The Design of Online Advertising Markets

Edelman, Benjamin. “The Design of Online Advertising Markets.” Chap. 15 in The Handbook of Market Design, edited by Nir Vulkan, Alvin E. Roth, and Zvika Neeman. Oxford University Press, 2013.

Because the market for online advertising is both new and fast-changing, participants experiment with all manner of variations. Should an advertiser’s payment reflect the number of times an ad was shown, the number of times it was clicked, the number of sales that resulted, or the dollar value of those sales? Should ads be text, images, video, or something else entirely? Should measurement be performed by an ad network, an advertiser, or some intermediary? Market participants have chosen all these options at various points, and prevailing views have changed repeatedly. Online advertising therefore presents a natural environment in which to evaluate alternatives for these and other design choices. In this piece, I review the basics of online advertising, then turn to design decisions as to ad pricing, measurement, incentives, and fraud.

The Online Ad Scams Every Marketer Should Watch Out For

The Online Ad Scams Every Marketer Should Watch Out For. HBR Online. October 13, 2015.

Imagine you run a retail store and hire a leafleteer to distribute handbills to attract new customers. You might assess her effectiveness by counting the number of customers who arrived carrying her handbill and, perhaps, presenting it for a discount. But suppose you realized the leafleteer was standing just outside your store’s front door, giving handbills to everyone on their way in. The measured “effectiveness” would be a ruse, merely counting customers who would have come in anyway. You’d be furious and would fire her in an instant. Fortunately, that wouldn’t actually be needed: anticipating being found out, few leafleteers would attempt such a scheme.

In online advertising, a variety of equally brazen ruses drain advertisers’ budgets — but usually it’s more difficult for advertisers to notice them. I’ve been writing about this problem since 2004, and doing my best to help advertisers avoid it.

In this piece for HBR Online, I survey these problems in a variety of types of online advertising — then try to offer solutions.

Accountable? The Problems and Solutions of Online Ad Optimization

Edelman, Benjamin. “Accountable? The Problems and Solutions of Online Ad Optimization.” IEEE Security & Privacy 12, no. 6 (November-December 2014): 102-107.

Online advertising might seem to be the most measurable form of marketing ever invented. Comprehensive records can track who clicked what ad–and often who saw what ad–to compare those clicks with users’ subsequent purchases. Ever-cheaper IT makes this tracking cost-effective and routine. In addition, a web of interlocking ad networks trades inventory and offers to show the right ad to the right person at the right time. It could be a marketer’s dream. However, these benefits are at most partially realized. The same institutions and practices that facilitate efficient ad placement can also facilitate fraud. The networks that should be serving advertisers have decidedly mixed incentives, such as cost savings from cutting corners, constrained in part by long-run reputation concerns, but only if advertisers ultimately figure out when they’re getting a bad deal. Legal, administrative, and logistical factors make it difficult to sue even the worst offenders. And sometimes an advertiser’s own staff members prefer to look the other way. The result is an advertising system in which a certain amount of waste and fraud has become the norm, despite the system’s fundamental capability to offer unprecedented accountability.

Pitfalls and Fraud in Online Advertising Metrics: What Makes Advertisers Vulnerable to Cheaters, and How They Can Protect Themselves

Edelman, Benjamin. “Pitfalls and Fraud in Online Advertising Metrics: What Makes Advertisers Vulnerable to Cheaters, and How They Can Protect Themselves.” Journal of Advertising Research 54, no. 2 (June 2014): 127-132.

How does online advertising become less effective than advertisers expect and less effective than measurements indicate? The current research explores problems that result, in part, from malfeasance by outside perpetrators who overstate their efforts to increase their measured performance. In parallel, similar vulnerabilities result from mistaken analysis of cause and effect–errors that have become more fundamental as advertisers target their advertisements with greater precision. In the paper that follows, the author attempts to identify the circumstances that make advertisers most vulnerable, notes adjusted contract structures that offer some protections, and explores the origins of the problems in participants’ incentives and in legal rules.

