Cloud Native

Q&A: Eventbrite’s Pat Poels on Building a Rapidly-Scalable Events-Driven Architecture

27 Apr 2018 10:11am, by

Pat Poels winning large at the 2005 World Series of Poker (Flipchip / LasVegasVegas.com)

Summer is almost here (at least for the northern hemisphere), and with the warm weather comes the festivals and outdoor events. In other words, it’s crunch-time for the event management and ticketing website Eventbrite. Or one of the many crunch times for the global company, which has hosted event info and ticketing for gatherings across 180 countries.

Eventbrite’s Senior Vice President of Platform, Pat Poels has been with the company for a little over six years, leading the software engineering team, and the management of its platform, which not only supports EventBrite’s website and mobile apps but also natively supports many third-party applications as well through an API.

Prior to EventBrite, Poels spent 15 years at Ticketmaster, helping that company move through the many phases of digital transformation. In between the two gigs, Poels has played tournament poker for a living, where he learned a lot about a lot of lessons about strategy and logic. So we had to start our Q&A there.

So you’ve both built large IT platforms and won major poker tournaments. Was there a math connection there, or what was the appeal? 

Yeah, there was. Definitely, math has always been a strong suit of mine, but I also loved puzzles from when I was very young. It was something that I did a lot of to pass the time, jigsaw puzzles, logic puzzles, whatever.

Well, there certainly are a lot of puzzles to think about when it comes to the scalable computing, that’s for sure.

Yeah, 100 percent. It’s not probably a big surprise those pursuits are actually pretty closely connected.

Yeah, so over the course of time here at Eventbrite, I think we attacked the industry in a bit different way than what I had experienced in my past at Ticketmaster. And I think that we’ve had to spend a good amount of time getting to a place where we can scale what we do in a very big way, which is a part of the reason why I moved over to this platform role. But, over the course of time, having a company that’s been focused on self-service, and a company that services just the biggest number of organizers with the largest inventory of free and paid experiences in the world, those are areas where scalability is really a big challenge.

That’s a real big interest for us. But, before that, perhaps maybe you could sketch out at a very high level, the platform for Eventbrite?

Yeah. We leverage Amazon Web Services for the majority of the things that we do. Like a lot of people, we have different tools for different applications, so most of what we’re doing is AWS-based.

A big part of what we’ve been doing is trying to understand how to satisfy a really large market with events that are both small and large, events that are self-service or events that are coming from organizers, and demand that can happen really in a planned fashion, where you know it’s going to be a festival where you’re going to sell 50,000 tickets over the course of a few minutes, or maybe something that someone’s put on the platform that actually isn’t even a real event. And you have to make sure that you’re able to tell the difference between those things, satisfy the stuff that has to happen at scale and make it really simple for the people who want to use it in a self-service fashion, and also find the things that aren’t real on the platform and block those things out.

The path that we’ve taken has really required us to be very focused on building in modular fashion, building in a more service fashion and leaning into APIs and service architecture. I think our first publicly available APIs we launched in 2009, so it’s been something we’ve been leaning into for a very long time.

We know Eventbrite mostly through the local user groups. They’ll set up an Eventbrite page. That’s a great benefit to local communities. But you guys are also doing commercial events as well.

Yeah, absolutely. We’re satisfying those events for much more of a self-service, smaller organizer who does it because they’ve got a group that they need to try and get together and it’s not necessarily their occupation to sell tickets to things. They need that as a part of what they do, versus all the way up to the enterprise level company that’s actually making it their wherewithal. Their business is selling tickets to events that they put on all the time. The platform and our products need to be able to satisfy all those different kinds of organizers.

Last year, we’ve processed more than three million tickets per week. If you think about all the different inventories, this platform has more inventory than any other, for free and paid experiences, than any other in the world.

The modularity part comes in in that we have people who are building a lot of things inside the building and we have those developers and engineers who are building as partners outside. And we want to make sure that the things that we build allow us to be able to facilitate both.

You’ll put tickets on sale for something where there’s 50,000 tickets for a festival or something, and you’re going to sell those tickets out in a matter of minutes. And you won’t have 50,000 requests, you’ll have 5 million requests or 10 million or 50 million and it happens very, very spiky.

We have a service that we built internally that allows us to help facilitate selling tickets. It’s at the core of how we monetize is selling tickets. And we use that service to allow us to sell tickets on our website and we use that same service to allow us to sell tickets via our mobile products. We’ve got a mobile product for our attendees that allows them to self-service buy tickets, and then we have a product for our organizers that allows them to sell tickets for people who walk up at the door. And the same technology, the same service, is used in both cases. But, externally, we’ve started leveraging this service to allow us to have native purchase experiences on distribution partnerships. One of the biggest ones is Facebook, but we do it with Bandsintown as well and we’ve got other things that we’re building out there.

And what that allows us to do is it allows us to get those great experiences in front of consumers wherever they are.

What does your platform encompass exactly?

I think that platform’s such an overloaded word, that it’s difficult sometimes to understand exactly what’s meant by that. From my standpoint, now being in this position of being SVP of platform here, what I think of is the services that drive the products that we build and the services that drive the products that other people build externally. I think of it as being our internal APIs that we extend to our own developers and to our external developers. It’s also the design system that we put together that allows people to build products with our platform.

