Hi, everyone. Thanks for joining us today for the CEO connection this quarter. Before we get started, I want to note that today's discussion may include some forward-looking statements, and the risks related to those are listed here and also in our filings with the Securities and Exchange Commission. Today, we're going to go inside Tinder's product and engineering teams to discuss something that you've heard us discuss quite a bit, which is our product velocity. It's a very important part of the Tinder story and something that investors are keenly focused on, and I get asked about a lot, for good reason. Today I'm joined by Mark Kantor. Mark is our Chief Product Officer at Tinder. He leads our cross-functional teams of designers and product managers who help determine what we build into our product. Mark has more than 20 years of experience as a product builder and an entrepreneur and a founder, including roles at Zynga and his own startups. Also with us is Vinay Kuruvila, CTO of Tinder. Vinay leads engineering, AI, and product innovation at Tinder and also leads our central AI teams at Match Group. Vinay has 20 years of engineering and product leadership experience at companies like Amazon, Venmo, and Brightwheel. Before we get into that discussion, we're going to start with a quick look at how much the Tinder experience has evolved over just the past 18 months. Many of you probably aren't active Tinder users, so we wanted to bring the product to you by highlighting where it was, where it is today, and just how much has changed in a relatively short period of time. Let's take a look. It's pretty cool to see. A lot of blood, sweat and tears went into 18 months. Hopefully, the video gives you a more tangible sense of what we mean when we talk about product velocity. I've been a tech executive for 28 years, and I have never seen a company change its core product so much in such a short period of time. What you just saw is the output, which I'm very proud of, and it's impacting our users and our business results every day. What I want to spend today focusing on is how we get it done. Mark, I'm going to start with you. We just showed how different Tinder looks today than 18 months ago. Let's start with the obvious question. What's changed, and why are we able to move at this pace today? As we just saw in that video, nearly every part of Tinder has improved. Not only does it look different and it feels better, but more importantly, it actually works better for our users. We've had major strides in trust and safety, bringing down the prevalence of bots and bad actors by more than 60%. We greatly improved recommendations, driving better outcomes and sparks for our users, and we built new ways to connect, like Double Date and Events. There are a few things that are really driving that change. The first is the org. The org has changed dramatically. We have smaller, more autonomous teams making much faster decisions. We've really sharpened our understanding of who we're building for and the problems we're solving with the use of new personas and a lot more time and focus directly talking to our users. Lastly, we have reoriented the entire team around one North Star metric, Sparks, which is our term for a multi-way, six-way conversation. It really aligns everyone around driving better outcomes. The result of that is not only a product that works better, but a team that moves a lot faster. Yeah, we just finished a three-day offsite, with products and engineering and marketing leadership from around the world coming together here in L.A. It was amazing how often personas and Sparks came up. In every conversation, people were talking about the personas. So these are the archetypes of whom we're building for. Sparks, obviously, is the KPI that we focus on. Vinay, take us under the hood from an engineering standpoint. How does Tinder actually build differently today? Yeah. So Tinder's entire operating model around how we build has completely changed. We've got fewer linear handoffs, really tight collaboration with product design engineering, faster experimentation and iteration. Our engineering team is actually shipping at a velocity that's twice as high as it was a year ago. Faster shipping matters because it creates faster feedback loops where we can build, test, learn from our users, and improve based on the data and the metrics that we're seeing. We've made a number of investments in our technology stack. For instance, we've rearchitected parts of our code base which had a lot of technical debt, and were slowing us down, like our chat system. We've invested in our infrastructure, so teams like our recommendations team and our machine learning team can move a lot faster. We've invested in our experimentation platform so that running and reading the results of experiments can be very cheap and fast. Vinay, you mentioned tech debt, and this is a term that I know investors are familiar with because other public companies talk about it from time to time. What is our philosophy for paying down tech debt and maintaining pace of innovation even as we are kind of retrofitting old parts of the product? It is a great question. Our philosophy has been to incrementally eliminate the tech debts and looking at the areas of the code base where the teams are getting slowed down the most, as well as the components in the product where we have the most potential to improve User Outcomes. Chat was a great example, right? Our chat team was really getting slowed down because of tech debt in the code base. There was so much potential to modernize chat, add a lot of new features. We rewrote and rearchitected our chat system. We did it in a way so that every other part of our ecosystem could move forward while we were rearchitecting our chat system. We are going to continue to do that. Our next step would be onboarding. That is our approach to tech debt. One for both of you. How does AI fit into all this from a product ideation standpoint and then from an engineering standpoint? AI plays a critical role both within the