slide 30 to talk about our balance sheet. Our balance sheet remains strong and resilient with no debt. Cash on hand at year-end increased by $5.5 million- $66 million through effective cash collection management.
The decrease in trade receivables of $ 9.5 million results from lower revenue and effective conversion of invoices to cash. Contract assets represent work completed at half-year end, pending satisfaction of customer billing milestones. Invoices have subsequently been invoiced in respect of this work, and the majority of these invoices have been paid. Non-current assets comprise mainly goodwill and identifiable intangible assets, mostly arising through acquisition. Following a detailed impairment review, we report adequate headroom in the carrying value of these intangibles. An interim dividend of AUD 0.045, in line with the 2020 interim dividend, has been declared and is franked to 50%. On to the cash flow on slide 31. Cash flow from operations reduced in the first half of 2021 due to lower revenues and higher costs. The comparative 1H 2020 cash flow from operations benefited from favorable receipt timing differences.
However, cash flow conversion continues to be effective, with 101% of EBITDA being converted into operating cash flow in the half. The cash balance of $66 million included a repayment of borrowings of $23.5 million in August 2020 in respect of the Figure Eight earn-out payment. Cash has been effectively deployed for product development, tax, dividends, operating expenses, and growth investments. I'll now hand you back to Mark. Thank you.
Thanks, Kevin. Over to our outlook slide now on number 32. First of all, our full-year underlying EBITDA guidance will be impacted by investments in Quadrant to the tune of $2 million, adjusting our current range of $83 million to $90 million to $81 million to $88 million. That aside, we're maintaining our guidance range, but expect to be at the lower end of the range due to the ad-related project impacts. Our order book, comprised of year-to-date revenue plus orders in hand, is at circa $360 million at August 2021.
Now, this is 10% above the order book last year of $328 million, and that's in US dollars, and that was 79% of last year's full-year revenue number. The forecast is supported by a stronger order book, higher confidence in the pipeline, and the flagged second-half skew, which is weighted to Q4 due to our customer delivery schedules for e-commerce, digital ads, and search programs. Note that the order book numbers do not include any revenue from Quadrant. It's also worth noting that the H1 to H2 revenue split is in line with historic splits, 2020 excluded, and gross margins are expected to improve in H2 to levels consistent with last year. On expenses, we see modest expense growth in the second-half of 2021, with the savings from the first-half restructure flowing in 2022, and we expect those to be reinvested.
Finally, we expect full year underlying EBITDA margins to be in line with last year. To conclude, we have a leading position in a dynamic market with strong long-term tailwinds. Please turn to the final slide, number 33. We are the world's largest player in the dynamic and growing market for AI training data. Our market leading crowd and technology combine to give us an unrivaled set of capabilities to respond to an increasing variety of use cases and opportunities. We're responding to and navigating the evolving needs of our major customers with agility and pace, albeit with some near-term impact. We have confidence that their wide-ranging investment in AI and their need for high-quality training data, and our track record of delivery and strong relationships will drive future revenue and demand for us.
We continue to invest in new technology and market expansion and are seeing the fruits of those investments in areas from machine learning-based productivity improvements to rapid sales growth in China. We welcome the team from Quadrant to Appen and look forward to working with them and winning in the location data market. We're pleased to be able to support our global crowd and are pleased to be involved in important initiatives such as the World Economic Forum's promotion of Responsible AI standards. Finally, I'd like to thank everyone at Appen for their passion, expertise, and friendship. Thanks and well done. I'll now hand it back to the moderator to take questions.
Thank you. If you wish to ask a question, please press star one on your telephone and wait for your name to be announced. If you wish to cancel your request, please press star two. If you're on a speakerphone, please pick up the handset to ask your question. Your first question comes from Garry Sherriff from RBC. Please go ahead.
