Keynote Presentation: - Boris Mouzykantskii, Founder, Chief Scientist & CEO, IPONWEB

Thanks a lot for having me, Kieran. Thanks for the very warm words about us. A bit about us: we built various custom media trading platforms for many years for many clients, and have an interesting inside look from technology outwards on what is happening in the industry, how it works, and how it sometimes doesn't work. So the topic we talked to our clients more often than we wanted is that the clients want many things at the same time. Very often we come back and say, you can't have everything—what's your real goal? A campaign very typically wants all of this and more. You work very hard, you deliver a thousand conversions, and they say, 'Yeah, terrific, but you know, click rate wasn't good enough.' And you say, 'Okay, fine, here is the click rate.' And they say, 'Not sure the quality of the traffic was good'—which is a polite way of saying that maybe some of those clicks were not real ones. And at the end of the day, when you deliver everything, they say, 'Why are your services so expensive?'—which is a polite way of saying computational efficiency wasn't there. So for a while we thought it's all about clients just being human beings, wanting everything, not terribly rational. But as we spend more time, we realize there is some underlying reason for those conflicts, and there are actually technology challenges in how those can be addressed.

One thing you realize is that on the demand side, on every single campaign or at least most campaigns, there are at least three parties involved: there is an advertiser, there is agency, and there is DSP. There could be more parties, but those three are almost always present. Out of those things I mentioned, different parties want different things. Advertiser wants quality, wants some KPIs, clicks, conversions. Agency is very sensitive about margin on their media buy. And DSP is measuring computational efficiency. To simplify: advertiser wants best impression, agency wants cheapest, and DSP wants as many bits per server as they can get away with.

Now the reason for this is not so much technology; the reason is because the underlying business model behind all this is somewhat broken. Advertiser wants best impression, and that makes total sense. Still, they go and contract agency to deliver impressions at three dollars per thousand. They don't really say, 'Agency, please give me only the best.' They say, 'Any impression, no matter what, as long as the pass is targeting, it's three dollars.' So at that moment, at least immediate interest for the agencies to get as cheap impressions as possible, because the difference—at least some agencies, at least in some cases—the difference between three dollars CPM and whatever they buy media at goes as a margin for the agency. Now agency goes to DSP and says, 'Look, I'm paying you 10 percent of media cost,' which again is misaligned. The underlying immediate reason to get cheapest possible media buy. The DSP doesn't really—the cheaper the media cost, the less money DSP gets. So the DSP doesn't struggle, doesn't put too much resources into making it as cheap as possible, they just do a decent job.

In the middle of that, there is a technology owner—could be DSP, could be somebody else—who needs to balance all those competing and contradictory requirements. So at the end of the day, if you build the technology, you need to buy an impression, and this impression will have some cost and might or might not generate a click. So you play sheriff and try to balance those interests of different parties which are all talking to each other during media trading. These days, this job of balancing their interests is really difficult, and it's not so much a technology problem; the problem is that the parties sort of don't trust each other too much. And that needs to stop. The technology can only do that much; you need some sort of policing, some rule of law, some way of making it—technology speak for this is limited transparency. So this transparency needs to be better. After all that is fixed—and I believe it will be fixed one way or another—but after everything is transparent, everything is clear, the technological problem is still there because at the end of the day, most of today's algorithms can go and deliver on one goal. This could be multiple different goals, you can maximize margin, you can maximize number of clicks, but it's this single goal, and the algorithm just goes and does it. However, in this future fully transparent world, we can see that what's required is not a single goal but a balance. Today you can get away with single-minded algorithms because you just do whatever you feel is right and you get away with it because there's no transparency. But once the transparency is there, it's important that you explain why you've done this, what else was possible, and so on. I'm trying to explain how that future thing actually works.

By the way, just to emphasize the fact that those goals are complicated, competing, and difficult, let's rather than talking in general terms take a specific example of an agency and look at how a very transparent agency might formulate their goals. So they say: I must spend the budget, and then I must hit the advertising KPIs. If I can do both, then actually I can take as much money as I can from the media buy—which is goal number three—and show their margin. That's relatively transparent; I can see agency coming and defending it in front of an advertiser as a fair set of objectives. How should it play together? Now the first thing you must understand is what is possible—not what you achieved as a result of running this campaign, but what was possible before you even started. So let's take a specific example: we spend a single day trading media in Berlin on AdX. All possible outcomes of this trading are captured with this reachable zone, this thing in blue. For example, I could have just spent the entire day doing absolutely nothing, in which case I'll get zero media cost, zero impressions—I'll be here. That's definitely possible. Alternative: I can buy at 10 CPM and buy everything which was available on this day in Berlin, which will be this point on the reachable zone. For this particular day, 96 million impressions—quite a lot for Berlin—you would have spent that much money and generate about 60,000 clicks.