product, how it benefits our users, but also how we build those products. Within the product, we use it to reduce friction during onboarding, helping people make better profiles, selecting better photos. We use it to improve trust and safety through features like Are You Sure? and Does This Bother You? in chat, as well as Face Check. We use it to improve our recommendations. We also use it in terms of how we actually build better. We have seen that AI essentially accelerates the entire product development life cycle from research and ideation, rapid prototyping to development. It essentially compresses the work that used to take months now into weeks and sometimes from weeks into days. The result of that is it essentially has let our teams have far more high-quality shots on goal. We could take an example of events, which we're all very excited about. This was an idea of [Spencer's] back in January. We had our first meeting in January. We had rapid prototypes days later. In March, we launched our first MVP publicly to the people of L.A. Just eight weeks later. Which is amazing to go from first meeting in January to public launch just a couple of months later. Vinay, how does AI impact your world? Yeah, and like Mark said, this is all possible. This pace is all possible because every engineer is using AI coding tools every day. Nearly every aspect of our product development life cycle has been completely rebuilt with AI at the center. Over 90% of all the new code at Tinder is AI-generated. We have built agents that now write our tests and verify our code. We have agents that can fix simpler bugs with human verification and oversight. We are prototyping and developing much faster because AI is compressing the entire software development life cycle. Sometimes people use the term AI slop to sort of derisively note low-quality product coming from AI. How do we guard against that? Mark, are the robots just coming up with all of our ideas? Vinay, is the AI just writing bad code that makes things worse? How do we guard against quality issues? Yeah, we have a policy in place where every engineer has to carefully review the code that AI has generated. In fact, it is reviewed not just by one engineer, but two engineers. Then we spend a lot of time codifying what it means to write high-quality code within our code base so that agents have really clear instructions and they are producing high-quality code and not AI slop. We are also spending a lot of time on verification and testing, which are again done by AI agents, so we know that the features we produce are high quality and do not have bugs in them. I think on the product design side, we have got a very talented team that I would say our ideas to date are coming from them. But really, also from talking to our users. We collect a lot of user feedback, a lot of research. We have used AI to help us synthesize that big corpus of data into really actionable insights. We are definitely changing the way we work because of AI. Has it also changed our talent recruitment, retention, and engagement strategy? Are we looking for 100% different things from people? Yeah. We are leaning very heavily into early career AI-fluent talent. We are very deliberate about evaluating AI fluency in our interview loops. It is something we take very, very seriously. We ask every engineering interview candidate to solve multiple tasks with AI, and then we talk to them about exactly how they used AI to solve those problems. We are having a lot of success with this strategy, right? We are pairing this early career AI-fluent talent with some of our very deep domain experts in recommendations and machine learning, and that combination is proving to be very powerful. Have you changed what you are looking for in a designer and a product manager? Yeah. I think I am really looking for curiosity, hustle, and initiative. With these tools now, there is no reason for people not to be at home prototyping and building their own things. So the first thing I ask is, "What are your personal projects?" I think if they do not have anything, then it probably means they are not going to be that creative here. So I really want to see what it is that gets them excited and makes sure that they are able to do it. It is interesting how the world has changed. It used to be like that litmus test of what are you tinkering with on your own nights and weekends. Now that is a positive sign. Yeah of somebody that is intellectually curious and pushing the envelope of how they can use AI to solve problems that they are fascinated with. It used to be like, "Well, hold on, why are you distracted on other things? That might be the most important question now. Yeah. Totally. What are you playing with? What are you tinkering? How are you learning? Vinay, some of the most important changes that we've made in the product have been around recommendations, which I usually describe as the beating heart of any dating app, trying to figure out whom to show to whom. Ultimately, that's the core thing about a dating app. So what inning are we in our recommendation improvements and what have we changed with rec? Great question. We're still in the very early innings on recommendations, maybe the third inning. Today, every major release that we do still moves our core very substantially. That's what early innings looks like. As you can see on this chart, we've made a series of major recommendation updates over the past 18 months. Sparks, which is, as Mark said, these six-way high-quality conversations, they move meaningfully higher as the system has evolved. Queue Unification V2, which is this big launch that we did in July, produced a single unified, coherent recommendation system that optimizes for one clear objective, which is Sparks, and Spark Coverage. When these very mature systems get tuned, the gains are very incremental, but the fact that we're seeing these step function changes every time we do a major release, shows that we're still in the early innings. This term queue unification is something we kind of throw