APXYY - Appen Limited (APPEF) Q2 2023 Earnings Call Transcript
2023-08-28 13:52:09 ET
Appen Limited (APPEF)
Q2 2023 Earnings Call Transcript
August 27 2023, 09:00 PM ET
Company Participants
Armughan Ahmad - CEO & President
Justin Miles - Deputy CFO
Conference Call Participants
Wei Sim - Jefferies
Siraj Ahmed - Citigroup
Josh Kannourakis - Barrenjoey
Ross Barrows - Wilsons Advisory
Presentation
Operator
Thank you for standing by, and welcome to the Appen Limited FY '23 Half Year Results. All participants are in a listen-only mode. There will be a presentation followed by a question-and-answer session. [Operator Instructions] I would now like to hand the conference over to Mr. Armughan Ahmad, Chief Executive Officer and President. Please go ahead.
Armughan Ahmad
Thank you very much, Melanie, and good morning, everyone. Welcome to Appen's first half FY '23 results presentation. Today, I'm here with Justin Miles, our Deputy Chief Financial Officer, who will present the financials. I'm also joined by Rosalie Duff, our Head of Investor Relations.
Turning to the agenda on Slide 3, please. There are two main sections in our presentation today. Justin will present our first half financial performance. After Justin, I will provide an update on our turnaround and outlook for FY '23. We will then go into a Q&A session.
Before I hand over to Justin, I want to provide some brief introduction comments, please. We are in an exciting time for artificial intelligence, and Appen continues to play an important role in our industry. Our data and services power the world's leading AI models, and we are seeing growing demand in generative AI. However, our results are far from satisfactory. They reflect the ongoing global macroeconomic pressures and continued slowdown in tech spending, particularly amongst our largest customers.
We remain laser focused on resetting the business. This includes instilling operational rigor across the business, releasing new generative AI-focused products, refreshing our go-to-market and sales capabilities, establishing meaningful ecosystem partnerships and continuing with our AI for good focus. The benefits from our turnaround have yet to show meaningful results. We are seeing early green shoots in our generative AI product offering. However, the revenue growth does not offset the declines we are experiencing in the remainder of the business.
We believe that our go-to-market strategy, along with our strong AI capabilities and market momentum related to generative AI will enable Appen to return to growth. In addition to our turnaround focus, we are continuing to evolve our business model to improve the quality and predictability of our revenue. Today, the majority of our revenue is services-based, as reflected in our global services revenue, we also have a renewed focus on our tech-enabled services business, that includes projects completed on our ADAP software platform, as a lot of our generative AI work is being delivered through this ADAS software platform.
I would like to reiterate that these results are not satisfactory. However, I feel confident that the progress we are making to turn around the business and Appen as a whole, will enable us to capture value from the growing generative AI market and return Appen to growth. I'll expand more on the market opportunity and give an update on our progress, after we walk you through our first half financial performance.
Before I hand it over to Justin Miles, our Deputy CFO, I would like to say a few good words to introduce him. Justin has been with Appen since 2016 as our Vice President of Finance and has over 20 years of experience in finance and accounting. During his seven years at Appen, Justin has worked very closely with our former Chief Financial Officer, Kevin Levine, and has been a key leader in the finance function and in the overall business. As we continue our search for our new Chief Financial Officer, Justin continues to be an invaluable partner to me and our business during this transition phase.
I will now hand the call over to Justin to go through the financials. Justin?
Justin Miles
Thank you, Armughan, and good morning, everyone. I've not met many of you, so I look forward to connecting with some of you over the next day or two. A reminder that we report in US dollars and that all comparisons are to the half year ended 30 June, '22, unless stated otherwise.
Starting with the financial summary on Slide 5. Total revenue decreased 24% to $138.9 million. This is mainly due to a lower contribution from our global customers, reflecting lower volumes as our customers optimize their spend and reduce costs in response to the challenging external environment. This also flows through to the new markets business, where we saw a decrease in global product revenue. Excluding global product revenue, the New Markets business recorded a 5% revenue reduction on the prior corresponding period.
Underlying EBITDA and associated margins were significantly impacted by lower-than-expected gross margin, as well as a proportionally higher cost base versus FY '22. We are also yet to see a material benefit flow-through to EBITDA from our cost-out program announced in May. Primarily due to the decrease in EBITDA, we reported an underlying net loss of $34.2 million. Our statutory net loss after tax of $43.3 million included one-off restructured costs of $6.3 million. $5 million of this relates to the cost reduction program announced in May. The remaining $1.3 million relates to one-off costs associated with changes to the leadership team to align with the strategy refresh and turnaround focus.
Turning to the revenue performance on Slide 6. At the group level, revenue was down 24%. Most of this decline can be attributed to Global Services, as well as the impact from Global Product that I described previously. Global Services revenue declined 27%, impacted by reduced volumes as customers look to optimize and reduce costs. New Markets revenue declined 14% and was primarily impacted by lower contribution from Global Products. Excluding Global Products, New Markets revenue declined by a more modest 5% on 1H '22. There was 10% revenue growth from enterprise Quadrant and government combined, which was offset by lower contribution from China.