Services for Advertisers – Avoiding Waste and Improving Accountability

In the course of my research on spyware/adware, typosquatting, popups, and other controversial online practices, I have developed the ability to identify practices that overcharge online advertisers. I report my observations to select advertisers and top networks in order to assist them in improving the cost-effectiveness of their advertising including by flagging improper ad placements, rejecting unjustified charges, and avoiding untrustworthy partners. This page summarizes the kinds of practices I uncover and presents representative examples drawn from my publications:

Services for Advertisers – Avoiding Waste and Improving Accountability

Measuring and Managing Online Affiliate Fraud with Wesley Brandi

Affiliate programs vary dramatically in their incidence of fraud. In some merchants’ affiliate programs, rogue affiliates fill the ranks of high-earners. Yet other similarly-sized merchants have little or no fraud. Why the difference?

In Information and Incentives in Online Affiliate Marketing, Wesley Brandi and I examine the impact of varying merchant management decisions. Some merchants hire specialist outside advisors (“outsourced program managers” or OPM’s) to set and enforce program rules. Others ask affiliate network staff to make these decisions. Still others handle these tasks internally.

A merchant’s choice of management structure has significant implications for both the information available to decision-makers and the incentives that motivate those decision-makers. Outside advisors tend to have better information: An OPM sees problems and trends across its many clients. A network is even better positioned — enjoying direct access to log files, custom reports, and problems reported by all merchants in the network. That said, outside advisors usually suffer clear incentive problems. Most notably, networks are usually paid in proportion to a merchant’s affiliate channel spending, so networks have a significant incentive to encourage merchants to accept even undesirable affiliates. In contrast, incentives for merchants’ staff are typically more closely aligned with the merchant’s objectives. For example, many in-house affiliate managers have stock, options, or bonus that depend on company profitability. And working in a company builds intrinsic motivation and loyalty. In short, there are some reasons to think outsourced specialists will yield superior results, but other reasons to favor in-house staff.

To separate these effects, we used crawlers to examine affiliate fraud at what we believe to be unprecedented scope. Our crawlers ran more than 2 million page-loads on a variety of computers and virtual computers, examining the relative susceptibility of all CJ, LinkShare, and Google Affiliate Network merchants (as of spring 2012) to adware, cookie-stuffing, typosquatting, and loyalty apps.

We found outside advisors best able to find “clear fraud” plainly prohibited by network rules, specifically adware and cookie-stuffing. But in-house staff did better at avoiding “grey area” practices such as typosquatting — schemes less plainly prohibited by network rules, yet still contrary to merchants’ interests. On balance, there are good reasons to favor each management approach. Our advice: A merchant choosing outsourced management should be sure to insist on borderline decisions always taken with the merchant’s interests at heart. A merchant managing its programs in-house should be careful to avoid known cheaters that a savvy specialist would more often exclude.

Our results clearly reveal that networks take actions that are less than optimal for merchants. It’s tempting to attribute this shortfall to malicious intent by networks, but the same outcome could result from networks simply putting their own interests first. Consider a network that receives undisputed proof that a given affiliate is cheating a given merchant. Should the network eject that affiliate from the entire network (and all affiliated merchants), or only from that single merchant’s program? The former helps dozens or hundreds of merchants, but with corresponding reduction to network revenues. No wonder many networks chose the latter. Similarly, when networks decide how much to invest in network quality — engineers, analysts, crawlers, and the like — their incentive to improve quality is tempered by both direct cost and foregone revenue.

Incidental to our analysis of management structure, we gathered significant data about the scope of affiliate fraud more generally. Some differences are stark: For example, Table 4 reports Google Affiliate Network merchants suffering, on average, less than half as much adware and cookie-stuffing as LinkShare merchants. I’ve been critical of Google on numerous issues. But when it comes to affiliate quality, GAN was impressive, and GAN’s high standards show clearly in our large-sample data. Note that our analysis precedes Google’s April 2013 announcement of GAN’s shutdown.