Morning, Mark, Kevin, Linda. A few questions. The first one in relation to ad-related revenue. Can you maybe just clarify for us how much of the ad-related revenue was down in the first half of 2021? I look on page nine of your slide. It looks like it's down about 25%-30%. Just wanted to confirm that. Secondly, what have the big tech customers said to you in relation to ad-related revenue, in terms of when it might shift back to historic growth levels?
Yeah. Hey, Garry. You're correct, yeah. That drop is clearly visible on page nine. The projects that are skewing to the second half are generally related to ad and search programs. A big part of that reason, and this is why we see some seasonality year-on-year, is retail towards the end of the year, especially in the U.S. E-commerce ramps up, traditional retail ramps up as customers look at events like Black Friday and Christmas and the like overseas. We see a seasonality in general, and this year we're going to see a little bit more of that. That's been advised by our clients from the beginning of the year.
Okay. Understood. Investment. You flagged high levels of investment continuing. How do you give the market confidence on those spending levels? Just given the customer base typically don't give you much revenue visibility. Put another way, what gives you the confidence that spending more will lead to more customers and revenue?
Yeah. The macro trends give us a lot of confidence, Garry. I mentioned in the presentation the incoming executive for product management, Sujatha Sagiraju. She's been involved in artificial intelligence for many years, and she jumped at the chance to work with us because we're at the beginning of the AI journey in her view. And that's just from tech to across the spectrum of opportunities. And she sees data labeling as the, or training data rather, as the biggest challenge to solve, sorry. And I recall the same conversation with Wilson, our CTO, when I first met him. He was very enamored with the fact that we were solving the training data problem because it's at the heart of every piece of AI, and that's going to be the case for many years to come.
Garry, the other thing as well is the investments are geared towards the new markets, and obviously we're showing strong growth in there, and look to entrench our market position and to continue to grow the revenue there through the investment.
Okay. Final question on Quadrant. Apologies, I haven't been all through the pack yet, but I couldn't see any financials on Quadrant. Could you maybe give us a sense on what revenue they did in the last 12 months, what their cash burn was over the period? The next question's in relation to Quadrant around use cases, just to maybe facilitate understanding.
Yeah. The numbers are small. It's an early-stage company, but it's got some really exciting technology. A couple of use cases, the one I alluded to in the presentation. The delivery person that wants to pick up the restaurant meal, they park their bike and dash inside to get the meal. Knowing exactly where to pick up that meal, it could be at the back of the restaurant, it could be at the side of the restaurant, speeds up and streamlines the delivery process. The other application of location data is just knowing where people are and where they move. If you're an advertiser and you want to put a billboard up and you want to know how many people move past this street corner versus how many people move past that street corner. That's another area that they work in as well.
We're pretty excited about it. We see a lot of demand for it by our customers that produce map bases, for example. It feeds into many industries, e-commerce, supply chain and logistics, even autonomous vehicles. We've done some work in this area around the exact location of electronic vehicle charging points, for example. As we demand more and more from our technology for many reasons, knowing where people are and where they need to go is super important.
When I think about that geolocation data, Google and Apple already have got that information. I'm just trying to figure out how, and I assume they sell some of that information to businesses already. How is it different in terms of what Quadrant are providing? Is that level of geolocation more granular or something? What's the difference?
Yeah, that's precisely it. It's more granular. On a map base, you can get the address of a particular business. Just say you're a car dealership. You get the address of the car dealership, the more granular data might be, well, where's the driveway for returns? Sorry, for maintenance. Where are new sales made versus used car sales? Think also about internal location data. In shopping malls and the like. That doesn't appear by and large on current map bases. We're getting strong demand from our largest customers for this sort of data.
Okay. Yep. That makes sense. Last one from me, and sorry to take up so much time, but the integration of Quadrant, how are you thinking about that into existing platforms or will it be standalone?
It'll be standalone in the near term. The integration is a pretty light touch initially. We get our sales teams working together so we can get them inside of our customer base and we invite, or they rather invite members of our crowd into their system as and when required. Over time, we'll look for tighter integration, but the beauty of it is that it'll work as is with a very light integration right from the start.
Helpful. Thanks very much.
