Behavioral Ad Targeting Not Paying Off for Publishers, Study Suggests
wsj.com
wsj.com
But in one of the first empirical studies of the impacts of behaviorally targeted advertising on online publishers’ revenue, researchers at the University of Minnesota, University of California, Irvine, and Carnegie Mellon University suggest publishers only get about 4% more revenue for an ad impression that has a cookie enabled than for one that doesn’t. The study tracked millions of ad transactions at a large U.S. media company over the course of one week.
That modest gain for publishers stands in contrast to the vastly larger sums advertisers are willing to pay for behaviorally targeted ads. A 2009 study by Howard Beales, a professor at George Washington University School of Business and a former director of the Bureau of Consumer Protection at the Federal Trade Commission, found advertisers are willing to pay 2.68 times more for a behaviorally targeted ad than one that wasn’t.
Much of the premium likely is being eaten up by the so-called “ad tech tax,” the middlemen’s fees that eat up 60 cents of every dollar spent on programmatic ads, according to marketing intelligence firm Warc.
Having said that, 2009 is a long time ago, and I would like to see this study repeated before drawing conclusions about the current ad landscape.
Without the ability to retarget, advertising revenue at both Facebook and Google would plummet. The majority of conversions for advertisers come from retargeting (behavioral) campaigns. Advertisers run cold traffic campaigns (non behavioral) just to be able to cookie a group of users interested in their niche of product or service. They then show a different set of ads to those that clicked on the initial ad and met some behavioral criteria - for example they read to the bottom of the landing page or watched more than 1/2 of a sales video. It is this second set of ads that results in most conversions from new customers.
So the entire model falls apart if we lose the ability to run retargeting (behavioral) campaigns. Most advertisers lose money on cold traffic campaigns, and make up for it with behavioral campaigns. When behavioral advertising stops, so does non-behavioral, and the whole Facebook and Google house of cards comes crumbling down in a hurry.
Of course Mercedes wants to reach people shopping for a BMW and will pay substantially more for it. They are paying extra to reach a small subset of users.
No conflict here at all
Note that this is how targeted advertising currently works, I see ads for products I have already purchased, which is just wasted money.
See Coke (and what follows is overly simplified).
Everyone already knows who they are and their opinions on them. You might ask "Why does Coke advertise?" and it'd be a good question. Sriracha doesn't and they are a household name, surely Coke has more standing than them. Coke advertises not only to influence you from a young age (getting new generation of coke drinkers) but to ensure that you associate a good feeling with your purchase. This is apparent in car commercials too (especially in America). That BMW commercial isn't to convince people to buy a new BMW, kinda ridiculous when you think about it and the money that you have to spend on a car like that (consider how often you see commercials for high end cars that the general public can't afford). Luxury car makers sure do want you to feel superior about your decision to buy their car though (which there's connection with "my next car will be X so I'm superior).
Compare this to McDonalds, who knows you'll be hungry soon and wants you to visit them for your next meal. These companies have different goals, and of course their advertising will be performed in just as differing ways. Not all advertising is the same. Though I'm not sure this is the best strategy for Amazon when I buy a deck of cards, I don't need 20 more types. But maybe it is effective and that's why they keep doing it.
But also, these ads create a general impression of high quality for a brand, and for this to be effective, everyone needs to know about this. Status symbols can only be status symbols if they are recognized by both the people who can and cannot afford it.
Therefore ads of luxury items viewed by people who will never buy it is still valuable to the company because they set the stage for the status associated with the brand.
This is a second order effect of crucial importance that people often overlook when only looking at first order effects.
For standard playing cards, probably not, but for collectible card games, tarot cards, the blank cards used by game designers when building prototypes, card driven board games, etc. it might be significantly more effective. Those are all "cards" things where customers are likely to be repeat buyers or interested in trying other options.
They were laughably bad about distinguishing those sorts of different types/audiences in the past, and as in your example still aren't always good, but anecdotally seem to be gradually getting better. Maybe the gradual investment in development of machine learning and accumulation of data over time will result in less ridiculous suggestions in a decade or two.
The tracking isn't perfect and not every targeted ad you see is intentional.
Admittedly based purely on anecdotal evidence; this is something of an understatement.
Since you are likely already aware of Mercedes and have opted for BMW instead, I'd question the logic of Mercedes spending good money to continue advertising to you?
You see a lot of comparison ads at this level, trying to sway purchasers.
Are you sure about that? That would widen the tracking scope by a lot, and would surely make adsense/adwords completely GDPR-incompatible.
