How Amplitude is becoming an AI-native Product company (Spenser Skates, Amplitude) - #149

Behind every great company is a great product. And behind every great product are passionate, ambitious people. Clé de Voûte is the podcast for founders and product builders who are shaping the must-have products of tomorrow. Each week, I sit down with the people building Europe's next generation of tech companies to uncover their real-world strategies, battle-tested methods, and the lessons they've learned along the way. My goal? To help you become the keystone of your own company. I am Timothée Franck, and this is Klaidwood.

Today, I am really excited to welcome Spenser Skates on Klaidwood. Spenser is the CEO and co-founder of Amplitude, the digital analytics platform that helps more than 3,000 companies around the world understand their users and build digital products that matter. He started Amplitude just 2 years after graduating from MIT. And 12 years later, the company is a global reference in the product world, listed on the NASDAQ with over 700 people in the team. In our conversation, Spenser will share what it really takes to grow from these scrappy early days to meaningful enterprise scale, the trade-offs, the tough choices, and how you keep moving fast as the company gets bigger. We will also talk about building real product differentiation with AI, how to see through the hype, spot what's real, and focus on long-term value rather than chasing trends. And finally, Spenser will give us a look at Amplitude's roadmap for AI, where the company is headed next, how they are planning to deploy AI, and how it will positively impact their customers. I won't spoil you more. Let's jump in.

Spenser, thanks so much for coming all the way to this studio in Paris. You're actually coming from very, very far away. Really happy to have you on the podcast. You are the co-founder and CEO of a company called Amplitude that's actually very well known in the product and tech world, even here in France. And today we're going to cover quite a few topics around Amplitude from the early days of the company. It's been like 10 years, more than that, I think.

Yes, 13 years.

13 years.

Yes.

I thought it was founded in 2014, right?

2012.

2012.

So we were kind of 2 years without having any success, but launched in 2014. But yeah, I like to say we're 13 years old.

Nice. Okay, so 13 years old. From the early days of the company to how have you scaled it over the years? And of course, maybe I will talk about the role of AI, what AI is actually playing as a role in that journey. But before anyone maybe who might not be familiar yet with Amplitude, and I'm sure there are quite a few, how would you describe what Amplitude is today and what's your vision for 2025?

So we do product analytics at Amplitude. We help you track data on what users are doing in your website or app so that you can understand their experience and translate that into a better product. We help you find things like where users are stuck, what they like, what they don't, what keeps them coming back so that you can ultimately build a better product. We've been doing that since, as I mentioned, since 2012, and we've added a lot over the years. We have analytics, we have experimentation, A/B testing, we have session replay, we have guides and surveys, and we have a lot more on the AI front that I'll talk about in a little. But, um, hey, I just want to say I'm actually very, very excited to be here, Tim, because as I mentioned earlier, the French product development community, so like the French tech community, is one of the leading ones worldwide. Um, they totally get what the future of technology is going to be like. France was actually our earliest market here in Europe.

Oh wow.

Um, and what made us decide to create an office here in Europe for the first time, which was our first office outside of the US, because there were so many great French technology companies that understood what we were doing with Amplitude. So anyway, I'm very, very excited to be here in Paris and to get to talk to you about everything going on with product and tech and everything else in France.

Happy to. How do you explain that, by the way? Because when you have a look at the numbers, I would say that London is actually quite bigger. I think the numbers I saw were like something like 180,000 developers in France and actually twice as much in London. So how do you explain that? You actually, you just said that the French tech world was actually probably one of the biggest for Amplitude in the very beginning. I don't know. Is there like a reason for that?

Um, that it's, it is, it is interesting. I didn't actually know that. Um, I think there's more of a homegrown startup ethos here, I would say, in Paris, at least from the founders and companies I've met. Uh, there's obviously a large tech scene in London too. Yeah. Um, but that's more international or larger companies, whereas I think there are a lot of great French entrepreneurs. Um, we actually acquired a company, uh, uh, uh, June. So,, uh, Enzo and Ferruccio recently, uh, because they were— had a similar vision to what we were doing at Amplitude, and it's been phenomenal. Whereas, you know, we haven't really acquired any, uh, any equivalent, uh, company or seen any equivalent analytics company in, in the, in the London area. So it's, uh, I, I don't know what it is, but maybe you could call it the French entrepreneurial spirit or something. Uh, there's always been a very great, uh, technology community that's at the bleeding edge of what, what is happening.

Nice. And we'll talk about June for sure. If we maybe look back at Amplitude's journey, what were the most critical inflection points that changed the company's trajectory? Maybe we could start off with that and see what we could actually dig in.

So many. Do you want me to go? I can even go before Amplitude if it's interesting. How far back do you want me to go?

Okay, let's start with before Amplitude.

Before Amplitude. Okay, so before Amplitude, Uh, so I actually started the company. I graduated from college in 2010, uh, with, uh, from MIT and went to work in finance for a year first and then left that to start a company, um, and ended up co-founding, uh, this company called Sonolight, uh, with one of my close friends from MIT, Curtis. And Sonolight, we didn't— I mean, we didn't know any. We were just new graduates out of school, not familiar with the business world, you know, how anything ever, or how organizations worked, how, you know, to sell the business is nothing, and had never built anything before. And we wanted to work on something bleeding edge, and so we ended up starting with Sonalight. And so Sonalight was a voice recognition app that helped you send and receive text messages to your phone. It was like an— it was actually a Siri before Siri. And actually one of the really cool features that we had built, and this is before anyone had built wake-up activation, was we had the phone listen in the background to your voice all the time. So you could say, text by voice, and then the app would talk back to you without any having to press any buttons. It was like a, it was a really, it was a really, really cool demo that was, that was ahead of its time.

It was in 2011 then.

This is 2011, yeah.

Wow, it was early.

