High-Speed Ad Traders Profit by Arbitraging Your Eyeballs(bloomberg.com) |
High-Speed Ad Traders Profit by Arbitraging Your Eyeballs(bloomberg.com) |
I am an engineer who spent 12 years in the ad tech industry. This problem is especially acute in the online video ad side where dollars are exchanged based on views and CPMs, not someone buying something online (this being more accountable is less open to abuse). In my old job, we tested what % of traffic are fraudulent. These traffic are daisy-chained from one ad buyer/seller to the next. Inevitably it will hit someone unscrupulous. In aggregate we found anywhere from 20% to 95% of the traffic we see for video ads to be fraudulent.
There are some very sophisticated bot farms out there that gets around detection, mostly operated out in Eastern Europe and Asia. If you look at Comscore 100 video sites, you can always tell who's gaming the system when from one month to next, an unknown brand just jumped high in the top 100.
This is the reason Facebook had shut down Liverail that they spent $450M on. Super high percentage of ad fraud.
In my mind there's only two things that will really change this...better viewability/inventory auditing and attribution technology.
The larger, more reputable inventory sources (Google, FB, etc.) have a VERY vested interest in pushing quality and using whatever data they can bring to the table to build that trust, and then use that trust as a moat against their competition. Viewability providers likewise have an easy sell once they can improve the accuracy and reliability of their offerings (they have a long ways to go based on general sentiment in the industry). This will put pressure on other pubs to up their game or be left out. Why bother with a crappy news site that has 50%+ bot views that you're being billed for if FB or Google can prove that their traffic quality is much higher, better targeted, and infinitely higher reach?
On the buy side, the pressure will come in the form of attribution. Right now, display performance, video in particular, is a super murky area in terms of measuring success. What is the value of a view-through? Most advertisers couldn't tell you. And the ones that can probably have a very fuzzy picture of it that varies based on the attribution model they are using. However the space has improved dramatically, and at the end of the day, performance marketers will be able to do a much better job of telling whether traffic is crap or not based on whether it backs out. Bots don't buy things (although they do sometimes sign up for lead forms now). So if someone with a solid analytics stack sees a ton of impressions and no sales through their various attribution lenses, guess what? They'll stop buying those impressions. And the more big buyers that get smart about attribution and stop buying based on impression counts, the more pressure will be exerted on the sell-side to clean up their game.
So while the problem is "self-correcting" in the sense that there is big money with very vested interests in solving for this, unfortunately it is a big ship to turn and will take time.
Articles that talk crap about the ad industry rarely get into the nuances which is horribly infuriating and does a lot to give the industry a bad reputation. Like any space, there are bad players. Legit players in this industry want the bots and garbage out of the picture yesterday. They make our jobs harder, and reduce our performance.
Everyone else who's peddling non-logged impressions with BlueKai data, good luck - first, Bluekai is only accruage 1/3 of the time. Second, majority of those are bot views.
Attribution like you describe is notoriously hard to measure. As views get to mobile, FB wins again since most apps use FB auth.
The similarities lie in the real-time nature of trading systems in both sectors.
The world is a curious place, full of contradiction and wonder. We only like the free market for some activities.
As we should... the free market is not a cure-all, it's good for some things, poor for others.
People buy up all the houses, "meh"
For example, you say "If reselling tickets was impossible and the price remained the same, then people who didn't manage to get one would be out of luck; now, they can pay more and still get one.”
But if there is a fixed supply of tickets, then there are a fixed number of people getting tickets. All you are changing is which people get the tickets.
Yes, people with more money and less time to stand in line or whatever will get a ticket. Yay! But people with less money can no longer stand in line and get a ticket. Boo?
No
There are two reasons why people can't get a ticket:
- Fans bought all the tickets, limited places so the only way you could get them is if someone gives up. Nothing wrong in reselling a ticket for the price you paid for them or less
- Scalpers bought all the tickets. This is the main reason why today you can't get a ticket. Because they get on the queue and buy all of them before the fans had a chance to even get it.
So the problem of "you can't get a ticket" is not solved by scalpers, it is caused by them
You could play with https://contributor.google.com this works though Google's exchange and the gist is they bid on your behalf for ads that will get shown to you, you can then put arbitrary HTML in those ads. I saw one guy that was learning Japanese so he used it to display vocabulary words in place of ads. The major downside is it only works on ads that go to Google's exchange which is kinda random on most sites because of stuff like the article above. It ends up being a couple dollars a month to replace random ads on websites.
Traders buying and reselling at a higher rate “could be distorting the markets and removing the efficiency that we’re supposed to see through real-time bidding,” he said.
By definition successful arbitrage makes the market more efficient - it brings prices closer together (making the cheaper exchange more expensive and the more expensive one cheaper).