Thanks.
Thank you. Your next question comes from Lucy Huang from Bank of America. Please go ahead.
Hi, Mark, Kevin, and Linda. Thanks for taking questions. I just have three. First in terms of gross margin. How should we be thinking about gross margins in the medium term, particularly as new services becomes a bigger contributor to revenue versus ad revenue? Do we think that gross margins will tighten or is there scope for kind of expansion over time? Secondly, you mentioned that you've won 34 new customers in the new services business. Just wondering if you can give us some color around the average contract sizes of these contracts. Have they been growing in size more recently? Thirdly, you mentioned that there's about 31% of committed revenues in global product. Just wondering, what are these commitments like? Are they annual commitments or multi-year commitments? Just some color would be great. Thank you.
Yeah. Thanks, Lucy. I hope I remember all of those. I'll start with the last one first. Typically annual contracts and typically with an automatic renew. To the point on gross margins, we point out in the outlook slide that the gross margins are expected to improve in the second half, consistent with FY 2020. Going forward, gross margins depend upon a couple of things. They depend upon the extent to which we can deploy technology to enhance the work of the crowd worker. It can also sometimes depend upon the nature of the project and the mix of locations. For example, some locations yield higher gross margins than others. There's a variety of factors. Overall, our strategy is to get gross margin expansion by using technology to enhance the crowd worker. The second question was? New customers. Sorry. Oh, right. Yeah.
Look, new customers typically start off small. They can be tens of thousands or hundreds of thousands. That's the nature of the development of AI. It's somewhat experimental in its nature. A customer might have a concept for a product. They come to us, they collect some data, they build a model, they see how the model performs, they may keep collecting the same sort of data, or they may collect additional data, or they may collect different data altogether. Until that product starts to deliver the performance they expect, the projects tend to stay relatively small in the tens or hundreds of thousands before they grow beyond that into the millions.
Great. Thank you.
Thank you. Your next question comes from Quinn Pearson from Credit Suisse. Please go ahead.
Hi. Good morning. Thanks for the time. I guess just firstly, with regards to what is a pretty substantial second half profit skew. There's some bridging items on the outlook slide, which is quite helpful, but I guess the year seems to be setting up quite similarly to last year, which resulted in a fourth quarter downgrade. I guess my question is just kind of in simplistic terms, I guess, how do we get more comfort that, I guess the second half bridging items come to fruition and I guess, what's different this year than last year? Thanks.
Thanks, Quinn. There's a few points. Most importantly is the strength of the order book. The order book is stronger this year than last year in terms of the quality of that order book, the orders that we have in it, for example. Also there's strength in the pipeline, which is the gap from the order book to the full year number. Our deals are in certain stages, if you will, and we have more late stage deals in that pipeline this year versus last year. Those two things are super important. The strength of the underlying order book and the gap to the full year number is much more late stage in its development. The second thing that's important is the split from the first half to the second half is very consistent with prior years.
2020 aside, it's very consistent with the years prior to that. The third thing is the conversations we're having with the customer. Clearly last year we were a bit surprised by some movements late in the year from the customer. We've been very close to the customer throughout this year, and we're tracking the work that we're doing with them very closely, and we're monitoring demand on a very, very granular level to make sure that we don't bump into the same problems as we did last year. Overall, we have strong confidence in the pipeline. We have strong confidence in the order book. We're very close to the customers in this regard. The things the customers tell us are very consistent with the way the year's played out to date.
Very mindful of last year's experiences, but there's a number of factors that go into our confidence for this year.
That's very helpful, Colin. Thank you. Secondly, with regards to the cost out, the $15 million of cost out more kind of on the services side, has that been fully actioned? Are we now at that run rate? Can you talk us through the deployment of that more into the product side? Related to that, I'm just wondering if there's a little bit of kind of mismatch that will occur in the second half such that some of that will maybe be backed from a profit perspective. Thanks.
