Watching Users So You Don’t Have To: Human Behavior, YC, and AI

Automatic transcript of the recording (English); speaker labels are automatic. Timestamps open YouTube at that moment.

0:00Sacha: Welcome back to AI with Sach. Today, I'm with Amogh, builder and co-founder of Human Behavior. Amogh, you do a lot of different things. So I'll let you introduce yourself with my first question, which is, who's Amogh?

0:12Guest: Yeah, I'm Amogh. I just dropped out of Stanford. I'm doing YC. Yeah, that's about it. I've just been working full time.

0:19Sacha: That's awesome. Let's start from the top. What is human behavior? And what is the core problem that you guys are solving? So,

0:28Guest: Human Behavior, we're building an AI that watches your session replays. They're basically videos of actual customers using a product to help you understand why they're staying, why they're leaving, why they're paying. Just like what they're doing on your product itself. That's kind of the general gist, like the why behind analytics. What

0:42Sacha: was the kind of like, aha moment that led you to this idea, right? Did it come from a personal pain point or was it something that you guys observed in the market? But

0:53Guest: the way we thought of this idea was like kind of funny. We got into YC and our partner, like during the interview, I think I'll get into this later. It was just like, our partners did not like the idea at all. Like I was celebrating my friend after getting to YC. We're just like getting boba. And so I was there for like three hours. I was just like tweaking. I was like, we have no idea. Like, what are we going to build? So we were throwing ideas for like three hours on the board and like randomly, like he brought up session replays. And he's like, I'd pay you like a bunch of money if you could build an AI that like analyzes for me. So that's kind of where we left the conversation. We didn't like have an idea. Texted all my startup friends from Stanford was like, hey, what are your thoughts on an AI that analyzes your session replays? And that was like the first idea that we had where everyone was just like, yes, yes, yes. So I was like, okay, maybe I should like look into this a bit more. We spent like a week just like cold emailing every single person we know. Got as much validation and conviction on this idea that we could. And then we're like, okay, let's just go with this. Can

1:42Sacha: you go like a little bit deeper? What do you mean by session replays? Just so that it's very clear to everybody. So

1:48Guest: a session replay, the way it works is that they're like recordings of customers using your product. So the way we do this is like a standard library or a web. And we recreate the DOM elements, the X, Y positions of like the cursor of like the user movements and recreate the video. So then we have, it's like basically a video for customers using products. Like that's at the core. Yeah.

2:06Sacha: How did you figure out that there was room for something like this in this space? I

2:10Guest: think there was always like, this was going to be like eventually made. We just decided to do this right now. The way it currently worked was just like manual instrumentation. Eventually this was like, everyone kind of already knew that this is like kind of what we have to do. It's just a matter of like Gemini wasn't good enough. Vision models just weren't there yet. I think the why now was just like really compelling is like models just got good enough to do this. Let's try it. And yeah, it was kind of, it was kind of it. Yeah. That's

2:37Sacha: pretty cool. And you mentioned that you guys got into YC. So was what I understand is this was not the first idea, but did the team pivot from something else or how did it come about?

2:48Guest: We pivoted a lot. So we were basically working together as just like, not even like co-founders, we're just like friends, like just hacking together, like side projects. We made a bunch of stuff, like test my hinge, like random projects just for fun. And we decided to apply to YC on just like a whim. Like we kind of always knew that we wanted to do like an official startup. Yeah. Like, do you want to know that, like the ideas that we applied with or? Absolutely. Yeah. So we applied to YC with like an AI accounting thing. And like, what do I know about accounting? It just seemed like something like, oh, it's like the old space, like QuickBooks sucks. So we were just like, my co-founder was actually like a cracked accountant. So we're like, okay, product like founder fit, accounting sucks. Turns out that was not a great idea. Like even though he was an accountant, first of all, accounting is just not like, I'm not passionate about it. But then we applied to YC a second time with like, like a neobank for e-commerce or whatever. Again, what do I know about fintech? We just were trying to find ideas that we never had any personal connection to. It was just like, these are like old spaces. This is what we can build. And so both times YC was just like, these ideas suck.

