Appen Limited (APX) Earnings Call Transcript & Summary

Well, good morning, ladies and gentlemen. My name is Chris Vonwiller. And it's my pleasure as Chairman of Appen to welcome everyone attending the meeting in person and those of you participating online to our Annual General Meeting. I'd like to begin today by acknowledging the traditional custodians of the land on which we gather, the Gadigal people of Eora Nation. I also acknowledge the traditional custodians of the various lands on which you are joining us from and the First Nation's People participating in our meeting. I pay my respects to their elders past, present and emerging. We've provided multiple ways to attend this meeting to encourage greater participation and engagement among our shareholders. It's now just past 10 a.m., the nominated time for the meeting. And I've been informed that a quorum is present. I note that the meeting has been validly constituted, and I'm pleased to declare the meeting open. The Notice of Meeting was made available to all registered shareholders within the notice period required. And with your consent, I'll take that document as read. As set out in the Notice of Meeting, voting on all resolutions will be decided on a poll. In order to provide everyone with an opportunity to vote and in case anyone cannot stay for the whole meeting, I will now formally declare the poll open. There are 3 components to today's meeting. First, I'll provide an update on the business. And this will be followed by a presentation from our Chief Executive Officer, Mark Brayan. We'll allow time for general questions on the business at this time. The Chairman of our Nomination and Remuneration Committee, Bill Pulver, will speak to the remuneration structure prior to the item seeking approval of our 2020 remuneration report. [Operator Instructions] We acknowledge that if you are listening to the meeting via your phone, you won't be able to see the slides that will be spoken to during the meeting. You can see the slides by joining the online webcast, of course. For shareholders in the room who wish to ask a question, please raise your hand at the appropriate time and I'll call on you. There'll be a microphone which will move around so questions can be here heard by the online participants as well as people in the room. The online platform and the phone line are now open for shareholders' questions. And we encourage shareholders to submit their questions as soon as possible. All questions submitted via the online platform or by phone will be read out to the meeting on your behalf. We've received a number of questions prior to the meeting, and we've sought to address these in the upcoming presentations. So thank you for those. We've thought to address them coming up. Following the general business questions, we'll progress to the formal business of the meeting where the resolutions provided in the Notice of Meeting will be put to shareholders. We'll allow time for questions and answers regarding the resolutions before proceeding to vote on them. So I'd like to begin by introducing my fellow directors that are present with us today. Bill Pulver, the Non-Executive Director and Chair of the Nomination and Remuneration Committee; Robin Low, Non-Executive Director and Chair of the Audit and Risk Management Committee; Deena Shiff, Non-Executive Director; and our Chief -- Robin Low is here also, Chairman of the Audit and Risk Management Committee; and our Chief Executive Officer and Managing Director, Mark Brayan. Our other 2 nonexecutive directors, Steve Hasker and Vanessa Liu are attending by phone as they're based in the U.S. We also have various Appen executives present, representatives from the company's auditor, KPMG, Cameron Slapp, the audit partner is here. I also see Iain Laughland from Mills Oakley, our legal adviser is here also. So -- at Link Market Services, who are also providing the online meeting platform today. So in about 7 weeks, our company will celebrate the 25th anniversary since its foundation in July 1996. In these times of the digital services revolution and with strict shorter corporate life cycles, 25 years of continuous and profitable growth represents an impressive journey. Appen was founded with the early recognition that the implementation of intelligent machine learning algorithms, in our case, initially speech recognition systems, needed large volumes of training data delivered to demanding quality standards and with highly reliable delivery times. In the subsequent years, we've built upon this to become the AI industry's largest provider of intelligent training data with global scale. We work in over 235 languages and support AI applications in all data modalities: speech, natural language, text, image, video and beyond. Our success derives from the culture of customer focus and quality, our agility and can-do attitude to addressing novel challenges, talented individuals and relentless focus on innovation. Our role at the center of artificial intelligence, machine learning is exciting but requires ongoing evolution. Such evolution happens more quickly than in most other industries. The Board is ensuring that Appen remains strong, nimble and well positioned to take advantage of changing market dynamics. The announcements we made 9 days ago reflect this. Through investment in technology, Appen is moving to a product-led company compared historically to our revenue being derived from projects and services. So Mark Brayan, our CEO, will speak to this further shortly. Turning now to the financial results for 2020. Total revenue grew by 12% to $599.9 million. Statutory EBITDA increased by 23% to $107.9 million. Underlying EBITDA increased 8% to $108.6 million. This includes growth investments we made, $12.7 million in sales and marketing, technology and our China business. We maintained healthy profitability on sales and achieved an underlying EBITDA margin of 18.1%. Statutory NPAT was $50.5 million, increase of 21%. Underlying NPAT was $64.4 million, which was a small decrease of 1% from 2019 due to the impact of the growth investments we made and higher amortization. The Board declared dividends totaling $0.10 per share, 50% franked through the year. We've spoken in the past of the impact on our customers and on our business of the COVID-19 pandemic. These factors contributed to lower-than-forecast growth in the last quarter of 2020, which resulted in us missing guidance. The Board is oversighting positive actions to enhance our performance. Investment in new technology is continuing. As our business evolves and pivots towards a product and technology platform company, we're seeking greater proportions of contracted revenue streams. And already, we're seeing a significant expansion in our number of customers and the diverse range of industries in which AI is being deployed, ranging from automotive to retail, fashion to health care. In future reporting, we'll be seeking to provide more transparency on the drivers of our financial performance. And in alignment with the sources of our revenue, we'll be reporting in U.S. dollars and we have introduced new segment reporting, reflecting our product-led growth strategy. So a few words on our people. We employ staff in many countries with the highest group in the United States. As you might imagine, there is fierce demand for top talent in AI. And if we were to continue to succeed, we must remunerate fairly and competitively. Our remuneration principles are heavily weighted towards performance-based pay and drive long-term sustainable outperformance. Since our company's IPO in 2015 under Mark Brayan's leadership and under our performance-based reward structure, we've achieved growth in earnings per share and share price far exceeding the ASX norms. The Board believes that this framework is appropriate and it has delivered for shareholders. The Chairman of our Nominations and Remuneration Committee, Bill Pulver, will speak to this later in his report. It's clear, however, that there are shareholder concerns about the structure of Appen's remuneration. The Board will consider this carefully and take this into account. Social and environment. The nature of Appen's business means that we have a relatively low environmental footprint within our own operations. Nevertheless, we continued to reducing the impact of our operations, including buildings, power consumption, travel, water usage and to achieving net zero emissions by 2050. AI opens attractive opportunities but must be implemented ethically. This is an issue of focus for the Board. Appen seeks to assist customers to implement AI solutions, which are fair and unbiased. We've been helping to develop responsible AI standards through our multiyear partnership with the World Economic Forum. In strengthening our governance framework, we've been sensitive to feedback from our stakeholders, including our crowd, employees, customers and shareholders. Our policies and practices have been guided by external frameworks, including the fourth edition of the ASX Corporate Governance Principles and Recommendations, the Sustainability Accounting Standards Board Standards the Integrated Reporting Framework and the Task Force on Climate-related Financial Disclosures. A valued attribute of Appen is the cultural and linguistic diversity of our workforce. Our inclusive practices are guided by our diversity policy. We also seek to maintain Appen as a great place to work with a high-performing culture. And I take this opportunity to thank our employees who have responded impressively to the many challenges through the year. The coronavirus pandemic has changed the way in which we work with our customers and within the company. To keep them safe and informed, we established a COVID response team to define safety protocols for all our offices globally and to update them frequently. Mark Brayan and his executive team have also shown commendable leadership through their quick and effective response to the pandemic and in the strategic realignment of our company. Our more than 1 million crowd continues to be one of Appen's most valuable assets. Our ability to support remote working in a secure environment has been a success factor in the new work environment. We've implemented policies to manage the risks of modern slavery and human rights abuses and we work with our customers to ensure ethical sourcing. Our crowd operations are governed by Appen's Crowd Code of Ethics. On governance, the Board composition requires a mixture of industry knowledge, international business experience, diversity and governance competencies. The Appen Board was strengthened by the appointment during 2020 of Vanessa Liu. Vanessa is based in New York and brings a deep understanding of digital technologies and AI. At this meeting, the Board is recommending the reappointment of 2 directors, Stephen Hasker and Robin Low, and you'll hear from them a bit later. At the same time, we intend to carry out a process of renewal in board composition. We are seeking amendments to Appen's constitution to facilitate a period of smooth transition by providing the flexibility of having more than the current limit of 7 directors at any one time. So in conclusion, ladies and gentlemen, we see the future potential for Appen as attractive as ever. We operate in the most exciting sector of the technology market, artificial intelligence and machine learning. The strengths we have built in our scale and depth and our continued investment in technology, are a foundation on which we can build the next phase of our growth. We're grateful for your support as shareholders. So now I'd like to hand over to Mark for his CEO presentation.

