Appen Limited (APX) Earnings Call Transcript

Thank you for standing by, and welcome to the Appen Limited Strategy Refresh and Capital Raising. [Operator Instructions] I would now like to hand the conference over to Mr. Armughan Ahmad, CEO. Please go ahead.

Thank you, Rachel. Good morning, everybody. Thank you for taking the time for joining us. Today, I'm here with our Chief Financial Officer, Helen Johnson, along with our Senior Vice President for Strategy and Innovation, Ryan Kolln. Thank you for taking the time. I want to just run through some opening comments, and then I'll walk you through some slides, if that's okay with you. I will call out the slide numbers so that you can all follow along. Just opening comments, if I have not had a chance to meet with you but meeting with various investors, the last time I was in Sydney, which was in February and that now over the last few days, meeting with various different investors. I look forward to meeting some of you in person in the coming weeks. I just want to give you a little background on me. I've been in the tech industry for the last 27 years. I have worked for companies like Dell Technologies, Dell EMC, we took Dell private and then public again, with some private equity, also I was that HP before that. Previous to that, I was with 3Com, with Bain Capital. We also turned around 3Com with a Huawei-3Com joint venture and then sold that to HP. So I've done a few of these turnarounds in the past, and I feel like this is an exciting turnaround and a very exciting time for generative AI. I really think that Appen, as I joined Appen 4 months ago and Helen, you joined Appen 2 weeks ago -- I believe 10 days ago, along with Saty Bahadur, our Chief Technology Officer, who has just recently joined as well. We're all coming from the tech industry, and we all felt that there's a phenomenal opportunity here, especially with the advent of generative AI along with our current deep learning AI work, that's important. But first things first, when I joined here 4 months ago, I really looked at the broader business and felt that the cost base was not under control, and we had to act very quickly. The first 4 months have been pretty busy, I would say. We have made sure that we have tried to, one, ensure that we cut the fat out of the OpEx, which we have now taken $46 million worth of OpEx actions. And when I say actions, this is not just something that we're saying we will do. We've already done it, we're not cutting into the muscle. So we're just cutting into the fat. And what I mean by that is that we had invested ahead of revenue in many of the areas, and we now need to make sure that we are bringing our OpEx to revenue ratio in check for our global business, our enterprise business, our China business, our federal business, along with our product divisions, and engineering divisions. We want to make sure we are running a profitable business. I'm thrilled that with the actions that we have taken on the cost out and now the equity raise announcement, we really think that we are going to finish the year with cash EBITDA positive, nothing, we will, I should say. The last few weeks have been really positive for Appen. I'm starting to see really great green shoots coming from our deep learning AI-type type global customers. So those are the big Apple, Microsoft, then Google and Amazon of the world. They make up almost 80% of our revenue. If I take out our #1 customer, our revenue over the last 4 years has grown at 10% CAGR growth. We had seen some challenges, unfortunately, over the last 2 years with our #1 customer with their shift into newer [Audio Gap] generative AI, along with search relevance, which has now started to see our revenue stabilize with them and now as of late we've started to see great green shoots where they're -- our #1 customer is now giving us newer revenue and newer projects, also consolidating some really great business to us away from some of our competitors. So I feel like that's a very positive thing that we're driving. We also feel that as part of our cost out. Unfortunately, we had the impact on nearly 300 people out of our 1,504-person organization, along with our hosting cost and so for our recurring costs on our engineering assets. We've also made a decision to move our -- a good chunk of our engineering to India and Hyderabad to ensure that we have lower cost areas to ensure that we're keeping our costs under control. I'll also say that our business is moving in the right direction with the right capitalization that we needed. That's helped us to really move our relationships to the next level with some of the largest system integrators in the world, but also some of the big tech companies like NVIDIA, for example, a few weeks ago, we announced NVIDIA, which has USD 700 billion market cap. They represent almost 80% plus of the AI chipsets in the market, have decided to go ahead with Appen and our data sets in their conversational AI solutions that they are providing with their own large language model, which I'll refer to later. We've also announced a partnership with Reka, Reka AI, if you're not familiar with the AI researchers from -- for Google DeepMind as well as Meta's LLM, creators of Google Bard, creators of Meta's LLM. They have gone out and started their own on-premise large language model similar to ChatGPT, and they've decided to also partner with us. So our goal, as I said earlier, is to finish the run rate cash EBITDA positive. And even if, let's say, we don't foresee that but if the revenue volatility remains, we are committed to adjusting our cost further. Helen and I call this a dynamic approach of making sure that we keep ourselves in a healthy area. So we're excited to go on this journey. Before I go through the slides, I just wanted to tell you, I made a decision to join Appen to run a $300 million circa company. There was a reason for it. I left KPMG, as a Senior Partner and President level and worked at all of these other companies