TaskUs, Inc. (TASK) Earnings Call Transcript & Summary

All right. Let's get started. Good afternoon. My name is Puneet Jain from payment processors and IT services team at JPMorgan. Glad to have here with us TaskUs' Co-Founder and Chief Executive Officer, Bryce Maddock. Welcome, Bryce.

Thanks, Puneet.

So the format of this presentation is going to be fireside chat. I'll start with a few questions, and I'll open the floor for questions from audience. And for those folks who are listening to webcast, feel free to use the online portal to send questions our way.

So Bryce, you recently reported your Q1 results. Maybe if you can briefly review your Q1 results outlook as well as how TaskUs is differentiated in the marketplace.

Yes. Sure. So for Q1, we reported $235.3 million in top line revenue, adjusted EBITDA margins of 23.5%. And that's against the guidance of $233 million at the top end and 21% adjusted EBITDA margins. We updated our guidance for the year to $925 million to $950 million in top line revenue and increased our adjusted EBITDA guidance to 23.5% from 23%. And we confirmed our free cash flow guidance of more than $100 million in free cash flow, excluding the payout to heloo that's associated with the earnout in that acquisition. As far as TaskUs goes, we provide customer experience, trust and safety and AI services to rapidly growing technology businesses and enterprises that are looking to transform the way that they deliver services. We've got nearly 50,000 employees across the globe in 13 different countries, and we're coming up on 15 years in business.

No, that's great. So 1 of the 2 headwinds that you faced last year and continuing to face on a year-on-year basis are high exposure to fintech and U.S.-based delivery as the clients are pushing work more offshore and to lower cost locations. So can you talk about like how large each one of those buckets are right now? And what do you expect for second half of this year?

Yes. The Fintech business, excluding crypto and equity trading clients, actually continues to grow quite healthily. But in Q1 of last year, just crypto and equity trading clients represented 15% of our total revenue and that dropped to 4% of our total revenues in Q1 of this year. In the U.S., we made 33% of our total revenues from U.S. delivery in Q1 of 2022, and that was down to 20% in Q1 of 2023. And so obviously, that creates a substantial headwind to overcome. Fortunately, we have been able to drive significant growth in our offshore and nearshore markets. Our revenue from our offshore and nearshore markets grew 19% year-over-year in Q1 of this year.

And how low the U.S. mix can go to from the current 20%? Like what's the absolute bottom.

Yes. This is a question we've gotten asked a lot. The reality is the majority of what remains in the U.S. needs to be done in the U.S. for regulatory reasons, strong client preferences or because the end customer is the U.S. government. We recognize that we've seen more downside in those U.S. revenues than we had forecast initially. And so in our initial guidance, we assumed that we would make about 15% of our revenue as a company from the U.S. in the back half of this year. In our updated guidance, we took that number down to 10% and so we think that we're approaching a floor on that portion of our business.

And as you are one of the vendors with huge capabilities, delivery capabilities across the world, especially in Philippines, in India, in Eastern Europe, are you seeing like more clients, new clients coming to you looking for lower-cost solutions, looking for more efficiency in their business processes?

Absolutely. I mean, particularly amongst our existing clients, we've seen very robust growth offshore. We're continuing to grow that business, as I said, 19% year-over-year in Q1. If I look across our top 10 clients and we exclude our largest client who went through the onshore to offshore shift and our largest crypto and equity trading clients, I mean, we've got 2 clients who have tripled revenue from 2021 to 2023 and 2 that have doubled revenue over the course of that period of time. And that's much faster growth than their internal revenue growth, which I think is indicative that we're taking share from the competition. And in some cases, we're actually taking share from internal teams. Across our client base, we're seeing clients look for more efficient solutions. And at this stage, our offshore and nearshore solutions are the most efficient way of delivering work.

And let's take a step back. Like TaskUs was set up like as a pure play providers to high-tech clients. And that has been your mode -- your differentiator for so long. Last year, you started talking about that you're going to offer services to enterprise customers in digital areas but going after like a different type of customer. So talk to us like how do -- how large that segment is right now? And what is your strategy to penetrate or to win more of those clients?