Thanks, Chris. Good morning, everybody. Welcome to the meeting. Good morning to those that have made the effort to come in to see us and hello to everybody online as well. My presentation today includes a recap of last year, some important market trends and what we're doing to remain at the forefront of what is a very dynamic and exciting space. So we are the world's leading provider of training data for artificial intelligence or AI. And AI comprises products that mimic human functions like speech or sight or complex decision-making. AI has developed with a technique called machine learning, which relies on many examples to learn and perform these tasks like humans. The examples are the training data, and that's essential to the development of AI. And it's our mission to ensure that AI works well for all users. And we assist our customers in this pursuit by providing them with data that not only covers all of their possible use cases, but all possible uses as well. And that's important to produce responsible and unbiased AI. So we maintained solid growth in 2020. As Chris said, revenue was up 12% to $599.9 million. Underlying EBITDA up 8%, and that was inclusive of a number of substantial growth investments and provided a margin of 18.1%. As we said at the full year results in -- at the end of February, the year started very strongly for us. But our performance was impacted by a few COVID-related factors in the second half, including COVID's impact on online advertising, which is a major source of revenue for some of our large clients. Correspondingly, the deferral of some of their major projects as a result of that online advertising impact and interruptions to our face-to-face selling motion. I was asked by the fin review at the time, isn't it great that you can run your business remotely over Zoom? And my answer was no, it's not great. Customers and staff benefit from face-to-face interaction. And whilst the technology has been very beneficial for us, we miss that human touch. Despite that slowdown in the second half, we did have some good wins last year. We won 136 new customers, and you can see the impact of COVID in that chart on the left-hand side. So the number of new customers we won declined Q1, Q2 through Q3 and then ticked up nicely in Q4 as the world got used to working under the conditions of the pandemic. We also added 34% projects to our major customers. And this is largely because we're expanding beyond the core areas that we provide for those customers into new product developments, of which more later. And on the right-hand side, revenue in China is growing very nicely at 60% quarter-on-quarter. And China is arguably the second biggest AI market in the world, so we're very pleased with the progress there. We saw an increase in our growth of committed revenue. So our annual contract value, or ACV, was up sharply year-on-year. And our total committed revenue increased to 31% of total revenue in the second half of the year. Committed revenue is driven by our annotation platform, how customers use the platform and stick with the platform for multiple AI product developments. And that was a key component of the acquisition of Figure Eight in 2019. I'd like to turn now to a few important trends in the world of AI. The first is that data continues to be a major obstacle for the development of AI. A survey this year by O'Reilly shows that 18% of respondents see a lack of data or data quality issues as the major bottleneck for AI development. As I said earlier, the training data are the examples that the AI learns from. And without correct or sufficient or accurate enough examples, it learns something different than the intent. The data issues are just 1 percentage point below the top bottleneck, which is the lack of skilled people. And these training data challenges see many data scientists spend a lot of their time on training data, which is on the right-hand side, a recent survey by Airbnb who use a lot of AI in the way that they serve up properties to users, et cetera, found that their data side spend 70% of their time working on data rather than the actual modeling or the creation of AI. Data scientists are unique and expensive individuals. And this is not a good use of their time or company resources. Hence, the importance of a training data provider. The next shift that we see is the move from what's known as model-centric to data-centric AI. So machine learning models are the engines underneath the AI application. So if you use a virtual assistant on your phone, there will be a model that recognizes speech, which is one component of that application. Models are built with 2 things, a set of instructions and the training data. And the question between model-centric versus data-centric is do we improve the model? Do we improve those instructions to get a better outcome? Or do we improve the training data? This particular study, which is run by a company called Landing AI, which was founded by a gentleman called Andrew NG, who is a leading thinker in the world of AI, he tasked his team to see if they could improve the model by either enhancing the instructions or enhancing the data. The model in this case was one computer vision model that detected defects in steel. And you could see from the results that the base model was 76.2% accurate. That is detected defects 3/4 of the time, more or less. The team that tuned the model found that the code were unable to improve that accuracy. However, the team that improved the data grew that accuracy to 93.1%, which may be as good as humans. I'm not an expert in steel defect detection, but that may be a very usable model. So you can see the importance of data in the AI process. And there are similar studies that provide the same results. So this shows that AI yields a better result with improvements to the data, the underlying data in the model. And hence, the challenge of getting data, which was in the first slide, is very real. And this leads to our third trait. That's the automation of -- sorry, the importance of automation in data preparation. So there's a bottleneck around data, but data is important. The other factor is human label data can be very expensive. So companies like ours are investing in techniques to automate the labeling process. We'll talk a little bit more about this when I get to product. But at a high level, we're doing things like using AI to pre label data. So if the task is, for example, to identify the cars in an image of a roadway, and we can use AI to identify, pick a number 3/4 of those cars successfully. And then we sent it to humans who then identify the rest to give human levels of accuracy, but far quicker. The other technique we're using is speed labeling, much like when you use a browser and it fills in the words or the search that you think you're going after. We do a very similar technique with data labeling. Just say the task is to label all of the words on an image of a page of writing, the human can start typing the words in and then the AI will guess what it is and the human can accept it and move on. Finally, we use AI to validate the label quality and return it to the human for work instantaneously. So it doesn't have to go through a laborious rechecking process. All of these techniques lower the cost of producing high-quality data and our product suite employs many of these techniques, and we'll talk more about that later in the presentation. The next trend in AI, and this reflects the question received prior to the meeting. Thank you for that question, is that the developers of AI rely on a lot of different techniques to build an AI product. So most AI products combine a number of models. Back to the personal assistant example. For example, that may have a dozen models alone just for recognizing the speech. It could have a model that recognizes the sound adhering, a model to comprise words from those sounds, a model that builds sentences from those words and a model that works out what the sentence is saying. So there's a lot of science in just that speech recognition, which we take for granted. And then there could be another collection of models to determine an answer or a response to what the assistant has been asked. And then there could be another collection of models to turn that into a response that the person can hear. All of these models may be built with different techniques, a combination of techniques, And of course, the different techniques rely on different amounts of data. For example, if we were building a chatbot for use in a U.S. bank, so it's U.S. accented English, which is fairly common. A chatbot in a bank is a fairly common use case. There may be existing models that we can start with. So in the -- at the top of the slide on the left-hand side, there's a technique called transfer learning. So we start with an existing