that I told you, managed hundreds of millions to billions of dollars P&L. If you take a look at my compensation and Helen's competition and others, we are all very clearly geared towards improving the share price. That is what our KPI metrics are set by the Board. And almost 50% of our stock will only vest if we improve the stock price 190%, which takes us to $7.63. And if we improve it by 320%, then it takes us to $11.25. So I would say my compensation and my leadership's composition is fully aligned with the investors who decide to invest with Appen. I would like to now walk you through a few slides with the pack that we have all provided you. If I could have your attention, go to Slide 10, please. I find that in the key findings of my operational review over the last 4 months, these are the areas that I've really touched in the business across sales and marketing. I felt that we were a very reactive sales culture, and we needed to pivot ourselves to proactive sales culture, especially going after our big 5 global customers that we have, along with the Fortune 2000 large enterprises. And we felt that we needed to move away from selling to a data scientist, which is 10th level below the CEO to really go and start selling to the C-suite and make sure that they understand what's required. In the coming days, I'll be announcing a new Chief Revenue Officer for the company. We haven't had a role like this at this company because we have relied on our customers, and we feel this is a perfect opportunity for us to provide that opportunity. I also feel that the leadership and the culture is very important. People like Helen, who ran a $9 billion business for a very large solutions integrator in North America and Europe and Asia inside which -- and then she ran the North America CFO role there of $9 billion out of $11 billion P&L revenue company as well as Saty Bahadur who left Upwork, who's got a good relevance to our crowd type of platform. He was running a 1,000-person engineering team there as their Head of Engineering, has left there and joined us to lead a 130-person team, which I've now asked him to reduce and then grow in the right areas. He's super excited. He was previously with Amazon Alexa AI along with -- he built -- he was the Head of Engineering there for Alexa AI platform and then previously at Microsoft and Intel there. We have Sujatha Sagiraju, who was 20 years Microsoft and she's MLOps, she ran MLOps, so machine learning operations there. And she had joined us almost a year ago, I believe, to lead our Chief Product Officer and she's amazing at driving that. You will see a Chief Marketing Officer announcement as well in the coming weeks that we're going to drive towards, which is important. I also want to just highlight on this Slide 10 around transformation, which is I really have felt that our product engineering has not delivered the road map that we needed to deliver. And I hope you will join me on May 26, at the Barrenjoey offices, where we will be hosting an Investor Tech Day to showcase some of our products, and you'll get a chance to see at Saty, Sujatha, myself and our leadership team there. That's how fast we're moving. So you'll get a chance to see some demos. And as I've said, cost controls are going to be important. Globals have been -- global business has been profitable for us. China is getting to profitability. We feel by the end of this year, we're going to be positive in China as well. The areas that we need to fix is really our enterprise area. And then as I mentioned, if we take our top customer out, we've been growing at 10% CAGR growth and even the first customer is now adjusting. So we feel that our second half will be better than first half. I know we've said this to you, where second half is always better, but in this case, we're actually seeing how we have readjusted the revenue base for the company that we feel like we're now seeing green shoots over the last few weeks in these newer areas that I'm really positive about. Even in this generative AI area, which I'll walk you through, we've seen 32 new deals in our pipeline, which is a very positive side. So if I take your attention to Slide 11. Slide 11 is what I told many of the investors when I first met them, when I first started, and in our earnings announcement, and our annual results, I told you, these are the few areas that I wanted to focus on. And I'm huge believer of say-do ratios and we started with operational rigor. We said we were going to put a business management system in place. We said we were going to do a zero-based budgeting exercise from bottoms up. We have done that. At that time, we told you we were going to take out $10 million. Since then, we've been really busy. We felt that we really needed to get our costs under control. So we have taken out $46 million, that's nearly 300 people. In product velocity, I think in February, when I told everyone that we're going to be moving into generative AI and we announced some of the generative AI products like RLHF, reinforcement learning with human feedback. A lot of people were questioning what that meant. And I think thanks to ChatGPT and everyone talking about generative AI, they now understand exactly how Appen is relevant there. Those products that we launched have actually given us really good -- those 32 deals that I talked about is exactly that from our global accounts, along with our enterprise accounts. We're also, as I mentioned, reducing our cloud vendor spend as well as our engineering offshoring to ensure that our product velocity continues to remain fast. World-class go-to-market, that's the area where I give myself a yellow, if you like, the last 4 months have been full on trying to get this business under control with the right type of cost base as well as the newer partnerships and newer customers. But now we need a CRO to build a consultative sales and go-to-market and elevate our brand for both from a B2B perspective as well as the B2C