Furthermore, arbitragers in ad exchanges are causing more ads to be sold, providing a valuable service to buyers and sellers.
Suppose there is a sell order on exchange X, a buy order on exchange Y, and these orders are compatible. Unlike public equities markets (which have RegNMS) this order may NOT be routed from X to Y. The result is inventory is wasted or put to a lower value use.
If an arbitrageur notices this he can cause the transaction to occur which would not otherwise occur.
Agreed. As header bidding and arbitrage proliferate, we also see ad tech companies targeting the supply side of the equation: with services that help publishers understand how well the header bidding product is performing, which ads work and which dont, and how viewers engage with their site. As pubs are getting more information into the ad selection process, losing exchanges will have to adjust the quality or lower the prices of their ad supply.
All of the exchanges (on the buy and sell side) know it's going on, and they generally don't try to stop it provided that it doesn't become news. A large number of well-known, venture-backed ad tech companies make money from this and they're not incentivized to stop it.
It's rare that the companies actively encourage arbitrage, but some do.
1. Get access to a DSP (either AppNexus, AOL, Doubleclick itself or a smaller DSP--there are literally dozens). DSP stands for demand-side platform, but in this case we're going to use them as a supply source.
2. At the same time, get access to a demand source, either AOL, Google, SpotXchange or someone similar. Someone that a reputable website publisher would use to fill ad spaces on their website.
Both steps one and two can be difficult to obtain, as every major player is on the lookout for fraudsters and arbitrageurs and doesn't more crappy, re-sold demand/supply on their platforms.
3. Place demand-source tags (AOL, Google, SpotXchange) in your DSP, so as soon as you buy an impression, you sell at almost the exact same time.
You make a profit when you amount your from demand-source tags (net costs + rev share) is higher than the cost of the ads you're buying (net costs + fees).
Even though the article is from 2014, there a still ton of people still doing this and making money, though the real money is running botnets and buying botnet traffic (which I know how to do but have never done).
All Advertising is arbitrage, exploiting a market inefficiency between the cost to reach an eyeball and the value of that eyeball.
Whats the difference between a high speed trader and a mortgage broker advertising? Whats the difference betweem a retailer advertising to sell a product they bought wholesale and a high frequency trader reselling an impression in real time?
Im not saying we are not all better served by closing this information gap and finding market efficiency, but lets not pretend that this is any more unscrupulous than any other business that exploits information asymmetry...and its definitely not the same as high frequency traders...especially because these ad buyer may never have gotten the impression if the arbitrager didnt bring it to a different exchange or support it with a different data set.
I don't understand how this is possible. The original advertiser has to trust that the first exchange is reporting accurate demographic info. How does the exchange know the demographic info is accurate if anyone can buy the ads and sell them on another exchange?
I can't feel too much sorry for the ones paying more in this case.
> People buy up all the houses, “meh"
1. It's unlikely that the price is set at $X, and there are precisely enough tickets to satisfy everyone who wants one at $X, but no more. Either the tickets won't sell out at $X, or there will be those who can't get tickets at $X.
2. If the original sellers knew how much demand there was, either they'd set the price to be as close to the balance point as possible, or not. If they did, then there should be no opportunity for scalpers unless the price is very unsticky upwards; but if there are a lot of people willing to pay well above the original price, then it's again unlikely that demand is equal to supply.
3. If the original sellers didn't try to set the market value, then there would have been a shortage anyway. Again, the fact that scalpers can extract a large amount from each resold ticket means there's enough demand at high prices, and therefore likely to be too much demand at low prices.
2 and 3 might be solved if (first) ticket sellers would set auctions for ticket prices, not sure if it's possible legally though
Is that about right? What information does the demand source tag include - is that basically a placeholder indicating you bought space for an ad on a website? I assume the arbitrageur's edge comes from finding traffic that can be bought cheaply from the demand source and sold higher on the DSP?
This is generally something that only online ad professionals can pull off, but if you have enough of a bankroll, you can try it as well. It's a very saturated space and the amount of knowledge that goes into it is much more than what I'm able to include in a forum such as this. Good luck.
To be precise, you're allocating the tickets to those who value them the most, under the objective criteria of "who's willing to pay the most for them" (which may not be fair, but in the same way that capitalism favors those capable of getting money).
>But people with less money can no longer stand in line and get a ticket. Boo?
Those people value their tickets less than the people who actually get them. If you're only willing to pay X for a ticket, and someone else will pay Y>X, then the situation where you get a ticket for X and they don't isn't Pareto optimal; you'd prefer to sell your ticket for Y, and they'd prefer to buy your ticket for Y. (This does assume something about utility of money, which isn't strictly justified. To wit, I'm assuming that an unwillingness to pay more than X<Y for something implies a willingness to sell it for Y after you've gotten it. For X and Y sufficiently far apart, this should be true, but it might not be if they're close. On the other hand, if they're close then the harm here is very low.)