A natural DSP strategy will be: okay, I go in and get cheapest 30 million impressions for this day in Berlin. That's this point. I need to spend about $25,000 and we'll be paying about 83 cents CPM. Now the interesting thing about the reachable zone is that it's a convex bounded body—at least it's convex on AdX because they run a true second-price auction. If it's not true second price, then this needs to be modified. The other important thing for multiple goals is that it's actually multi-dimensional. Here I just drew a projection onto media cost and impressions, but if you care about clicks, that will be three-dimensional. If you care about clicks and unique users, that's a four-dimensional body, of which I'm showing a two-dimensional projection. To emphasize the point, let me rotate this body around the media cost axis. Hopefully it might even work. What you see is that now I'm basically turning it and I'm seeing clicks. So that's the same reachable zone but I'm looking at a different view. On that view, my 'buy all' strategy is still up here with 60,000 clicks. I also realize now that my cheapest 30 million impressions actually generate about 13,000 clicks. For that money, I could have bought way more clicks—I could have been here click-wise—but then it would mean I get less impressions. This shape can be properly measured without spending $274,000. You run a very small sample campaign, you get all this, and then you understand what's possible.

Now how do you use it? What you can do now is formulate two things: restrictions and prioritized goals. Restrictions are hard restrictions; the idea is that you must absolutely stop the campaign unless those restrictions are met. Very typically it's a budget restriction—there is no way on earth this campaign can stop with more than 30 million impressions. Often it's CPM restrictions: on average the CPM should be less than three dollars. Sometimes there's margin restrictions and sometimes more. Then you define not a single goal but a sequence of prioritized goals. So in the example we just discussed, first goal is impressions—get 30 million impressions. If you can't get 30 million impressions, don't bother with this, don't bother with that, just spend all your resources to get this goal. If this goal can be reached, then you're allowed to consider this one, and this one will be hitting some KPIs on the advertiser side. Go do it, and only after that is reached can you go and start further tinkering with the margin. The combination of restrictions and the sequence of goals we call strategy cards. We discovered that even relatively small agencies generate dozens of these, and they have hilarious discussions between their sales team and operations team about which exact cards to assign to which campaign. For example: do you want to spend the budget even without meeting advertiser KPIs? Yes or no? Terrific discussion. That's how it should happen—they need to make campaign-by-campaign choices. Nothing to do with technology, everything to do with running a transparent business.

How does it look with the reachability zone? I said two restrictions: less than 40 million impressions—actually 48 million impressions—and cheaper than 2.85 CPM on average. That means out of this big cigar shape, which was all that was possible in Berlin, I'm now only allowing myself to consider this piece because I want less than 48 million and cheaper than 2.85 media CPM. Now what about the sequence of goals? My second goal was to reach a particular CTR, and my last goal was to minimize media cost. So you cut this shape with a horizontal plane, kill everything above because you don't want it, and focus on this area where 48 million impressions could be delivered. Everywhere on this plot you get exactly 48 million impressions. This part is not allowed because of the CPM restriction. So remember that CPM restriction which looks like a line on one view, in the projection it looks like a vertical line and cuts out this. Everywhere in this blue shape the first goal is reached. What about the second? Because you can reach the first, you can proceed with the second. The second goal says CTR should be this point, 0.62 percent. That's this line. Everywhere between here and here your second goal is reached. You have an entire line to yourself. Now the agency is allowed to go and proceed with their third goal, which is to get onto this at the cheapest possible media, which ends up being 65.8...

So that's how it looks overall. The agency hit the restriction on media cost, hit the restriction on CTR, managed to deliver all this at 65.8, which in this particular example means taking 62 percent margin. It's transparent; we explain why it's 62 and not some other number, and generated about that many clicks. That's the sort of technology which enables you to run a proper, fully transparent media buy, explaining to all parties involved why you've done this and why it's fair. To sum it up: conflicts right now are being gamed, and you just get away with what makes sense for you and ignore some other parties. This needs to be fixed. The technology to find a fair balance is out there working, and I hope it will be used more widely. Thank you very much.