perspective. Ecosystem partnerships, I said I was going to do. And we have delivered NVIDIA and Reka. Those are announced. PwC, Deloitte, Amazon, AWS, Google Cloud Platform, Azure have also become partners and customers who are already delivering work with us, which is great. We're not at a point that we can start announcing those. But we're now seeing those as our customers and partnering with us in the right direction. We're mixed, I would say, on the progression of the certain verticals. Automotive vertical is advance, but we need to really get going on financial services, retail and other areas. And then finally, AI for Good. I think the Crowd Code of Ethics and ensuring that I'm leading now the chair of our AI for Good committee to ensure that we're driving responsibility by design is important. If I take your attention to Slide 13, please. This is just explaining that how Appen is focused on deep learning AI, which is our traditional side of the business, along with generative AI. I would tell you that there is a bit of a misnomer. I've met with a lot of investors, and I think the market feels like we're an annotation company or data labeling company, that could not be far from the truth. Almost 70-plus percent of our revenue comes from relevance side, which requires human in the loop, and that is what's delivered through our Crowd, but also delivered using our Eight app and Appen Connect platform, such as relevance, irrelevance, search relevance, content moderation and evaluation work. Data collection is a smaller portion of our revenue. Data annotation is a much smaller portion. Data relevance is the biggest portion. That relevance portion is exactly what's needed in generative AI now. The relevance is done by human feedback that is what's required now in fine-tuning of these large language models, LLM, that -- we came from the language background with Julie Vonwiller founding the company 25 years ago as a linguist researcher at University of Sydney, that language piece is super important to our customers. Insertional data sets, model evaluation, all of that is done by our customers wanting to have different type of crowd, people who are teachers or people who are in financial professional or people who are gardeners, for example, how do they create those prompts, so those prompts can then provide the fine-tuning of a LLM. You may think what an LLM is. ChatGPT is a brand name version of an LLM. There are other LLMs. NVIDIA makes their own LLM. Cohere has one. Reka has one. Bard, Google Bard is an LLM. So all of them are now providing -- are asking us to now do a lot of that fine tuning work, which comes from our relevance area. The other area that I'm super excited about is the far right of this is our assurance work. So we're now finding that a lot of the top enterprise clients, customers of ours are looking for assurance services from us. This is after they have trained or fine-tuned the LLM. They're asking us to certification and monitoring, benchmarking and AB testing of that work, and we feel that allows us to further entrench ourselves. If I take your attention to Slide 15, we will go into a lot more details on this, but the main moat is -- for Appen has been our Crowd and our Appen Connect and Appen [ AI Data ] platform that came through our Figure Eight acquisition. But I feel like to leverage these large language model, base models, our customers are calling us to do a lot of this, not only fine-tuning but also assurance. And this is a great flywheel effect that we have created where it doesn't matter if it's NVIDIA, Cohere, Reka, OpenAI, Anthropic LLM, all of our customers are in need for this block that comes on top of it, which is us providing first instructional data sets, RLHF. That's the first stage. With the customer on one project, the second stage becomes RLAIF and model evaluation, which is reinforcement learning with AI feedback and then model evaluation is, for example, if you need NVIDIA to do the contact center work LLM, you need Reka to do a KYC type of LLM at a bank. You could use Anthropic, Cohere, OpenAI in different ways. You can put it on a public cloud like AWS Azure or Google Cloud or you do put it on-premise. Again, you would need fine-tuning and you would need more importantly assurance products. So we have now seen PwC and Deloitte, that their risk and compliance teams approach us and not only the compliance of ours, but also are now using our services to provide assurance, risk assurance on LLM and compliance assurance, which is to remove the bias, toxicity and hallucination. If you go to Slide 16, please. Slide 16 provides you how we are planning to go to market. So before, we were waiting by the telephone, if I can still use that analogy of having that telephone to ring and for orders to come in, now we're becoming a lot more proactive with our enterprise customers. That's why we're building an enterprise sales team. We're working with our ecosystem partners to get out to the market so that when I got the call for Appen, I did not know who Appen was. When Helen got the call for Appen, she didn't know what Appen was or Saty, when our Chief Technology Officer, got that call, that's what we need to fix. So that Appen becomes much more of synonymous name in the enterprise. And that also helps us unlock our TAM. We really feel that it's in generative AI, alone, it's an $8 billion TAM that's growing to $110 billion by 2030, and we feel that's going to be super relevant to us. If I take your attention to Slide 17. Slide 17 is an example of what I mentioned that we had -- and I stand corrected, I think, I said 32 deals, that's actually 36 deals that are in our pipeline. And these deals look something like that. So global account, a global client of ours is giving us opportunities to now not only code but also we're now delivering evaluation model performance for text-to-image generative AI models for many of the very big customers of ours, so which has been a great green shoot for us. We're