Behavioral Advertising may account for only 10% of Display Advertising which itself accounts for only 23% of the overall online ad market. This means that cookies may be used in as little as 2% of online advertising to achieve to behaviorally targeting! And of that 2% only a very small portion would need to use a cookie to identify you for purposes of such targeting. Alternatively it would be difficult to find any online ad from Search to Lead Generation that did not greatly depend on cookies for reporting and unlike targeting which generally seeks to find broad groups to which to present the same creative, reporting often seeks to know precisely who interacted with a given ad. It is on the reporting side where most of the need for true personal identification takes place, and cookies are integral in providing this functionality. In fact cookie are so frequently used for identification in reporting we will bifurcate this discussion into cookie reporting uses not involving personally identifying data and those directly involving personally identity.
Personal Data Use in Ad Reporting
Why would an ad need to know that I, John Quincy Adams, clicked or interacted with it, when it didn't need to know who I was when it was shown to me? To answer that question we need look at the three primary ways which ads are sold online today: CPM (cost per thousand), CPC (cost per click) and CPA (cost per action). The amount an advertiser is willing to pay increases with the risk the publisher is willing to take on the advertisers behalf, so while an advertiser may be willing to pay on 1/20th of a penny for a single ad display, he may be willing to pay $1.50 for a click on the ad or even 10% of an eventual sale for a single ad view that leads to a purchase. Unfortunately as the dollar values increase per individual event so does the need to prevent fraud or otherwise accurately account for the transaction. While a publisher could potentially print money by just visiting his own site and reloading pages to view ads, at 1/20th of a penny per ad, this is not a very efficient means of gaming the system.
Alternatively, clicking on ads who's CPC price may range from pennies to up to $100, either to make money for your own site, or to make a competitor pay for your malicious clicks, sometime called click fraud is certainly a potentially more effective gaming of the system and to combat this CPC sales organizations like Google's Ad Sense and Ad Words run a number or programs to attempt to weed out such activity. The chief defense has been that under CPC clients often name their own price and therefore should fraud enter into the system, they will decrease the amount they are willing to pay to accommodate the portion of their clicks that are given to fraud. Implicit in this self correction assumption is that bidders are able to ascertain the true value of a click event. Without commenting further on the state of click fraud, it is important to acknowledge that the assumption is true, that in fact the success of CPC is in no small part directly attributable to the fact that buyers can directly measure the efficacy of their ad programs and cookies are integral to this measurement.
How does this measurement take place? Let's take a hypothetical advertiser who wishes to pay on a CPC for his lawncare service. If such advertiser were to cookie everyone who clicked on a CPC advertisement and later cookie all those people who converted on his site as paying customers he would have a direct feedback mechanism to know directly how much money they ad campaign netted him. If for instance he knew that an ad buy for which he paid for 100 clicks yielded him 2 customers, Tim Smith, profitability $200 and Susie Allen, profitablity $450 the math becomes straight forward. 100 clicks yielded $650 in profit ergo pay up to $6.50 a click. If fraud is introduced into the system (say 100 fraudulent clicks), now 200 clicks would have yielded the same $650 and each click would have declined in value to $3.25 per click. Importantly what makes all this work is knowing WHO specifically purchased as a result of the click as the feedback mechanism for doing the analysis.
CPA requires the same type of ROI analysis but even more explicitly. Under CPA a publisher is typically provided a direct bounty as a percentage of the sale that the publisher's ad directly lead to. If for instance a banner ad for a book seller lead to the purchase of $211 in books, the publisher may be paid $21. In the event that the ad serve did not yield a sale, the publisher is paid nothing. In the event that it did lead to a sale, there is a significant payment. This makes specific attribution the single most important aspect of the process, attribution which is possible because of the cookie's ability maintain state from the publisher's ad to the advertisers conversion.