China revenue decreased 15% to $15.3 million, due to the protracted impact of the COVID-19 pandemic, which occurred in Q4 '22 and continued into the first half, as well as challenging external conditions. Despite the challenging conditions, Global Services recorded 45 new projects and new markets secured 89 new client wins. And of our non-global deals, 24 were greater than $250,000.
Over to Slide 7. Group underlying EBITDA loss before the impact of FX was $15.7 million, impacted by lower gross margin and a proportionately higher cost base coming out of FY '22. Our operating expenses were up 9% on 1H '22, but was stable compared to 2H '22. As expected, in 1H, we did not see a material benefit flow through to EBITDA from our cost-out program announced in May. This applies to both the Global Services and New Market segments.
The Global Services division reported EBITDA of $8.7 million, down 67% on the prior corresponding period. This reflects the impact of reduced customer spend on revenue and gross margin that I mentioned previously. The proportionately higher cost base coming out of FY '22 also impacted EBITDA. Global Services costs grew 2% in 1H '23 compared to 1H '22 and were down 10% on 2H '22. The decrease on 2H '22 is predominantly due to a lower allocation of indirect costs. New markets reported an EBITDA loss of $21.8 million compared to an EBITDA loss of $15.6 million in 1H '22. This reflects a reduction in global product revenue, increased costs to support Quadrant growth, a lower contribution from China, and as already mentioned, a proportionately higher cost base coming out of FY '22.
Slide 8 shows our investment in product development. Investment of $20.5 million for 1H reflects the continued focus on product development and maximizing customer experience. 52% of spend was capitalized in 1H '23. This is slightly lower than historical levels, predominantly due to the strategy refresh and update to the product roadmap during the half. 14.7% of revenue was reinvested in product development for the half. As a percentage of revenue, this is significantly higher than historical levels, due to the lower-than-expected revenue. As mentioned, there is no material benefit to our cost reduction program in the half year, and we expect to see the full benefit of cost reduction realized in FY '24. By the end of Q3 this year, we expect the annualized investment in product development to be circa $29 million. This is down 29% on 1H '23 annualized.
Turning to Slide 9. We've made strong progress against the $46 million cost-out that we announced in May. As planned, at the end of 1H '23, approximately 63% or $28.9 million of our annualized cost savings target has been achieved. And by the end of Q3, at least 80% of the $46 million cost-out program will be implemented. As already mentioned, the first full year benefit of the cost savings will be from FY '24. We continue to focus on exiting FY '23, with a return to underlying EBITDA and underlying cash EBITDA profitability on an annualized run rate basis. We remain diligent with respect to cost management and will manage our cost base in line with our revenue opportunity and market conditions. With this in mind, further cost savings have been identified and are expected to be realized in FY '23, with the first full year benefit of the cost savings in FY '24.
Over to Slide 10. The cash balance at 30 June was $55.2 million and includes net proceeds from the equity raise of $38.2 million. Cash flow from operations decreased due to lower trading volumes, partially offset by positive working capital movements. There was positive cash flow from operations despite the EBITDA loss, due to strong Q4 '22 customer invoicing, with receipts flowing into Q1 '23. Cash was used during the period to pay CapEx and fund operations while our turnaround remains in progress.
On Slide 11, we present the balance sheet, noting the comparison is to December '22. Our cash balance at 30 June was $55.2 million, and there is no debt at 30 June, '23. The decrease in receivables and contract assets is due to lower revenue volumes. Our noncurrent assets includes goodwill of $53.1 million and identifiable intangible assets of $53.3 million. The increase in current liabilities and decrease in non-current liabilities is mainly due to the reclassification of the Quadrant earn-out liability. We note the Quadrant earn-out remains subject to performance metrics being met. If the metrics are achieved, the earn-out is payable in cash or shares or a combination of both, method of payment is at Appen's discretion and the payment date would be in February '24.
That concludes the financial slides. I'll now hand back to Armughan. Armughan?
Armughan Ahmad
Thank you, Justin. On to Slide 13, I'll now talk to our turnaround progress and provide an updated outlook statement. There's a lot of hype about the potential impact of generative AI, and the level of change continues to accelerate. The market size data on the slide demonstrates the acceleration.
In December last year, generative AI market size was forecasted at $111 billion by 2030. A more recent update by Bloomberg and IDC now predicts that the market will almost be around $900 billion in 2030, that is an 8x market size increase in six months alone. My conversations with many of the CEOs and top customers and our industry experts validates the potential of generative AI. We are very bullish on the impact of generative AI, and we see it as a great opportunity for Appen.
On to Slide 14. For those of you who have attended our Investor Tech Day in May, one of the key takeaways is that humans are critical to the development of generative AI. The transformative technology used to develop generative AI relies on vast amounts of data to train these models. The base models work well out of the box. However, they're prone to error and hallucinations, and the accuracy is not very high. Human feedback, which is called RLHF, Reinforcement Learning with Human Feedback is required to improve the alignment of the base models with human values. This helps minimize the hallucinations, bias and toxicity. Also the base models out of the box often do not understand the specific domain or company context. Human fine-tuning and evaluation is required to provide additional context of these models.