now delivering that work. Another example of our customer is a very large -- the one of the big 4s in professional services, who has now given us a $2 million to $5 million opportunity that we're executing on, on creating large language prompts for them and then a very large bank has partnered with us and Reka to build out their knowledge management solution on-premise because for regulatory reasons, GDPR reasons, that they require that. If I take your attention to Slide 24 is the leadership team. I mentioned Helen, who's recently joined us. Sujatha, I mentioned earlier, building an Azure AI platform. Saty, who recently joined us from Upwork before that designed Amazon Alexa AI platform. Ryan Kolln, who is based out of Sydney is now moving. He's been promoted to lead our strategy and innovation for the organizations moving to North America. So that we have a whole brain trust that is executing towards where majority of our revenue comes from. We will also be announcing, as I mentioned, a Chief Revenue Officer and a Chief Marketing Officer. So that would be a team. I also felt that point of this, right, the tech pedigree, high-profile large companies, why are they choosing Appen, and I would ask you to maybe pause and ask that question, why did Armughan joined Appen, why did Helen join or Saty, why is Ryan Kolln, ex-BCG, ex-Telstra Executive, AT&T and others, why he's joining -- why is he saying that Appen, right? It's all because we see the opportunity, and we feel where the stock is at and where the opportunity is with the TAM is tremendous for us. If you go to the next slide -- hold up, and also Roc Tian, who is from China, and he runs our China business and our South Korea and Japan business, he's done a fantastic job. Our China business has become #1. Our main competitor is SPEECHOCEAN, who has similar revenue to us, has a $1 billion valuation. We now have all of the hyperscalers as our customers in China, most of the top enterprise organizations like China Telecom and others have become our customers there, along with some of the top 8 automotive manufacturers have become customer of ours because they need the data relevance work to provide self-driving solutions. Slide 25, I can go there. To me, I can say all of these things to you, but I would like to maybe share just a personal perspective. My background is I grew up in a very underprivileged environment in Pakistan until the age of 15. For me, being purpose-driven on solving income and the equality is a huge, huge passion of mine. And I feel like purpose is super important to every human that is working at Appen. It's -- if you're not looking at Appen, you're at investor, I'm sure purpose is very important to everybody. It's very important to me. And we felt that we needed to really set a culture code at Appen. So purpose is we are unlocking the power of AI for Good to build a better world. AI is going to take a lot of jobs away. We want to ensure that we're creating a lot of jobs using our Crowd, and we feel that the Crowd is going to morph into -- especially after COVID, so many people who are wanting to work from home and not wanting to come into an office, they'll have a great opportunity to use our platform in different ways. Perspective is we are learning all culture. And I fundamentally always tell my team that I don't want to be the smartest person in the room, I'm not a know-it-all. We are learn-it-all organization, and we embrace that comfort and growth do not coexist. And right now, Appen is going through some uncomfortable moments. And -- that I think failure, struggle is formidable. The good news is we've got a great team, not just my leadership team, but the team that is here at Appen are really embracing this culture. The organization has become a bit lethargic and it's now moving into this next direction, which we're really excited about. And if we do purpose and perspective really well, we feel that leads to prosperity. And these value that unite us are, being customer obsessed, thinking about the customer day and day out. The customer says, get it done for me, let's not do that in 2 weeks, let's do it tonight or tomorrow morning. Courage to innovate. This RLHF, RLAIF, many of the products that I mentioned earlier, those are super important. That's how our engineering and our delivery organization is executing. Being very action-oriented. I'm a big believer in initiatives. Like I find initiatives are interesting. Finitiatives are much better, to finish the initiative. And then finally, winning together and not being silent. And then finally, if I can take you to Slide 26, I'll end with this and turn it over to Helen to talk about an important slide for financials is the turnaround scorecard going forward. So I told you what my say do ratio was back in February and now what you should be measuring me over the next 6 months. It gives you a very clear road map on operational rigor, exiting the year, underlying EBITDA, cash EBITDA profitable, reducing our cost, if we need to, we'll continue to adjust our cost base, sales product delivery structure, we're moving away from a business unit structure of globals and enterprise, and we're integrating into a much more of a sales product delivery structure. Product velocity. These are the products that we're going to announce. You will see that on the 26th of May when Saty and Sujatha are here to showcase that to you. I hope you will join us. You want us go to market. My aim is to have the CRO here as well, and I introduce you to him proactively going after the market and building pipeline. Our ecosystem partnerships. You should check to see how we're doing on that. That's going to super helpful on how we're going to drive this change. And then finally, AI for Good, which is this is not just anything. To me, it's very near and dear to me on ensuring that we focus on the SMG part of ESG, which is purpose perspective and prosperity that I mentioned. With that, if I can maybe turn it over to Helen, please.