The bulk of that will be actioned next year or will come to fruition next year. The bulk of that will be reinvested into the business, most probably into product development. The technology base that we're building is super important for the growth of the business in order to, as explained to Lucy's question, automate many of the functions that we have internally and with our crowd to expand our gross margins. All of that, not all of that, sorry, the bulk of that benefit will flow through next year and the bulk of it will be reinvested into product.
Okay, just to clarify what I heard. The bulk of the actual realizing of the cost out will be the next year as well?
Yes.
Yes. Okay. Thank you. With regards to some of your new markets, to me, it seems like the quality of the platform itself is kind of key to winning and gaining share. I guess, can you talk us through how you feel your current tech platform and tech capabilities compare to key peers? I'm cognizant one angle is that you have, I think, a competitive advantage by being able to access a high-quality crowd, just particularly on the technology side of things, I guess, how do you think you compare to peers and what, if any, gaps are there to fill? Thanks.
Yeah. We run a fairly regularly updated matrix of features and functions across our competitors to the extent that we can find information about them in public sources. Our view is that we compare very well with all of our competitors. Some of them have particular strengths in particular areas. Typically, they start off in an area such as computer vision, and they build strong capabilities in those areas. Where we have the advantage is that we can cover multiple areas, and we're adding to those areas such as the addition of Quadrant. Our ability to go across such a broad variety of data types and use cases is extremely important because the use cases are expanding as we explained in the presentation.
There are very few, if any, opportunities that we have to walk away from, compared to certainly our view of our competitors, is they tend to be focused on a particular area and a particular niche. We feel pretty good about our technology, and you're absolutely correct that our crowd capabilities are pretty extensive as well. From our perspective, we have a really good set of capabilities, the combination of technology and crowd to tackle many, if not all, use cases that we're seeing.
Helpful. Thank you for your time.
Thanks, Quinn.
Thank you. Your next question comes from Josh Kannourakis from Barrenjoey. Please go ahead.
Hi, Mark and Kevin. Thanks for taking my question. Just firstly on the new product and investment, could you give us a bit of an idea of how you're thinking about that level of investment as a percentage of sales moving forward, please?
Hey, Josh. We call out 10.8% of revenue in the first half, and that's been edging up, which I think is visible on one of the finance slides, 28. We've seen that pick up as a percentage of revenue over time. I expect that we'd continue to invest into product. We haven't disclosed the exact amount over time, but product investment is super important.
Yeah. Obviously, with the introduction of a new SPP product, there'd be a lot of new initiatives and so obviously a lot of direction that we'll work with her on into the future.
Okay, great. Just further into that, I guess it looks, when you look at the new products, some good granularity around, I guess, getting further up the value chain in terms of that relationship on the build of the algorithms as well. When you compare it to peers, say, for example, like Scale AI, what do you see as some of the roadmap of new products or services that you see selling to your customer base over the next two to three years?
Yeah, as Kevin says, Sujatha's going to have a big influence over where we head in that direction. As we identify in the deck, there's a big opportunity beyond the collection and labeling of data into all other facets of helping people build AI. That roadmap's still under development, Josh, but there's plenty of directions that we can go.
Great. Just obviously, you've mentioned a bit of detail around the acquisition Quadrant. If we look at other adjacencies and how you're thinking about the sort of M&A strategy, is that a fair assumption that you're also looking at things upstreams in terms of the value and services around the algorithms themselves, or is it more so around the data provision? Thanks.
In the near term, it's more around the data provision. We see that as a very valuable and very doable thing for us. Heading up stack into the world of models and other things is something that is going to be firmly on Sujatha's desk once she starts, because there's a lot of exciting things in that space. There's a lot of early things in that space. We think our most exploitable near-term opportunities are around data sources. That's our focus for now, and we'll tackle up stack areas as Sujatha comes on board and gets a feel for things.
Great. Just final one, just on China, just the lazy 5.8x growth on the period, and I guess we're keen to understand a little bit about what you're seeing from the end customer trends, the expansion within the existing customers, just maybe a little bit more color on the growth opportunity there.
