Playing the Online Adblocking Game, Part 1
Written on 23 June 2015
by Ruth Fisher, PhD
A copy of the full analysis can be downloaded by clicking on the link at the bottom of this blog entry.
One of the biggest issues currently in the media is the subject of online adblocking. Adblocking involves the installation and use by Internet Users of adblocking software on their web browsers, so as to prevent advertisements from appearing on websites. Many are predicting that adblocking software will seriously hamper the provision of content on the Internet by preventing the use of the highly popular, ad-based, Internet business model. For example, recent headlines include:
A recent announcement by Apple further heightened the controversy over the use of adblocking software. In early June the Company announced that it would provide adblocking software for the next version of its Safari browser for use on mobile devices (see for example Joshua Benton, “A blow for mobile advertising: The next version of Safari will let users block ads on iPhones and iPads”).
What’s even more interesting is that an entire ecosystem is popping up around the use of adblock software, for example as different players introduce new adblocking software for Users and new Service Providers introduce analytics to help Content Providers understand the extent of adblocking use by Users.
This analysis examines the dynamics involved in the Online Adblocking Game, which includes such players as Online Users, Content Providers, Advertisers, and Adblock Software and Services Providers. The first part of the analysis will examine trends in online ad revenues and ad pricing models, as a backdrop for analysis of the Adblocking Game. The second part of the analysis will introduce adblocking and describe its use. And the third part of the analysis (or perhaps third and fourth) will discuss the adblocking game.
Trends in US Online Ad Revenues
This first section describes recent trends in Internet advertising. It’s important to understand what’s been going on in this area, since these activities serve as a backdrop for the analysis of the Adblocking Game.
Figures 1 – 5 were taken from the “IAB/PwC Internet Ad Revenue Report, FY 2014,” which reports advertising statistics for the US through 2014.
Figure 1 displays the amount of money spent on advertising for each media segment in 2014. As seen in Figure 1, the largest portion of advertising revenues in 2014 was spent on Internet ads ($49.5B or 29% of segment totals), with Broadcast TV coming in second ($40.5B or 24% segment totals), and Cable TV coming in third ($25.2B or 15% segment totals). These three segments together account for two-thirds (67%) of total advertising revenues ($172.1B)
Figure 2 shows how the amount of spending on Internet ads has changed over time. As seen in Figure 2, mobile (e.g., smartphones and tablets) ads started appearing in 2010, and since then, funding for mobile ads has increased much faster than funding for non-mobile Internet ads (110% CAGR for mobile vs. 10% CAGR for non-mobile). By 2014, mobile ad revenues had reached 25% of total Internet ad spending.
Figures 3 and 4 indicate which formats of Internet ads are most popular. With a 2014 share of 37%, ads associated with Internet search dominate. However, the share of Internet search ads peaked in 2009, and has been decreasing ever since, as the share of mobile ads grows. In 2014, mobile ads accounted for the second largest share of Internet ad revenues (28%), followed by banner ads, which accounted for 16% of Internet ad revenues. These three largest segments together (search, mobile, and banner ads) accounted for 81% o 2014 Internet ad revenues.
So, online advertising has increased in quantity over time, but so too has the number of Internet users. The real question then is: how has the number of ads per Internet user been changing over time? Figure 5 shows the trends in Ad spending and Internet users over time, and Figures 6 shows the amount of Internet ad revenues per user over time. The data in Figure 5 indicate that Internet ads spending has increased at an annual rate of 17% between 2010 and 2014, while the number of Internet users has increased by 9% over the same period. And Figure 6 shows that between 2010 and 2014, Internet ad spending per user increase by 7.4% annually.
As an aside, it originally occurred to me to also look at trends in advertising revenues per website as an alternative indicator of how Internet advertising has changed over time. Based on the information presented in Figure 5, one can easily deduce that ad revenues per website have been decreasing over time, since growth in the number of websites has far outpaced growth in ad revenues. However, one must also consider that the popularity of websites follows a power law (see, for example, Lada A. Adamic and Bernardo A. Huberman, “Power-Law Distribution of the World Wide Web.”). What this means is that a relatively small number of websites account for the vast majority of visits by users. Presumably, ads will only be placed on websites that attain some minimum level of popularity; otherwise, the ads wouldn’t generate enough views to be cost effective. The trick is then to determine the threshold level of viewers per website that would make ad placement a worthwhile prospect and trying to determine how the number of websites with that minimum level of visitors has changed over time, given the power law distribution of websites. Unfortunately, this would be is no trivial undertaking, so I will take the easy way out by focusing on trends in ad revenues per user.
Trends in US Online Ad Pricing
This section describes recent trends in pricing for Internet advertising. These trends will also serve as a backdrop for the analysis of the Adblocking Game.