Yes. Thank you, Armughan. I'll direct you to Slide 22. I really wanted to recap what Armughan said a couple of times during the call and see if I can frame up, crisply, where we are on our cost reduction efforts. So we announced last week that we're undertaking a cost-reduction program of $46 million. And I wanted to take you through the components. As Armughan said, there's about 300 positions in our organization that are impacted by this action. We started that program just last week. The majority of those actions have occurred in -- across our footprint, across North America, China and our various business units. And the balance of them will happen over the balance of the year, and really completed by early December. And the pieces that are yet to come are the strategic decisions that we noted around consolidation of our go-to-market and delivery organizations in the United States. That, we believe can provide an opportunity to get both synergies in the way we think about sales process and onboarding and productivity of our delivery resources. And then the second piece is standing up our engineering support resources in Hyderabad, India. So that's yet to come over the next couple of quarters, and we wanted to make sure that you had a path, so thinking about how we're realizing against this program. What we did say is that we expect to exit the year at $113 million of cash-based expenses in the business. And in the short run, we believe that these cost-reduction efforts, one, they rightsized our cost structure to our revenue plan for the year and for the intermediate term. And then the capital raise that we're doing today really supports fortifying our balance sheet so that we can execute against the strategy that Armughan just laid out. So these cost savings were identified as part of our strategic review of the business over the last handful of months. It was not -- it was a zero-based budgeting exercise that the organization undertook to really rationalize the level of cost that was deployed in the business to support the very specific revenue streams that we have today, and at the same time, make room so that we can invest in the future where we're headed with generative AI and really how we intend to invest in our sales and marketing organizations to make sure we realize on the opportunity that we have ahead. So I think with that, I will open it up to the audience for questions. Operator...