model, add some of our own data and it builds a model suitable for our purpose. We may be able to, in the center of the screen on the top, use off-the-shelf data because it's a fairly common language. Overall, that particular use case may require less bespoke data to provide a successful outcome. At the other extreme, a chatbot in Arabic, for example, for a marine insurance company, will require a lot more bespoke label data. So in response to the question, which was around one of the techniques, which is self-supervised learning and the question was specifically, will that undermine your business? It adds to the number of techniques that are already in use. And it helps people build more AI. And all of that AI, depending upon the technique, will require some training data in greater or lesser amounts. And the developers use a variety of these techniques so they can produce a high-quality product. And the good news from our perspective is that the techniques on the right-hand side, so supervised learning with off-the-shelf data sets, supervised learning with AI-assisted annotation and then 100% manually labeled data is what we do. And hence, we see a long runway in the business that we have. The other trend is that all of the popular AI development techniques rely on a human touch. So another study by O'Reilly found that supervised learning, deep learning, which is a very data-intensive learning technique, human-in-the-loop or active learning when -- is a technical where if the AI can't find an answer, it goes to a fleet of people. And then finally, knowledge graphs, which is used in search engines. All of those techniques, 4 of the top 5 techniques, are human-heavy and data-heavy techniques, which, of course, is also good for our business. In response to this, our company continues to evolve. In the last 5 years, we've evolved from a company that provides language data to one that provides data in all data modalities from speech, to text, to relevance, to image, to video, most recently in satellite imagery and also in LiDAR. We are shifting our delivery from service-led to product-led, of which more later. We're increasing our committed revenue per the earlier slide in the presentation, which goes to the quality of our earnings. We're also adding more customers, 136 new customers this year to diversify our customer base. Very recently, we've restructured the business to align with our customer cohorts, and we'll be reporting in U.S. dollars for our upcoming half year results given the dependence on the U.S. market and to remove the uncertainty of currency exchange. We recently announced 2 important changes. A focus on product to respond to the market and drive scale, repeatability and margins. And a customer-centric restructure to align us with the needs of our 4 target customer cohorts. Our suite of products is growing to meet the needs of the market and give us the scale and the automation we need to remain at the forefront of the market. Our crowd management platform, Appen Connect, is used by our crowd and our project managers to work with our 1 million-plus skilled contractors. The platform gives us scale, automation and cost-effective management in the large and important resource. Importantly, it also shields our customers from the risk of managing large cohorts of contractors. This is one of the reasons why our customers work with us rather than in-source the capability of training data collection and labeling. Our Annotation Platform, second from the left, acquired with Figure Eight, is the portal through which our customers request and receive their label data. It also provides the tools that our crowd uses to collect and label the data. We also have some new products. Appen Intelligence is a suite of 20 machine learning models, currently 20, it will grow, that automate internal processes and data labeling processes. And recall the importance of automation in improving the unit economics and the speed and quality of data. We use AI to match crowd workers with tasks, much the same as Netflix throws up recommendations for you or your favorite e-commerce site throws up recommendations. We put recommendations in front of our crowd workers. We also use AI to detect potential fraudulent activity amongst our crowd community. Our In-Platform Audit, second from the right, is a product that assesses the quality and completeness of training data sets to help our customers assemble the right data they need before they invest in model development. This is especially important because it ensures that the data sets are complete, representative and unbiased. There are many examples of biased AI where it may be biased against a particular gender or a particular demographic. And having complete data sets is really important to make unbiased AI. Building models is expensive, so this tool helps them assess the quality and completeness of their data set before the expense of model development. Finally, on the right, Appen Mobile, provides a user-friendly mobile-native experience for our workers, many of whom are mobile-first individuals. And it also expands the crowd that we can market to mobile-first economies around the world. We also announced to restructure our business into 4 business units, each with customer and P&L responsibility. Our global team is responsible for our 5 largest customers, the U.S. tech giants. Our approach to them is bespoke, consultative, collaborative and, in many cases or in some cases, uses their platform to do the work. Our enterprise business unit serves all commercial and -- noncommercial or nongovernment customers in every country, except China. Our government unit is purpose-built to serve the emerging needs -- the emerging AI needs of government. And our team in China is rapidly expanding our footprint in the second largest AI market in the world. Supporting our customer-facing BUs are our product, engineering, crowd HR and corporate teams. Note that we are currently hiring an SVP of Product to ensure that our products remain relevant to customers and are market-leading. Our new structure enables us to report on these customer segments. Segmenting by customers shows that our global customers are growing solidly at 14% compound annual growth. And that the combination of enterprise, government and China is growing more rapidly at 25%, albeit off a smaller base. Over time, this higher growth outside of our global customers will help with our customer and revenue diversification and segmenting our teams to give them this focus is very important to achieve that goal. Please note also that all the figures are in U.S. dollars, in line with our change in reporting currency. Segmenting our revenue by the work we do for our global customers on their platforms and the revenue from our new offerings, which includes product-based revenue and our new markets, shows that our new offerings are growing very quickly at 41% compound annual growth. This underscores -- sorry, this includes use of our product by our global customers. The services revenue on the left are at high single-digits growth is impacted by many of those projects running on our platform. And you may recall earlier we said a 34% increase in projects from our major customers. That's because many of them are using our platform to do things other than their core relevance programs. The revenue on the right would not be possible without the products and shows the value of our products. Looking at earnings according to these segments shows that work for our global customers on the platform provides a solid earning base that self-funds our high-growth new offerings on the right-hand side, and it also enable further investment for growth. Looking forward, we expect revenue from services on major customer platforms to grow steadily at mid- to high single digits and product-based and new market revenue to grow in line with the AI market at circa 25%. Our restructure provided some cost benefits that will start to flow in the second half of the year. Finally, our full year guidance remains unchanged. Our order book comprised of year-to-date revenue plus orders in hand sits at USD 260 million. Full year EBITDA of USD 83 million to USD 90 million will be skewed towards the second half due to the project delays that we've talked about previously, the cost base in the first half and the restructuring benefits that will flow in the second half. In summary, our recent moves to be product-led and customer-centric accelerate our transformation to be an AI-powered provider of AI data and solutions. We are well positioned in what is an exciting and forward-facing market. And our investments in technology and AI will ensure that we maintain our market leadership. Thank you once again for joining us at the conference. I'll now hand it back to Chris to take us through the rest of the agenda.