Yeah. It mirrors what we've seen in the U.S. in terms of the trends and demand for data. We get into a customer with a sort of an anchor project, and it grows from there, and then we start growing across the customer into different data types and different modalities. One thing we've been very careful to do in China, is to add a lot of customers outside of the tech giants. We've got a really good foundation in autonomous vehicles, which is very exciting. We're also pleased that we're winning. I think we talked about this in prior periods. Clearly, we bring a strength in our international capabilities, but we're winning a lot of local language work as well.
The summary is it's not dissimilar to the U.S. pattern of sort of land and expand, if you will, but we're also making sure that we win plenty of new customers to grow our customer base.
Great. Thank you.
Thank you. Your next question comes from Michael Aspinall from Jefferies. Please go ahead.
Yeah. Mike, Kevin and Linda. You've touched on it a couple of times, but I don't think I've quite understood just on the data collection. Sorry about that. I would've thought lower growth in global services and higher growth in new markets had a positive mix effect.
Sorry. I'm gonna have to either ask you or Fido to repeat the question.
Yeah. It's Ziggy. I'll go again. I would've thought that lower growth in global services and higher growth in new markets would have had a positive mix effect on gross margins, but that's not coming through. Can you just talk to that a little bit more? I don't think I quite understood what you mentioned before.
Yeah. No. Sorry. Thank you. Some of the global services projects have been going for some time and have very good gross margins. Many of the new projects are in startup mode and have lower gross margins. The projects in general tend to improve, sorry, the gross margins of the projects tend to improve over time, as we get more practiced in doing what the customer is after and deploying more technology to help them, et cetera. Your intuition is correct in one sense in that product-based stuff is gonna have higher margins, but there's the timing elements of the projects that means newer projects have lower margins and more established ones have higher margins.
Okay. No, that makes sense. Following on from that, on slide 10, you show revenue from projects starting before 2021, and the revenue split. Are those newer projects structurally smaller than those started prior to 2021? Could the average size grow to that of what you see in the older project bucket? I think it's about $800,000 versus $200,000.
Yeah. In general, yes. The only thing I would call out is, you know, a handful of extremely large projects. We have a set of very large projects, and not all projects are going to get to that very large stage. It's not unusual for projects to start in the tens of thousands, get to the hundreds of thousands, get to the millions. Getting them to the $10 million, $20 million, and above million range is probably more the exception than the rule, but it's very easy to get them into the millions.
Okay. Yeah, that's very useful. Year-to-date revenue and orders in hand +10% so far, which compares to the first half of -7%. That implies 20% revenue growth in the second half. If I just put the pipeline and the order book aside for one minute, have you seen that in the first two months of the second half?
Our confidence in the second half is very solid. As explained earlier, the order book is better than last year. The pipeline, which is the gap between the order book and the full-year number, is more advanced than last year. Our confidence is high. We are very aware of last year's experience, so we've spent a lot of time with customers, a lot of time monitoring projects, making sure that we don't run into the same issue as we did last year.
Yeah. I mean, just on what happened so far, are you able to comment on whether you've seen those improving trends in July and August?
Other than what I've said, Michael, no.
Okay. That revenue growth that we're expecting in 2H2021, should we expect that to flow through in the first half of next year? Is that larger 2H skew just going to be something that exists from here on out?
The trend into the second half of this year is due to a very firm second half skew that our customers called out earlier in the year. Whether that will repeat itself next year is too early to tell. It's based on the advice from our customers and the projects that they had ongoing and the work that they needed to do on new projects, they wanted to have a heavy skew to the second half, in 2021.
Okay. Great. Thanks for that, and sorry about the dog.
Say hi to Ziggy.
Thank you. Your next question comes from Siraj Ahmed from Citi. Please go ahead.
Hi, Mark and Kevin. I just have 4 questions. This first one, on the work in hand, you refer to the last year conversion ratio. Is that the right number to look at? Given last year does not have a fourth quarter skew and this year you do have it, shouldn't we be going back to FY 2019 before, which is around 70%? Just can you try to tell the mismatch?