Yeah, so this is before Siri launched or any of the voice-controlled technology like Siri or Alexa or any of the other stuff ended up coming online over the next few years. So we were one of the very first. It's actually a funny story. The voice recognition technology was dominated by this company named Nuance, which frankly was more of a patent troll than a technology company and would constantly sue other companies for trying to develop voice recognition. So it was a very, very different industry and very, very different time. So anyway, but so we made Sonolight, we got into Y Combinator as part of that. We had this amazing demo on stage at Y Combinator where I had my phone in my pocket, like I literally put it in my jacket pocket and I talked to it on stage and I was like, oh look, I'm from the future, this is so amazing. It's like I'm in Star Trek. The Demo Day crowd went wild. We got the most coverage from a press standpoint from our batch. So it was 60 different companies in our batch, and we had the most press coverage, something like 12 articles coming out of the batch because the demo was so cool. And, you know, we raised a little bit of money, but we had this issue where our users were not sticking around. They would try out the app, it seemed cool, They would send or receive a few text messages, but they wouldn't keep coming back to it. So one of our questions was, okay, is the— how much does the quality of the voice recognition matter to whether someone comes back? Intuitively, you think it should, but until you look at the data, you don't really know. So we ended up spending all this time building our own in-house analytics, probably about a third of our team's time, which is a lot for a 2-person company. And to answer this question. And through that process, we actually figured out it mattered massively. So a user who got a successful match in their first attempt at using the app was twice as likely to retain long-term than one who wasn't. So it was like, okay, this is a really good signal. That means we can improve the quality of the voice recognition and get more users to stick around. So we spent a bunch of months doing that. Ultimately, we realized we couldn't do it fast enough to get to a high enough quality where it would be regularly used, so ended up giving up on it.

But— Just like the, at that time, what were the existing solutions? Were there any like great solutions?

Ah, for analytics, right. So why did we build it ourselves? Well, partly because we were engineers with a lot of hubris, but partly a lot of the other stuff on market could not do this. So there was Google Analytics, there was Mixpanel, there was Kissmetrics, there was Flurry, which was big in mobile analytics, there was Appsalar, there was a lot. With a lot of players. So we tried, you know, 3 or 4 of these guys, and they all would give you very high-level stats. So they would say, here's how many users did this action. But if you wanted to connect that to, okay, how many of those users did the action ended up going on to retain, none of them could do that. And so it was like, all right, we gotta go build this. And the good thing about that, it's funny, so voice recognition, You know, now it's hot with AI again, but it's a probabilistic problem. You don't know with certainty whether someone will— whether what someone is saying and be able to match it. And so it's hard. There is no perfect answer. And whereas what's funny on the analytics side, even though that was a hard technology problem, it was like, thank God, this thing has an answer. If you want to ask how many users did an action, there's an answer. If you want to ask how many users did that action then retained versus those who didn't, there's an answer, and there's a concrete number. And turns out it's just an algorithms and distributed systems problem that's very, very solvable. So anyway, so analytics ended up being a much more better fit for us as founders than the voice recognition app.

So you killed Cinderlight?

Yes, yeah, so we killed it. We looked, Considered different ideas, but we kept coming back to analytics because we, one, we had the problem ourselves, and it was obvious to us this was the right way to build product. You should look at what your users are doing. And then the other part of it was that so many other companies that were successful and were kind of taking off had no idea about any of this stuff. And we're like, what? How are you guys building product with no idea of any of this stuff? This is crazy. And they're like, well, you know, you just put it out there, people use it, it's great. Like, you know, they'll tell us if it's not working. And I'm like, okay. But when they saw what we had done, they were like, holy smokes, this is so valuable. I really, really want to get the same type of insight for my own product. And that was enough of a signal for us to say, there's something here. Let's go further and dig into it.

So you pivoted at the time? Yes. So this is the first inflection point, if we can call it that way.

Yes.

Could you think of another, like, other inflection points?

Yeah, so that was June of 2012. At that time, we added a third co-founder, Jeffrey. And the next big one was when we launched the product in 2014 and I switched full-time from the engineering to the sales side. And we went from $0 to $1 million in revenue in the first 9 months. And that was a huge, huge, huge freaking deal because we had been wandering. Yeah. Do you want me to go into that or should I go into the other inflection points first?

A quick break to tell you about Stellar, the team behind this podcast. At Stellar, we help tech companies turn their product into a true growth engine. Because your software isn't a cost center, it's a profit driver. We've brought together 30 top CPOs from companies like Zenly, Photoroom, and YouSign, and supported over 70 clients, from scale-ups like Fountain to innovators like Phenix. If you are a founder, CEO, or leader in product and engineering, you can start with a free 30-minute assessment tool. The link is in the description. Now back to the show.

How did you end up getting that result? It's actually very, very— we see a lot of— we are very biased because now we see a lot of companies such as, I don't know, Loveable, Cursore, and all these companies that get— they have crazy revenues within a couple of months. But how do you get that amount of revenues within a couple of months at that time?

I mean, it was 9 months, to be clear. So it wasn't like a few months like some crazy companies are these days. But we There was actually a lot of demand for this understanding of users. So, okay, let me back up. There were a lot of companies out there that really wanted to do the same thing. The internet was changing from a marketing channel to a product channel. And what I mean by that is if you look at the internet 2010 or before, it was very much a content-centric internet.. And that means how can you write long-form content and have it be indexed by Google and find it? And that was when you wanted to look at data, those were the kind of questions you were asking. What's happening now was there was the whole rise of web applications and mobile applications, and so people had very, very different questions. And so, you know, we, I guess you could say, end up getting lucky that the timing of our problem happened to coincide with the same timing of the market problem. Or I guess our problem wasn't— it was an instance of the whole market's problem of not being people that have these insights. Um, and, uh, yeah, so they were willing to use a company that was crappy with a minimal product and had all sorts of bugs and no brand and no track record and, you know, nothing that seems particularly trustworthy about it. One of our, um, one of our early, um, uh, ideal customers was— we had actually come across this community of ex-Zynga product managers. So Zynga was a, are you familiar with Zynga? No, not really. So they were a mobile, they were a, they make Farmville and like all these other social and mobile apps. So they were like a mobile app and social app developer. And they were really good at making these kind of free-to-play apps that would monetize a user base over time.

And it was a studio basically.

This is a studio. Yeah, they're still around today.

We've got something similar in France, which is called Voodoo.