Our full analysis is under review by an academic journal.

(update: published as Edelman, Benjamin, and Wesley Brandi. “Risk, Information, and Incentives in Online Affiliate Marketing.” Journal of Marketing Research (JMR) 52, no. 1 (February 2015): 1-12. (Lead Article.)

A Holiday “Top 10”: Rogue Affiliates at Commission Junction and LinkShare with Wesley Brandi

Our automation continuously scours the web for rogue affiliates. In our query tool, we provide a basic sense of how much we’ve found. We have also written up scores of sample rogue affiliates, but the holiday season provides an impetus for more: Thanks to high online spending, affiliate fraud at this time of year is particularly profitable for perpetrators — and particularly costly to merchants.

In today’s article, we report the ten Commission Junction affiliates and ten LinkShare affiliates most often seen by our automation. Our findings:

Twenty Oft-Found Commission Junction and LinkShare Affiliate Violations

Affiliate Fraud Litigation Index

Some analysts view affiliate marketing as “fraud-proof” because affiliates are only paid a commission when a sale occurs. But affiliate marketing nonetheless gives rise to various disputes — typically, merchants alleging that affiliates claimed commission they had not properly earned. Most such disputes are resolved informally: merchants withhold amounts affiliates have purportedly earned but have not yet received. Occasionally, disputes end up in litigation with public availability of the details of alleged perpetrators, victims, amounts, and methods.

In today’s posting, I present known litigation in this area including case summaries and primary source documents:

Affiliate Fraud Litigation Index

Flash-Based Cookie-Stuffer Using Google AdSense to Claim Unearned Affiliate Commissions from Amazon with Wesley Brandi

Merchants face special challenges when operating large affiliate marketing programs: rogue affiliates can claim to refer users who would have purchased from those merchants anyway. In particular, rogue “cookie-stuffer” affiliates deposit cookies invisibly and unrequested — knowing that a portion of users will make purchases from large merchants in the subsequent days and weeks. This tactic is particularly effective in defrauding large merchants: the more popular a merchant becomes, the more users will happen to buy from that merchant within a given referral period.

To cookie-stuff at scale, an attacker needs a reliable and significant source of user traffic. In February we showed a rogue affiliate hacking forum sites to drop cookies when users merely browse forums. But that’s just one of many strategies. I previously found various cookie-stuffing on sites hoping to receive search traffic. In a 2009 complaint, eBay alleges that rogue affiliates used a banner ad network to deposit eBay affiliate cookies when users merely browsed web pages showing certain banner ads. See also my 2008 report of an affiliate using Yahoo’s Right Media ad network to deposit multiple affiliate cookies invisibly — defrauding security vendors McAfee and Symantec.

As the eBay litigation indicates, display advertising networks can be a mechanism for cookie-stuffing. Of course diligent ad networks inspect ads and refuse cookie-stuffers (among other forms of malvertising). So we were particularly surprised to see Google AdSense running ads that cookie-stuff Amazon.

The 'Review Different Headphones' ad actually drops Amazon Associates affiliate cookies.
This innocuous-looking banner ad sets Amazon Associates cookies invisibly.
The Imgwithsmiles attack

We have uncovered scores of web sites running the banner ad shown at right. On 40 sites, on various days from February 6 to May 2, our crawlers found this banner ad dropping Amazon Associates affiliate cookies automatically and invisibly. All 40 sites include display advertising from Google AdSense. Google returns a Flash ad from Imgwithsmiles. To an ordinary user, the ad looks completely innocuous — the unremarkable “review different headphones” image shown at right. However, the ad actually creates an invisible IMG (image) tag loading an Amazon Associates link and setting cookies accordingly. Here’s how:

First, the ad’s Flash code creates an invisible IMG tag (10×10 pixels) (yellow highlighting below) loading the URL http://imgwithsmiles.com/img/f/e.jpg (green).

function Stuff() {
  if (z < links.length) {
    txt.htmltext = links[z];
    z++;
    return(undefined);
  }
  clearinterval(timer);
}
links = new array();
links[0] = "<img src="http://imgwithsmiles.com/img/f/e.jpg" width="10" height="10"/>";z = 0;timer = setinterval(Stuff, 2000);

While /img/f/e.jpg features a .jpg extension consistent with a genuine image file, it is actually a redirect to an Amazon Associates link. See the three redirects preserved below (blue), including a tricky HTTPS redirect (orange) that would block many detection systems. Nonetheless, traffic ultimately ends up at Amazon with an Associates tag (red) specifying that affiliate charslibr-20 is to be paid for these referrals.

GET /img/f/e.jpg HTTP/1.0
Accept: */*
Accept-Language: en-US
Referer: http://pagead2.googlesyndication.com/pagead/imgad?id=CICAgICQvuXgahDQAhiYAjII3bQHU19r_Isx-flash-version: 10,3,183,7User-Agent: Mozilla/4.0 (compatible; MSIE 8.0; ...)Host: imgwithsmiles.comConnection: Keep-AliveHTTP/1.1 302 Moved TemporarilyDate: Wed, 02 May 2012 19:56:59 GMTServer: Apache/2.2.21 (Unix) mod_ssl/2.2.21 OpenSSL/0.9.8e-fips-rhel5 mod_bwlimited/1.4
X-Powered-By: PHP/5.2.17
Expires: Thu, 19 Nov 1981 08:52:00 GMT
Cache-Control: no-store, no-cache, must-revalidate, post-check=0, pre-check=0
Pragma: no-cache
Set-Cookie: PHPSESSID=174272468a212dd0862eabf8d956e4e0; path=/
Location: https://imgwithsmiles.com/img/kick/f/e.jpg
Content-Length: 0
Keep-Alive: timeout=5, max=100
Connection: Keep-Alive
Content-Type: text/html-

HTTPS redirect decoded via separate manual request
GET /img/kick/f/e.jpg HTTP/1.1 Accept: text/html, application/xhtml+xml, */* Accept-Language: en-US User-Agent: ... Accept-Encoding: gzip, deflate Host: imgwithsmiles.com Connection: Keep-AliveHTTP/1.1 302 Moved Temporarily Date: ... Server: Apache/2.2.21 (Unix) mod_ssl/2.2.21 OpenSSL/0.9.8e-fips-rhel5 mod_bwlimited/1.4 X-Powered-By: PHP/5.2.17 Location: http://imgwithsmiles.com/img/t/f/e.jpg Content-Length: 0 Connection: close Content-Type: text/html-GET /img/t/f/e.jpg HTTP/1.0 Accept: */* Accept-Language: en-US x-flash-version: 10,3,183,7 User-Agent: Mozilla/4.0 (compatible; ...) Connection: Keep-Alive Host: imgwithsmiles.com Cookie: PHPSESSID=174272468a212dd0862eabf8d956e4e0HTTP/1.1 302 Moved TemporarilyDate: Wed, 02 May 2012 19:56:59 GMT Server: Apache/2.2.21 (Unix) mod_ssl/2.2.21 OpenSSL/0.9.8e-fips-rhel5 mod_bwlimited/1.4 X-Powered-By: PHP/5.2.17 Expires: Thu, 19 Nov 1981 08:52:00 GMT Cache-Control: no-store, no-cache, must-revalidate, post-check=0, pre-check=0 Pragma: no-cache Location: http://www.amazon.com/gp/product/B002L3RREQ?ie=UTF8&tag=charslibr-20 Content-Length: 0 Keep-Alive: timeout=5, max=99 Connection: Keep-Alive Content-Type: text/html

If a user happens to make a purchase from Amazon within the subsequent 24 hours, Amazon will pay a commission to this affiliate — even though the affiliate did nothing at all to cause or encourage the user to make that purchase.