Yes. So I mean, when we look at the history of the business, we started by focusing on rapidly growing technology businesses and helping them to scale their outsourced service footprint. We've increasingly begun to focus on big tech customers. These are the largest technology clients in the world, and we made good progress there last year, landing 3 of the world's largest technology clients where we continue to grow. In the traditional enterprise space, last year, we talked about wins that we had in the travel space, including one of the U.S.'s largest airlines and a large retailer. And those relationships continue to expand as well. We don't break out revenues based on sort of tech -- high-growth tech and traditional enterprise. But it is safe to say that the traditional enterprise continues to make up a larger portion of our overall business.

And what is your sales strategy? Like do you have to do something differently in sales as well as in delivery when you service those clients?

The sales cycle tends to be slightly longer. Usually, though, we're talking about different opportunities than their core commoditized customer care work. Usually, these clients are pulling us into their network because they're looking for a vendor that can reimagine their customer experience in a more digitally native way. And so those contracts when we're being pulled into the vendor network can contract faster.

And can you also talk about like the large technology companies that are in your backlog? Like how are they ramping up? How large they can become as part of your overall revenue?

Yes. On the earnings call, I gave an example of a -- one of the big technology companies that we had scaled for the holiday season. And typically, these companies would cut volumes pretty substantially. In Q1, we held our volume. And then at the end of Q1, we actually got double the volume. And so we're continuing to grow pretty aggressively into the next holiday season with that client. Across our big technology clients, we also see opportunities to sell into their subsidiary companies and those conversations are progressing nicely.

And we often get like lots of investor questions on generative AI and what it means for customer care. And thanks for doing that call with us 2 weeks ago?

Yes. Yes, it was last Friday, Friday before.

It feels like million years ago actually. But no, thanks for doing that call. So let's look at it like as 2 ways, like generative AI, its applications and the benefits that you can get. So when we think about like the applications of generative AI to customer care, there is like a growing view that a lot of customer care work can get automated, that the productivity savings, we have heard estimates as high as 30%, 40%, 50%, 70% even using generative AI. So talk to us like what's real, what's high and which areas are more susceptible to get automated or to see more productivity gain using this technology.

So I think if we look at our business, first take where I think there's going to be clear benefit. In our Trust and Safety services, it's very obvious to me that this technology is going to be used to create lots of content. You only have to look at the most recent deep fakes to understand the implications that this has to democratic elections, the creation of offensive content, impersonations of people. So I think there will be an explosion in demand for trust and safety services, and I think we're well positioned to benefit from that. Inside our AI services, we see a huge amount of interest from the generative AI companies themselves. We're supporting the development of 2 of the 3 leading large language models. We're also building adversarial testing teams, insider trust and safety space, and prompt engineering teams to help these businesses build the next generation of these models. So I want to put that portion of the business aside because I actually think that there's a real tailwind to those sides of the business. Now 2/3 of our revenue comes from customer service. And I think, here, the conversation has to be a little bit more nuanced. The simple Tier 1 support work is going to be increasingly automated. Now because of our clients being high-growth technology customers, the vast majority of them have put huge investments in automating as much as they can of their upfront customer workflows. And that automation usually looks like some sort of a chatbot, which is based on a decision tree. We think those decision trees will be ripped out and they'll be replaced with large language models that are more dynamic and potentially could solve more problems. At this stage, we have not seen, what I would consider like a real killer application that's facing the customer, but I'm sure that those will come. Fortunately, for TaskUs, the vast majority of the customer support work that we do is either highly technical in nature, requiring lots of research across multiple different domains or it is like multiple points of contact. So you can imagine in a food delivery use case, you might have a hungry customer text in, but that we need to pick up the phone and call the delivery person or call the restaurant. While that seems simple, it's actually very, very difficult to automate that type of work. We've built a platform called TaskGPT, and this platform is being used by our teammates today to research answers to our clients' questions. So for our highly technical programs, we're able to load the entire knowledge base and repository of training materials and historical customer answers and it makes our teammates faster and more effective at answering customer questions. At this stage, we're building agent-facing tools that will make our teammates more productive and augment their work, but I don't think we're anywhere close to fully automating these workflows.

A hungry customer trying to order food, that was me last night. So let's talk about like TaskGPT. Talk to us like about this platform or 2. Like how does it help your agents and your team members become more productive? How much effort was put in creating this? And is that something that your peers cannot replicate?