Yeah. Siraj, we specifically call out and help guide in terms of that reference point to last year, because it can be a bit confusing. Remember, though, that what we're talking about here is annual numbers. It's kind of very independent of the skews between H1 and H2, hence the reference to FY 2020 there.
Got it. Second thing on ACV, the committed revenue, that's actually declined from the February levels that you disclosed. Can you understand what's happened there, given that's been a key focus?
Yeah. As explained in an earlier question, there are annual contracts and there are automatic renews on those contracts. Not all contracts renew. There's a churn rate in ACV that every business has to a greater or less degree. Yeah, your observation is correct, and it's due to not all projects renewing in the period.
Mark, would that be the major customers or the newer, smaller customers?
It's a variety of things.
Right. Third thing, this might be a bit early. Given you are reinvesting the savings, which makes sense, how should we think about FY 2023, the data margins sort of indicating that it should be in line with this year? Is that how we should think about it?
Yeah. Siraj, we're still obviously working through all our planning and initiatives together with our investment and how that's deployed together with obviously new product initiatives. I think we'll come out with further guidance into the full year. Obviously, it's going to depend on basically the levels of reinvestment and the areas that they deploy to, and then the revenues that we expect to derive and how much in 2022 and how much of that's building revenues beyond that.
I think also, Siraj, we've been through a couple of rocky halves, second half of last year, first half this year. Our confidence around the second half of this year is very high, and once we get into the second half and towards the end of the second half, we'll have a better view on how things are going to go next year.
Understood. Just last one. Just on the acquisition. Am I right in this not being really an AI data play? It's more of data, but not AI data?
It's definitely data. That data will definitely go into AI in various ways and forms, but it could also be deployed in other ways. For example, some of the algorithms that drive supply chains, they're not AI algorithms. They're different sort of algorithms, they also require vast amounts of data. Having said that, search algorithms, for example, that could also rely on location-based data use AI models. It's a bit of both. It's definitely a valuable source of data. It's very hard to collect source of data, and it's a very reusable source of data. Whilst our overarching mission is to deliver training data for AI, if we've got data that can be used for other technology purposes, we're happy to participate in that as well.
Got it. Thanks, Mike and Kevin.
Yeah. Thank you.
Thank you. Your next question comes from Bob Chen from JP Morgan. Please go ahead.
Hey. Morning, guys. Just a few questions from me. You've obviously called out sort of the second half skew as well as sort of the fourth quarter skew. Can you talk a little bit about that sort of fourth quarter skew and how that compares to some of your historical periods?
Other than the reference to the first half, second half skew that I talked about, that is that the skew this year is comparable to prior years before 2020. We always have a degree of the skew into the fourth quarter, and a lot of that's driven by the U.S. retail market. Online advertising ramps up into the fourth quarter. Searches ramp up into the fourth quarter. E-commerce activity and regular retail activity ramps up into the fourth quarter, and that drives a lot of the work that we do. The skew from a half-on-half basis is comparable to pre-2020 years. The fourth quarter pardon me, has been a feature of the business for some time.
Okay, sure. Just referencing slide nine in the pack. Obviously, you called out the decrease in advertising-related revenues there. You sort of also mentioned that you expected a bit of growth coming through in the second half. Do you expect that advertising revenue stream to ever get back to sort of your historical level?
The background to what's illustrated on slide nine is, if you cast your minds back to the beginning of the pandemic, we pointed out that online ad volumes had taken a hit. It was perhaps a realization for many companies in the online ad world that their hyper-reliance on that platform was something they needed to address. There were other factors as well, privacy, regulatory scrutiny, et cetera. From that point onwards, many of the companies in the ad world started investing very heavily in other projects. They were already doing some non-ad development prior to that, but it ramped up quite considerably. That had a knock-on effect to our business, which is very visible in the slides on page nine. Having said all of that, this is not a sort of a switch from ad revenue to other revenue.
