Using Real-Time Data to Power Billions of Conversations with John Kim, CEO of Sendbird

Using Real-Time Data to Power Billions of Conversations with John Kim, CEO of Sendbird

Hi there, and welcome to the Canvas podcast, where we bring business and data leaders together to talk about how to make data easier for everyone. Today, I am super excited to have one of our investors, and CEO of Sendbird, John Kim on the show.

John, tell us about yourself!

I am the founder and CEO of Sendbird. We are the world’s number one conversations platform for mobile applications. We power user-to-user communication as well as brand-to-user communication within your mobile app.

For instance, we have customers like DoorDash, who uses us for communication between the delivery person and the customer. We have a lot of healthcare customers connecting doctors to patient, like Walgreens. Also a lot of dating apps. One of the world’s largest online communities, Reddit uses us for their community chat as well. A bunch of games and live streaming sports use us for user engagement plus conversion for some commerce businesses.

We power over a quarter billion users on a monthly basis sending 7 billion-plus messages. We raised $200 million in funding and are valued at slightly over a billion dollars. We employ about 340 employees globally in seven different countries.

This is my second startup. My first company was way back in late 2007, the perfect time to start a company with the financial crisis waiting around the corner. This was a great experience for the second time around with Sendbird. For the first couple of years, it was a social gaming company, it grew to about 5 million users, and got acquired in 2012.

A side fact about me is I used to be a professional gamer. I’m the undefeated champion in the Unreal Tournament. I played a lot of Quake, Duke Nukem 3D, etc. So that’s a little bit about me.

What drew you to technology and data specifically?

Technology goes way back. I have a fake memory about being a really well-rounded child that played on the playground and did all these fun things but my mom recently found a diary of mine from my childhood and every single page was about computers, using PC tools to edit new games, hex editing, using GW-BASIC, and using programming language to create very basic elementary games. So pretty much all my life, I guess, I’ve been in love with technology and science. I think my dream was actually to become a scientist one day.

After working at a couple of technology companies kind of saw this potential to build tools that increased productivity across humanity. I wanted to do that for the rest of my life. So that’s kind of when I pivoted from trying to become a scientist to becoming more of an engineer. I decided that I should figure out how to build tools that I can actually really see the world get feedback and iterate, and really feel that meaning and purpose from it.

I went into university as an electrical engineering student, but kind of pivoted to computer science because wanted to build a lot of stuff myself as an engineer and programmer. The experience that kind of really drew me into data and productivity, working for a company called NCSoft which is one of the largest gaming companies in the world. Previous to that I worked as an engineer and software engineer at a company that shut down but raised $10 million dollars. I was the 100th employee in. That was around the .com, Bubble days. When they started to crash and they couldn’t fundraise anymore they went from 100 employees to 10, I was the last engineer. I inherited the work of 15 engineers.

With the next company, I wanted to try something else. I actually went to the other sites working for a business unit at NCSoft and quickly realized a lot of people were using doing manual work like using word processors and spreadsheets to crunch the data. They use macros in Excel spreadsheets to increase and build kind of less hacky productivity tools. Because I had a background in engineering, I started building and observing what people did to build an internal operating tool and that was actually not part of my job description, but I kind of did it as a side gig in the evening, after work, normal working days was over. I started spinning up my own Linux servers on my local desktop. Then started programming these operating tools that I wanted to use myself. I quickly realized how much of that can actually benefit the rest of the team, my manager found that I was like, “Hey, what is the tool that you’re using?” I’m like, “Well, it’s kind of a tool that I want to use for myself.” And he said, “Well if it’s good, can our team use it for our team?” Our team was like, 160 people large. We did a rollout within our own team, and still, sort of status, like the actual officially became the official tool within the company. So that really gave me a positive experience, I learned that if I build something and get feedback and iterate on it, there’s a lot of value that I can create. A lot of that has to do with, handling data, and running automated reporting, increasing productivity for the organization. So that’s kind of how I guess I've got it as entrepreneurs' journey of building software and then turning it into something that’s useful.

What led you to start Sendbird?

I would say, it’s not a fun story. I have had a lot of experiences with B2C companies, my first startup was a B2C. Then when the financial crisis hit, all of a sudden, nobody wants to invest in B2C companies or Web 2.0 companies anymore. But because I have a background in gaming work for a gaming company, and am a software engineer, our investors, were like, “Hey, if you build games, we’ll invest.” I’m like, “Okay, now I have to make games!” because I didn’t want to fail. So we pivoted to games. But instead of just normal games, I built social gaming.