Taken too seriously, the result of ignoring this issue is absurdities like rich people caring more about anything they happen to spend money on than poor people care about eating.
Generally society's answer is to have the government give necessities to the poor, and otherwise let the rich buy what they want.
Edit: if you really want to untangle the money abstraction, look at it like this: person A has $X. They have $X because ultimately, other people gave them $X because A provided value to them worth at least $X. If they had a job, they got wages; if they sold a product, they got the sale; if it was A's ancestor's money, A's ancestor provided >$X value to someone or someones. (This is an abstraction, and obviously not always true; but we try to outlaw ways of making money that don't provide value to anyone, like stealing, or colluding with competitors. To the extent someone earned money illegitimately, that's a problem with how they got the money, not how they used it; the solution is to take away that money or prevent them from getting it, not prevent money from being used altogether. Also ignoring rents, interest rate/inflation, etc.)
So A's desire to pay $X to get a ticket is actually cashing in on society's "debt" to them of $X or the equivalent in happiness/utility. That's what needs to be compared with someone else who's only willing to pay some fraction of $X; the value they provided to society, that they're willing to give in exchange for the ticket, is less.
Again, obviously the abstraction is imperfect, but that's the ultimate justification for capitalism.
> which may not be fair
I don't think this is true. Without scalpers, a sold-out concert ends up with basically every seat filled. Since services like StubHub have gotten big, it is now more common for there to be many empty seats because those were snatched up by scalpers and never got bought.
The scalpers are often acting rationally here by holding out for high prices until the end instead of dumping their tickets because they make more overall that way even if some inventory goes unsold.
What you suggest is certainly one way to measure "better" objectively, but that doesn't magically make it the best way.
On a large scale, you can see this in the field of health care. Some people argue as you do, and this leads to a fully free market for medicine. And the principle is that the most good will be done, because the people who value their health the most will pay the most for their services.
But whether we agree or disagree with the principle, surely we must acknowledge that there are entire countries full of people who have a different belief about the greatest good with respect to health care: In Ontario, for example, it is usually illegal for a doctor to charge more than a set amount for a certain type of procedure, and thus the people able to pay more are prevented by law from doing so.
That doesn't make Ontario, Canada "right" and Ontario, California "wrong," but clearly the idea that "it is better if the people willing to pay the most get the most" is not a universally held belief.
But if you're going to criticize capitalism, maybe you could offer an alternative that does better?
Like evolution, capitalism exists so we need to describe it and understand it. It's definitely a force to be reckoned with.
It's not a system for approximating justice. It doesn't result in people getting what they deserve, even approximately.
The problem isn't only the cheating but also the randomness, which basically gives outsized rewards for being lucky enough to be in the right place at the right time. Anyone watching the tech industry knows this.
If the people who end up with the money choose to spend it to pursue justice, sometimes the world becomes a better place. But nothing in economic theory guarantees it will happen. It's up to people to do it.
There are several different possible justifications for such a law:
1. We think the amount people can pay for some services should be capped, someone willing to pay more doesn't mean they value it more 2. We think that people are uninformed about the prices of services, and/or are extorted by time pressure, and/or the medical profession is extracting rents
If it's 2, then they can still agree that people should be able to pay what they want to, but there's an informational asymmetry because you don't have time to shop around when there's an emergency, or there's a monopoly that needs price controls to break, etc. None of those invalidate my point.
Does any of Ontario's rules apply to discretionary medical services?
Do you have any other example that applies without falling into one of the categories above?
As a society, we value the idea of everyone having access to health care, and if we make medicine a free market, the only people who get it will be those willing to pay the most.
Thus, we specifically reject the argument that providing health care to fewer people who pay more results in a greater benefit.
The reasons for our belief vary from completely subjective moral arguments to arguments that society overall benefits from having a healthy citizenry, just as society overall benefits from having a healthy infrastructure.
I don’t need to debate the specific reasons with you, I’m just pointing out that it is neither #1 nor #2. We certainly think there are people willing to pay more, and that those willing to pay more are correctly valuing it form themselves. We don’t believe that those unwilling to pay more don’t understand the value of medicine.
In the end, not everyone agrees with your perspective on value. Many do. I understand there’s this gigantic country south of us who believe medicine should be a free market, and that many of its citizens think its health care system is providing better value than ours.
But I insist that your views, while valid from your perspective, do not represent a universal truth.
Um what? If you artificially reduce the price, you reduce the supply, and thus decrease the number of people who get it.
My model of free market here is not supposed to affect the total supply, just the limited supply is divided based on money. If price controls increase supply, then there was a market failure of some sort, which probably counts under reason 2.