3:49I think now that I'm like doing the, like the bash right now, I kind of understand why they hated those ideas. But like, we were just in like the mindset of just like trying to find an idea instead of an idea organically coming to us. While we were building these things up, we installed like post hoc, all these session replay softwares. So at that point, when like I was brainstorming for this idea with my friend, it was just like, I knew that this was a pain point. This is something that came organically. It wasn't us trying to look into like construction websites and like trying to build like, I don't know, like AI construction, like a system or whatever. Like this was something that like came organically. Yeah.

4:19Sacha: That's pretty cool. Walk me through the earliest version of human behavior. What did like V1 look like if you have a V1 already? I

4:27Guest: mean, this is the earliest version. I mean, the way it works is the core tech is just, we have an AI like Gemini. Watches your session replays, creates tags, creates events. And based on that, we can do a bunch of cool stuff, right? Which is like funnel analysis, see where people are dropping off from one point to the other. We can use natural language to query up sessions. So like some people already have hypotheses about why people are dropping off. So you can just like search up like, okay, show me videos where my customer did X, Y, and Z. See those videos, see those problems. Another thing is just like tagging like 404 pages where things crashes. So our AI, like once it starts tagging those things, we can create issue tickets. And that's kind of like, that is the V1 right now. Like from there, hopefully we build like a better product.

5:03Sacha: What is like technically the hardest part of building a system that watches user sessions with AI? I

5:09Guest: think the scale, like a lot of like architectural decisions and like infrastructure that we have to build out. To like handle like millions of sessions coming in per like some of these customers have like a million sessions a day. Especially these like e-commerce sites that have a lot of traffic. That's been really hard. Gemini hallucinates quite a bit. I'd like to give a shout out to another YC company in our batch, BlackSail. They kind of sat down and wrote a bunch of like our architecture and gave us a bunch of advice on that. So yeah, me personally, building an app to like that level of skill, I've never done that. So it's been really, really hard getting that right. I

5:41Sacha: can only imagine. How did you guys approach that problem differently than, you know, the traditional even based analytics, if you will? So

5:48Guest: traditional analytics, the way they're doing like, for example, like post-hub, mix panel, especially mix panel, right? You have to set up like, like instrumentation yourself, have like event listeners, trackers. So a lot of time when your app changes, you have to like basically re-engineer your like analytics, which is quite annoying. Our whole approach was just like, we want to basically have a human. Normally people watch your session replays, right? We just want to get rid of the human part and just have an AI do it. So I don't know if that answered your question, but. No,

6:15Sacha: absolutely. It does. How did you, or how are you guys validating early on that this is actually useful to product teams, right? Any insights that already surprised you from people you've talked to?

6:27Guest: Wait, actually, yeah. The way we validated was just beg for calls. Just like, please talk to us. Please talk to us. I think even before we decided to build even a fraction of the product, we talked to over a hundred people easily who were not just like random people. They're like a hundred people at like product teams at like fortune 500 companies, unicorns. Like quote emailing, begging, like, please respond to me. Like follow up, follow up, follow up. It was really annoying. I feel bad for those people, but we just needed to get those conversations in. And I think at first we came in with like a very like rudimentary idea what the product would be, which is an AI that watches the session replays. What do we do with that? Right? So these product managers are like, okay, I spent two hours a day creating these graphs. If you can make those for me, that'd be helpful. I spent a few hours a day looking at individual user journeys. If you can cluster them for me, that'd be like very helpful. So a hundred conversations, we sat down and we made like, we just have like a huge whiteboard and still there. After every call, I used to go on the whiteboard and just write like four or five bullet points of what we learned. Four or five bullet points and then tally them based on like what other people were saying. So like if I sat on five calls and all five of them said the same thing, that thing was just like tallied up.

7:27We know that this is like the one thing we need to work on. So that's kind of what we did. And

7:31Sacha: I know that YC is very, very strong about talking to users before you even start building, right? And this is what I'm hearing. And you said it's called outreach. Can you go a little bit deeper into that? And how did you find the right users and ensure that you're talking to the right people?

7:48Guest: So if you want to get like customers, like the best way to do that is just like friend of a friend of a friend. You just got to hustle that. You just want to do like mass outreach. Apollo has just been great. Like we spent so much money on Apollo. Just like the premium plan. Just email like 500 people a day. I have like four inboxes and like every day it just sends out 500 emails. Yeah.

8:08Sacha: Makes sense. You've mentioned on, I think it was on your website, startups like Dove and Conduit as users. How did you land your first customers?