Thanks, Mark. I'd now like to open the meeting for general discussion on the business. You'll have opportunities to ask questions pertaining to each of the resolutions when we get to them. There are 7 items on the agenda. We're going to move through those. Just to remind you, this is a shareholder meeting only. And therefore -- sorry, shareholder meeting and only shareholders are able to ask questions at the meeting. So let's open now to any general questions on the business, not on the specific resolutions, we'll come to that later, but general questions of Mark and the team on the business itself. We'll do that in an orderly way. We'll take questions submitted online before the meeting first. Then questions online during the meeting. And finally, third, we'll move to the audience floor here, and we'll ask questions -- questions can be asked by people in the room here today. So Leanne Ralph is moderating and keeping log of the questions coming in. Leanne, have we received any general business questions as succinct from on the resolutions?

We have, Chris. And the first question comes from Charles Palmer. Can you confirm what investments or decisions Appen is making to combat the increasing competitiveness in the market of AI and to ensure Appen maintains a strong footing in the industry?

Okay. So Mark, can you respond to that?

Okay. Should I stay in my seat? Sorry, take some direction for the experts. Thank you. So we detailed some of those investments in the presentation. AI needs training data. Training data can be expensive. So automating the production of training data is crucial to our business and our success. And some of the automation that we're enacting is, for example, the use of machine learning to pre label data. I used the example of identifying vehicles in an image for autonomous vehicle training. We also use speech recognizers to pre transcribe speech data for our customers. In addition to the automation, our product suite makes it easier for our business and for our crowd to engage with us. Appen Mobile, for example, means that people can -- our crowd workers can do work when they're commuting, for example, which makes them more productive and speeds up the production of training data and lowers the unit economics of training data. Just more a byproduct of the automation. I hope that answers the question.

Leanne?

Yes, Chris. The second question we have comes from [ Shimming Zu ]. What percentage of the autonomous vehicles market do you think you will be able to secure? I ask this as it's a next big thing in AI.

Mark?

Yes. Very exciting opportunity and autonomous vehicles require a vast amount of data. Just think of your own decisions when you're driving a car or, well, still in your own decisions when your 17-year-old daughter is learning to drive a car. And it brings into stark relief the amount of data that an autonomous vehicle may need for even the simplest of conditions. It is, however, hard to put a figure on the size of the market. We have autonomous vehicles customers in the U.S. We have a nicely growing business in the AV market in China. So we think we are well positioned. We have good technology. We're certainly competitive, but it's hard to put a figure on the size of the market and correspondingly our share, but it is a focus for us.

Okay. So Leanne, online questions?

No more online questions at this time, Chris.

All right. So open now for questions from the floor. There are microphones which could come around. Are there any questions on -- these are general questions on the business.

[ Peter Greger ] here, the Australian Shareholders Association. I'd appreciate it with the recently announced restructure of the business into 4 separate business units, each with separate P&L responsibility. How you'll report to shareholders on the results or progress of each of those businesses.

Okay. So again, I'll pass that to management, to Mark, and if necessary, Kevin.

Yes. Thanks, Chris. Thanks, Peter. The restructure gives us a couple of important things. It gives us P&L accountability for each of those teams, and we'll be reporting as we have done by the global team, and we'll combine the others for now until they get to scale of their own reporting, but we'll be able to report on their revenue and their EBITDA, effectively a contribution to the business. And that will help us see how they're growing. It also gives the teams hyper focus and agility on those customers. So for example, rather than -- if they're growing quicker than budget rather than coming to Kevin and I for new headcount, they can look at their P&L and make those decisions on their own. And we found this to be very important for our business given that we're scattered around the world in different time zones. So investors will get to see the progress of those business units, combined enterprise, government and China for now until they grow. And clearly, the global segment as well. And what you should see is high growth for the new markets and solid earnings for the globals. Does that help?

Yes. No, thanks for that. And we look forward to the scale happening quickly so we can see what's happening with each of the business units to grow. Can you just outline the challenges that you see within each of the 4 business units?

Yes. So the global customers, the major challenge is -- or 2 challenges, maintaining growth. Our global customers, which are the 5 big tech companies in the U.S. and for those potentially new to the company, the work we do is in very forward-facing product development. So they prefer that we don't mention them by name or mention the products that we're working on. But our global customers, we want to maintain our share and grow our share amongst those customers. And we also want to use our technology to improve our margins. So that's a revenue growth and margin expansion story. For our new markets, enterprise, China and government, it's much more a high-growth story. So we'll be investing into those markets to get high revenue growth. Those markets will be much more product-led as well. So we need to invest in our product at the same time to make sure that we've got features that outstrip our competitors.

So you aren't able to share, for example, the challenges you face in China, government or the enterprise separately?

Yes, I can. So enterprise, probably the characteristics I just shared, which is high growth and investing into the product. Government, there's strong demand in government, but there are -- there's a different cadence, if you will, in the way that government purchases new technologies. We have to help us in that arena in particular. We have a number of partnerships with many of the big government technology integrators. In China, It's growing very rapidly. So growth and management of the business is less the issue. We have to keep our eye on the geopolitical situation, which is sort of above our pay grade. But nonetheless, something we have to respond to. Currently, we don't see that as an issue given the appetite that China has for AI data, but we have to keep our eye on that.

Okay. Can I get going?

Yes, please do, Peter.

You've got the mic.

In your recent restructuring announcement and also today, you commented on the expecting increase in revenue from product. And you alluded to, I think, a margin increase from that. Are you able to give a little bit of color as to the extent of that and what you're projecting for the next 2 to 3 years in margin improvement?

So historically, we've shown margin expansion as we've achieved scale and relied more on our technology. And we've used that margin expansion to fund growth. In 2019, it was about technology, building the bedrock of our technology with the acquisition of Figure Eight and hiring a lot of engineers. In 2020, it was more about sales and marketing. So we will continue that general approach as we -- as scale and technology gives us margin expansion, we'll reinvest that margin into the growth of the business. So we haven't forecast what that margin expansion will be. But historically, you could see it. And you can also see a little bit in the presentation the margin of the global services business has improved, and that gives us more funds to invest into growth areas.

Okay. And one final question for now. You're moving to measuring the business in U.S. dollars, which I think is good, reflecting that's where the business is. But as Australian shareholders receiving dividends, we're also concerned about the -- what that means in terms of money coming back to Australia, reporting to the ATO and, of course, paying us dividends. Can you give me some indication of how you're protecting that the currency -- against currency fluctuations in that respect?