And finally, measuring the effectiveness of generative AI models is a challenge for many enterprises. Humans play a critical role due to the subjective nature of the generative AI outcomes. We mentioned these points in our last investor presentation, and since then, Meta have released a responsible use guide for their Llama 2 open-source model, in this document that's on the slide, they refer to the importance of human feedback for fine-tuning, evaluation of the models and red-teaming, which is a security and monitoring of the model. It's a great validation of Appen's strategy from one of our largest customers, that humans will remain critical for the development of generative AI model.
On to Slide 15, please. And we talk a lot about the potential of generative AI, but clearly, this is a new and -- in addition to our strong capabilities in deep learning AI. We have been providing deep learning data services for 28 years at Appen across relevant data collection and annotation services. Our work is underpinned by software platforms for data and our cloud services. As we expand into generative AI, we are repurposing our data and cloud platforms, to fine-tune and provide these assurance and monitoring services. Our software capabilities, along with our expertise and AI data and global crowd, gives us a very strong head start in the generative AI space, and the ability to provide a full suite of AI data services for our customers. I will provide more details on the LLM customer wins in the later slides.
On to Slide 16. While we have defined our growth strategy, our performance does not reflect the AI market potential. The broad tech slowdown continues to impact the performance -- our performance, especially our customers reduce their spend across most areas. So we have seen a reduction of engineering teams, resulting in lower level of activity in converting opportunities into spend commitments at the customers.
As the expectation of generative AI accelerates, many of our customers are in the phase of evaluating their AI strategies, which has caused some projects to be paused or delayed. Consequently, we are seeing many of the enterprise large Fortune 2000 customers piloting large language models. However, there are very few customers using these models in full production. The combination of market headwinds in some of our large deep learning customers and nascent nature of generative AI market, has resulted in declining revenue. While we can't control the market conditions, however, we can control our operations and therefore, continue to focus on our turnaround. I will now provide an update on our turnaround progress.
On to Slide 17, please. The turnaround started in the first half, not long after I joined the business in January. Since then, we have created a meaningful impact on the operations of the business, with a focus on operational rigor across all of the business functions. We've launched our cost takeout initiatives in the first half, improved product velocity. We are creating a world-class go-to market. We have launched new ecosystem partnerships, and we have made significant progress on our AI for Good mission. In March, we welcomed our new Chief Technology Officer in Saty Bahadur, who has done a great job improving our product velocity. We have achieved a lot in the first six months of our turnaround journey, and continue to now run the business with key metrics and KPIs.
On to Slide 18. In the second half, we're continuing our focus against the same five turnaround strategies as I laid out in the first half. We've made some great inroads here with 63% of our $46 million cost takeout program complete. And as Justin mentioned, we would achieve at least 80% by the end of Q3 FY '23 of that cost-out initiative. We've also deployed a new target operating model, and streamed our crowd onboarding process. I'll talk more about our crowd experience further in a moment.
Product velocity has also improved and so far this year, we have launched new product offerings, including the RLHF and model monitoring solutions, we call LLM benchmarking. In terms of go-to-market, in May, we hired our new Chief Revenue Officer, Andrew Ettinger, who is building out our sales and marketing teams, including a modern sales operations function.
We are also excited to have a new Appen brand refresh coming soon. I will provide a sneak peak in a few slides for your reference. We've also increased our ecosystem partnerships and have joint go-to-market approaches in place. We have hired an experienced partnership executive, who is focused on maximizing the opportunity from our partnership.
And finally, the most important for me, from a values perspective, is our AI for Good focus. We have several important initiatives in place, including the active recruitment of impact sourced crowds from Africa, where we provide work opportunities to refugees and other disadvantaged people. We are also making great progress against our Net Zero pathway. As part of our turnaround strategy and continued cost management, we are also flattening the org structure from 12 layers to six layers of the organization, and a spans ratio from an average of one manager managing four people, to one manager now managing eight people, for each people manager, increasing the accountability throughout the business. This will help us increase our velocity of decision-making and compete in a fast-moving set of start-up competitors.
Related to this, I'm making changes to the executive team, with the following new appointments and departures. Ryan Kolln, our current SVP of Strategy and Innovation, will take on the new position of the Chief Operating Officer of Appen. He will lead the crowd and delivery functions, in addition to his existing role. To create a leaner and more efficient organizational structure, Appen will consolidate the leadership of its product functions. Mike Davie, who is the current General Manager of the Quadrant business, will be taking on the role of our Chief Product Officer for Appen, in addition to his current responsibilities. We are also combining our sales and marketing functions under the leadership of our Chief Revenue Officer, Andrew Ettinger. Sujatha Sagiraju and Fab Dolan have decided to leave the company for personal reasons. We thank Sujatha and Fab for their valuable contributions to Appen, and we wish them all the very best.