around internally. Let me make sure that I'm understanding it and also take a pass at explaining it to viewers. The way I think about it is we have maybe 10 or 15 different decks of cards, and maybe one deck of cards is sort of optimized to maximize revenue to the company. Another is optimized for the new user experience. Another is optimized for, I don't know, retention of an existing payer, et cetera. When the user comes in, let's say Mark comes into Tinder and we're trying to decide what recommendations to show him, we decide based on what we know about him as a user, which deck do we want to pick up. We pick up this deck, and we start showing him card one, card two, card three. Queue unification is basically taking all of those decks that are currently independent and combining them and reordering the sort of that much larger deck of cards based on what will maximize Mark's chance of not just matching. So not just saying, "Yes, I like this person," and she says, "Yes, I like Mark," but actually post-match them arriving at a spark, a six-way chat. Have I described queue unification appropriately? Yeah, I think you are spot on. Previously, every deck of cards had its own objective, as you said. One deck of cards was to maximize revenue for users that maybe were at risk of churning, and then one deck of cards was to help get new users a bit of a boost in the system so that they are retained. Now we do not have all these 15 decks. We have a single deck of cards, and the only objective that our machine learning algorithms optimize on is sparks. That has been a complete game changer for us. It has really driven Sparks significantly higher for straight women, and we have got many other, I would say, segments of our ecosystem which are not yet rolled out, and we are still continuing to roll it out. The other aspect of our current system is that once you pick up the deck, whether it was one of the 15 or now the combined unified set of cards. Right now, when Mark says, "Swipe right on this person, swipe left on this person, right on this person," the sequence of cards in the deck does not currently change real-time. Can you describe this phenomenon? Yeah, absolutely. Today in the Tinder ecosystem, if your swiping behavior changes, if you are looking for something different today versus what you were looking for last week, it can take up to four hours for the system to update and to start to reflect that change in preference. Whereas real-time adaptive recs, which is something we are working on, it is going to launch in late Q4, is going to start having the system react in seconds to those changes in behavior. We think that is going to be a big step change in outcomes for our users. This is, of course, what we see in other services. If you are on LinkedIn and you are watching a video about a particular topic, you will start seeing more videos about that topic. If you are on TikTok, Instagram, even Spotify, if you are skipping certain types of songs, the recommendation algorithm changes real-time, and you sort of feel it as a user. That is coming. Yeah, absolutely. There have been people working on these problems for 15 years or so, kind of the whole history of Tinder. What is it about this moment in time in 2026 that's created this unlock, these step changes that are available to us? Why now? Previously, we were optimizing for likes, for revenue, and the big shift that's happened over the past 18 months is we've started to optimize for Sparks and User Outcomes. We 100% believe, and we know that Sparks and User Outcomes translates into MAU growth. In fact, we believe that there's even one step further we can go, which is targeting not just conversations or six-way conversations, but really high-quality conversations can drive even better MAU growth. In fact, this year we have a user give back budget in place where teams can make changes that maybe drive engagement at the expense of revenue. In practice, we've seen that most of the changes that we make that drive engagement also drive revenue. So in practice, we've seen we don't really need to use that user give back budget. But the fact that we have that budget in place sets the right culture for our team so they know they need to optimize for User Outcomes. Investors hear us talk about this user give back thing all the time. It's great to hear how it actually goes from the board to the investor community to where the real work gets done with engineers actually making changes to the recommendation algorithm and the freedom and latitude and therefore innovation and positive results that have come from it. Okay. Mark, users are, of course, getting more sophisticated, and they're starting to understand all these different algorithms. Even the way users think about these LLM models now. I'll hear random people, not even engineers, just people saying like, "Oh, are you using Sonnet? Are you using Opus? Are you using Fable?" These different AI models. It seems like people's sophistication about the way they interact with technology has increased. How are we taking that insight and bringing it into how we talk about our recommendations with users. Yeah. We learn a lot about what a user wants based on their swipe behavior and other things they do in the app. Previously, that was all kind of hidden to the user. But what we've realized is that we want to expose that. So we're working on different ways of presenting that information, telling them, "Hey, this is what we think you'd like based on your behavior. If we've got it right, great. Let us know and we'll keep on doing more of what we're doing. But if there's changes you'd like to make, let us know, and we'll adjust very quickly. Mm-hmm. So we're trying to give users more transparency and control over the product themselves. This obvious pushback to this product velocity initiative is that just shipping more is not always necessarily better. How do we make sure that we are solving the right problems, not just shipping more stuff? Yeah, that is a great question. Velocity, it is not just about