Ah, Voodoo. Yes, yes. Exactly same as Zynga. Exactly same. Yeah, it's kind of like probably an earlier version of Voodoo. Yeah. And, um, yeah, so they kind of pioneered this model, um, and they were so dialed in on the analytics front. They had this huge internal system. Now what happened was a lot of people who were in product there ended up leaving to start their own companies. And when they did that, they're like, well, obviously you're going to do this by using data. Yeah. But they didn't have anything to use it. So, uh, we happened to run into a few few of them, and they were like, oh my God, this is the best thing. I'll tell you that the first time we sold the software, this guy named Brett Terrell, who was a former Zynga product leader, uh, and was the co-founder and CTO of this company named Super Lucky Casino. So it was like a mobile slots game. And I go into the meeting, um, and we had no idea how to sell. So we kind of went through— he, you know, he asked some questions, we went through the demo, we showed it off. I remember actually at one point, one of our early employees, Alan, he had like a pen in his pocket and I guess he was playing with it and somehow the pen exploded and he got like ink all over his hand in the meeting. And I'm like, oh my God, we look so amateur hour right now. But they were, you know, this guy was cool on it. He's just like, oh yeah, bathroom's over there, you know, no worries. We just laughed and we laughed it off. So anyway, we get to the end of the meeting. And, you know, normally this is the point where I'm thinking, okay, I'm going to have to beg this guy to use, use the product because, you know, nobody— yeah, they say they want data and they say they want analytics, but they're not really willing to work with somebody.

Yeah.

Crappy, you know, early stage, you know, company. And he asked a question I'd never been asked before, which is how much does this cost? And I was shocked to hear this question because I was like, what? Like, you're asking how much you want to spend? Like, what? And I'm like, I didn't even plan for what it cost. And in my head, the first number that popped in my head was $50 a month, because that's what SaaS costs, $50 a month. But I was like, wait, wait, no, no, no. I got to think of a bigger number than $50, because the advice is always ask for more dollars when you're being asked how much something costs. And I'm like, all right, let me double that to $100.

Yeah, exactly.

And so I was about to say $100, but I'm like, wait, wait, wait, that's not big enough. I'm like, well, let me think of the biggest number I can possibly think of. I'll put another zero on that. I'll say $1,000. Wow. Which I'm like, you know, I don't even know if he's going to go for it, but like, fuck it. Like, we have nothing to lose. So I'm like, $1,000 a month, which is much confidence I could say. And to my absolute surprise, he was like, wow, that's so cheap. And I was like, oh my God, this is incredible. This guy wants to buy my software. And like everything I had ever wanted to do as a software entrepreneur was validated in that moment. Because you had this person who was willing to spend money and not just a tiny amount of money, like $1,000 a month. Was more than I had made, we had ever made with Sonolight. Um, and I was like, oh my God, this is amazing. So he ended up becoming our first customer. Um, and more importantly, he ended up introducing us to a lot of other, uh, founders who were also former Zynga product managers. Um, and that kind of, kind of kickstarted our growth.

That's so cool. So what, it was in 2013?

Yeah, so this was mid-2013. Yeah. And then, um, what happened was in, at the start of 2014, we officially launched the company with a TechCrunch launch, and then I switched full-time from building the product to sales. And I was like out there and like just selling the, you know, just talking to customers all the time. And I had no idea what I was doing. You know, I got a sales coach, this guy Mitch Morando, who was incredibly helpful. By the way, if you're a founder that's an engineer and you go and you decide you want to learn how to sell, get like a sales coach who can help you because it is, yeah, there's so much to learn and it's just as deep and important and complex a problem to the success of your business if you're a B2B business than the engineering. Engineering. And then we skyrocketed from 0 to 1 million in that first year. Excellent.

You said in the podcast that I listened to before, after preparing this podcast, that Amplitude is around now like $300 million.

$335 million in ARR as of Q2. Yeah.

So it got actually big. I won't talk about the headcount, the size, because I know this is not a great indicator, as you said in the podcast. Terrible indicator. Exactly. But I guess that you guys are not like only two. Now.

It's always a trick. We always ask our people, are like, how big are you? And it's like the right answer to that is your revenue or your number of customers, not your headcount. Any idiot can hire a lot of people.

But I didn't want to ask a question around that. I just wanted to give a piece of context before asking my question, which would be around how do you actually— so when you get bigger, how do you actually maintain product speed? So I mean, like, overall velocity, product velocity. Exactly. And experimentation while growing the org and serving bigger customers. Because I guess you didn't start from scratch with enterprise customers. I guess you started with smaller customers.

Yeah. Well, there's so many things in that question. The whole shift to enterprise, we started that in 2017, and that's a whole topic in itself. The innovation, that is extraordinarily hard as you get larger for a lot of different reasons. And that has been something that as co-founders, me and Curtis and Jeffrey have been paranoid about making sure we keep in place from the very start. And it was, it's not been easy. Like we spent a lot of effort into doing this. Where to, where to even start? I think the first most important thing is a high engineering bar. And so we have a very rigorous engineering interview process that we were deliberate about keeping from the start because no system or no process can fix if you do not have the right talent in the building. Like nothing, nothing can fix that. And so being fanatical on that, and there is pressure for all sorts of reasons. Hey, you want to go higher and you want to lower the bar and you want to add headcount. You know, like this is why I don't like the headcount question. It's not a good indicator of success. You know, in fact, sometimes the opposite. And so we just were fanatical about that. And the other flip side of it, it was just being very aggressive about performance management for people who are not the right fit, moving them out of the company. And it, you know, it's kind of a tough and ruthless thing to say, but also incredibly important to creating a high-performing team. So that is the foundation. You have to have that. Once you've done that, then it's about How do you set people up with the right mentality around how to solve problems? And so one instance of that is we are a very bottoms-up culture in terms of product development at Amplitude. So, um, we leave it to individual teams as to what to prioritize and what to build. And, you know, I'll have opinions on that sometimes too, and I'll come in, but it is We are looking for people to own that prioritization decision of what is going to make an impact. I was writing a post about this recently. We did this big change in product management leadership at Amplitude, and we were talking about what makes for great product leadership. And one of the things I say in it is your job is to invent the future. And it's almost an impossible task because it's extraordinarily hard. If any of us knew the future, we'd be already rich. But that's the job. And there's a lot of different methods you can use to figure that out. Like, you can be a customer yourself. You can talk to customers. And that's— we're very big on talking to customers at Amplitude. One of our things is product managers should talk to customers on average once a day, if not more. That's kind of once a day is the minimum, 5 days a week. And even engineers, we want them to talk to customers at least 1 day a week. And so, you know, that's a really good signal of, you know, your customers are living in the future already and they're kind of seeing what's broken and they're going to give you really good feedback. So anyway, so, and that is the, that gives you a ton of perspective on what to build and what the, you know, what is likely to lead to success. We also do like lots of things culturally with rituals to help enforce that. So, One thing is we have hackathons twice a year where people just build whatever they want, and they do some amazing stuff. They absolutely amazing stuff that comes out of those. Like so many of our innovation comes from stuff that originates in a hackathon at Amplitude.