Let's look a little more closely at how this works. By way of example, a banner ad for a book seller, Big Book Seller may be displayed on a publisher, Home Town Newspaper. When a user first sees this ad he is assigned a cookie id=abc123. It is recorded by the ad server that cookie abc123 has seen (and perhaps clicked on) the ad for Big Book Seller on site Home Town Newspaper. The cookie is anonymous at this point. At some time in the future (and in accordance with the terms of the agreement between the advertiser and the publisher) the same cookie makes a purchase at Big Book Seller. During this conversion process personal data (name, address, phone, credit card # etc) is collected about he purchaser. This data is understood by the seller and referenced by either an Order Id or a Customer Id which is associated with cookie through a pixel on the conversion page. This effectively retrofits the identity of the purchaser to the previously anonymous event of having seen an ad on Home Town Newspaper.
Seen on Home Town Paper <-> Cookie ABC123 <-> CID XYZ456 <-> Tim Smith
Importantly, because the dollar values are relatively high, the seller wants to pay the bounty only on collected revenues and should there be an offline or subsequent order cancellation it is vital that a percentage not be paid on the on collected revenue. Without this protection the entire model would break down as entirely anonymous users would purchase products through the web and then cancel by phone all while generating large bounties for their websites.
To Get ROI You Need PII
As this analysis shows, the more you need to specifically attribute revenue accurately the more you need to know WHO purchased. Unless you know who purchased it is impossible to know with certainty if the revenue was actually collected, how much the customer was really worth, etc. Regardless of the specific model, Sponsorship, CPA, Affiliate, CPC, Search or CPM, the ability to calculate ROI has been the defining characteristic of web advertising, and this ability is made possible by knowing the identity of the purchaser and attributing that identity through a cookie to the place where they first encountered the ad.
Having the gory details of personal data use taken care of, it should also be pointed out that there are indeed aggregate or non-personal uses of cookie data in ad reporting. These typically fall into two categories - reach and frequency.
By assigning each user a unique cookie advertisers are able to learn a great deal about how many people saw their advertising. If an advertiser, for example, purchased 1 million ad impressions on the home page of Home Town Paper they could verify this by looking at server logs and seeing that indeed the creative for their ad was requested 1 million times, but the question then becomes was that 1 million people looking at it once or 1 person looking at it on million times? By looking at the unique cookie associated with each ad serving event it is easy to derive that the on million ad impressions were consumed by only 500k unique cookies (meaning that on average each cookie saw the ad twice). We'll ignore questions of cookie deletion or cookie blocking here, which while very significant to this entire discussion complicates greatly the math of a simple example!
By looking only at aggregate numbers, total impressions and total unique cookies, a great deal of important information can be derived, particularly for brand advertisers who's intent may not be to immediately drive measurable online sales but rather to understand something of the size of their audience.
Frequency is the logical corollary to Reach. While serving a million impressions to a million unique cookies is almost certainly better than serving a million impressions to one unique cookie, it may not follow that it is the optimal solution. Indeed some marketers have a sweet spot as to how many times they would like each browser to be exposed to their message. It may be one it may be three or four. Measures of frequency can be combined with other measurements to derive important information such as how many exposures it typically takes to generate a click (or a purchase), how many exposures it takes to generate brand recognition etc.
Cookies are indeed vital for both online ad measurement and targeting. The targeting component is perhaps better recognized because we see the visual clues of behaviorally targeted ads which reflect not the content on which they appear but our previous web activity, but it should not be left unrecognized that cookies play an even more important role in ad reporting as nearly online ad uses a cookie in some way shape or form to measure its efficacy. It should also be recognized that whereas the goal of targeting is to target a finite set of creatives to a finite grouping of audience segments, the goal of measurement is often to understand exactly who interacted and how much they were worth.
Again this should not be seen as a condemnation of the process, for without this type of measurement we may well assume the value of online advertising would decrease and as a result we would see either less sites and services or would be asked to pay for those sites and services in another more direct manner. If, as I suspect, this analysis is new to most readers I would suspect that a better job need be done in making the trade off more explicit.