shipping the most stuff. It is about shipping the right stuff. The way that we make sure that we are doing that is to make sure that we start with a real consumer need grounded in research and a ton of direct member feedback. We take all of that information, and then we are very liberal with what we prototype. We try a lot of things internally. We see what feels good. When we see something that we think has a high likelihood of improving the outcomes for our users, we will then test it publicly. Then if we see that it drives Sparks, that is when we decide to scale. If we launch something and it is not actually driving those positive outcomes, we do not move forward with it. We also regularly do an assessment of everything in the product, and we ask ourselves, "Is this feature driving the impact that we thought it would?" If it does not, we take it out. Everything we saw in that video earlier today really solves a very clear human need, whether that is better discovery, helping people express themselves more authentically, lowering the pressure with every connection. We also just have this tremendous design team that really works tirelessly to make sure that all of the work that the product and engineering team is doing fits together nicely. What stood out to me in the video is how much more social Tinder is becoming. Can you elaborate a little bit on that strategy? We have spoken to thousands of our members and thousands of singles off the app, and we're asking them, what do they want, right? I think the thing that comes up time and time again is they want to bring their friends into the experience. We believe that friends will lower the pressure, improve safety, and really make everything more fun, and we have a lot of evidence that supports this. We launched Double Date last year, and in the U.S., more than one in five singles between 18 - 22 on Tinder has a Double Date pair. When we talk to them, we hear they're having more great conversations. The conversation tone and vibe is more fun. When they meet, they feel safer. It's lower pressure. There's lower expectations. They've even come to us and said, "Hey, look, it's fun to Double Date with one friend." We actually are a Double Date pair. "But let's add more people." So a little bit of a sneak peek that we can maybe talk about in future events is that we're working on group hangouts now that supports more people. We also have our events feature that we've talked a lot about. A great learning there is that people bring their friends to those events, right? Very rarely does somebody come alone. So we're continuing to layer social into that experience, too. The whole point of this is that we're trying to talk to our users, understand what it is they want, and then give it to them. I think by doing things like building Double Date, testing groups, building events, we're giving people new ways to connect that will drive reconsideration and bring new people into Tinder. Tinder is better with friends. All right, so last one. Beyond more real-world connections and recommendation improvements, what are the other big things that you are both focused on? Let's start with you, Mark. I think one of the big pushes right now is really making sure that our matches become Sparks. The fun really begins after the match. People come to Tinder because they want to have great conversations and ultimately a great time together in real life. I think really this post-match experience has been pretty underserved in years past. We're really excited to have, as Vinay said earlier, completely rewritten our chat infrastructure, which is letting us now build on top of it these great experiences to help spark conversations, help people plan meetups. We've got a lot of momentum going here, and I know that as people use the app over the next couple of weeks, over the next couple of months, they're going to see a huge difference. Vinay, what are you focused on? Historically, a lot of our brands of Match Group have operated very independently, and I'm excited that we're starting to take more of a one Match Group approach to a lot of our tech services. The central Match Group AI team pursues longer horizon bets like conversational and agentic AI that all brands benefit from. The infrastructure that we're using, the GPUs, the machine learning platform, is increasingly shared across our brands. Our AI investments are reflected across the whole portfolio. When we build trust and safety features, we build it once, and we deploy it everywhere across all our brands. Things like age assurance, verification, AI moderation, they serve every brand. The learnings that we get from Tinder scale across the whole portfolio. Of course, Tinder's AI dev tooling and agent platform will soon be used by other brands at Match Group as well. Yeah. Again, we just had this product leadership offsite with leaders from around the world, and it was just amazing seeing so many people working on so many similar problems. We had people from our Tokyo office at Pairs and our Paris office who work on Meetic, and our Vancouver office who work on recommendations, and our L.A. office who are working on Tinder. They are all focused on a lot of the same problems. How do you improve trust and safety? How do you improve recommendations? How do you let people show up more authentically? How do you bring more friends into the dating experience? So getting more knowledge sharing, and then in some cases, more than knowledge sharing, actual shared technologies, has been a big priority of mine, and it is great to actually see it starting to have an impact. That is all the time we have for today. I hope the product video helped bring to light just how much Tinder has changed, and more importantly, I hope our discussion helped explain how we are making it happen. We are now understanding consumers better than ever. We are building faster and more thoughtfully. We are using technologies like AI to accelerate that. I am going to let Mark and Vinay get back to building. Back to work, guys. Thank you. Thanks, everybody.