In the following sub-sections I first define the different Internet advertising models. I then go on to describe trends over time in (i) prevalence for specific ad models, (ii) pricing and effectiveness of ad models, and (iii) patent activity associated with ad models. I finish the section by collecting together the information on changes in ad model effectiveness and pricing over time and try to make sense of the various trends.
Internet Advertising Models - Definitions
Kohki Yamaguchi’s “Pay Per What? Choosing Pricing Models In Digital Advertising” provides an excellent primer on Internet ad pricing models. Yamaguchi defines the following models:
CPM: Cost per 1,000 Impressions
CPC: Cost per Click
CPF: Cost-per-Follower/Fan for Social Media
CPV: Cost-per-View for Internet Video
oCPM: Optimized CPM for Facebook
CPI: Cost per App Install
Some additional important Internet ad-related definitions from Google Ad Words:
Click-Through Rate (CTR): A ratio showing how often people who see your ad end up clicking it.
CTR = Clicks / Impressions
Conversion: A conversion happens when someone clicks your ad and then takes an action that you’ve defined as valuable to your business, such as an online purchase or a call to your business from a mobile phone.
Conversion Rate: The average number of conversions per ad click.
Conversion Rate = Conversions / Click
Yamaguchi indicates that the various pricing models “are translated into a single metric on the publisher side: eCPM, or effective-cost-per-impression,” where he defines eCPM as:
eCPM = CPC x (Predicted CTR) x 1,000
= CPA x (Predicted Conversion Rate) x (Predicted CTR) x 1,000
He goes on to compare the relative risks associated with the different pricing models. He indicates that CPM ads are the least risky ads for Content Providers (and the most risky for Advertisers); that is, under the CPM model ad revenue that will be generated for Content Providers is the most predictable of all the pricing models, while being the least indicative of actual ad value for Advertisers. At the other extreme, CPA ads are the most risky (least predictable) ads for Content Providers (and least risky for Advertisers), since click-through and conversion rates are so difficult to predict.
… [C]onversions are difficult to predict because they happen less often, after more touchpoints, and are dependent on many different factors. Because of this, CPA pricing is more risky for the publisher compared to CPC pricing, which in turn is riskier than CPM.
For non-auction media, the result is an often-seen and rather straightforward order of preference, where a publisher may first sell as much inventory as possible on a CPM (preferably premium) basis, and then any remnant inventory on a CPC basis, and moves on to CPA (affiliate) sales as a last resort.
The end result is that there is a risk and effort vs. scale tradeoff for the advertiser, where low-risk options such as CPA-priced campaigns have limited scalability; whereas, CPM or CPC campaigns can potentially attain the same CPA target at greater scale, but requires greater optimization effort to reach that same level of efficiency.
It comes down to which of these three factors the advertiser is concerned with the most: Effort, Efficiency, and Scale.
• Effort is the time and knowledge required to optimize the performance of an ad campaign.
• Efficiency refers to the ROI for the campaign objective (need not be actual revenue)
• Scalability is how much volume you can obtain for the objective.
The overall trade-off between effort, efficiency, and scale for different pricing options is shown in the chart below [Figure 7].
Source: Kohki Yamaguchi, “Pay Per What? Choosing Pricing Models In Digital Advertising”
Internet Advertising Models - Prevalences
So which pricing models are most prevalent? Figure 8 shows that over time, performance-based ad models (CPC) have become twice as prevalent as impressions-based models (CPM).
Internet Advertising Models – Prices
In “The Cost of Pay-per-Click (PPC) Advertising – Trends and Analysis,” Jonathan Hochman provides ad pricing and click-through rates based on “average results for a group of approximately fifty advertisers on the Google AdWords ad network.”
Hochman’s survey provided the following results (see Figure 9):
• Click through rates dropped precipitously between 2005 and 2007, remained stable through 2009, jumped between 2009 and 2010, dropped between 2010 and 2011, increased somewhat between 2011 and 2012, then leveled off through 2013.
Hochman explains the 2005 through 2007 drop as being due to an increase in Google’s content network. Presumably, a broader network might very well have lower response rates than those in a more concentrated one.
The decrease in click through rate (CTR) from 2005 onward is attributable to the growth of Google's content network. Content network sites usually have lower CTR's and low cost per click.
• Cost per click increased significantly between 2006 and 2010, decreased between 2010 and 2012, and then roughly leveled off through 2013.
Continuing on with Hochman’s survey results (see Figure 10):
• Cost per impression decreased between 2005 and 2006, decreased slightly between 2006 and 2007, then increased gradually between 2007 and 2013, with the exception of an anomalous peak in costs during 2010.
• Cost per conversion and conversion rates generally moved inversely with one another. Cost per conversion (conversion rates) decreased (increased) between 2005 and 2007, increased (decreased) significantly between 2007 and 2012, then dropped off (jumped) between 2012 and 2013.
An important point to note about conversion rates is that while better targeting leads to higher click rates, it doesn’t necessarily lead (in and of itself) to higher conversion rates. Conversion rates also depend on conversion marketing, that is, customer service.