What is the development pipeline for getting new components or new functionality onto the platform?

If we’re talking about the workflow for making updates to the platform, we have continuous integration that we do within the organization to try and help update all those different services. We definitely have that distributed in a way where it allows us to be really nimble in how we distribute these things.

And, obviously, when you have APIs and you have a system for interacting with a platform, you have to make sure you’re doing that in a very smart way because you make changes and you potentially break things that are in production, not just your own things, but everywhere else. Doing those things in a coordinated fashion is really important to us, but it’s also critically important that we’re able to do it quickly.

Now when we’re talking scalability, I imagine Eventbrite, it’s a large operation, it’s going to continually be demanding of AWS resources. Is it very spiky or is the traffic more regular, but just high-volume?

 It is incredibly spiky. When an event goes on sale that has limited inventory and a lot of people want to go to that event, you have incredibly spiky nature of the public wanting this thing that’s a very scarce resource and it all hits at the same time. But there’s a constant flow of transactions that are happening because we’re a global operation. You have a flow to what the day looks like in the U.S., but Australia has theirs that looks a certain way for a certain timeframe, but it’s offset based on their time of day.

You’ll put tickets on sale for something where there are 50,000 tickets for a festival or something, and you’re going to sell those tickets out in a matter of minutes. And you won’t have 50,000 requests, you’ll have 5 million requests or 10 million or 50 million and it happens very, very spiky. And there are lots of companies who have to deal with this sort of spiky nature of the kind of requests that they get, but the reason why it’s so much more challenging for us is that it has to happen in a very, very compressed timeframe.

Do you routinely overly provision then or is there some other secret to always being ready?

I think it has to do with smart provisioning. It has to do with looking at the traffic patterns you have and making sure that you’re ready for the things that may show up. But it also is really important that we cache really smartly and that we do things that allow us to make sure that we’re never in a situation where the platform falls down completely.

Are there any plans to open source any components that you might have built in-house?

 Yes, absolutely. I think that we, just like everybody does, we take advantage of some great tools that we use from an open-source standpoint and we also contribute back to the open-source community. And I think it’s something that, when engineers are your customers and you realize how much it helps them to be able to get their hands on great technology that they don’t have to build themselves. We do open source a portion of what we do and I’m sure that that’s going to be something we continue to invest in over the course of time.

How would you say your mix is between open-source technologies? Are you still using commercial software at all or is it pretty much a fully open-source stack or where are you on that spectrum?

I would say we’re more towards using open-source tools than not. Yeah. We write so much of our code ourselves and so much of what we’re using and building and have built are things that we’ve written ourselves, but we also definitely take advantage of open-source tools wherever we can.

What technologies do you see there on the horizon that you’re maybe kicking the tires on? Are you looking at AWS Lambda now or what sort of things potentially might have interest for you?

Yeah, we’re definitely looking at Lambda. I think that our data teams, in particular, are spending a lot of time looking at that as a potential path for us.

The place where we’re doing something that I think is really pretty interesting is RFID. Because a big part of our operation is getting people in and out of the building, that’s the place that we started was extending and expanding that RFID capability to make it really easy for event holders to check people into events and easy for consumers to be able to just use a bracelet with an RFID chip in it to get in and out of events.

But I think the things that we’re doing past that that are really interesting have to do with cashless. I think there’s a lot of opportunities for us to leverage that RFID technology in better ways. We’ve done it with a few festivals to test this out already.

So if you get drunk at a festival, you don’t have to worry about losing your wallet. As long as you have a little RFID wristband, you can still get more beer.

I’m so glad you said that and not me.

We’ve been doing a lot on AI research and AI model development, that sort of thing, but how do you get AI into a production system? 

One of the big areas for us is fraud detection. We want to be able to determine whether an event is real and we want to be able to determine whether a transaction is fraudulent or not.

And the best way for us to do that, or one of the most effective ways to do that in real time, is to have a lot of information about what real transactions look like and a lot of information about what fraudulent transactions look like. And when you have the kind of depth and breadth of inventory and transactions that we have on our platform, you have a wealth of data to look at and to analyze and to put models against so that you can determine, okay, with a very high degree of certainty what are the characteristics that are going to point you towards a transaction that’s not real or that’s fraudulent.

And so we have been investing down this path really for as long as I’ve been in the company, for a good six years. We’ve had dedicated resources on the engineering side and on the data science side getting ourselves better and better at machine learning around our transactional data and around our event data to help make sure that the transactions we process are real and that the events we put on the platform are real.

And when I talked earlier about a modular platform, this is one of those pieces where we’ve invested a lot internally. And the modularity of what we do, we believe that we’re going to be able to bring that to bear for external developers as well and be able to share that knowledge that we’ve acquired. Not just inventory management, which we’ve spent a lot of time at and gotten very good at, but also fraud detection as well.


A digest of the week’s most important stories & analyses.

View / Add Comments

Please stay on topic and be respectful of others. Review our Terms of Use.