Do you have any examples that you'd like to share here?

Yeah, I mean, a lot of our integration with data warehouses came out of a hackathon. A lot of our coolest visualizations came out of hackathons. Chart visualizations. A lot of, like, our, like, people built early versions of new products like Session Replay and Experiment in hackathons. So it, like, there's so much, so much good stuff that came out of it. And then, uh, what else, what else would I say kind of culturally in terms of what makes Amplitude's engineering and product development really strong? I think So we also have a strong bias. We have this thing called ship fast, very similar to Facebook, move fast and break things. And the reason we do that is because one of the, or the best predictor of the quality of any piece of software is the number of iterations it's been through. And that is a better predictor than the— how good the people were that wrote it. Number of iterations is a better predictor than how long you've spent writing the code, than what sort of methodology you use. You kind of look at any sort of heuristics. Number of iterations is the number— is the best predictor by far of the quality of any product or piece of software. And part of that is we embrace that mentality aggressively with, you know, ship fast as a mantra. And there's this whole false dichotomy where people are like, well, that means you don't focus on quality. Absolutely not. Shipping fast is what leads to high quality because that is what is going to make something really, really good. So anyway, there's a lot of kind of cultural things like that that we've learned over the years from our own experience, you know, partly inspired by stuff that Paul Graham and the Y Combinator community has written about. Um, you know, and we've just been fanatical about continuing to drive that even at scale. Like today we're a 200-person product development team, uh, between engineering, product, and design. Uh, but we still embrace all of that mantra. Now it has trade-offs, like especially in enterprise software where, you know, stuff will break more often and that's frustrating for customers. And yeah. Um, but that's a very intentional trade is like, okay, well, in exchange for that, you get a much higher quality product shipping you much faster.

Would you say that Amplitude is a product company? Yes, absolutely. In the DNA. How do you, like, what cultural or org mechanisms help you keep this DNA at scale? I think you actually gave a bit of the answer right before, but I wanted to ask you this question.

It's hard. I mean, I'm ultimately, as companies are kind of reflections of their founders in a lot of ways, and so it's a reflection of —of me, it's a reflection of Jeffrey and Curtis, it's a reflection of a bunch of the founding team, folks like Jin and Nirmal. It's, you know, it kind of is just how we think, and that manifests in the product and, you know, how we build it and deliver it. You know, one other thing I'll comment on is, like, I think people like to overcomplicate product. Like, sorry, it is, and it is very, like, trying to predict the future, It's incredibly difficult. But it doesn't have to be complicated in the sense that you don't need like a— like one of the anti-patterns I see in product is like, well, you know, if I just use this framework for building product and I embrace this religion of how you're supposed to do it, I will necessarily invent the future. And it's like, no, not really. Like, you know, because, you know, you do it and then people are disappointed that it doesn't actually take. And it's like, well, what did I— I just need to follow the process better. And it's like, no, it's like inventing the future is intrinsically intrinsically hard. It's an intrinsically unpredictable problem. And so you're going to make a lot of mistakes in doing it. What's important is that you're constantly reevaluating what is you're doing and adjusting and learning, and then ultimately, you know, kind of heat-seeking to where it is that you have. And so I don't think of our success at Amplitude as like any magic insight we had about the market or, um, the product or the engineering. It all comes back to the fact that we have focused on building a machine that is just great at churning out lots of stuff all the time. Um, and so it's not about any one bet succeeding. It's about, hey, the velocity of this is going to get you more practice and more shots on goal in it than anyone else out there.

As the CEO of Amplitude, what's your take on— I'd love to have your take on how you actually balance data and intuition when it comes to product building, product decisions, basically. How do you approach that? Because you could actually be biased since you're totally there. So I actually got this question this morning.

I was like, hey, how do they do that? Yeah. So I'm going to say something very controversial for someone who sells data for a living. Love that. Which is, I think another, in addition to process or frameworks being a kind of false god, I think data can be a false god too. Data is merely a tool to help you understand what your users are doing better so that you can make a better experience for them. It is not the— I see one of the mistakes I see people making with data is that they will say, okay, hey, our goal is number of active users. Oh, how do we drive more active users? Well, we get more people to convert. Well, how do we get more people to convert? It's like, okay, let me just follow that blindly and see and like try to up those numbers without thinking through, is this a better product experience? Am I driving more value for customers? Is this something that people are going to keep coming back to? Instead, data is very much a tool. It's a— that can help inform what is going on with the usage of your product. And it's— there are things you can There are tests, like there are things you can validate or invalidate about your thinking, right? So like we found exactly when we were doing Sonolight, exactly how important the quality of voice recognition was. But it's not the end in itself. The end in itself is, the end is inventing the future, not the data in itself. Now, that said, there are enormous, I think there are so many great examples of enormously powerful things people have found out through data. Facebook had a very famous study in its early days where they found the best predictor of long-term engagement was how many friends you added. So if you added 7 or more friends in the first 10 days, you had an 85% chance of sticking around 2 months. And if you didn't, it was less than half that. And they— now, that may sound obvious. The whole category is named social networking, but it was not obvious to any social network before them. MySpace, Friendster, Orkut, none of them figured this out. It was Facebook that really figured this out and oriented the whole product experience around adding friends. Airbnb had a famous finding in the early days. Listings with professional photography converted, I think, at a 5x higher rate than ones without. So there are— DoorDash, another great example. DoorDash figured out on-time delivery was critical to getting people to do repeat deliveries, repeat food orders, and operationalized their entire business around that, whereas a lot of their competitors did not and ended up losing out to them. And now they're the largest food delivery. So there can be lots of amazing insights you can uncover, but data is not the end in itself. Your job is just to invent the future. There's actually a really good segment. I don't know if you know Nate Silver. I know. Yeah, yeah, yeah. So he predicts. He's election forecaster in the US and does a lot of other data and data-like things. One of the— he had this really good segment recently on how do you— how do you— like all the different fields where you have to predict the future. So like finance or technology or poker is another good example where you have to make big decisions under large uncertainty. One of the attributes he said, which really resonated with me, is that you have— people have a— you have a lot of different models of the world., and when a lot of them are telling you something contrarian, that's a strong signal. So here in product, what do we have? Well, we have, uh, the users you talk to, you have the data you look at, um, you have your own experience using the product, uh, you have what your competitors are doing and what the broader ecosystem is doing. So there's a lot of different signals that, that you can see, um, and these inform when you should make a that to say, okay, here is where I think the future is going to be.