It's the addition of other product revenue alongside the ad revenue, which continues to grow strongly for our customers. We anticipate that ads will be a feature of our business and part of our revenue for some time. It's likely that the nature of the projects will change because the underlying solutions will change. Their solutions are going to be reliant on different types of data. They're going to be solutions that are more cognizant of data privacy, et cetera. We do expect that those solutions will be data-based and will be machine learning based, and they'll require training data. All of that, Bob, is to say that there's some sort of underlying things that go into what's on page nine.
The reliance that our customers have on advertising revenue makes it extremely likely that ad revenue will come back, and that's what our customers are telling us. The extent and nature of that comeback is still to be seen.
Okay, sure. I think you also called out in the pack that the opportunities with government have been a little bit slower than expected. Can you talk a little bit about how big that opportunity is, and how that sort of compares to what you were previously thinking?
We don't see any change in the size of the opportunity. It's been called out in prior packs. It's just the speed with which that is developing, is what we're calling out at the moment. We've got a bunch of projects ongoing inside that market. Some of them are very interesting and in some areas that present substantial scale. It's just not moving as quickly as the commercial side of our business. We're just calling that out.
Okay, great. Thanks.
Thank you. Your next question comes from Paul Mason from E&P. Please go ahead.
Hey, guys, just three from me. The first one, I was just wondering if you could talk about slide 21 a bit more, the ML-powered automation projects. Are these in commercial production, like for client use at the moment? If so, are they sort of doing anything on your gross margin performance, or are they still sort of in the R&D phase?
Hey, Paul, some of them are being used in commercial projects, and they are going towards the efficiency of those projects. With any new technology, we start small and grow from there. Not material numbers at this point. The two down the bottom are being used across a number of projects, and they go to our internal efficiency. For example, the fraud detection, we get a lot of people applying for crowd work under multiple names, sometimes hundreds of names that they generate automatically. We're now able to find that extraordinarily quickly, and that benefits us and the customer in two ways. One, it makes us more efficient. Two, the quality of the work for the customer is far higher. The short answer is, we are using some of these in production, but it's early days and entry-level projects at this point.
No material impact to gross margins.
Okay, great. Thanks. The second one, I just wanted to sanity check my interpretation of your guidance on revenue. I know you haven't provided these numbers, but if you can just point out if I've made a flaw in my logic. You're calling out similar EBITDA margins to FY 2020, and the low end of your guidance is $81 million EBITDA. If I use 17.1%, that's about $475 million revenue. In terms of your guidance around the global services revenue, you're still pointing to mid to high single digits. If I use 7%, that's about $350 million. If I'm thinking that it's $350 global services and about $125 to new markets, is there anything wrong with my logic there?
Paul, I'll tell you the one thing, though, is when you're taking the $81 million, you're taking the effect of Quadrant there. Essentially, we haven't factored in Quadrant into any of the numbers that we've talked about in terms of the revenue outlook. I suggest when you're doing that work on that original guidance range, which has not been reduced by the provision, the call-out for Quadrant. I think that's probably the main thing to talk about. I think essentially, if we think about the clues or the guidance element that we provided here, essentially, if we break it down, we're basically saying, look, the order book is up 10% on the prior year. If that's a proxy for full year, we'll then apply that percentage to last year's revenue. All right?
The other guidance point, which is around the EBITDA margins in line with last year. If you do that, then you'll get to numbers that are obviously supporting where we're coming out with the guidance, which is at the lower end of the range. Obviously excluding the Quadrant business.
Okay, great. Thank you. Just the last one from me, just on the revenue plus work in hand number. I'm translating the February numbers that you used to provide in AUD to US, but it looks like the actual dollar gap in February was about $45, and in May at the AGM, it was down to $20, and now it's up to $30 again. Is that reflective, assuming I've got the February numbers accurate, of how your revenue and contracting accumulation has gone? That it sort of dipped a bit in March, April, and now it's ramped up again in terms of your order book?