On to Slide 19, please. Those who attended our Investor Tech Day would be familiar with the tech stack on the slide. This diagram represents the building blocks required to develop a generative AI application. Appen plays a key role in fine-tuning and monitoring and assurance of the various models in the market, as I mentioned earlier. Fine-tuning aligns the generative AI models with human values to minimize hallucinations, bias and toxicity. Assurance provides model monitoring and evaluation solutions, to ensure that the models meet performance, risk and regulatory requirements for our customers. I will now provide an update on our fine-tuning and assurance offerings that our product and engineering teams have recently launched.
Turning to Slide 20, please. As you know, human feedback is critical to the performance of large language models for the reasons I've just explained. Fine-tuning a model requires a process called reinforcement learning with human feedback. We have launched that RLHF product earlier in the year, and we have now updated it, to include multi-turn conversations. This allows model developers to capture feedback across an entire conversation, rather than a single interaction. It sounds simple, but there is a complex set of product requirements to achieve the routing of different conversational elements to different crowd workers. We're seeing strong customer demand and have recently signed a very large deal with a leading model developer using this specific product.
On to Slide 22, please. Turning to our benchmarking solution. Benchmarking is a methodology for customers to evaluate performance of different generative AI models. And just as a reference, in January, when I started, there was one model, we now have over 60 models in the world ongoing. So these types of solutions will become even more important. So for example, if a customer wants to compare performance of open AI ChatGPT 4 versus Anthropic’s Claude model, they need a way to gather performance metrics against a variety of tests. Our benchmarking solutions enable customers to test different models with our crowd to capture feedback against different dimensions, including model accuracy and toxicity, et cetera.
On to Slide 22, please. Our generative AI products and solutions are showing strong momentum, and let me try to provide some details to you. To date, we have delivered 42 large language model projects and pilots and have 40 deals in the pipeline. I am pleased to say that all of our top hyperscaler customers have completed a generative AI project with Appen or have a project in the pipeline with Appen. On the right-hand side of the slide, you see the examples of various large language model projects we have delivered. Some of these projects have been for generative model builders, but we also count global financial services firms as our customers, along with social media companies as our large language model customers.
As discussed earlier, the market is early in generative AI. Therefore, the revenue impact is small, but the customer deals are moving in the right direction. I'm very pleased with the progress we are making as a team, and we will continue to evolve our products, services and software to capture the growth in generative AI market.
Now on to Slide 23. Our crowd is clearly extremely important to Appen. We love our crowd, and we continue to invest in improving their experience with Appen. That's why I'm very excited to be announcing a new crowd interface that we have launched to improve our crowd members interface with Appen. The big focus here is making it easier for our new crowd members to join Appen and easily identify tasks that are best suited for their needs. It also helps the business by reducing project delivery time and improving fraud signals through the mobile telemetry solutions. This is just a start towards a set of improvements that we will be making to our crowd experience, and it's something that I am closely working with our engineering team and I’m very passionate about.
Now on to something exciting on Slide 24. I would like to share and provide you a preview of our upcoming new Appen tech forward branding. Appen's a technology company, but in the past, our brands become dated, and our website didn't have the look and feel of a modern technology company. I'm very excited to be launching our new brand with a much cleaner and tech-focused look and feel. On the slide in front of you, you can see the various applications of our new Appen branding to be unveiled in September starting with our new website.
Moving on to Slide 25, please. Now on to the progress against our ecosystem partner strategy. As we outlined in May, we have been focused on increasing our channels to market. As discussed, we have hired an experienced ecosystem channel leader, who is driving the value of our partnerships, along with our sales and delivery teams. We are delighted to be collaborating with large partners like NVIDIA, Reka and other cloud hyperscalers, and along with other professional services firms, this provides important access to enterprise customers for Appen, and will enable Appen sales organization to reach more customers. We have made some really strong early progress.
We have made a small minority investment in Reka, which is a model builder, along with Snowflake and a large venture capitalist firm, DST Global Partners and have a good market plan in place with Reka to promote the best practices for deploying generative AR models, including the involvement of humans to reduce bias and toxicity. We have signed our first $1 million deal through the NVIDIA partnership to service a global Fortune 500 company together, and we have many more in the pipeline. Finally, we are in multiple advanced competitions with global professional services firms, cloud hyperscalers, and we are excited to be closely collaborating with them on joint customer opportunities.
On to Slide 26 for the 2023 outlook. I will now share some views on our outlook for the remainder of 2023. As I commented earlier, headwinds from the broader technology market slowdown are persistent, and customers continue to evaluate their AI strategies. Due to ongoing uncertainty across all customers, we now expect the second half FY '23 revenue to be closer to our first half FY '23 revenue. We continue to focus on exiting FY '23 with a return to underlying EBITDA and underlying cash EBITDA profitability on an annualized run rate basis. We will achieve this by prioritizing our growth investments into a smaller set of higher potential areas. In return, this will simplify our business and deliver incremental cost savings, but have -- but may have a negative impact on 2024 revenue. We now expect to exit FY '23 with an annualized run rate operating cost base lower than $113 million.