AdX is Google's ad exchange. Over there, you do see a lot of data. However, I would bet that the personally identifying data is mostly scrubbed in the EU. User agent and page URL are not identifying enough, so they're probably included in the data.
Sometimes campaigns are bought on a flat-rate basis for a certain amount of impressions but as of 2019, these are just treated as any other bid in the auction and only override the bid price where there's danger of not meeting the final quantity.
The issue for publishers here is lack of scale, no cookies, and 2nd-price auctions, meaning they don't get enough high bids and don't get the full price for those bids when they do happen.
I don't agree that it's value add since the advertiser pays the publisher directly, and the advertiser then slaps on all of the extra bells and whistles, so the impression isn't modified to be more valuable before it gets to the advertiser.
However, the comparison is not disingenuous. Even though I disagree with the parent poster, there is a clear perspective where the two are comparable.
> There is no "ad tech tax"
That is correct. You might get better revenues (as a publisher) dealing with SpotX (an exchange) over say...Viant (a reseller who acts as a publisher to SpotX), but probably won't do better on volume, which directly correlates to payout. Then again, SpotX may not pay you as much for your average demographic and volume as Viant, who wants your business more than SpotX. None of this is supply chain. The chain is not a chain, but more of an amalgam of situations that is affected by the market, as much as personal relationships. Conceptually, an ecosystem where anyone can consume from anyone (as a publisher) and feed from anyone (as an exchange).
> The issue for publishers here is lack of scale
You don't know that. In 2009, the economy was still recovering and ad-tech had been slapped hard in 2008, leaving banner ads at fractions of the previous years. They have slowly floated down a few more cents to where they are today, primarily due to better targeting, ad fraud, video, mobile browser support, and clickbait (er native) ads.
> no cookies
This was the comparison that led to the 4% revenue bump. If platforms do not pay forward to the publishers (because why bother, the platforms set their own cookies anyway), it suggests that publishers shouldn't bother with targeting. That's exactly what has happened. No platform consumes publisher targeting data in arbitrage, because it might be faked. Using something like Zvelo for context and your own cookies, you get a much better view of the client.
> 2nd-price auctions, meaning they don't get enough high bids and don't get the full price for those bids
Whatever "high bids" or "full price" means here, is nebulous. How an exchange decides to run auctions is often negotiated up front but is, ultimately, black box. What this has to do with publishers is confusing. Publishers rarely get paid off what the platform makes. You sign an order for a set period of time or slice of impressions unless you have traffic that is hundreds of thousands per day (an exeedingly small number of sites). In the case where you run a Real Time Bidding auction (header bidding), you might make some variable money. It's rare and wasn't even seen in the wild (ie IAB) till well after 2009.
See this more recent breakdown: https://www.emarketer.com/content/why-tech-firms-obtain-most...
Scale is the main issue because publishers need to have users with cookies, then match cookies to high-value bids, and then receive enough competing bids from advertisers to see a real difference in rates. This is fundamental market mechanics. Publishers getting flat-rate deals is rare, not common. Most of the time they get paid on variable programmatic basis with some take rate by the upstream SSP.
It's a supply chain because these vendors facilitate the buying of media inventory. That's the definition of a supply chain. You can build your own car after buying raw parts but that doesn't mean that there's no supply chain that lets you buy the finished product from Ford. The article you linked shows exactly where the money goes to in the chain. DSPs, DMPs, verification vendors are used to buy ads and they need to be paid. You can easily avoid it and deal with the publisher if you want, but advertisers don't because of the value they get from those vendors.
If the untargeted ads lead to less valuable placements, then it will hurt the revenue of the advertiser.
Short term, both the ad network and publisher can get away with bad placement and make the same amount of money. It will take longer for the advertiser to see the loss in conversion and reduce their spend accordingly.
How strange would it be for me to write the same article, wondering why elite athletes go through all the trouble they do to get an 'only 4% advantage'.
The reason "targeted" ads exist is not because they work better, but because they create more inventory that would otherwise not have been sold.
TLDR - they analyzed logs provided from an “ad exchange” in 2016 (as exchanges or SSPs were still “remnant” inventory players at the time and this is pre header-bidding, so lowest quality of the available inventory went there), of the logs they only looked at open auction (lowest quality of the already low quality inventory self-selected by using the SSP logs).
RTB Deal ID, private marketplaces and other higher-rate products all use behavioral/audience data and generate the lions share of the revenues. We are talking impressions being boosted to between 10x and 100x in value easily if they qualified for one of these higher-value channels.
By focusing on open auction that was preferred for valuable behaviorally targeted impressions, the papers authors basically rigged the results.