8:18Guest: Like a friend of a friend. Honestly, like even like right now, I think like a lot of inbound isn't like targeted. There's like a very specific ICP that we have in mind, like certain size that we want to hit. And like huge unicorns or like huge like Fortune 500 companies aren't reaching out to us. We still have to go to them. The best way to do this. And also like we haven't like gotten like insane customers or anything. We're like barely into the batch. But the customers that we have gotten so far right now are just, hey, do you know this guy? Go to that guy. Sit on a call. He connects it as his boss, his boss. As long as it's just a chain. That's how we've gotten like someone for like bigger customers. But again, like if you, we haven't gotten like insane customers or anything so far. What

8:52Sacha: does onboarding like your first or a new customer look like to you guys? And how long until they start getting value?

8:59Guest: Onboarding takes like five seconds. Just connect your postdoc ID right now. Install our SDK. Like actually it takes five seconds. Like and then how long it takes to get value. I mean, it takes us a little bit of a while right now just to start ingesting the data. So once we do get the like start ingesting data, like probably like two to three days, we get like your first like we sent out like an email. We're like, hey, this is what we analyzed. I think that's kind of how we're trying to prove that we get value. I don't know if it works, but I mean, people have like been pretty perceptive to like, okay, this is interesting. This is some value that we immediately gave within two to three days of them like paying and onboarding. But onboarding itself takes five minutes. And then everything, I think the thing that we're betting on is just like we have a Slack channel. We're up 24 seven. So if they have any questions, we try to answer them. Do they have any like analytics, like stuff that we could do for them? We're sitting down and like working side by side with them as basically consultants right now. And then also they have access to the platform. So like, it's been like, that's how we've been like able to prove our value. Just like 24 seven availability and then like an email within two to three days.

9:53Sacha: That's awesome. So you guys are almost co-creating with the users is what I'm hearing. Have you found that any specific, any specific types of product or teams or human behavior works especially well or not at all yet?

10:07Guest: Yeah. I mean, I think it should work pretty well for like almost all websites. I'm going to always going to say that I'm going to plug that in. The ones that, okay. So I'll just be very transparent. The ones that it doesn't really work well with is just like one of our customers are like not customers. It connected to the postdoc ID and had a website was just like a single page, like not even like sub tabs. It's just like a single landing page. I feel like at that point there isn't that much information. Like AI is only as good as the information you feed it. They had like a bunch of customers, like people like viewing their page. All we could do was just like figure out to the depth they scrolled and whether they clicked the button that led to the Stripe link, which also wasn't connected on their website, so we can't even tell whether they checked out or not. So at that point it was just like, it was a single website. There's nothing much we could have done, but yeah, at that point it's not that helpful.

10:49Sacha: How do you, because I'm sure there is some privacy, you know, side to this. How do you handle the privacy and compliance side of this? And given you're, you're watching your user sessions, right?

11:00Guest: I mean, technically we're not watching any users. Like we're recreating the dumb elements, but I mean, you mentioned Delve. We already have software compliance, follow industry best standards, right? Like we have like, like if you read like our like terms of service or whatever, it was like, we had like lawyers and stuff like data retention policies and all that like fancy stuff, how we're storing data. Like there's a bunch of different stuff that we have to do, but yeah, I mean, it's like we've gotten like, we went through all the checks and follow best practices. Yeah.

11:23Sacha: Awesome. Let's talk a little bit about YC. Why did you apply personally and what was it like to interview for YC?

11:31Guest: I mean, I applied to YC just because like the YC is YC. Like there's no point in not applying. Everyone applies. Like it's great. I'm a first time founder, like drop shot at college and stuff. So I don't have like huge, like industry backing or whatever. Right. Like YC puts you in that like Silicon Valley circle. So like obviously makes sense to apply to YC. How was the interview? It was stressful. I'm not even gonna lie. That shit was stressful. It was like the way this interview is structured. Do you want to get into that or? Yeah, absolutely. So the way the interview was structured is like it's 10 minutes and they don't ask you anything like insanely difficult. They want you to succeed. They only ask the basics. Like what are you building? What are the competitors? Like genuinely the bare basics for you, like for them to know that like, you know, your shit, right? They want to like, you can't be an idiot. But the thing is the way they ask it's like 10 minutes. If you're rambling too long, they will like cut you off and be like, okay, next question. Next question. I think the environment they create is really stressful. The questions they ask are actually like looking back, like very easy. Like what are you building? Like if you can't answer that question in two sentences, like, like it's, it's going to be a problem.