Rachel, are you able to open it up for questions, and let's see who has the first question, please.

[Operator Instructions] Your first question comes from Bob Chen with JPMorgan.

A few questions for me. Just on the cost base, obviously, significant cuts this year, but you've also made some comments about reinvesting around the go-to-market and the sales and marketing. So how should we think about that reinvestment going into next year?

Well, we intend to keep our cost base aligned to the revenue, the revenue trends that we're seeing in the business. And so the cost takeout contemplated some reinvestment in the business, and we'll accelerate that opportunity as revenue performance improves. So you should expect that it's incorporated.

Okay. Cool. and then I think Armughan on your comment earlier around seeing signs of stabilization with your top customer. Can you give us sort of a rough split of how much basically they contribute to your revenues now? And what signs you're seeing that gives you that confidence that the sort of decline in revenues is stabilizing with them?

Yes, great question. Fortunately, I'm not able to give you exactly revenue numbers, but I'll give you just directionally the comments that I made earlier is that we're seeing our revenue stabilize there. We're seeing newer generative AI as well as deep learning AI-type solutions from them. And we're also seeing that we're now starting to win business away from our competitors.

Okay. Great. And then just finally, with some of these newer relationships with Reka and NVIDIA. I think you outlined a couple of sort of green shoot programs, but can you talk a little bit about sort of the revenue model or the business model there? Is it sort of incremental projects that you're working on? And what's the revenue contribution we might see from these opportunities?

Yes. Great question. I think I would like to direct you to that slide that I showed you, it was the Slide 15. So if you think about how that works is that we see different stages with different customers. So just like when we do work with Google, Meta, Apple, Microsoft and others, right, and Amazon is they do multiple projects with us. Just like that, we're now seeing enterprises do multiple projects. So for example, they'll start with the structural data sets in RLHF, then they'll move to RLAIF, then they'll ask us to do the model evaluation work. Then once that's done, that becomes fine-tuning phase. So that's like 2, 3 phases just to do fine tuning. And that's just in the contact center area of a bank, for example. Then you have the KYC area, then you have retail banking, then you have open banking, open payments, right? So there are areas that we can go after. So there's multiple stages. And then there is a stage of assurance and how do you do the monitoring and AB testing. So we feel like enterprises, if we want to win a customer for 10 years, and in AI, I don't believe AI becomes a recurring revenue business. I believe AI becomes a repeating revenue business where you go in and win a customer and you do a really good job in these areas. And that's how we have done it. It's been a repeating customer, some of our top globals have been our customers for a decade plus. And that's what we want to do now in enterprises as well.

Your next question comes from Josh Kannourakis with Barrenjoey.

Just one regarding the new set of LLM data products. So obviously, if we go back in time, figure out, I suppose that was targeted at the enterprise segment, obviously too early. Just to help people understand what is the step change here in terms of, I guess, the market's readiness to take on these products? Like maybe you can just reference in terms of the open-source nature of these algorithms, like how that's leveling the playing field for guys like you to be competing with guys like Reka and things like that in the market? Can you talk about the opportunities there, who some of the key players are and your competitive advantage?