Sure. I might ask my partner, Kevin, our CFO, to respond to that one.

Yes. Okay. Thanks, Peter. Yes, so there's actually no real change to that in the context that the business today generates more than 90% of its revenues from overseas, right? And so essentially, costs that we incur in Australian dollars are funded by receipts that come in U.S. dollars. And essentially, that's the earnings, and that's the way we've been paying dividends to date. What we also do as a company, though, is to the extent that we have Australian expenses incurred in Australian dollars, but obviously being covered by funds coming from U.S. dollars, then we do have a hedging policy in place to cover a proportion of those expenses. So that's a process that we have in the past, we will continue to do so. And as such, we will protect Australian shareholders from currency risk around the translation of those funds in order to settle Australian dollar expenses. Thanks.

Thanks, [ Peter Gregory ]. Mr. [ Hebard ]? Sorry, we've got one over here, too.

My name is [ Allan Thompson ]. Just wanting to ask a question on -- you spoke about environment and governance. And you've already restructured Appen in the social area. How much investment are you putting in ESG in terms of creating a facility or a business that will draw other businesses involved?

Let me have a go that first, I guess, Allan, I'll ask by management colleagues to add to it. Obviously, that whole ESG issue is an item for board attention, and we do spend a lot of time on it at a board level. It covers emissions, 0 carbon emissions. We have a pathway through there to -- as soon as it's reasonably practical to get to 0 emissions. So we're moving through metrication, to the Scope 1, 2, 3 types of emissions and trying to calibrate that. We've put a stake in the ground and said 2050 zero emission. But that's only, I think, an interim position. We'd be looking to accelerate that obviously, in timing. It covers ethical AI. That's a big area of attention for us, how artificial intelligence is implemented by our customers without bias, transparently, fairly. So that clearly is something that it's a bit over the horizon for a lot of applications, but it's important. And there are standards now being developed. There's a government framework of principles, which we're following. There are international principles. So it gets a lot of attention. Another area is because we employ a lot of people, either within the business or within the 1 million crowd, how do we remunerate, how do we treat them fairly, how do we handle work opportunities for people with disabilities. So all of those issues, ESG is a very wide ambit. And we've tried to define in the annual report, the key drivers that drive business performance and how that relates to -- these are input factors to eventually providing business performance. So it's on our focus, clearly, it's a recurring item at all meetings. The annual report addresses it in some detail, but that's only a stepping stone. You'll get next year's annual report even more detailed about that ESG topic.

[ John Hebard ]. Mark, can you -- you mentioned about the risks with China. Can you elaborate a little bit more on those risks? And how are you planning to manage those?

Yes. Thank you. So there's -- I guess the risk that we keep our eye on all stem from the current relationship with the U.S. and China. We sit outside of any immediate risk. There are, for example, export controls of U.S. hardware firms into China. We're not subject to any of those. We -- our business in China is run by a local management team. So from that perspective, it appears as a Chinese business locally, we've also established good relationships with local government, which is important in China, not just for doing business, but managing risk. And the size of the business, although growing sharply is not one that is on the radar at any point. We've consulted a China expert in this field. And the general view is that because we are providing a service, we're using our expert knowledge to provide a valuable service to Chinese companies and that we're employing people in China. The general view is that we present very little risk to our business. However, that could change and it could change quickly. And so we just keep an eye on it. So in general, for now, we don't see a lot of risk, but things can change quickly, and we just have to keep an eye on that.

Just one moment, [ John ], we'll just get the mic.

Yes, for the viewers at home.

What about the loss of any intellectual property?

Sorry, I should have covered that. We have a completely separate tech stack in China. So we have an engineering team that's building a tech stack That is not -- so none of our U.S. developed technology is imported by China or exported to China. China uses their own tech stack. So we protect our IP in that regard. The other important thing is data protection. Both China and other countries have strict data protection laws. So all of the data that we work on for our Chinese customers stays in China. And correspondingly, the data for our non-Chinese customers stays outside of China.

[ Kerry Bible ]. You mentioned that in the second half of '21, we're likely to see an improvement. Does that mean that the share price might go up? Because in actual fact, it's gone down by more than 50%. Now, I know you missed guidance, but not by that much. And if you listen to what you're saying, everything is rosy and now everything is going to increase. But some things happen, the market doesn't seem to like you.

Well, I can start, I guess, to respond to that. Mark, you can add to it. Look, the company is profitable and growing and has had a great record of growth, earnings per share growth and profitability since we floated in 2015. Clearly, having missed guidance because of, I guess, slowdown in the last quarter of the year due to the pandemic, that hit the share price. And we can't control share price. We can control the drivers of growth. Other factors, the general, I think, move away from tech stocks happening in the U.S. coming through to here to move across to value stocks and whatever, that general climate would affect the share price to get us beyond our control. The Aussie dollar strengthened against the U.S. dollar. So because we are a U.S. dollar earnings company, that can affect share price as well. So in the end, I mean I think the analysts can speak for themselves. But I suspect there was a climate of expectation that we will beat guidance 6 months after 6 months and kind of that -- despite the fact that we try to give guidance without any kind of bias up or down, but is a straightforward statement, people made assumptions about growth and having then missed guidance in 1 half year, that caused an effect on the share price. But we can all speculate what drives share price. In the end, what we could affect -- what we can drive is the growth of the company, the earnings performance, and that's what we're focused on. Mark, do you want to add to it?

I'd add. Yes, clearly, there are some external factors running into the -- or at the outset of the pandemic, investors pulled into tech as great growth opportunity, and we were part of that momentum as the pandemic matured for one of a better reason, investors piled out of tech and we copped a bit of that. In terms of the ferocity of the swing, as Chris said, we provide guidance a year ahead. Investors sort of just played that -- sorry, the analysts, sorry, played that forward, very high rates forever. Now we report growth that's lower, and they play that forward. So on a sort of future value basis because they play forward such a high rate of growth versus a lower rate of growth, it swings the value and the share price responds accordingly. What I can assure you though, is that, as Chris says, the fundamentals of the business are strong. Yes, our performance last year wasn't to our expectations, and I'm a shareholder. And my wife reminds me of that on a regular basis. But we can't control how the market responds to it. Our focus is on the underlying financial performance of the business, investing to make sure that we can maintain strong financial performance into the future. And if we do all of that well and predictably, then the market responds accordingly.

Microphone here. Yes.

[ Curtis Larson ]. You talked in your presentation about self-supervised learning, you gave the example the chatbot and which parts they use all different types and your also talked in terms of models, deep learning and human in the loop. Could you talk about maybe going forward from a data type point of view, which ones you see as more maybe where self-supervised learning could probably take over a lot bigger percentage of that type of data and which ones are perhaps more immune to self-supervised learnings? For example, like Facebook has the image, which they've said that they don't need any annotation or maybe in Tesla and LiDAR, that sort of thing, what you think going forward?