I'm not 100% sure of the numbers you're quoting. I don't have everything in front of me. I think that the order book reflects the business in a sense. There's a skew from the first half to the second half, as we've explained. That could explain your conclusions there.
Okay, cool. All right. That's all from me. Thanks a lot.
Thank you.
Thank you. Your next question comes from Ross Barrows from Wilsons. Please go ahead.
Hi, good afternoon. Thanks. Just 2 super quick ones, both on Quadrant. It's been explored quite a bit already. Just in terms of team size, looks like it's in the mid-twenties, but any color you can give around that?
Of that order, Ross. Yep. It's a small business.
Yep. Then just lastly, the earn-out where it does mention it's got 2022 and 2023 revenue milestones. Just being particular around that, will that be paid at the end of 2023 once both are achieved, or is it kind of half after 2022 and half after 2023?
It's spread across both.
Yeah. Okay.
Yeah. Ross, just to clarify on that. There's measurement periods, and then the shares will be issued around the time of those measurements, based obviously on the share price around that time.
Yeah, that's great. Thank you.
Thank you. Your next question comes from Conor O'Prey from Canaccord Genuity. Please go ahead.
Hi. Maybe just a quick question for Kevin on share-based payments. Apologies if I missed anything you said there, Kevin. Just they were obviously a contributor of again in the first half, are they going to revert to more like -$5, -$6 in the second half? Like they would have been in the halves of FY 2020?
Sorry, Conor, can you just repeat those numbers again in terms of just the comparative metrics, and then I'll be able to respond?
Yeah. I think they were a net contributor to EBITDA in the first half.
Yeah.
Last year, I think it was -$10 for the full year. Will they revert to an expense or will they continue to contribute in the second half?
No, it will be an expense in the second half. Essentially that trip adjustment is a point in time in terms of expense up to that point in time. Definitely expect that it'll be an expense in H2.
Of a similar magnitude to previous?
Yes. Around those levels. Obviously, with the exclusion of this particular plan that obviously we've adjusted for.
All right. Thank you.
Thank you. Your next question comes from Siraj Ahmed from Citi. Please go ahead.
Mark, a quick follow-up. I know you're not talking about better-than-expected revenue for the second half, but in previous years you've had a fourth quarter surge, which is better-than-expected. Now, given your more confidence in the pipeline, should we be thinking that there may not be a surge if there is one, that you actually have more work than you expect in the year?
That'd be a happy problem. At this stage, we've got a lot of confidence in the numbers that we're calling out, and there's a lot of things to support it, including the strong order book and the strong pipeline. We've seen historically, sometimes the demand goes beyond our expectations, but at this stage, we're calling it the way we are.
Yeah. The other thing, obviously, it's reflective of demand volumes that we've been told about, as Mark talked about, in terms of the later stage quality within the pipeline. Essentially, what we're saying is it's largely based around known demands from what we're seeing right now. Obviously, if there's any change to that up or down, well, obviously that has to be factored in. If there is change to Q4 in terms of significantly higher than what we've been told right now, then obviously there'll be movement there, and similarly the other way.
Just to remind you, Siraj, the delivery requirements or our ability to deliver, sorry, tends to be around the crowd. The crowd can flex up and down pretty quickly.
Yeah, got it. Okay. Thank you.
Thank you. There are no further questions at this time. I will now hand back to Mr. Brayan for closing remarks.
Yeah, thank you very much. Thank you everybody for attending, including kids and pets. A pleasure to hear everybody and glad to hear that everybody's safe amidst the pandemic. Again, just to reiterate the closing remarks, we're very pleased to be the largest player in what is a dynamic and growing market of AI training data. Our leading crowd and technology combined to give us a tremendous set of unrivaled capabilities that enable us to respond to an increasing variety of use cases and opportunities. Thank you to everybody at Appen. Thank you to Kevin and Linda for their work in preparing the presentation and the numbers today. Thanks to everybody for attending and asking questions. That's it for now. Thank you. Speak to you all soon.