Today, every company is actually rethinking their strategy, how they are, like how they differentiate. And AI, I think, has completely changed the playbook in some ways. So yeah, let's talk about how you're approaching that shift at Amplitude. Could you maybe give us a quick overview of how AI has changed Amplitude since maybe this AI wave that came out 3 years ago, I think now, or before, because I guess you guys, since you're based in the US and you see a lot more stuff that we do in France, maybe you actually started working with AI before that.

Well, we were late to it. I think we didn't go aggressively on it until about a year ago. And that was— it's a long story. But what's interesting, we've actually seen is a lot of people who are closest to the technology are the ones who are most skeptical of what it can do. And that included us because we have a lot of top engineering talent at Amplitude. You were skeptical? Yeah, I was. I was. My co-founders were skeptical because one of the things we saw was a lot of magical thinking with AI, which is like, well, you know, here's this hard problem I don't know how to solve, so let's throw AI at it. You know, AI will do something. You know, it's like, guys, like, this is not how this technology works. It's not a substitute for thinking hard about the problem. That changed though a year ago where we went much more aggressively against it. So we did— so first was the team and then how we've approached it. By the way, I will say that the key difference on AI is you— it's so different from building SaaS because in SaaS what you do is you talk to the customer, you get a description of the problem, and you prioritize it and you build it and you just go do that again and again and again. And it's relatively straightforward. With AI, customers do not know what is possible and what's not. And so they're not gonna be able to tell you. They may have some idea of what they might wanna build, but it's not gonna, they're not gonna connect it to the reality of technology. And so what's more important than listening to the customers, which feels like blasphemy to say as a product person, is understanding the technology and what's possible, what's not. Andrej Karpathy had this great thing on it, which is the capabilities of AI are very jagged. And so there's some things that it does extraordinarily well, and there's some things that it's terrible at. And understanding that edge and how that edge is changing is actually key to understanding what's possible to build from a product standpoint. So because of that, you have to, I think most SaaS companies are not going to successfully make the change and going to be dead, even though there's so much hype and so much focus around it. Two big things that we did. First, we changed the team significantly. So we had a new engineering leader join us in October of last year, and this guy named Wade Chambers, and him and I spent a lot of time focusing on hiring great AI engineers. So people were familiar with the technology and he had worked with a number of great people from before. The other thing we did was we acquired a number of companies starting with Command AI back in October of last year. We've also acquired June, as I mentioned earlier, as well as Craftful and this other company, Inari. And so that team, those individuals are incredibly important because they were already very familiar with the technology and has spent years understanding how to build with AI. And what we did was then we took those, that expertise and broadcasted it to the rest of the company. So in June of this year, we had an AI week where the whole goal was to just get everyone learning about AI. So we had a number of sessions where we had different people familiar with it do talk about the technology. We brought a few people in from other companies. And then we just had people build the entire time. And, uh, so the first 2 days were some training, and then the last 3 days were just building. And at the end of every day, uh, on those last 3 days, we did a show and tell. It wasn't even kind of like a demo thing. It was just, hey, show what you've been working on with this. And people came out with some amazing, amazing things, uh, that they had built with AI. And the point was that they would just learn and embrace the technology. You know, we got everyone repping on a cursor. We actually did this really cool thing where some product managers and designers actually started submitting pull requests. So it was awesome. And then after that, it was kind of— we were off to the races as a company on AI. People started coming up with all sorts of ideas for how analytics would look different with AI. So we had automated insights. One person wanted to do like SEO except for LLMs. We had someone got really excited about MCP servers. Someone ended up building a You know, like we got a bunch more paid people that were going deep on the agents team. We had Automated Insights. We had all this amazing stuff that was very bottoms up. And then we looked at, okay, how do we— so we kind of went on, did these side projects for a little bit, and then we reorganized the teams to focus people so that they could go after these problems full time. And now we're in the midst of launching a whole bunch of amazing AI features. Next week we're going to launch our MCP service. Server. The week after that, we're going to launch what we call AI Visibility. The week after that, we're going to launch AI Feedback. Early next year, we're going to launch Agents. We're going to launch this AI chat functionality. We're going to launch Automated Insights. We're going to rehaul Amplitude's navigation— or sorry, interface with a global chat experience so you can talk to your charts and have it make changes to them. And so it's a total grounds-up bottoms-up rebuild of Amplitude and analytics from an AI lens. So I'm incredibly excited about it. Now, we still have a lot of work to do to prove that people are going to adopt and use these features, but we have some promising data on that so far.

That's super interesting because you said that you shifted from kind of a skeptical mindset around AI in the beginning. You told me that you started to actually work hard on it a year ago. Yes, one year ago. And now you're going— you're very bullish, actually.

Very aggressive on this. Yeah, aggressive on this. There is no way analytics doesn't get transformed by this, because you think about what analytics is. You're looking through massive amounts of data, and you're trying to find insight and value in that. And it's such a human manual process today that just even a little bit of automation in the right way and a little bit of thinking can lead to amazing results.

I think that every analytics company today claims to be AI-powered, and not even analytics. Every SaaS product claims to be AI-powered.

Yeah, you look at any of our competitors, it's AI this, that, and the other thing on the website. We joke, we call that like Salesforce AI at Amplitude, where it's like, Salesforce is like, and Marc Benioff is hyping up AI like crazy. But if you ask anyone, we're Salesforce customers, and you ask anyone, it's like, hey, have you used any of this stuff? And it's like, no, not really. Doesn't work. Doesn't actually do anything. So yeah, there's a whole spectrum from the foundational model companies, which are legit AI, to Salesforce. And I think a lot of companies have branded like Salesforce has, but haven't actually done the product work to transition.

How do you cut through the noise and focus on real sustainable differentiation with AI?

I mean, there is no substitute substitute. I mean, you have to change the team fundamentally and how they work. And this— yeah, yeah. So all the stuff I went through where we brought in people who had known how to do it, both through hiring and acquisition, as well as retraining our entire organization on it, like there's no substitute for that. Yeah, you have to do that. And frankly, what I see is a lot of SaaS companies, they're just— they aren't going to do the that change. Yeah, that organizational change, because it's hard. There's a lot that's required of it, and it's much more comfortable to do the same thing you've already been doing.