Yes. That's a great question. Thank you, Josh. I would tell you that these LLMs are fast becoming available by everybody, right, because it's a foundation model, and it's a large language model and then how do you train it, it becomes your differentiation. And a lot of our -- we obviously have these public LLMs that people have LLM models like NVIDIA has that, that they're now shipping in their conversational AI box with GPUs and their own software. They need our data sets as part of that, instructional data sets. And then as the customer deploys it in the contact center, then they'll need to fine-tune and assure it. In Reka's case, Reka is more going after a market where people, even if they have a VPC, a virtual private cloud, think of a bank, think of a large retailer who has their own private cloud who're using Azure or AWS or GCP if they use open AI with Azure on it, their data still uses the base LLM, which makes OpenAI better, right? So Reka is going after that to say we want to keep it on-premise, either on-premise on an NVIDIA box or on-premise on a VPC, private cloud, and they get to own the LLM, meaning the customer gets to own the LLM. Then we're seeing that there are public LLMs, open source LLMs that are now coming out, right, where they're again going to need fine-tuning products and assurance products. So our view is that we become relevant regardless of which LLM you want to use, if it's OpenAI or Reka or NVIDIA or Cohere, Anthropic. That's our differentiation. And I think, Josh, it's also important to note that when I said something earlier where almost 70-plus percent of our revenue, in deep learning AI comes from the relevant side and that requires a human in the loop. And in this generative AI in order for you to do the fine-tuning, you need the human in the loop, and that is our differentiation. You obviously need a much different subset of different humans, and that's why our Appen Connect and Appen platform that came from the Figure Eight acquisition works really well. And then after that, assurance products is what we are now building. We will show you a few demos on the 26th, where we're doing a tech day and an Investor Day to show you how assurance products will work. So it's not just fine-tuning. It's also assurance products. So that's why we feel it's much more relevant. Does that answer your question, Josh?

Yes. No, that's really helpful. And just as a follow-on to that, if you think about -- because as you mentioned of 70% of the relevance work with the Crowd already, if we look at the sort of the economics almost of some of the work within that category versus the generative AI category, how do you describe that in terms of comparing, I guess, the economics of those projects as you sort of obviously are just starting to price some of them up now?

Yes. I'm going to have Ryan start who's here with me, and then I'll add on to that.

Josh, so we're saying, we see some greater specialization and the need for the type of crowd and that comes with a different set of economics that's going to be favorable for us. We're also seeing like what relevance has been for a very ongoing in the nature of the type of work. The need for generative AI, we're starting to see that ongoing needs. So we expect it to be a very, very positive business for us, like relevance has been for a long time.

Yes. And I would just add on to that, Josh, that based on what Ryan said, I think it's -- we're seeing -- we're winning generative AI projects and deep-learning AI projects in the global accounts. We have lately seen a lot more generative AI projects in addition to our deep-learning projects. Our #2 customer has also continued -- it grew for us 20% last year. It's continuing to grow. And in the areas of generative AI, pretty much the top out of the 5, I would say, the top 4 have given us business in generative AI. And then in the areas of enterprise, as I mentioned, we've got almost 36 new LLM-based generative AI projects across different sectors. So we're seeing the demand move really fast. We're also working very closely with IDC, which is a very large tech analyst firm, industry analyst firm, IDC, Forrester, Gartner, they're all in the same bucket. And I think this area is moving so fast that they're starting to sort of put their numbers on it to see what the TAM looks like. So you'll see more from that as well, upcoming on the 26. We'll share more about that, but we're -- how a third-party industry analyst looks at this space.

Your next question comes from Darren Leung with Macquarie.

I've just got 2, please. The first one was just an extension of an earlier on, still on the cost base. It's obviously been well documented around softening macro conditions and the amount of cost reductions that your largest customers are going through. Can you give us a feel as to how much flexibility is left in that $113 million cost base, if I suppose the go-to-market strategy, doesn't come to fruition or if there's other sort of costs that need to be reduced, please?

I'll have Helen take that one.

Sure, sure. The cost base is largely variable to the delivery of the revenue. So as we are seeing revenue trends decline, we'll rightsize, if we need to, our cost base. I think that this was the appropriate action at this time, given current visibility in the pipeline and plans for positioning in the generative AI space. But certainly, this business has operated at a level of profitability of 7 or 8 years ago at $60 million business was running at 16% to 18% EBITDA margin. So we know that we could operate this business at a smaller scale with more profitability, and obviously have a bit more scale as we grow. That is not what you saw over the last couple of years. We grew notably, but the cost outgrew the top line. That will not be the model [indiscernible] going forward.