12:32But yeah, it's stressful for sure.

12:34Sacha: Very interesting. What has been the most valuable part of YC for your team so far?

12:40Guest: Okay. The advice is great and all, but I think the most important thing is the community. I think building a startup is like quite lonely. It's just me and my co-founders most of the days. I think YC, the greatest asset is like the community, the network. Like, for example, I mentioned the Blackstone guys were stuck with a lot of like infrastructure stuff that we didn't get. Another YC company came in there, like they didn't even charge us. They're like, yeah, we'll just like help you cut this out. Right. Yeah. It's just been like the community. And also it's like not lonely. They're events. Like you get to meet people. Other YC companies help you out a lot. YC deals have been quite helpful. Like you just get like free AWS credits, like free open AI credits. Like, yeah, I'm just going to say no to that.

13:14Sacha: How is, because I know you're, you're a first time, first time founder. Sorry. How has been in YC change maybe your approach to, you know, product or company building, if at all?

13:25Guest: Wait, it's actually YC. That's like, they've been huge with that. Like, I think YC has like changed me as a person, honestly. I'm not even joking. So YC has like this thing. Like, I think they, a lot of people underappreciate this, but they focus on the fundamentals. Like that's all YC, right? This sounds so simple in retrospect, but build something that people want, talk to customers, focus on product. These seem like very obvious things, but people don't actually like implement it. Like our, even we sometimes just like lose track. Our YC partners like build something people want. Why are you building out this feature? Right? Why are you getting stuck in this? I think it's like running lean experiments is something huge that I learned in YC, which is like, we used to waste so much time trying to like build out one small feature. Instead have build it, give it out to the customer. Do they even like it? Because running lean experiments, build something what people want. What else? Yeah. Because when I was at like Stanford, when you look at like startups and you talk about them, a lot of times, like what we talk about is like launch, how much you're raising, like branding, all like the sexy stuff. But in reality, like no one really cares. Like if you're not solving a real problem or whatever, like it doesn't matter how cool your website looks or like whatever, like it doesn't matter.

14:27It's just, yeah.

14:28Sacha: You're talking about lean experiments. For those who don't know what that is, could you, could you explain a little bit?

14:33Guest: Yeah. It's just like shipping really quickly, like talking to customers. They say build something up instead of building something that's perfected. I also, I could be wrong about this, but this is my understanding. Instead of building something that's perfect, build something that kind of works, give it out to your customer. Do they even like the feature? Is this something they want? Iterate quickly with them. I think that's kind of what I mean by like, that's like my understanding of Linux. Love that.

14:53Sacha: Did you guys raise money before YC, during, or like, is it going to be after? And what is your fundraising strategy and what worked for you guys?

15:02Guest: Honestly, right now, I'm probably like the best person to ask about it because we haven't started fundraising. Like, actually, we did raise a little bit for YC and got like an angel check during YC from like someone. But the way we raised before was just like, I ran a hacker house my freshman year summer with just like me and my Stanford friends. The guy who funded the house has just like been a real one since day one. He like introduced me to all my co-founders. I'd been meaning to do a startup. He was like our first believer. He gave us conviction. I think at that point it made sense. Like I didn't know anything about equity or startup. And he was like another Stanford alum, like runs a legit fund. I was like, you know, I'll take money. We'll see how this goes. The entire fund has been very supportive. They do a lot of customer intros. Yeah. I mean, like I'm happy to give whatever equity of them and like helped us find our initial team. First believer, been supportive since then. But yeah, we did take money before. You

15:48Sacha: mentioned he helped you found your co-founder. How do you see that people, do you think that people usually struggle to find a co-founder? And if so, how would you advise them to go out and find one?

16:00Guest: Again, I'm not the best person to ask about this. I'm like pretty introverted. I don't meet that many people. I think I was very fortunate to meet my co-founders. They're like the best people in the world. Genuinely, like whatever the positive superlative you can think of, like my co-founder is even better than that. I think the co-founder dynamics are like the most important thing. Like I think you'll see like some statistics. Like most startups fail not because like the idea sucks or the product sucks or whatever. It's just co-founder disputes. And that's the main reason. I got really lucky. I don't even like, I can't be more grateful for my co-founders, but I just got lucky. I don't know how you make, this is just asking how you make friends. I don't know how to make friends. I just got lucky.