Does Amazon know?

The available information does not reveal whether or not Amazon knew about this affiliate’s practices. Nor can we easily determine whether, as of the May 2, 2012 observations presented above, this affiliate was still in good standing and receiving payment for the traffic it sent to Amazon.

On one hand, Amazon is diligent and technically sophisticated. Because Amazon runs one of the web’s largest affiliate programs, Amazon is necessarily familiar with affiliate fraud. And Amazon has ample incentive to catch affiliate fraud: Every dollar paid to fraudulent affiliates is money completely wasted, coming straight from the bottom line.

On the other hand, we have observed this same affiliate cheating Amazon for three months nonstop. All told, we’ve seen this affiliate rotating through 49 different Associates IDs. If Amazon had caught the affiliate, we would have expected the affiliate to shift away from any disabled affiliate accounts, most likely by shifting traffic to new accounts. Of the 28 Associates IDs we observed during February 2012, we still saw 6 in use during May 2012 (month-to-date) — suggesting that while Amazon may be catching some of the affiliate’s traffic, Amazon probably is not catching it all.

A further indication of the affiliate’s earnings comes from the affiliate’s willingness to incur out-of-pocket costs to buy media (AdSense placements from Google) with which to deliver Amazon cookies. As best we can tell, Amazon is the affiliate’s sole source of revenue. Meanwhile, the affiliate must pay Google for the display ad inventory the affiliate receives. These direct incremental costs give the affiliate a clear incentive to cease operation if it concludes that payment from Amazon will not be forthcoming. From the affiliate’s ongoing actions we can infer that the affiliate finds this scheme profitable — that its earnings to date have exceeded its expenses to date.

How profitable is this affiliate’s attack? Conservatively, suppose 40% of users are Amazon shoppers and make an average of four purchases from Amazon per year. Then 0.4*4/365=0.44% of users are likely to make purchases from Amazon in any given 24-hour period. Suppose the affiliate buys 1,000,000 CPM impressions from Google. Then the affiliate will enjoy commission on 0.44%*1,000,000=4,384 purchases. At an average purchase size of $30 and a 6.5% commission, this would be $8,547 of revenue per million cookie-stuffing incidents. How much would the affiliate have to pay Google for 1,000,000 CPM impressions? We’ve seen this affiliate on a variety of sites, but largely sites in moderate to low-priced verticals. At $2 CPM, the affiliate’s costs would be $2,000 — meaning the affiliate would still be slightly profitable even if Amazon caught 3/4 of its affiliate IDs before the first payment!

We alerted our contact at Amazon Associates to our observations. We will update this post with any information Amazon provides.

Search My Logs of Affiliate Fraud

Since 2004, I’ve been tracking and reporting all manner of rogue affiliatesusing spyware and adware to cover competitors’ sites; using trickier spyware and adware to claim commission on merchants’ organic traffic; typosquatting; stuffing cookies through invisible IFRAME’s and IMG’s, banner ads, and even hacked forum sites; and the list goes on. I now have automation catching these practices in ever-increasing quantities.

While I’ve written up dozens of rogue affiliates on this site and in various presentations, today Wesley Brandi and I are introducing something better: query-based access to our records of affiliate fraud targeting top affiliate merchants. Enter a merchant’s domain name, and we’ll tell you how much affiliate fraud we’ve seen targeting that domain — handy for merchants wanting to check whether their program is clean, and for affiliates wanting to confirm the trustworthiness a program they’re considering promoting. We’re not currently posting details of the specific perpetrators, but we have affiliate ID numbers, domain names, and packet log proof on file for each violator, and we can provide these upon request.

Take a look:

Affiliate Fraud Information Lookup
(2015 update: service no longer operational)