Then with the second company after selling my first company, I started this company with the buddies from my first startup. So we’ve been working together for 12 or 13 years now. We started out building this social network for moms basically find other moms in your area with similar-aged kids, to buy and sell used baby products, do Q and A’s, set up playdates, and things like that. It was around 2015 when the entire world was talking about messaging apps and conversational UI chatbots. Remember everybody started talking about those things. And so WeChat took over the world, Telegram, WhatsApp, everyone’s talking about messaging apps. We wanted to add a missing feature to our own application. But we had built chat four times in our gaming company before. So our CTO said, “I’m not gonna do this same thing over again. Why reinvent the wheel?” So we went out on a buyer's journey, first, we tried all the free stuff like open source, etc. A lot of open source for optimized for the desktop experience, like the JavaScript plugins that you find on random websites. It wasn’t really mobile optimized. So then we went on to use Firebase, we built our version 0.8 on top of Firebase, it was kind of easy to build a POC and prototype. But once we got to the really sophisticated user experience, there’s a lot of workarounds, hacky things layered on top. So we ended up ripping that out again, and then built this entire thing from the ground up. Then we were kind of running out of money. But I saw this notification from YC. It said to apply to YC. Now, it was like, a week left? Oh, no, actually, it was a day left. I was helping out another friend to apply to YC. So I thought maybe I should apply too. So we applied to YC with that idea. And then our partners turned out to be Michael Seibel and Justin Kahn, obviously, as co-founders of Twitch, they know a thing or two about chat. So they had a lot of questions about chat. We had scaled to about a million users back then. We solved this. Here’s how. So that’s how we kind of got into YC. That’s 6 and a half years, here we are with a quarter billion users.

Did you have to record your videos back then and submitted the same day for the last-minute YC application?

We actually applied to YC with the mom’s social network app, once we got to interview but we didn’t get in. But the second time I was actually trying to help out my friend to apply to YC. I basically a copy pasted my mom's app application, but just changed the business part and the team and everything else was the same. So we got an invite to the interview again, but this time with just a different idea. My friend who I was helping actually did not get in and we got in. So because of this kind of awkward moment, I offered to buy beers. But, yeah, so we were pretty lucky there.

What are some of the unique data challenges that you’ve run into over the years at Sendbird? How did you solve them?

There are so many hurdles that we have to jump through. You’re absolutely right. I mean, the Data Challenge is real and also, we are dealing with real-time socket connections. Scaling a concurrent connection is nontrivial compared to scaling, you know, HTTP, stateless connections. So there’s that. And then the way the data flows in, if you think about the moderation capabilities, everything has to be real-time. Then there’s a ton of web posts related to the messages, for every single message that gets sent. There’s so much metadata, right deal receipts, typing indicators. Plus, if you’re sending a message, there’s a pre-event webhooks and post-event webhooks that trigger certain amount of data, then there’s the audit logs, a lot of analytics, that get piped to other systems. We never had an issue of lacking data, we have an abundance of data. So we’re dealing with a lot of different infrastructure. Trying to be super transparent, we kind of stopped stumbled into the Sendbird idea, right?

Our main business was consumer application, a social network. Sendbird was kind of like a side gig almost within the same entities and members and co founders. So internally, we had a much more sophisticated data, structure, database architecture, all those things. For the consumer business Sendbird was more of a hacky product that we’ve kind of did on the hackathon positive test case, or selling a sideline. So we had to really rebuild our architecture multiple times throughout the past six and a half years. Now, I think we’re pretty good state, we have a data team that, you know, uses BigQuery, you know, Fivetran all those things. And then we use like Looker for the dashboard side of things. But initially, it started out as purely a database and what we call the insight is an internal tool was like operating tools, a little bit of data.

And then obviously, there’s always that business guy asking the engineer, “Hey, can you like do a database query and pull this data out?”l, and you have to keep harassing them, basically, to give us a data or wait like 24 hours, and then copy paste it into spreadsheets, of course, spreadsheets, runs out of memory, stuff like that. So there’s a whole bunch of messy stuff that we had to go through. Now, obviously, at our scale, that does not work. So we have a data warehouse project, that instead of running, a lot of them, obviously are connected to different parts of the system wouldn’t be a Salesforce, Looker. And, of course, sometimes there are some manual spreadsheets. So it’s still to this day is kind of messy. I mean, obviously, we have gotten way better at it compared to six years ago. But still, to this day, there’s a lot of manual work, a lot of kind of collaboration that needs to happen. Some missing data that we know is in the logs that we have to then pull it out and connect it to more of a sophisticated system like Looker, etc. So that took good years, and we’re still working through it. But it was never like a straight line. Because he kind of accidentally stumbled onto this timber as an idea. It just might actually be my first time doing B2B business. So what do I know?