Yeah. So you're releasing a lot of AI features now. How do you make sure that these features drive measurable business impact for your customers rather than just like becoming vanity addition. Yes. Yes.

So this is actually really important. It's— I was going through our— talking with our AI agents team the other week about this. It's very easy to trick yourself that something is valuable, especially when you're innovating in a category for the first time, because you'll just be like, look, we have all these people using it. But there is a difference between usage and getting deep value out of, out of something. Usage is just like, hey, I have a daily active user, you know, and it's like, okay, yeah, someone might be trying it out or who knows. But it's, again, that's not the same thing as are they getting massive value out of it. I remember like probably the worst example of this, if you remember Google+ way back in the day where they're like, wow, we have hundreds of millions of people using it. And it's like, guys, like, no, no, no, you don't. So I think a lot of companies, especially when they innovate, trick themselves on this. Now, there is no hard criteria for like, is this deep value? But there are some tests. There are some signals. If someone is using it a ton, that's a great signal. If someone is coming back day after day, that's a good signal as well. If someone would be very upset if you took it away, that's also a good signal. If someone would pay dollars for it, that's also a good signal. There's not any one magic like, "Oh yeah, this is for sure valuable," but there's a lot of heuristics you can use to know it. The most important thing, when you're innovating, particularly on this AI stuff, is not to deceive yourself whether it's valuable or not. So anyway, I made this whole distinction to the team about, okay, there's usage and then there's deep value. You guys have some usage, but show me the people that are closest to deep value, because that's what we need to go prove when it comes to AI agents, as well as all of our other functionality.

That's funny that you're talking about Google+ and such product that maybe didn't kind of had the right north star metric to follow.

It didn't get to deep value.

It never did. Exactly. What's your opinion on what does Loveable, for example, a lot of people are actually saying that they have a strong, high growth, but that this revenue could be kind of the curiosity revenue, like people coming, just paying and sort yeah, the time to value is actually so short that they actually pay very quickly. But after that, maybe they would actually not be sticky. What do you think about that?

I candidly, this is one of these predicting the future things that I don't know. So let me tell you the signals going back to my models earlier that I see. First is they have 100, they're beyond $100 million in ARR, and that's incredible. In general, if you're a SaaS company and you get to the $100 million ARR mark, it's basically impossible to kill you. There's enough people getting enough value, and even if there are problems with it, you'll probably stick around in some form. Now, in addition to that, if you look at the retention metrics for any of these companies, whether it be Lovable or Bolt, or if you go on to the coding side, like the full IDE side, like Cursor, which has gotten to a billion, The churn of these products is incredibly high, like much, much higher. I remember talking to one venture capitalist about, they had looked at one of these very hot AI companies and they're like, wow, they have 30% churn. And I was like, man, 30% churn, that's like, that's pretty high. But we were kind of looked like that when we were early as a company, where like, yeah, you'd only have after the first year, 70% of your initial user base. So you'd have to grow out of it. And we've reduced it and we've gotten closer to, we've gotten to 80 and then 85%, you know, and we're trying to get to 90%. So, um, you know, so I was like, oh, that's not that bad. Uh, he's like, no, no, no, Spenser, 30% churn a month. And I'm like, what the fuck? 30% a month? After a year, you basically have none of your users left. What the fuck? Like, how can you possibly run a business like that? Anyway, so I— so now, you know, so you have this irresistible force, which is these incredible growth curves, and then you have this immovable object, which is these massive sky-high churn rates and how those net out, I don't know. And so, and so it's, it's going to be a to-be-seen. Now, I will say, if you look at Cursor, it's clear to me that thing is sticking around. Very clear to me. Like, we're using it internally with Amplitude. We're all in on it. Yeah, we're trying out other things like Cloud Code and Gemini and others, but like, as an IDE, it works very well. It's replaced our existing setup there as a company. Like, As long as they continue being on the bleeding edge of innovation, we're going to stick with them. We're going to continue to pay them money. So even if a lot of their customers churn, I know there's a lot of customers that won't, and there'll be a real business there. Now, with something like Lovable, I'm less sure.

So again, you say that you are working like crazy on integrating AI into the product. Would you say that you're trying to make Amplitude look like or not look like, just building around being an AI-native company, if I can say so.

Absolutely. Absolutely.

We're all in on this. Okay, cool. I've heard on one podcast called, oh, How I AI, that Wen Chambers that you— Oh, yes.

Yeah, Claire's podcast. I'm a huge fan of hers.

So you guys apparently have built an internal tool, like AI tool called Mona. Moda. Moda. Yes. Do you still use it?

We do. Yeah. So it's a Slack bot that is hooked up to a number of different sources in Amplitude. It hooks up to Glean, it hooks up to our own Amplitude data, it hooks up to our CRM, it hooks up to all these places, and it pulls data from them and summarizes and aggregates it for you. And that is a great use case of AI, like being able to search and find information, great use case.

Do you think this is these kind of initiatives that actually turn a company to like an AI company, or is it like just, you know, one initiative among a lot of others that would actually make it an AI company?

The key thing is the talent. You have to get the talent right. And you can get people practicing it in different ways. And Moda was Wade's actually way of kind of kickstarting it because he shows like, look, this is actually possible. Let's show people within the product development team, get them using it and showing them. And that was actually, yeah, that was actually kind of a certain important thing. But the underlying thing was changing the team. By the way, on the team side, I will say the talent looks pretty different on AI. It tends to be younger. It tends to come from elite coding or math backgrounds. So you have a lot of people who have done the International Informatics Olympiad or the IMO. It tends to be like very engineering technical heavy as opposed to product or design heavy. Heavy. And so the— we've been lucky that this has been a little bit of our DNA at Amplitude. Like, Jeffrey was on Stanford's ACM team. Curtis and I did this competition at MIT called Battlecode back in the day, and we were very into this sort of stuff. So we've actually, you know, we actually, you know, we like to think we know a little bit about what we're talking about when it comes to good AI talent, because then you marry that background with a kind of focus on customer obsession and you get great results.

So you changed, so you hired that kind of profiles that were actually new to Amplitude?

So no, no, no. So we are that is what I'm saying. Amplitude was always like, we were always like this like top engineering team. Since the beginning. Coding team. And with this AI, that gave us the confidence when evaluating talent on the AI side about, hey, this actually is going to be a fit.