• Invalid click rates generally increased between 2006 and 2011, and then dropped off somewhat between 2011 and 2012, remaining relatively stable through 2013.
Internet Advertising Models – Patents
A source of information on changes in ad effectiveness over time is patent activity in the area. In “Web Based Targeted Advertising: A Study Based on Patent Information,” Nishad Deshpandea, Shabib Ahmedb, Alok Khodec analyzed patent activity for “internet based targeted advertising.” Figure 11 illustrates their findings for the total number of patents in the area over time. The study indicates that annual patent activity increased significantly between 2004 and 2010, remained constant between 2010 and 2011 then dropped off significantly through 2013.
Source: Nishad Deshpandea, Shabib Ahmedb, Alok Khodec, “Web Based Targeted Advertising: A Study Based on Patent Information”
Internet Advertising Models – Discussion
So, what to make of the information presented in Figures 8 - 11?
Changes over time in ad costs and ad activity (click through and conversion) rates are presumably associated with
• Changes in the supply of ad space by Content Providers relative to the demand for online ad space by Advertisers,
• Advancements in advertising effectiveness (e.g., targeted advertising) and user conversion (customer service), and
We know from information previously presented that
(i) Ad revenue per user increased between 2005 and 2007, decreased through 2010, then increased between 2010 and 2014.
Given the power law distribution of websites, I presume that [this is the most tenuous of my assumptions]
(ii) Ad revenue per website that hosts ads has been following a similar pattern to that for ad revenue per user.
Also, I assume the following points hold:
(iii) Impressions are easier to fake than clicks, which means performance-based payments are less susceptible to fraud than impressions-based payments;
(iv) Increasing amounts of impression fraud (i.e., visits to a website for the sole purpose of increasing the number of website impressions) lead to lower click through rates:
Click through Rate with No impression Fraud = Clicks / Impressions
Click through Rate Impression Fraud: = Clicks / (Impressions + Fraud)
< Clicks / Impressions
(v) Increasing amounts of advertising, all else equal, lead to lower click through rates (i.e., user saturation);
(vi) Increasing effectiveness of advertising leads to higher click through rates and lower cost per click rates;
(vii) Increasing amounts of click fraud and/or clickbait lead to higher click through rates and lower conversion rates:
Click through Rate without Click Fraud = Clicks / Impressions
Click through Rate with Click Fraud = (Clicks + Fraud) / Impressions
> Clicks / Impressions
Conversion Rate without Click Fraud = Action / Clicks
Conversion Rate with Click Fraud = Action / (Clicks + Fraud)
< Action / Clicks
So now using the seven points above, together with the other information we’ve learned about online advertising, we are left with the following patterns to explain:
• Move toward Performance-Based Models
The increasing prevalence of performance-based ad models is probably due in large part to the greater ability to fake impressions than clicks (point (iii)). At the same time, however, Yamaguchi’s information provided above indicates that performance-based ads are more risky for Content Providers, while being less risky for Advertisers. So the increasing prevalence of performance-based models could also suggest that over time Advertisers have gained more leverage in ad rate negotiations relative to Content Providers. I suspect, however, that the fraud issue is the larger driver towards performance-based models.
• Trends in Costs per Click and Click through Rates
Improvements in advertising technologies (i.e., targeted advertising) over time (point (vi)) should increase the efficiency of advertising, leading to higher click through rates and lower costs per click.
At the same time, however, the power law distribution of websites, together with the increasing rates of ad spending (increased demand = higher price) suggest that more popular websites are being increasingly inundated with ads, leading to User saturation (points (i) and (v)). That is, it’s been getting harder to get Users to click on ads, which means that click through rates will be lower and Advertisers will pay more per click.
Conversely, the incidence of ad fraud (click fraud and clickbait) has been increasing (it’s been easier to generate clicks), leading to lower costs per click and higher click through rates (point (iv). To the extent, however, that Advertisers have improved their ability to deter click fraud, its impacts would be attenuated.
• Trends in Costs per Mille (Impression)
Greater demand for advertising over time (point (i)) would lead to higher costs per impression.
• Trends in Costs per Conversion and Conversion Rates
Greater demand for advertising over time (point (i)) would lead to higher costs per conversion.
At the same time, the power law distribution of websites, together with the increasing rates of ad spending (increased demand = higher price) suggest that more popular websites are being increasingly inundated with ads, leading to User saturation (point (v). That is, it’s been getting harder to get Users to respond to ads, which means that conversion rates will be lower and Advertisers will pay more per click.
Furthermore, increasing amounts of click fraud and/or clickbait lead to (higher click through rates and) lower conversion rates (point (vii)).
• Trends in Invalid Click Rates
Greater demand for advertising over time (point (i)) leads to greater benefits associated with ad fraud (i.e., there’s more money to be made), which would increase the incidence of ad fraud in all forms.