ATS London 2022 – Building Next Gen Ad Tech Through Business Transformation: Dr Boris Mouzykantskii

Look, Kieran, thanks for inviting me and sort of embarrassing me a bit. Yeah, it was by far the most difficult times of my life. I mean, especially the last 112 days. So it's tough. Look, in terms of the transformation and everything, we learned a lot during the pandemic. And what we thought was a nuisance and something which slows us down kind of helped us a lot when we hit the next challenge. The key idea was always: when the communication is sort of degraded as it was in pandemic, you need to trust people and empower people. You basically need to move your decision downwards, move decision to the people who are closer to the real tasks in real life. And we started doing a lot of that through the pandemic, and then we used that template, used that operational capabilities, if you like, to really supercharge them and accelerate them and do the relocation things we were doing in the past three months.

Look, I mean, we were fortunate in a weird way. We never really had clients in Russia, we never had service in Russia, so we didn't really run a business operation per se. We had a team of extremely talented people there. So when this war started, we said, look, there are really two priorities for us: we need to keep servicing our clients and we need to look after our people. So initial ideas, which all were wrong, was that the biggest immediate risk is some sort of internet partitioning, and you know, we wouldn't be able to get those people to manage the service which were all in America and Germany and stuff like that. So we kind of went together and we said, okay, we need to have a small group of people out to a place which is less likely to be affected by internet issues, and then we need much better internet connections. And then we kind of leveraged the expertise we got from the pandemic. So instead of sort of obsessing at a central authority level when solving those problems, you basically form small task forces and you tell them, here is your task, just go do it. You give them money, give them decision power, you just get them to report every week about how they're doing, or sometimes every day. But generally, they have a well-defined task that just needs to execute that now. We got like 25 people out, not just random people, people who actually could manage our service if something, if their internet partitions itself, in Dubai in like seven days. And we didn't need them, so they kind of had a decent time. It was much, I mean, you're having your departure thing in March, right? Good timing, good weather. So they kind of... the internet didn't crumble. What did crumble instead is the world in a way. So it became very clear a couple of weeks into this that we can't really service American, European, Japanese clients out of Russia. The world just stopped being connected. So that thing hits us. My internet was fine, internet was working, the world didn't. So we said, okay, fine, we don't really have a choice. All our clients are outside. Most of our people are strongly anti-war, they don't really want to stay, at least long term they don't see themselves living in Russia long term. So we need to get them out. And that was when we ended up... well, the template was similar: a small team totally dedicated, empowered, just go figure out how to relocate people quickly. And yeah, that's when we learned a lot about the hospitality industry. It's a horrible business. I mean, it's not rocket science, it's not like machine learning, there is no math. But there is like... you would hate that, don't you? Well, yeah, there's no algorithm to optimize. I think initially we were somewhat, I should say, two kind of cocky. We said, okay, you know, it's what can possibly be very difficult? The difficult part is that there are millions of tiny things which you need to get right, and if you don't get them right, you devastate people. Like, I don't know, hamsters don't travel on planes. Did you know that? Sorry, what? Hamsters? I mean, hamsters and planes, not a lot of planes. Yeah, that's a devastating set of events for a couple of families.

Look, we were always delivering, or at least aspiring to deliver, sort of engineering excellence. And in a way, ten years ago, five years ago, it was more about building bespoke custom systems and going after problem after problem by just building a custom solution to address that. Now, these days, the industry gets so complicated and so kind of interconnected and commoditized that even if you have a ridiculously relevant and important piece of technology which you want to implement, it wouldn't actually work in isolation. You need connections to, I don't know, quality inventory, quality management systems, you need connections to anti-fraud systems, you need connections to data providers, and 25 other different things. So none of them are technically challenging, and most of those capabilities are treated as being standard. So the net effect is that yeah, there is still a ton of really interesting unsolved problems and you need brilliant engineers to solve them, but instead of doing it bespoke one at a time, you really need to put more of a product around to cater for problems which, if you like, are already solved by the industry. So you need all that available, and then on top of that you can innovate and you can deliver engineering excellence.

Look, I wouldn't describe it as compression. I would say that this path between supply and demand is just getting messier. So The Trade Desk cuts in and tries to work directly with pubs. They still work with SSPs. Those pubs are still letting other SSPs connect to them to provide demand. Look at a very high level, if you have a big pool of inventory, and it needs to be really big, your best possible outcome for a publisher is that you don't have either SSPs or DSPs. You just get the advertisers to come, or agencies more realistically, to use your systems directly. And Facebook got away with it, now YouTube got away with it, probably up to maybe who knows, Snapchat works like that. So if you're really big, that's what you really want. And I guess that's part of the consolidation drive between different publishing groups now. That's your supply side aspiration. But the demand side, or more specifically agencies, they don't really like that because if you think about 50 sort of huge walled gardens, that means a poor agency needs to work with 50 different interfaces to actually move the budget around. It's very difficult to automate, there are all sorts of training challenges. So they're kind of pushing back. And then another piece which also happens is that those smaller gardens are never perfectly balanced. You have your own sales team, that sales team brings money, and sometimes they sell stuff which you don't really have. So you kind of want a little bit of sneak peek of some of the adjacent walled gardens so that you can actually drive their inventory up. In the same way, you get too much supply, you can't sell everything. So you've got those imbalances. So net net, you probably have a situation when those 50 odd semi-walled gardens are kind of interconnected by some sort of systems which are trying to balance supply and demand, and then provide all sorts of yield services, probably some data services, and all stuff of that. So I wouldn't call it compression, I would call it just a more complicated set up compared to what we got used to.