With that, Melanie, the operator, I would like to open it up to questions, please.
Question-and-Answer Session
Operator
[Operator Instructions] Your first comes from with [Technical Difficulty] with JPMorgan. Go ahead.
Unidentified Analyst
Good morning, guys. A few questions for me. Just on the outlook. I think previously, you've made comments that you're seeing some stabilization with some of your core customers, but it seems -- the outlook sort of seems to suggest that the second half still looks pretty weak. Like, has anything materially changed in your discussions with your customers?
Armughan Ahmad
Thank you, Bob. I appreciate the question. So, your question was about stabilization of our core customers in the second half. We're seeing ongoing cost management at our large customer. We're also evaluating their AI strategies. They're also evaluating their AI strategies, as I mentioned earlier. So it's a base of -- we have a very large global services revenue, while we're seeing these green shoots that I mentioned earlier, Bob.
Unidentified Analyst
Okay, sure. And then your comment as well, on sort of reprioritizing some of your growth investments that might have an impact on '24 revenues. Is that just on the incremental stuff or do you expect that to impact your core '24 revenues as well?
Armughan Ahmad
As I made the comment, it's more about looking at what are the higher priority areas that provide Appen the growth that we need in the coming years. So we're looking at both our deep learning and generative AI areas, but to ensure that we're very focused on the areas where we do want to spend our dollars, and our OpEx dollars towards.
Unidentified Analyst
Okay. Cool. And then just a final one, obviously, the product investment as a percentage of revenues ticked up a bit. How are you thinking about this longer term, given the traction with some of these new products seems to be taking a little bit longer to come through on the revenue side?
Armughan Ahmad
Yeah. And the number you're seeing on product investment is the first half number. And as Justin and I both mentioned, our cost takeout strategies have been fully realized in the first half. So if you take a look at some of the information that we provided on our product slide that Justin mentioned, it will show you that it's not going to be ticking up as we are moving towards more of our July and along with going into August, we do see a lower spend on R&D.
Unidentified Analyst
Okay. Thanks.
Armughan Ahmad
Yeah. Thank you, Bob.
Operator
Thank you. Your next question comes from Wei Sim with Jefferies. Please go ahead.
Wei Sim
Armughan, just in regards to the OpEx comments that Justin made before, I'm just -- what's the OpEx split between Q3 and Q4 historically? Are you able to give us some sense on that?
Armughan Ahmad
I'll let Justin take that.
Justin Miles
Yeah. Thanks, Armughan. Wei Sim, we don't generally provide that split, so there's nothing kind of extra to give there. Historically, the business has been growing in Q3 and Q4, so we've seen an uptick and obviously, the focus now has changed around getting that cost base aligned with the revenue and the market condition.
Wei Sim
Okay. Great. And my next question was just really on the China business. That weakness that we're seeing -- we called out COVID. How do we think about the impact of COVID? When do we expect it to, I guess, start to ease off? And something that I did read about for the competitors was, they were also impacted by some of the changes in terms of data privacy laws which came through. Is that something that also impacted on our China business?
Armughan Ahmad
Thank you, Wei Sim. On the China business, as we mentioned, coming out of COVID that obviously had a downturn and you know the macroeconomic situation in China at this time. We are, however, seeing that we're taking market share away from our competitors there, and we continue to see green shoots, particularly in large language models in China. And we don't see much impact of data privacy. Most of our work is domestic in China. So other customers who are doing global work outside of China, are the ones experiencing the privacy issues.
Wei Sim
Okay. Great. Thank you. That’s it for me.
Armughan Ahmad
Thank you, Wei Sim.
Operator
Thank you. Your next question comes from Siraj Ahmed with Citigroup. Please go ahead.
Armughan Ahmad
Hi, Siraj.
Siraj Ahmed
Hi, Armughan. Just the first question, just in terms of the further cost-out and impact on your FY ’24 revenue, Armughan, just -- I mean, any way you can quantify -- I think it's early stages, but just how much we should be thinking about, and also the impact on revenue that we should expect for next year?
Armughan Ahmad
Yeah, we're not providing what additional cost-outs we need to provide. What we are providing at this time, Siraj, is that we're going to continue to commit to delivering cash EBITDA positive run rate as we exit December. As we do see any types of tech downturns and the tech spending downturn with our top customers, we'll continue to take action. But as I laid out in my talk that we're really investing in key areas that are now showing us those green shoots, while we are doing the spans and layers exercise of our organization that I also highlighted, so that we are fit and competing against much nimbler startup, and that's the process that we are following.
Siraj Ahmed
Maybe -- sorry…
Armughan Ahmad
Yeah. No, I was just going to say if we're [managing] (ph) to the EBITDA breakeven and no major moves that will impact revenue materially and also being much more diligent on the ROI in our investments as well.
Siraj Ahmed
Got it. Maybe I’m doing a fast math. Is there any sort of business that you think, revenue generating today, is not really core or critical that you think going forward? I'm not sure it's moderation or something like that or work are you doing today?