Yes. So the -- there's one important characteristic in the work that we do. The relevance work that we do, which is the majority of our revenue is effectively a human judgment. Is this a good ad for this web page? Is this a good response for this search? And that's really hard to get a machine to do, super hard. Wilson Pang, our CTO, his background is in search at eBay. And I said to him one day, if we can automate that, we're rich. And he kind of shook his head. He's very agreeable, but he wasn't agreeable then. So that means that we'll have that sort of base of work for some time to come. Other data types perhaps in order of difficulty, speech more difficult than image, lend themselves better to different techniques. Automation also lend themselves to the provision of off-the-shelf data. If somebody has a good corpus of U.S. accented speech data that can be used by many people. However, what we see is that there are increasingly more and more diverse edge cases, which per the -- if you recall the slide with the different data types down the page, everything is going down to the bottom. We were asked, for example, to provide images of images of patients getting out of bed in a hospital by age and gender. And that was they wanted to have cameras on people. And if they weren't meant to be getting out of bed, the AI would detect that they were getting out of bed, for example. There's not off-the-shelf data for that. We have to actually capture that data. We have to set up a little platform and people have to get on and off it, and we have to film them. For example, we had another one of people doing summersaults whilst skydiving, I don't know how we solved that one incidentally. But nonetheless, what happens is as AI solves problems, there is an increasing and a growing number of new problems, which leads to an increasing demand for data, for training data. The other factor to take into account is that the best AI is narrow AI. So the more specific the problem you're trying to solve, the better outcome you're going to get. And so if you sort of roll that logic forward, okay, for every problem for every part of the virtual assistant, there's another model, and that requires more specific data. And the last thing I'd say is that of the many data sets around and of a lot of the AI around, it has errors in dealing with a more diverse set of individuals. There are some sort of particularly egregious examples of this around but a simple one could be a chatbot that works well -- or sorry, the speech recognizer that works well with male voices versus female voices because there aren't as many female voices in the data set. So we're increasingly being asked to fill the gaps in our customer data sets. So overall, there's a -- there are, per the presentation, many different techniques and many different demand -- data demand requirements. But the growth of use cases, the specificity of those use cases, the requirement of all of those use cases to work for very diverse user sets means increasingly more and more volumes of high-quality data. Thank you.

Any more questions before we move on?

So let's move now to the formal part of the meeting. I progress now to the 7 items on the agenda. Just a word upfront. Shareholders attending the meeting online will be able to cast their vote using the electronic voting card received when their online registration was validated. If you have any questions about casting your vote online, please refer to the online meeting guide or use the helpline specified on the screen in front of you. The result of the polls will be released to the ASX shortly after the close of the meeting. I've been advised that all proxies received for the meeting have been checked, and I declare them valid for voting. I'll disclose the proxy votes prior to the vote being taken on each item on the screen. These figures are as at the closing time for receipt of proxies, which was 10:00 a.m. Wednesday, 26 May 2021. There are a number of voting exclusions that apply to the resolutions being put to today's meeting. These were outlined in the Notice of Meeting. As Chair of the meeting and as detailed in the notice of meeting, I will vote where authorized any undirected proxies in favor of each resolution. Item 1 is the financial statements and report. The first item of notified business is to receive and consider the financial report, the directors' report and the auditor's report for the year ending 31 December 2020. There's no formal resolution required for this item, but I invite shareholders to ask questions or make comments on the financial report or the reports of the directors and auditors. You could ask questions of or make a comment on the management of the company, ask any questions of the auditor, relevant to the conduct of the audit, preparation and content of the auditor's report, the accounting policies adopted by the company in relation to the preparation of the financial statements or the independence of the auditor in relation to the conduct of the audit. So I'll ask for questions and comments from those online or on the phone first, and then I'll ask for questions from the floor. So Leanne, any questions on the financial report, directors report, auditor's report?

We have no questions on this item, Chris.

Okay. So from the floor? Yes.

Nigel Burgess. Just looking at the intangibles valuations, the acquisitions mainly in 2019. And you don't -- you had a disappointing revenue performance from those acquisitions. Is that fair to say? And is there a sense that maybe some of these values need to be adjusted?

Yes. So twice a year, when we report, we look at the value of our intangibles for specific impairment. Particularly at the full year, we do very detailed modeling. As of the model that we prepared and updated for the 2020 results, that model showed that we had significant headroom in the value of our goodwill. And obviously, this was also very carefully and very methodically checked by our auditors. So off the back of that, we, the management, the Board feels very comfortable at the headroom that we have in that. In addition to that, the levels of growth that we put into the model for future years as expressed in the annual report of very low growth numbers, adding to the confidence around the valuation and the headroom.

I'm just trying to find out where it says in the annual report, but the Figure Eight's revenues actually declined for the year. So are you treating the goodwill as a whole rather than on an individual basis?

Yes. So when -- so first of all, the process we went through is we actually had an external comp firm doing the valuation work for us around the goodwill and the intangibles. Obviously, there's an excess of the purchase price over the net assets. And obviously, what has to happen then is you have to do very detailed work to identify what specifically intangibles all that you can identify. And then obviously, what comp identified is just general goodwill. So what we did was, obviously, the segments that we had at that point in time, we had a relevant segment. We had a speech and image segment. We actually ran through with the Figure Eight team. We got them to classify all of the projects, all of the work they had done on their platform for the year before. And they had about 10 or 11 categories. And we've got them to identify those categories into the 1 of the 2 categories that we had, either speech and image or relevance. And the overwhelming majority of that, I think, was greater than 90% of those tasks and those projects that were done on their platform aligned with our relevant segment. As a result of that, the earnings and the models that we have where we measure the goodwill for the Figure Eight acquisition is against the relevant segment. And that's how we did the modeling, and that was the allocation.

Any further questions? Okay. So let's move on to Item 2. The next item of notified business, which is the remuneration report. This is a voting item. Before we move to questions and a vote on the remuneration report, we'll hand over to Bill Pulver, the Chair of the Nomination and Remuneration Committee for his report. Then we'll take discussion and questions. Bill?