Yes. So there are 2 versions of TaskGPT: the first is effectively an interactive knowledge base. So in this case, we're taking an existing knowledge base in training materials, and we are embedding that content into either GPT-3.5 or GPT-4, and then we're able to allow our teammates to write prompts against that content and get answers back. The other version is one which sits on top of the CRM system, and we can integrate directly into Salesforce, Zendesk or any major CRM. We look at what questions are coming from our customers and then we were able to suggest the best response to our teammates. This is all done with a human in the loop so that the teammate can review and edit the answer, and it requires extensive training based on existing either answers or knowledge-based information. To your question, the reality is that because of OpenAI, all of the start-ups that were raising hundreds of millions of dollars to build chatbot technology in their own large language models are back to square one. Every single one of them is deploying ChatGPT or OpenAI's technology in the exact same way that we are. So this technology has really leveled the playing field for a service provider like TaskUs and where we think the differentiation now sits is in who controls the data. So we're sitting inside of our customers' CRM systems. And with their permission, we can use their historical data to train these models to more efficiently answer customer questions.

Got it. Got it. And then if you think about your AI data annotation services, should we expect like, in addition to the data tagging that you do, more like data modeling, more prompt engineering and some of those capabilities in there?

Yes. So I think what's happened here, which is interesting, is at the start of training these models, it looked a little bit like being an elementary school teacher or a preschool teacher. It was very simple data tagging and now we've moved up the value chain, and we're recruiting people who are experts in particular subject areas. And these experts are either free-writing responses to questions or they're rating the responses that the model creates. And the builders of all of the large language models are recruiting experts in every domain, from chemistry to popular movies. And so we're helping to recruit and deploy those experts for the development of these models.

And do you have to change your hiring strategy? TaskUs has always been very innovative in reaching out to potential employees, but as you change the type of people you hire, like you talked about like the industry experts, people with subject matter expertise, maybe more technical workforce, do you need to change hiring strategy?

I think we are an employer of choice in almost every market in which we operate. And so we use our brand's positioning to attract employees. Clearly, we're moving up the value chain and attracting a higher caliber of talent. And so there's some dynamic work there, but I feel pretty confident we're well positioned to build a very talented prompt engineering workforce.

And as you drive this adoption or some of these tools, generative AI tools within your project teams, within your workforce, do you have to change how you define success when winning a new customer or adding a new contract or when servicing existing customers?

Yes. I think the challenge here is to whom does the upside accrue, right? So if we're deploying one of these tools to a customer who we're billing hourly, we're going to charge a licensing fee for the technology. But ultimately, the tool could make us use fewer hours and make less revenue. So increasingly, we're pushing our clients into outcome-based agreements. On the Q1 call, I talked about our largest outcome-based agreement where from January to March, we improved our margins by 4%. That's a model, I think, for the future of the business. Historically, it's been challenging to convince our customers to contract for outcomes because they were so focused on growth. Hourly billing models are by far the easiest to put in place. But I do see an increase in demand for outcome-based agreements in this environment where cost focus has increased substantially.

And how ready your clients are to adopt these new models? In the past, we have heard like the clients resist moving to outcome-based or output-based because they want certainty of how much they're paying to a vendor. But do you think like clients appreciate the productivity savings through these tools and are more ready now to adopt these models?

I think given the cost pressure in the environment, we are seeing an increase in the appetite for these models. But again, in terms of fixed unit economics, there actually is no better model for a client because you're paying per outcome.

And I do not want it to be a generative AI call but last question. We often hear like there are constraints around governance control, privacy related to this technology. Talk to us, like you share your perspective, like how real some of those constraints are and how those constraints can become an opportunity for someone like TaskUs.

Well, I think the reason -- and again, I'm looking for examples here. So if anyone has one, please feel free to send it to me. But the reason I don't think we've seen a widespread consumer-facing deployment of this technology is exactly this, which is it's difficult to control. The historical approach to chatbots has been fixed decision trees and fixed decision trees are very easy to control. Deploying a large language model, you do not know what it's going to say. And so for that reason, we've already deployed this technology in many of our customer use cases on the agent side because you have a human in the loop who can review the content before it gets back to the customer. I think companies will be a little more cautious in deploying consumer-facing applications. They're absolutely going to do it, but I think it will be a little more slow going than the tech inside the contact center itself.

At this point, if there are any questions from the audience, feel free to ask. Just wait for the mic. Do you need to wait. Go ahead.