Armughan Ahmad
Yeah, we're evaluating many of the areas, Siraj. LLM changing daily these days, as I mentioned earlier, that they were one model six months ago and there's over 60 models now. I shared how the market has changed from $111 billion by 2030 to over $900 billion. So we're seeing much more nimble -- we're being much more nimble than our investments, so that we're staying ahead of a fast-growing market here.
Siraj Ahmed
Got it. Second one, in terms of gross profit margins, it seems like compared to your trading efforts in April, it’s actually improved in the last few months, margin is actually up year-on-year. Anything to call out there? And secondly, have you -- just in terms of outlook like, are you seeing any pricing pressure from your largest customers?
Armughan Ahmad
Yeah. So I would say, when I take a look at our gross profits and how we are delivering a lot of our solutions, we continue to look to optimize and look to automate a lot of our delivery capabilities. So we're not seeing much pricing pressures from our customers. We are also -- we have a long-standing price agreements with our customers that remains in place, Siraj.
Siraj Ahmed
Last one from me. In terms of competition, you sort of mentioned a couple of times, being nimble against competition. Are you seeing any increase in intensity competition from new players out there that's impacting you as well?
Armughan Ahmad
Yeah. Siraj, a lot of our competitors are not the legacy competitors of Appen anymore. As you can see, some of our competitors are also seeing a similar kind of downturn, who are focused on the deep learning AI side. A lot of the competitors that we're seeing on the generative AI side happen to be VC-backed startups that are moving a lot faster. But with smaller -- but the work is much more competitive, but it's, as I said, with smaller players. We're seeing more opportunities as customers look at scaling their large language models work globally in multi-country, and Appen has been doing this for a long time. And as I mentioned, majority of our hyperscalers have completed or have a pipeline, with Appen similar to the NVIDIA comment I made, where we have now closed a $1 million Fortune 500 customer deal with them, and we have a lot more in the pipeline, professional services firms, cloud hyperscalers. So this is where Appen, being a publicly traded, reputable company, moving a lot faster versus some of these private high valuation startups, that differentiation remains a differentiation for us.
Operator
Thank you. Your next question comes from [Andrew Giles] (ph) with Macquarie. Please go ahead.
Armughan Ahmad
Hi, Andrew.
Unidentified Analyst
Good morning, guys. How’s it going. First one from me. Just wondering if you could provide any color on the revenue guidance to be flat half-on-half? I appreciate this assumes no change in macroeconomic conditions. But can you maybe clarify, if it assumes any normalization in the trends you're seeing in your key customers, please?
Justin Miles
Yeah, I'll take that one, Andrew. Hi, Andrew, it’s Justin. What it does assume is that historically -- and when we gave that update in May, those assumptions built around in terms of the 1H, 2H skewing, and that's what Appen has seen year-on-year and particularly geared towards Q4. What has changed now, is there is this level of uncertainty around that skewing. And that's come about for all the reasons that Armughan has talked about before, right, is the customers optimizing their spend, and focusing where they put their money. And just through discussions with the customers, the signals are not giving us a great level of confidence that, that skew would be there. And that's why we've called out the uncertainty and I think it will be closer to H1. I don't think it's necessarily a trend forever, but that's what we're seeing right now.
Unidentified Analyst
Perfect. Thank you. And then just one more on gross margin. I was wondering if you could flush out a little bit more. Just in particular, crowd costs are going up and discounting. And if you can maybe give a feel for kind of the percentage changes in each of those elements, please?
Justin Miles
Yeah. So in terms of discounting, we're not seeing anything there. If you look at our crowd costs, they're pretty flat on the same period last year. So again, not seeing a great deal of pressure there.
Unidentified Analyst
Perfect. That’s it for me. Thank you.
Armughan Ahmad
[Technical Difficulty] Andrew.
Operator
Thank you. Your next question comes from Josh Kannourakis with Barrenjoey. Please go ahead.
Armughan Ahmad
Hi, Josh.
Josh Kannourakis
Hi, guys. Thank you very much for taking my questions. First one, just with regard to some of the larger customers. You previously mentioned that, some number two customer had been a bit more stable than your first one and continued to grow. Has there been any change in trend in the last three to six months in that customer, particularly?
Armughan Ahmad
Number two customer continues to stay strong. When we provided that information during the equity raise, we were just providing all of the context, and we'll look to provide a lot more information at the end of the year.
Josh Kannourakis
Okay. So that's great. So it's primarily at number one customer. Yeah, got it. Just in terms of second question around the working capital move, there is a pretty significant positive working cap benefit in the period. Could you maybe just talk through that and just give us a bit of context around maybe how much of that has unwound or we should consider as unwinding into the sort of second half?
Justin Miles
Yeah. Sure. Hi, Josh. Yeah, and a lot of it has to do with the strong Q4 where customer invoicing was quite high and that cash flowed through into Q1. The majority of it has unwound.