Thank you, Chris, and good morning, everybody. On behalf of the Nomination and Remuneration Committee, I will provide a brief overview of the 2020 Remuneration Report. At Appen, our goal is to ensure that the level and composition of remuneration aligns with the interest of shareholders and allows us to attract and retain high-performing talent. Our framework has been heavily weighted to performance-based pay to ensure that executives think and act like long-term owners of the company. By setting challenging financial targets that incentivize outperformance, we have delivered strong returns for shareholders. We've conducted independent benchmarking on compensation in our key markets to ensure that we remain competitively positioned. We've also kept the framework simple and clear and have provided transparency on targets, assessment and outcomes. Our remuneration structure has remained largely unchanged since Appen listed in 2015. And the company's extraordinary performance over that time indicates that it has been highly successful. 2020 was, of course, a uniquely challenging year for many companies. In Appen's case, our executive team continued to drive performance for shareholders with revenue up 12% and underlying EBITDA up 8% compared to 2019. The team also made progress on our long-term strategy, including expanding into new markets, winning new customers and delivering technology improvements to maintain and grow our competitive advantage. Although our business proved resilient during the pandemic, we were not immune due to the reasons Mark outlined in his presentation. This resulted in a reduction to our full year underlying EBITDA expectations. Despite the impact of COVID-19 on business development and growth, the Board chose to maintain the challenging remuneration targets we had set prior to the pandemic. The short-term incentive hurdles for revenue, underlying EBITDA and gross margin were set approximately 30% higher than 2019 actuals. Stretch targets were also applied to the equity-based long-term incentive plan with the hurdle set at 20% growth in underlying basic earnings per share each year for 3 consecutive years. In relation to STI, executives achieved 86% of the revenue target, 83% of the underlying EBITDA target and 79% of the gross margin target. When adjusted for the sliding scale it applies, the STI payment for executives was between 68% and 71% of the target. With respect to the LTI, the 20% underlying basic earnings per share growth hurdle was not achieved. Executives are now incentivized to work harder to recover the 2020 tranche by delivering the equivalent compound annual growth rate, namely 44% growth over 2 years or 73% over 3 years. We believe this incentivizes outperformance and is a powerful retention tool. Although our remuneration structure did not change in 2020, it is clear that we will have received a first strike in respect to our 2020 Remuneration Report. Prior to today's meeting, I engaged with many of our shareholders to explain our approach and to understand their concerns. The most common adverse feedback we received was an objection to the retesting we allow on the LTI. Some shareholders also expressed concern with the vesting schedule for STI. With respect to Resolution 5 and the grant of rights to the CEO, we also note shareholders' concerns regarding the rebalancing of the CEO's LTI and STI to achieve a pay mix that is closer to market. We appreciate the constructive way in which shareholders have engaged on these issues, and we respect their feedback. The Board will consider shareholder views carefully as we make decisions on remuneration going forward. We note our shared objective of having a remuneration structure that drives long-term sustainable performance. To conclude, we believe that our senior executives delivered solid performance for shareholders in a challenging year. And that the 2020 remuneration outcomes were fair and appropriate. Thank you. And I'll now hand it back to Chris.

Okay. Thanks, Bill. The resolution to the meeting is now being displayed on the screen. I won't read it out. So now open for discussion on this item. Firstly, Leanne, any questions online or by phone?

We have received no questions online, Chris, for this item.

Okay. So would shareholders in the room have any questions? Okay. So we'll move on then to the details of the votes received for this item. They're displayed on the screen. If you've not already done so, please now select either for, against or abstain for resolution 2, if you're online on your voting card. If you're attending the meeting in person or via electronic voting card, or if you're attending online. So we'll allow just a pause for voting, but it's electronic, so it should move pretty quickly. [Voting]

Okay. Thank you. Item 3. The next item of notified business concerns the election of Mr. Stephen Hasker as a Non-Executive Director. So I'll put the resolution to the meeting as displayed on the screen. And before opening the item for discussion, we're going to have a few words from Steve just about his background. So let's run that now.

Hello. I'm Steve Hasker. I've been a Non-Executive Director of Appen since 2015, seeking reelection at today's Annual General Meeting. I've been in the United States living and working for 26 years. First as a partner at McKinsey & Company in New York. Secondly, as Chief Operating Officer of Nielsen. And thirdly, with the alternative asset investment firm, TPG, based in California. During that time, I've worked very extensively with most of the leading players in Silicon Valley, in social media, in e-commerce and in the diversified internet sector and built expertise and exposure to those firms and their strategies and plans over time. I'm currently the President and Chief Executive Officer of Thomson Reuters based in Toronto and New York. And Thomson Reuters is one of the world's leading players in the world of AI and machine learning and its application to solutions and content in the area of legal professionals, tax and accounting and risk fraud and compliance. These experiences and expertise position me well in my view, to be a contributor to the Appen Board. Thank you.

Thanks, Steve. Now open for discussion. So Leanne, questions received before the meeting or online?

We have received no questions on this item.

So questions from the floor? Yes, Peter.

I assume Steve's online?

Steve's on the phone, online.

Okay. Steve, in the document, the Notice of Meeting document, you -- it refers you having deep expertise in information services, artificial intelligence and machine learning. With particular regard to artificial intelligence and machine learning, can you give us a little bit more information about the actual experience you've had?

Yes, Peter. Thank you for the question. So I'd point to a couple of things. First and foremost, I have worked as a consultant and adviser to and partner to the largest Silicon Valley firms in the areas of social media and e-commerce. And so including working with their data science and artificial intelligence teams and understanding in some depth the specific models that they run and the, most importantly, the objectives and outcomes of those models. So as a Director of Appen, I'm able to bring to life for the benefit of the Board and the management team, a perspective as to what some of the biggest customers are trying to achieve when they contract with Appen, firstly. Secondly, as referred to in the short video, Thomson Reuters is one of the largest appliers of AI and machine learning to -- at scale in business relation services. So our leading product, which is called the Westlaw Edge, takes 100 years of case law data predominantly in the United States, but also in the U.K. and markets like Australia, New Zealand. And applies machine learning and AI to predict the outcomes of legal arguments in matters of litigation. And this is one particular application with which I'm obviously very familiar as I'm the Chief Executive of Thomson Reuters. So Peter, I hope that gives you a little bit of a sense for some of the exposure I've had to AI.

Yes. No, that's good. Thanks, Steve.

Thanks, Peter. Thanks, Steve. Any other questions? Okay. So thank you all. We'll now move to the voting. On the screen, votes received so far. If you've not already done so, please either select for, against or abstain for resolution 3 on your voting card. If you're attending the meeting in person or via the electronic voting card if you're attending online. Again, we'll pause for a few seconds for voting. [Voting]

Now Item 4, which is the next item of notified business, the reelection of Ms. Robin Low as a Non-Executive Director, and I put the resolution to the meeting as displayed on the screen. And before we open this for discussion, Robin is here with us today. So I'll ask Robin to come to and say a few words about herself.

Thanks, Chris, and good morning to everybody in the room, and hello to everybody online, wherever you may be. I'm delighted to be standing for reelection to the Appen Board. I joined the Board over 6 years ago prior to the IPO, and it's been wonderful to see this company grow so much in scale and in product range over that time. In terms of my relevant experience, I had 28 years with PricewaterhouseCoopers, where I specialized in audit and risk. At Appen, I'd Chair the Audit and Risk Committee, and I'm also a member of the Nomination Remuneration Committee. In terms of other experience I bring, I'm on 3 other ASX-listed boards, AUB Limited, IPH Limited and Marley Spoon AG. I have some commitments for some not-for-profits, and I'm also on the Australian Reinsurance Pool Corporation, Federal Body. Previously, I was Deputy Chairman of the Auditing and Assurance Standards Board. I believe that the combination of my long professional career and now my quite substantial director career is of value to Appen. However, I know there have been questions about my workload, and I do want to assure shareholders that I have got all the time I need for Appen. There are formal meetings. There are committee meetings. But there's also ad hoc meetings this time with auditors, this time with Kevin, the CFO. And there's plenty of time for all of that. And I believe also that by bringing experience from other boards that actually helps you in being a director on a listed company. So I hope that answers that question, but I'm happy to take further questions. And with your support, I'd be delighted to serve for another term. Thank you.