Yeah. Okay. And I'd love to have your take on maybe the way AI changes the roles of maybe like engineering product or even like the GTM, like the go-to-market part of the side of the company. How do you see these roles evolving over time with what AI brings to you as opportunities? Just like in terms of strategy? I don't know. I feel like there is a strong change now I went to the US last week and I feel like there is a huge gap between the US and Europe when it comes to thinking of, sorry, how these teams will be like, yeah, will change in the future. So maybe you have one opinion about that.

Yeah, so first on the product development side, I think everyone becomes more of an engineer and more of a designer and more of a product manager because is the barrier to all three has gone way down. I mean, first on the engineering side, you know, anyone can, can code and submit pull requests now. And we have designers and product managers who do this at Amplitude. And then secondly, you know, if you're an engineer or product manager, you can create designs much easier than you could before in Figma or in other tools. And then same on the product management side, engineers or designers can get lots of look at customer feedback in a much more scaled and aggregated way. So it blurs the lines.. And instead of specialization, I think you're actually starting to see a little bit of convergence, and then it's just the skill of the people that matters in leveraging these tool sets. Yeah, we have engineers— we did a survey after our AI week, and I just asked, what is— how much has your productivity changed because of— from AI? You know, is it less? Is it more? Is it zero? Is it 10%? Is it 50%? Is it more than on that, median response was 40%, which is wild. I was like, holy shit, like I've never seen that level of productivity change. And then there were some people that, you know, self-reported, but would say it was like 2, 3, 4x more productive for them. So I was blown away by seeing that. So anyway, so I think the roles start to merge together and you don't need as much specialization, which I think is a good thing. And it's more about, you know, back to this invention in the future of building great product it's that skill gets crystallized. So whoever is able to best really understand the future and create it, like, that's who gets rewarded, whether you're an engineer, whether you're a product manager, whether you're a designer. Now, on the go-to-market side, it's interesting. I think some roles are going to change a lot, and some roles are going to stay the same. So I think the roles that are going to change a lot, a lot of— Ops roles and back office roles, I think, are— I don't think they're going to get fully automated, but I think people who know how to automate them are going to replace those who don't. I think on marketing, marketing AI looks very different. Like, if you look at, again, like Salesforce AI on this side and, you know, foundational models on this side, I think the old school Salesforce way of marketing is going to go away and you're going to a lot more like, hey, you know, how compelling content can you create on Twitter? And so that's going to be a reformulation. And I'm just in the process of starting to do that with Amplitude's marketing team. Now, that all said, I think the role of great salespeople will actually not change. Sales is about evangelizing someone a better future than they have today. And there is like, that process requires a lot of human connection and trust and authenticity that you're not going to replicate any time soon with AI. Now there's small stuff in it that can be tweaked, like in Moda, for example, you can pull instead of having to, you can really easily pull, you can say, hey, tell me the latest about this account and you can pull all this data on it. It's great. But so I think a lot gets more efficient and automated, but that core thing of evangelizing someone a better future and connecting with them to do that, like that, it's going to be a very human. We as humans are very wired for authenticity. Authenticity is incredibly important for us. And we can tell when we're talking to AI and interfacing with AI and we fucking hate it. Like when, and it pretends to be human. And so that, that, like, you look at the rise of TikTok, for example. What is TikTok? TikTok is an endless feed of just fucking people and their faces talking to a camera to you. It's wild. But we care about that because that's what we connect with as humans. And so I think that authenticity AI is not going to replace. And so I think great sellers will have a role. Now they're going to have to learn how to use a lot of this tooling, but you know, that, that comes with every technological change.

If we quickly go back to what you said about the product and engineering roles, you said that the boundaries between these roles are getting blurrier. So how these people actually know what roles they have, because designers can actually kind of code now, engineers can actually directly contribute to product development and product thinking, and actually PMs can actually do both. So how do they know? How do these people know exactly what role they have? I felt like in San Francisco last week, the gap was like all the companies over there, they are all using these new technologies, coding generating tools and many, many AI tools. But I think the huge thing now is how will they actually make these roles evolve in the future? Because everything gets blurrier, basically. I think, yeah, there is no boundaries between all these roles.

It's a great thing. It's actually a great thing because now you're no longer pigeonholed into a particular functional role, and you are just judged on can you predict the future and invent it? And so the quality of thought and the quality of leadership is the— that's the last barrier, for it. And so it's actually great. Like, we have designers who are leading some of these work streams at Amplitude. We have engineers that are leading stuff on the product side without PMs. We have PMs that are just kind of coding stuff themselves. It's awesome from my standpoint. Like, look, now we still have all the functional roles there. Like, people still have their titles and their roles, but it's just— it doesn't— it's great because it unlocks the ability of great leaders to be effective.

Yeah. And do you Do you have thought of an internal, I don't know, way of prioritizing AI initiatives since you're launching a lot of stuff now around AI? Is there something that is different to what you used to do in the past?

So with SaaS, when you are getting the requirements from the customer, It's rare. You don't have to do as much invention in SaaS. Most of the time, you are building something relatively straightforward that you know some number of customers want, or you're, say, entering a new category or competing with a bunch of existing companies, and you know it's pretty straightforward what roadmap you have to build to go do it. So the risk is more on the execution and delivery than it is on the innovation. You don't worry as much about, will people use this? You know they will. There's still risk on it, but you know they will for the most part. Whereas with this AI stuff, the risk is all on the innovation. You have no freaking clue if people are actually going to use this stuff because we're creating something that's literally never been invented before. So the big change is you have to be much more reactive to what works. And so you have to double down on things that that you see a lot of usage on and go big. On the other hand, you have to be more aggressive about cutting things that are not working and saying, hey, look, we've tried this thing, but it hasn't worked. And that's hard. And so we're going to get tested on that. And a lot of companies are going to— a lot of companies, I think one of the challenges that happens when you're large is you have all this planning set in place because there's lots of dependencies. For example, when you do a product launch for the first time, it's like, okay, well, you need to create a marketing plan for that. You need to enable your salespeople in the field on that. You need to have pricing for that. You need to have documentation on that. You need to make sure your support team understands. And it's long, long list. You need to make sure to manage the change with your customers, introduce them to it. You need to make pitch decks. So there's all this stuff that are dependencies for the launch of a new product. The problem is when you're doing something innovative, it's like you don't know if any of this stuff is going to work. So it's like, do you invest all of those resources? Yeah. And how do you get the right sort of touch on it? Because yeah, you can invest all of those resources for everything you try, but you also want to invest sufficiently such that something gets the chance to succeed. So you have to be a lot more dynamic to what you see working, and you have to do kind of smaller tests and earlier tests and all sorts of Yeah, you have to be willing to change your mind a lot, and that's hard. That's really hard as a larger organization because you're trying to coordinate all these people, hundreds of people in the case of Amplitude, to make your business successful. Would you say that you're killing more features than before, than you used to be? That's a great question. I'd like to say yes, but my guess is we're probably still not good at it.