Yes. That's something I've mentioned in my prepared comments earlier where we want to make sure that we are staying on top of our cost on a regular basis, right? Helen and I are used to running our 13-week shop, which is week 1 to week 13, our sales teams and others have to give us comments on week 6 and week 10, and we want to make sure that we're running a stabilized cost base. And if we have to take additional cost out, we will take additional costs out to ensure that we are comfortable.

Understand. And then my second question was if we think about what we've talked about so far, it looks like it's green shoots into second half in relation to those enterprise customers, new products. As you've rightly pointed out, the second half has a seasonally stronger skew and presumably revenue accelerates into next year. You've done well on the cost-out program. I suppose my question is, why raise equity now.

Yes, listen, as I take a look at the revenue base of where it is and where our profitability is and as we take the cost out, the cost is a run rate cost out that we're trying to get to, right? This is not all just coming out in one day. And as we do that, and we have to run our P&L. We have a debt facility available. We want to make sure that as we draw on to any debt facility and others that we're giving ourselves enough room to maneuver. And as I talked about, I need to invest in sales, marketing and others, but we want to do it in a prudent way to ensure that while we're doing all of this and ensuring that as we leverage our debt facilities that we stay positive. Anything you want to add to that?

I think I would only add that we did have a benefit in our working capital trends in the first 4 months of the year. We have a slightly inverted cash flow cycle and with the decline in revenue, we did see a benefit in cash growing despite the EBITDA losses in the first 4 months. As we revert back to growth in the second half, we have guided or given directional output -- I mean, input that we expect to be up in the second half. We'll use it for capital there. And so with that working capital benefit unwinding and just really ramping up to the run rate expense base at the end of the year, this is the right amount to position us solidly to get to that and have some cushion going into next year.

Next question comes from Garry Sherriff from RBC.

Two questions. One on the go-to-market and the second one on the pipeline and the revenue visibility. If we start with the go-to-market, it certainly sounds like that market strategy appears to be shifting. You talk about expanding the partner ecosystem. Can you maybe just remind us your direct sales force in the U.S., how much of your revenue is generated by them? What are the plans going forward from a growth or otherwise perspective from a direct sales force? And secondly, the shift to the indirect channel. I just want to maybe sense check who you're partnering with -- and how you get that indirect channel firing? How do you measure the effectiveness of those partners? Any color would be appreciated.

Yes, Garry, I would tell you that the sales organization that I've assessed here is not a sales organization that I'm used to in the tech sector. So it's a very reactive sales culture. It's more of the customers reaching out to us and us reaching out to customers directly. It's not a much more of a focused on how are we going after it based on different regions, what's the C-suite executive, what's their pain point and then driving it that way. So it's pretty much all direct sales at this time. I would -- there's no channel sales motion at this time. That's what we are trying to create, right, multiplying our feet on the street. Obviously, we can't build an empire of a sales organization. And the CRO that I'm hiring is very focused on -- he's done this for different companies in cloud native world and data world, where basically he built out revenues from 0 to upwards of $700 million, right? So without building out a huge cost base of sales, that's what we feel that even using a channel go-to-market. That's -- by the way, Helen used to be my channel partner in a way at Insight, right, when I was at Dell EMC, they were a huge systems integrator, solutions integrator now. I think that -- I know the current CEO of Insight, Joyce Mullen is a previous colleague of mine at Dell and then previously at [indiscernible]. Like they were huge channel partners. And then -- I mean, think about how big your sales force was, right? So we would multiply the sales force at Insight. So it's the same kind of model we want to create here. The sales organization here is what we have to upgrade to ensure that they have the credibility and the relevance of what I have done and what -- or what others have gotten used to like Helen in the past, and that's what we have to create here, Garry.