Thanks, Robin. So open now for discussion. Again, the usual way we'll ask for questions that have come in online. Leanne?

We have received no questions on this item, Chris.

So questions from the floor. Peter?

Robin, thanks for those comments. Appreciate your comments about your extreme workload of a number of other ASX companies, some nonlisted companies, I believe, and also not for profit. You're also the Chair of Appen's Risk Committee, which I think is a pretty critical part of this business. Can you give us some assurance as to how, if you had some competing demands on your time, you would be able to give absolute precedence to the demands that Appen placed on you were something to go wrong?

Look, I've been in this role as a non-Executive Director now for over 7 years, and I have balanced competing demands. I think it's part of the role. And as I said, I always make myself available for every company I'm involved in. I can say there have been times we've done transactions in Appen. There've been times of workload. And along with the other directors, we've all been there for that. So I can give you that assurance. I think I know the -- it's interesting with some of the views on -- are there too many boards? Sometimes, I think there's a qualitative factor as well as quantitative and I think that we have people -- directors may have 3 ASX 50 boards, and I would say that workload is far greater than my workload. So I think it's just I'm very confident of being able to manage it. And I certainly -- maybe Chris could comment. I don't think we've ever had any issues with that, and that's certainly my intention to supporting the Board.

We would strongly recommend Robin's reelection. She is an essential part of our audit and risk framework and a strong contributor to the Board. Any other questions, comments? Okay. So we'll move now to the voting, and the details of votes received are on the screen. If you've not already done so, please select for, against, abstain. Resolution 4 on your voting card if you're here in the room or via your electronic voting card if you're attending online. A couple of seconds to get that done. [Voting]

Item 5, the next item of business is the grant of performance rights to our CEO and Managing Director, Mark Brayan. I put the resolution to the meeting as shown on the screen. Now open for discussion. So again, online questions first, Leanne?

Yes, Chris. We have received one question online from [ Robert Hindman ]. Given that the actions of the CEO will have effects over many, many years, what is the incentive to make decisions giving benefits in 10 or 20 years from now? Examples of long-term decisions would include management succession planning, training, subsidiary acquisitions and product development. These activities may be negative for short-term results, but very good for long-term results.

Yes. It's a good question. Obviously, quite a wide ranging one. Let's ask Bill Pulver as Chair of the Remuneration Committee, to respond to that.

Yes. Look, thank you very much, and a very good question. Clearly, our long-term incentive structure is specifically designed to target this issue. And hence, we have 3-year deferred payment, if you like, from a long-term incentive structure. And I would also point to just the way we target the style and nature of the executives, which is to take senior positions at Appen. And they are people of good character who clearly, while the LTI plan has a 3-year horizon, their long -- their focus clearly goes way beyond that 3 years. And as a result of the LTI plan in normal circumstances, those senior executives become reasonably significant shareholders in the company. And so as such, are completely aligned with the interest of shareholders and the long-term performance of the company. Thank you.

Okay. Thanks, Bill. Any other questions?

We have no more questions online, Chris.

Okay. So questions on the floor. Peter, from the ASA?

Yes, thanks. This resolution includes a change in the STI proportion from 21% to 37% of total remuneration at target. We would like to understand the rationale behind this STI increase bearing in mind that particularly there's an increasing number of small retail shareholders on your share register and that we, in general, are long-term committed shareholders.

Bill?

Thank you, Peter. There was a change in the structure of the compensation of the CEO, which essentially saw a shift of $500,000 out of LTI and into STI as part of the total compensation of $3.5 million. And total target compensation of $3.5 million that actually has not changed for 3 years. We conduct research to get comparable market data both in the U.S. and Australia. And one of the outputs from that research was in the case of our CEO. He's waiting for base salary and STI or the cash component as we call it, was only at around the 37 percentile. We've actually known for a long time that we weight Mark's compensation very heavily towards long-term incentive. But over time, we've been slowly trying to shift it back into a more logical market comparable level. By making the change, it obviously had no impact on his overall compensation. And post change, it still leaves Mark's LTI at 43% of its total compensation. His STI is 37% and his total at-risk compensation, if you like at 79%, with only 21% of his compensation being base salary. So within market norms, that is still a relatively aggressive at-risk structure.

I might add, with a 20% hurdle rate earnings per share growth year-on-year to get the LTI, that's been an extraordinarily tough LTI environment, but one that we've lived with. And we sincerely thought that was best for the shareholders and best for the company, but we acknowledge that there are concerns expressed by shareholders about some aspects of that remuneration schema, and we are going to review that, as I said. Any other questions? All right, we'll move to the votes. Details of the voting is now on the screen. If you've -- it's time to vote. If you've not already done so, please select either for, against, abstain. Resolution 5 on your voting card if you're in the room or via your electronic voting card, if attending online. So time to vote on Item 5. [Voting]

Moving now to Item 6, amendments to the Appen Constitution. These are some updates, as I mentioned in my -- by covering address, we put the resolution to the meeting as shown on the screen. And the context of this was to allow process of Board refreshment by allowing some overlap of incoming directors with existing directors. So we want to increase the ceiling from 7 so that we can do that in a transitional way. Leanne, questions?

We have received no questions on this item, Chris.

Okay. Questions from the floor? Very good. Let's move now to the votes on that item on the screen. And again, time to vote on item 6 for, against abstain, on your voting card or your electronic voting card if attending online. [Voting]

So the final item now is item 7, the nonexecutive directors' remuneration. It relates to the pool of remuneration available -- the total pool of remuneration available to nonexecutive directors. And the resolution is now displayed on the screen. Now open for discussion. Leanne?

We've received no questions on this item either, Chris.

Thank you. In the room? There's no questions. So we'll put the votes that we've received so far on the screen. And we'd ask people now who haven't voted, and would like to vote to select for, against, abstain resolution 7 on the voting card or the electronic voting card if attending online. [Voting]

Good. So ladies and gentlemen, that concludes the formalities of the meeting. Shareholders in attendance at this meeting are asked to complete now your voting cards. To cast the votes for, against or abstain, please place a mark on the corresponding box for each item on your voting card. And then if you place a mark in more than 1 box in relation to a resolution, your vote for that resolution will be invalid. Shareholders online should now submit their votes. The online platform will remain open for 5 minutes after the conclusion of the meeting to allow you to complete your voting on the electronic voting card, allowing for any potential time delay between the physical meeting and the online environment. Before I formally close the meeting, are there any further questions from either shareholders online, by phone or from the room?

We have no further questions online.

Okay. And I see no hands here. So I now declare the poll closed. As I mentioned earlier, the results of the meeting will be announced to the ASX as soon as they've been counted and verified. I now declare the meeting closed. I'd like to take this chance to thank my fellow directors, Mark and his management team for their diligence and commitment to the business. And I'd also like to thank shareholders for your support and your participation today. I look forward to meeting you again at next year's AGM. Thanks very much.