No, it is very hard to do, actually. It's extraordinarily hard. There's a very strong emotional connection you have with what you build.

The problem with B2B is that you have customers who are paying you for a thing. And so if someone's paying you for it and relies on it, the stickiness is very high, which is great in a lot of respects, but it makes changing the surface area much harder. And so there are products we've deprecated, so we're not selling anymore, but still, like, we have a handful of customers that are paying and relying on it. And so we maintain it and we actually will continue to support it. So it's much harder. And I don't have like a magic, you know, this has actually been a long debated topic within Amplitude and we haven't figured out a great answer on how to approach it.

Yeah. You said earlier that you've acquired the team from June. So it was a, I don't know if it was a French startup legally, but at least like Enzo, the CEO is French. Yeah, Enzo and Ferruccio, they were based here in Paris. Yeah, exactly. We had him on the on the podcast a couple of times, I think twice a couple of years ago. And so it was basically like a simplified analytics tool. And you also acquired Command AI in 2024, I think late 2024, which was an AI-powered user assistance tool. Right. And so you're kind of building Amplitude by acquiring other products. Could you actually share some maybe internal principles or thoughts, frameworks, I don't know, that you use to decide whether to build, buy, or just partner for AI capabilities?

Yes, you're always going to be doing both. You want to have a team that is capable of building something, and you always want to be looking for great talent to acquire and bring into the company. Company. And I think a lot of companies get too religious on building everything in-house or not being able to and having to go buy, and it ends up hurting them. So you've got to be able to do both, and we've done both. Like, on experimentation and session replay, we built those in-house, whereas on activation or on guides and surveys, you know, we, we bought, uh, those capabilities. In the case of acquiring companies and teams, what we're looking for is a team that has spent many years thinking about this problem and is capable of customer development and is a good fit culturally with the Amplitude team. And that's actually quite a hard set of requirements. So, okay, well, let me go back to first principles and why do you buy something in the first place? The reason you buy, because there's a lot of downsides There's risk that it doesn't work out. You're paying a lot more money than you would otherwise. You have all this distraction, all this friction of integrating two organizations, two cultures, and two teams. Yeah, there's a lot that can go wrong. Most— the vast majority of acquisitions fail. So why do you do it? The reason to do it is because there is someone out there that has studied your problem and has a set of expertise that you do not. And so we'll accelerate your understanding of that problem. You pay for that. So that's what you're paying for. Yeah, yeah, yeah. So for example, in the case of Command, James Vinay and the Command team had spent many years thinking about how to create a great experience, and then specifically great guides and surveys product. And so they were still quite early on when we reached out to them, but we knew we wanted to get in the space, and they ended up being a phenomenal fit. They were based in San Francisco. They had a similar ethos to us about customer development, which is like continually talking to and building for customers. They had a, you know, they had the same kind of cultural attributes of humility and ownership and growth mindset, which we take very seriously at Amplitude. They had a very like scrappy, get-it-done startup mentality, which I loved. And like I mentioned, is I've been paranoid about keeping within Amplitude. So we ended up Uh, and then, so we figured out financials that would work for both sides. We did the acquisition that came out in October. It's been awesome. And then a lot of people on the Command team have gone on to take on bigger leadership roles at Amplitude. It's been fantastic. Like James, um, the, uh, CEO co-founder is now responsible for all new products on product management. Uh, so everything outside of analytics. Uh, Vinay is leading our AI engineering efforts. Tyler is now leading our design team. Um, you know, we have Lucas on product management who's leading a bunch of, uh, you know, he's now leading session replay and a bunch of other stuff. So we have lots of great, uh, we have lots of folks who have taken on bigger, more expanded roles. And I think that's also part of the key thing is that you're not just acquiring yet. Yes, you're acquiring for, hey, they're learning on the technology, but what you really, what you really want to do is acquire, um, people who have the same vision as you do and want to go take on on a bigger leadership role as part of a larger company.

I think we've covered quite a lot.

What do you think? I think so too. What would be most helpful? Yeah, hopefully we got to everything you wanted.

Yeah, sure, sure. Then let's move towards the end of the episode. If we zoom out a bit, because many founders and product leaders listening to this episode are trying to navigate a similar transformation that you're actually navigating now with Amplitude. What advice would you give to founders or product leaders, tech leaders, sorry, trying to make the same transition with their company from a, let's say, traditional SaaS to an AI-first company?

It's very hard. I know it is. And you won't be able to say everything, but I think talent is key. So you want to identify people who have embraced the technology and hire them or, you know, give them, empower them to lead on things. Um, and then the change needs to come from the tops down. So Wade, uh, did a phenomenal job at Amplitude leading the AI week and setting that up and really getting people bought into the technology. Um, he, as part of AI week, he got a lot of the engineering and product leaders and design leaders to, uh, start using the technology in the weeks before and then demo their usage during the AI week so that people could see it and get motivated. Innovated themselves. Yeah, I mean, my biggest thing, yeah, like, you need to get some talent that knows it, the technology, and then you need to, and you can do that both through hiring and acquisition, and then you need to go all in as a team saying, hey, we're going to go do this. And that's hard because, like, there's all sorts, you got a roadmap, there's stuff you got to do, you know, there's all sorts of distractions, but there is no substitute for doing that work and getting the team trained and embracing it. Thanks a lot for your time, Spenser. No, of course, Tim. Always happy. I love the French tech community and I love how many entrepreneurs are here. I'm very excited working with Enzo and Fruccio and a bunch of other folks.

And yeah, very excited to be here. Yeah, can't wait to see what's next.

And yeah, just let us know when you're back in Paris. It'll be good. It'll be good, man. No, thank you for having having me on.

That's it for today. I hope you enjoyed this episode. If you did, the best way to support the show is to leave 5 stars on Apple Podcasts or Spotify. It helps more people discover these conversations and grow from them. See you next week for another episode full of actionable insights. Until then, take care.