16:37Sacha: That's awesome. Zooming out. What's the long-term vision for human behavior and where does this go beyond just session replays?

16:46Guest: Okay. So session replays is there. It's a, it's quite rich data, right? The way you can go about this, by the way, it's like helping train browser use agents. By the way, we're not training on anyone's data, but like if you want to like training up thinking about this on the spot right now, like our whole thing has just been like focus on like the immediate value you can provide eventually. Okay. I'm going to take a, can I take quite like slight tangent here? Of course. So the slight tangent is like, and this is something that YC partners also tell you is just like take small incremental steps and eventually you're like company will take on like a very unique shape that you can't even think of. Like Google started off as a search engine. Now they're building like pixel. Now they're like doing everything. Amazon started as a bookstore. Now they do AWS. Now they do everything. So you don't even know the shape your company will take. Focus on like the next like three months, next month. How can you like slowly, slowly, slowly provide more and more value? However, if I'm just brainstorming ideas right now on what we can do with session replays, like AI IT would be huge, right? Like you have data on how people get stuck. Some guys rage clicking, like a cursor pops up and it'll be like, okay, hey, I think you got stuck here. Like, are you trying to find like this page or whatever?

17:47Like that could be cool. AI quality assurance, like recreating errors or whatever. That could be huge. Like training browser use agents. What else can we do? It's just, there's so much data. We could do like so much with this, but like, I'd have to like think about this more, but this is like off the top, like what we could do. I think. What about the gaming industry? I've been, that's gaming industry is like interesting. The thing is, I don't think we're good enough to handle that type of data or like give any meaningful insights there. Cause there's like the way Gemini works is like, we're taking like snapshots, like capturing individual frames and like video games. There's so much happening. So that would be like a quite an interesting engineering challenge. Hopefully we like, we will be able to do this down the line, but not right now.

18:23Sacha: What's a product decision or bet that you've made that felt risky, but it's paying off right now? I

18:29Guest: don't know. I think everything right now we're doing is like quite new. I think there isn't like any single bet that we've taken that's been like insanely risky. And

18:36Sacha: as we're slowly closing off here, if someone is building a startup right now or today, what would you tell them about how to understand their users better or how to even just get started from an idea to execution? I

18:50Guest: mean, if you want to understand your users, I mean, I'm going to plug human behavior. I just use us. How do you go from like idea to execution? Oh wait. Okay. So this is something that like, okay, first of all, I should not be giving any advice on like how to build a startup. Cause like I haven't built like a huge one myself, but like based on the advice from people who know more than me, it's like, you should not build until you've talked to a bunch of people until you have like already like validation through like contracts or money that people want to buy this thing. And yeah, I think that's like the biggest thing. So you shouldn't even start building until you have like insane validation. And then when you do start building, you should build in parallel with your like early adopters, keep iterating, keep getting feedback. But again, anything I say in this call, like I'm just going to put like a little asterisk. I haven't built like a huge successful startup. So like, I don't know what I'm talking about.

19:32Sacha: What does validation look like?

19:35Guest: Like for me, it's like overwhelming. Yes, yes, yes. We want this. We want this. Like, even if your product sucks, one thing we noticed is like, if you're solving like a hair on fire problem, people are willing to forget about all the bugs, like your onboarding, like stuff just fails, whatever. As long as they're like, okay, this is like a hair on fire problem. I have like a million replays. I can't get through all of them. They're solving a problem for me. I'm willing to do pay as like, okay, not pay as much as possible, but like I'm willing to pay for this, even if the product isn't finished. I think that's like something that's been quite validating for us. I'm actually bullish on my products, but yeah. Very, very

20:08Sacha: cool. Amog, I appreciate all of your insights and your time today. I want to give you the floor if you have anything to share that maybe you didn't have the chance to. So just the floor is yours.

20:18Guest: Yeah. I mean, if you're a first time finder, apply to YC. I think it changed my life. And if you're like an actual company who's watching this, check out Human Behavior. We'd love new customers. Yeah. Thank you so much, Amog. All right. Nice meeting you.