When was the moment that you knew that Sendbird should be the thing that you focus on?

I think there are a couple of milestones or episodes that I remember right? One, for our mom’s application we had like zero revenue for two and a half years. We had users, right, we had like a quarter million users, but we had zero revenue. Now because we’re running out of money, we had to figure something out. My only way to convince our investors in our own team was telling them there’s something here and how do I show that is by having either sign an agreement, something to do in the pipeline, or actually have revenue, right? So I started pitching random pricing points because I’ve never done B2B before, I started pitching random pricing points to my friends where we also tried to build chat and we started generating revenue before we even had a product. Okay, if you give me this SDK by then I’ll pay you this amount. I’m like, “Holy cow, we never seen these dollars coming to our bank account for 2.5 years, now we’re starting to see dollars, or at least committed dollars before we even have a proper product!” So I came back, like talk to our teams, and said that we have to build this now. So that was kind of like, oh, this is like, kind of how customer development process by Steve Blank or you know, the agile startups are built. This is cool.

So we did that. Then once we got into YC, there are moments when you feel like there are prospects or customers that come inbound, that you clearly know that you do not deserve. Right. They’re always this customer that you never heard of, but you look on CrunchBase, they raise hundreds, hundreds of millions of dollars, wait a second, we raise like a million bucks, we clearly cannot take care of their traffic nor their use case, we don’t even have a proper office. And yet we jump on a call with them. They know that we are a tiny little startup, but they’re willing to take that bet with us because it’s so critical that they get this with a builder with a team that’s super dedicated to this. So once you there’s a constant, like a step function increase in terms of like deal sizes, but also traffic also, like when we signed with Reddit, I think Reddit was probably about 250 employees back then. We’re like 25. So and they literally helped us walk through what is needed to sell to large companies. They told us we have to do penetration tests, and you gotta have compliance. Then they were willing to sign a deal. And it’s almost like after the fact, like, within X number of days, you will do this and this. And this. I’m like, so like, normally, you’ve asked all those things to be done in advance of signing a deal. But they’re like, we’ll sign this. And there was like, almost like a question that came system. And they also paid us annual upfront, I’m like, This is amazing. You paid us upfront, nondiluted capital. So basically getting a seed check with no dilution, plus, they gave us a clear path to success. So that’s kind of when you start to realize, okay, there might be a clear need to, build a solution like this. So, I think we had a couple of instances.

How would you describe the data culture today at Sendbird? And how has this changed over the years?

Yeah, to be honest, we’re still trying to figure this out. I mean, yes data is always important and we have no shortage of data. But I think it’s getting the data into the hands of the right people who need them at the right time, right? You don’t want data that’s too old that you requested, like, two months ago, you need things in a timely manner. So just like if you think about, you know, some engineers are perfectly fine using JIRA. Right. But if you give JIRA to a lot of nontechnical folks, they say it is too messy. Can I just use like Trello or Asana? So there are tools that are more accessible, that come with compromises or some people may prefer Notion over Confluence for several reasons. So how we give more data access to kind of nontech folks is always a challenge. I think there is a portion that we’ve done well like you’re getting the data through, whether it be Looker or Salesforce or Keysight. So different infrastructure ingest different kinds of data. That is kind of usable to nontech folks. But still, to this day, we’re still building it and building data, I think we got our data warehouse up and running, probably the fifth year in. So it’s fairly recent. And again, because our growth can kind of come from organic and kind of accidentally, but if I could go back in time, I certainly would have started differently. Assuming that we had intentionally tried to build Sendbird from the beginning, because my previous background was in social gaming, right? Social gaming is all about data, understanding user metrics, you know, user acquisition, you know, potential monetization, the AR framework

Yeah, it’s all about data optimization and everything, but we didn’t have that with Sendbird in the beginning, now we kind of do. So I would probably put the foundation up-front so that we could save years and years of pain. Now today, we have a real-time database that kind of connects to our system the shows our archived right path, you have different ways to measure it, we have a weekly basis of looking at Pipeline, looking at OKRs, looking at different kinds of metrics, product matrix, uptime data, all those things, right. So it’s now all of the data we have can be processed in real-time, if not, at least on a weekly basis. But it took us not a trivial amount of effort, I was still trying to figure out how to make it more streamlined.

To learn more about John and Sendbird, you can follow him on Twitter here or visit the Sendbird site here.