Last sentence should read s/preferred/prefiltered
My expectation is that if a big US news publisher tried this they would see much more than a 4% difference, but that's speculation on my part.
There is no shortage of companies getting megabytes of data points per person, yet they don't seem to be getting terribly ahead of ones who don't.
The a/b testing cult is a prime example. I knew companies who "optimised their websites to death" blindly following the deemed "signal." I knew of a guy who ran Borland's internet marketing and a lot of stories from him how hundred thousands in ad spend were ending up bidden on common sense defying keyword combinations.
Also, there are a lot of independent movies and TV shows on NetFlix. And they suck. Horrible acting and plot but the cover is sometimes good enough to trick you into wasting time taking a look.
Statistically, after viewing a great movie the next one will be rated as less satisfying. We build up our expectations to a level that is not in sync with reality. This psychological insight was used at one of the early Netflix challenges by one of the top contestants.
They never, ever admit that they don't have what I'm seeking and then suggest alternatives.
I get it free from Tmobile but hardly use it.
Amazon, meanwhile, has an incentive to show whatever ads pay the most. In this case, it's something of a winner's curse situation -- the ads that pay the most are often the ones that overestimate the value of an ad placement, not the one who correctly estimates the value of an ad placement. And if the ads don't generate the expected revenue, as long as Amazon can find more suckers, there's never a feedback loop occurring where Amazon's poor ebook ad placement ends up aligning with anything they care about. So I get ads for a lot of self-published trash (either in the Get Rich Quick With Bitcoin variety or the modern day romance novel sort), Amazon gets money, and most of these authors need a very low conversion rate so what do they care.
> You don’t need to pay any engineers to get behavioral tracking for your ads.
Google doesn't need to pay any engineers?
This is important. Its cost effective for _Google_ to pay big bucks for engineering labor since their network is used so widely. Its not cost effective for WSJ to do so.
Either they can't measure the real ROI (brand advertising) or it's hard to, or they are kind of lazy.
When big companies spend on ads, they often 'allocate' and the money just gets spent.
The efficiency of those ads is going to be very low.
The world of actionable ads is narrow. Google search ads are generally very specific and actionable, and are fairly well measurable. FB ads similarly (if you want them to be). In those contexts, the 'targeting' will definitely pop out at you and will make or break a campaign, so long as the advertiser is watching the numbers.
A lot of what one would think of as 'actually targeted advertising' gets lost in a wash of arbitrary ad spend.
It's not about whether the cookie is "enabled", but whether the cookie for a particular impression is available for matching.
The experiment looked at ad transactions of a single media company.
The first issue is that ad auctions are 2nd price, meaning that if an advertiser bids really high, they still only pay the 2nd-highest bid + $0.01. Of course the publishers wouldn't realize much from this. Second major issue is cookies and scale. You can't get high-value bids if advertisers can't recognize those users.
The industry is slowly moving toward quais-first-price but scale is a fundamental problem, which is again why Facebook, Google, and now Amazon get all the money and do really well with behavior targeting while even big publishers struggle. Add in the new privacy regulations and this will only further widen that gap.
It's not just oba... much of the non-oba demand running through programmatic requires a "cookie" for frequency capping or just basic anti-fraud.
Disclosure: work in industry; am biased.
That's why I said scale (reach + cookies) is the fundamental problem.
Are you talking about the article or the paper? The paper is published at the Workshop on the Economics of Information Security (WEIS) with the conference char being people like Bruce Schneier. Besides, the author is Alessandro Acquisti and the details of their experiments are well outlined in the paper.
I understand that you see challenges, issues with their methods but you need to perform your own experiments and make a sound argument to make any strong statement. I'd be very cautious before I make a casual remark at a well conducted study. The study may have limitations but this doesn't mean the conclusions of the study are invalid.
This wasn't a scientific study, nor is it well-conducted, because they looked at a single publisher over the course of a week in 2016 and are drawing conclusions for the entire market when they don't understand how big the market is and how the programmatic supply chain works. People in adtech deal with petabytes of data everyday. The fact that publishers don't see much revenue lift from behavioral targeting isn't a surprise, the whole industry knows it. So the study isn't completely wrong, but just looking at small picture that ends up aligning with the larger truth.
Why the rev is low isn't covered by the study, but the article tries to come up with an explanation. This is incorrect, which is what I cover in my first comment.
My hypothesis is that advertisers overpay for targeted ads, but this is not the right way tot test that. They would need to find data or experiment that correlates advertising spend to revenue generated.