Josh Kannourakis
Yeah. Got it. That's great. And just final question, just with regard to some of the green shoots you're seeing across the generative AI stuff, can you maybe give us some context around the enterprise customers that you've talked to the hyperscale projects, but in terms of the enterprise level, just where do you think people are at, in their sort of lifecycles for gen AI, how long, I guess, do you think in your minds before some of the contracts they start committing to are going to be a bit more material for the business?
Armughan Ahmad
Yeah. Josh, it's Armughan. So we find it very early at this time. We're working with financial services, top retailers, insurance companies, manufacturing companies around the world, and we're finding that there's lots of exploration going on, where they have a pilot going on. There's a lot of need for support to ensure that these models, as I mentioned earlier, they're all fine and dandy when it's -- you're using it for fun and games as a B2C approach for consumer use case. But as soon as you look to use it in an enterprise, especially in a regulated market, there's a lot more challenges for them to deploy it in production. So we think that next year will be where enterprises will seek to invest more heavily in generative AI, aligned to their budget cycles.
Josh Kannourakis
Got it. And final really quick one, just on the exec team and changes. Is the majority of that largely sort of positioned now like in terms of the go-to-market strategy in terms of the different segments that you've now aligned? Like maybe just to give us a bit of context on how stable, I guess -- hopefully, the HR team is or if there's further hiring you need to do, maybe just a bit more context on that comment?
Armughan Ahmad
Yeah. Thanks, Josh. As we're continuously adjusting Appen to become a lot more nimbler to compete against these startups, we're always looking to see how we can optimize ourselves. As I mentioned, I provided you some spans and layers numbers and how we're trying to adjust Appen into a much more of a startup like, much more nimbler organization. I feel that the current leaders that we have in place are solid, and we're going to continue to evolve. And as I mentioned that the CFO recruiting is ongoing, and that's probably going to be our next big hire.
Josh Kannourakis
Okay. That’s great. Thank you, guys. Appreciate the questions.
Armughan Ahmad
Thank you, Josh.
Operator
Thank you. Your next question comes from [indiscernible]. Please go ahead.
Unidentified Analyst
Good morning, Armughan. Thank you for having us.
Armughan Ahmad
Hi, Karan.
Unidentified Analyst
So my first question -- yes -- so my first question was, is China's stagflation a concern for Appen given that a lot of business is moving to India from China?
Armughan Ahmad
Yeah. Our China business continues to, as I mentioned, be a very strategic part of Appen. There are some macroeconomic conditions that China is going through, as I mentioned earlier. We do continue to see a huge potential in China. India, where we have recently opened our Hyderabad engineering facility there, and we'll continue to invest in that area. LLMs are getting big in China, and they have created a lot of great opportunities for us there. A lot of the hyperscalers along with enterprise and telco and others are working with us. And as I mentioned, we're actually taking share from our competitors there.
Unidentified Analyst
Okay. Awesome. Thank you for that. So my next question was, so we've seen the NVIDIA story, right? We've seen in the last few years that they've been achieving multibagger gains and Appen is a partner of NVIDIA, yet we haven't seen any of this trickle down and reflected in the share price. Is this possibly a one-sided partnership?
Armughan Ahmad
Well, if you take a look at our revenues and where revenues come from, it's very much a global services revenue. And as I mentioned, large language models are still getting up to speed, with a lot of large enterprises. So doesn't matter if it's NVIDIA or if it's any of the other LLM providers or hyperscalers, I think everyone is seeing a similar kind of process in customer adoption on large language models. So the benefits from our green shoots are not offsetting the declines in other areas, as I mentioned. Its early days in our partnership with NVIDIA. I did mention that we have a $1 million win, that we already have had with a Fortune 500 customer. We're very bullish on our partnership there, and particularly as large language model picks up.
Unidentified Analyst
Okay. Thank you for that. I just had a final question. So has there been any consideration that maybe the crowd model is dead? Maybe there's a disruption somewhere that Appen is not able to take advantage of? And we're just working with rebranding the old stuff, like maybe there's something new and maybe the old needs to be gone.
Armughan Ahmad
So, Karan, I would encourage you to take a look at some of the slides I shared earlier in terms of what Meta calls their Meta Llama 2 use guide, that was in particular to answer that question that you just asked, which is the human in the loop continues to be very important, which is through the crowd, to ensure that there is less toxicity, less bias. We still see strong demand for the crowd model. I speak to our top customers regularly on this, and we're exploring software opportunities that are less reliant on the crowd, along with the crowd capability. So almost think of it as our enterprise customers want internal crowd, as well as external crowd or a hybrid crowd kind of a model. Those are some things that we're seeing a big uptick on. So crowd continues to be very important for us.
Operator
Thank you. Your next question comes from Ross Barrows with Wilsons Advisory. Please go ahead.
Armughan Ahmad
Hey, Ross.
Ross Barrows
Hey, thanks for taking the call and the questions. Just a couple for me. In terms of where you're talking about the customers that are reconsidering the AI strategy, can you share a bit more color around that? Like you have mentioned there are some smaller engineering teams, but are they trying to do more in-house, is it just overall reduced spend. So maybe some color in that regard?