Well, our industry keeps delivering many systems when one can potentially do so. I don't believe that will be one system. A bunch of systems, and I think many of the existing SSPs and probably some of the DSPs will start playing more of that role, the role of balancing supply and demand across multiple semi-walled gardens. So in some sense, the SSPs will pick up some of the demand side capabilities, and the DSPs will pick up some of the supply side capabilities. Because at the end of the day, those semi-walled gardens are 60% closed networks, or maybe 80% closed networks. It's the remaining 20% which needs to be balanced. So those systems are not like network ad servers, they actually have a bit of open web built into them just because of this intrinsic imbalance between supply and demand on that smaller scale.

Look, I mean, the customization happens all the time. It happens around us. Sometimes it's difficult to see. For example, and actually part of the reason it's difficult to see is that a lot of it happens inside biggest platforms. So we have a bit of a sneak peek in what Google is doing from these court papers, and they've done a hell of a lot of customization for their biggest client, which happens to be Facebook. And I bet they're doing something, maybe not as profound and hopefully not as potentially law-breaking, but something similarly important for some of their other bigger clients. And it kind of makes sense, right? The industry gets more complicated. In order to win, you need to have something special, and you need to get that something special somewhere. Now, back 10 years ago, you could have just gone and built your own system which would have something special. Now that gets very, very expensive. So the bet on MediaGrid is: look, we'll provide the commodity, and then we'll try to democratize that custom building ability. So we can build pieces of more custom stuff for the important clients on top of the structure which basically would provide a decent commoditized set of services which you expect from other SSPs and curation providers.

Look, I don't have a recipe. I don't have a good answer. At a very fundamental level, there was a pact between kind of technology companies launching themselves on the internet and their consumers and users. And that pact is like 30 years old. That's basically: we will provide you a lot of value for free, and you give us some of your attention which we will gently monetize without actually disrupting you through the ads. And that actually powered that huge ecosystem for many, many years. I think that pact is sort of crumbling in more than one way. Everybody is upset. The users feel that they are being spied on. The technology companies feel that the politicians have stopped them from doing valuable business. There is also argument that because of that nature of the pact, it actually corrupts the type of value which is delivered. So instead of data and facts, you deliver opinions because you get more attention from opinions, and then instead of opinions, you deliver conspiracy theories because you get even more attention. So the economic structure which drives the web actually forces the tech companies to deliver anti-value, so to speak. And then on the very other side, which is even more fascinating, the companies who used to be pure, no ads, just payers will build content, they seem to be discovering that the cost of creating quality content like Netflix is actually too high. You need a bit of ads, and because people are refusing to pay for that cost. So I don't know, that whole thing is probably changing. And that's probably the most fascinating thing which happens in and around ad tech. And that better thing, so I feel that every time something changes in a profound way, the more independent smaller companies which basically are the bare bones of the open web, they get a chance. They react faster, they see what happens faster.

I mean, it wouldn't solve all of them. Some of the problems are deeply entrenched in keeping that same pact going. What I'm saying is the pact is changing. Probably even the problems we're foreseeing are not quite the problems which need to be solved. But it takes Google's of this world 20 years from launch to dominating. So if we are changing the rules, or if we're changing the basic underlying principle behind the internet, I feel the big companies will be on the back foot just because they're big, not because they're bad or anything, just because they're big. So you have a next wave of problems, and much smaller companies will actually have a lead, if you like. And that basically happens everywhere every 30 years.

Look, I have a set of weird answers. For me, it was always an excuse to do interesting data science and algorithmic research. And from that angle, things around supply path optimization are ridiculously complicated. It was complicated before because there were just many paths and the basic assumption was they're all roughly the same, just the fees are different. Now, different parties on those paths, different SSPs, they inject different types of data, so they're not actually different anymore. And if anything, there are even more of them. There is combinatorial complexity there. How on earth you find the best one? That's again, it's probably not super critical for the business, it's just personally a very fascinating example of a problem: huge data set, difficult science, and some economic value if you get it right. And there are probably two or three like that just emerging. Like another one which I know you're discussing a lot is: okay, you used to buy fairly efficiently because you knew what was the effect of your particular decision. You could literally trace, you showed somebody an ad, you kind of traced the chain of events after that, and you know that there was a conversion, there was some buy. Okay, terrific. Now, you lose all that. So you have sort of cohort-based reporting which is vague and unclear, but you still need to maintain a decently effective buy. So how on earth do you do it? You kind of throw away 20 years of machine learning research which is supervised learning, you start working on something totally different. How not to do that? Fascinating problem, a lot of money. Whoever solves it first, I hope it's not Google. But that's the sort of thing.

Thank you very much.