Armughan Ahmad
Yeah. Thank you, Ross. I think what I meant by that is, lots for valuation of LLMs going on, and considering the year of efficiency that a lot of these big customers of ours, especially the hyperscalers have continued to slow down their internal spending. That's one. The second part is really that evaluation that they're doing on large language models. That's what I was referring to that their engineering teams are focused on, looking at both deep learning and large language models that they make their decision. That's what I was referring to earlier, Ross.
Ross Barrows
Got it. Thanks. Just a question on staffing, and I know, I would likely get it later in the day. But just there have been a few hires that haven't really played out in terms of tenure, I guess, from the CFO to the CMO, case in point. Can you share a little bit of color or some other information around, I guess, those roles and I guess why they didn't quite pan out in terms of duration?
Armughan Ahmad
Ross, there are -- the changes that we're going through at Appen, as we obviously put our best foot forward to have the folks join for this transformation that we're going through. We don't always get the fit right, particularly during a turnaround, and we have to respect people's personal reasons to depart.
Ross Barrows
Okay. Thanks. And just the last one is around gen AI. Obviously, it's an incredible opportunity and you are very well placed to participate in that. I guess, balancing that against some of your comments where, some of your competitors there are probably newer, younger, PE backed, for example, and they might have more niche offerings as opposed to the broader skillset that Appen could bring. Maybe just help us understand the, I guess, your ability to -- and capacity to invest and grow into those opportunities, acknowledging that R&D had obviously come off considerably this year versus the PCP and then maybe just contrast that with maybe some of the competitors that are -- do have some capability to invest going forward either organically or acquisitively, and potentially go a bit quicker into that opportunity?
Armughan Ahmad
Yeah, thank you, Ross. So competitors don't have a unique advantage. It's because of a lot of our relevance work we've been doing for a long time, especially coming from this reinforcement learning with human feedback that's required. Relevance plays a very important role there. We were a bit earlier, but we are now -- comparable now, if not in many areas, ahead on fine-tuning and monitoring assurance capabilities in certain areas that was on announced some of the products. We will be increasing our investments in generative AI. And this is what I mean when I say we will be reviewing other investment areas. We will focus on high-growth, large language opportunities. And I'll also say our engineering teams were moving to that low-cost model in Hyderabad. So it's not like we're just reducing our spend, we're also being very mindful of how do we do more with less, right? And those other companies you mentioned who are start-ups that are mostly VC-backed, they're doing a lot of that stuff in very expensive areas, and we're just trying to make sure that we are getting fit to ensure that we have a long-term play here.
Ross Barrows
Very helpful. Thank you.
Armughan Ahmad
Thank you, Ross.
Operator
Thank you. Your next question is a follow-up from Siraj Ahmed with Citigroup. Please go ahead.
Armughan Ahmad
Hey, Siraj.
Siraj Ahmed
Thanks for that. Just a couple of quick follow-ups. First one, in terms of deep learning versus LLMs, has your view changed in terms of the shift -- the mix shift that it's going to be -- deep learning is going to be declining quite a bit faster. It's all going to be LLM. Just can you give your thoughts on it?
Armughan Ahmad
Yeah. No, our views haven't changed. We're just seeing -- deep learning continues to be in production for many of our customers, whereas generative AI is, mostly a lot of pilots going on, that will get into production eventually. But deep learning continues to be, as I shared last time when I met you in the Investor Tech Day, that according to IDC, almost 70% of the market -- or 70% to 80% of the market is still deep learning based, right? It's just a lot of our concentration with the large customers, and the tech slowdown that we're experiencing there, is the reason behind it.
Ross Barrows
Understood. And I guess a follow-up on that. In terms of kind of the comment on large projects with -- sorry, some of these LLMs are just committing to large projects, right, early days in there. Just your expectation -- I mean I know it's for -- it did not happen, but just when you reckon this enters production? Because you previously said you're working the second largest customer on their service, right? So just in terms of when you are looking to convert to some of those large projects?
Armughan Ahmad
So we're starting to see multimillion dollar opportunities in these large language models since the last time I was here in May till now. I mentioned that we had 40-plus deals in the pipeline. I've now come out and told you that we have closed 40-plus deals in large language models. I've also told you that our large language model deals with hyperscalers, we have completed either a project or have pipeline with them. So again, just think about end of May till now, how much this has progressed, right? So again, these are smaller dollar amounts, Siraj, compared to the bigger dollar amounts in deep learning. And again, we have very high expectations that this will expand in the future. Lots of human alignment is required to get large language models to work at scale. We have big opportunities for Appen, particularly in global large language model work, multi-country.
Siraj Ahmed
Understood. Thanks.
Armughan Ahmad
Thank you, Siraj.
Operator
Thank you. There are no further questions at this time, and that does conclude our conference for today. Thank you for participating. You may now disconnect.
For further details see:
Appen Limited (APPEF) Q2 2023 Earnings Call Transcript