Competition and Cannibalization of Brand Keywords

40
Competition and Cannibalization of Brand Keywords * Andrey Simonov University of Chicago (Booth) Chris Nosko University of Chicago (Booth) Justin M. Rao Microsoft Research September 4, 2015 We describe and quantify the effects of competition on search advertising brand key- words. In this market, firms compete over advertising on their own keywords as well as those of their competitors. Traffic often has a direct substitute: a similar organic link that appears directly below the paid advertisements. Given this potential crowd out, real and perceived (non-incremental) metrics of ad performance will diverge, complicating the op- timization problem for firms. We utilize a large-scale, fully randomized experiment on the Bing platform where paid search advertisements were selectively removed for subsets of users. The experiment gives us good data coverage for over 2,500 firms. We find that there is sub- stantial heterogeneity in the causal effect of one’s own brand keyword advertisements, which is driven in part by how well known the brand is. Competitors greatly impact the market. First, if the focal brand is not present in the top slot, they can steal a substantial fraction of clicks. Second, when the focal brand is in the top slot, the firms can steal a modest amount of clicks, but the primary effect is to dramatically shift the focal brand’s traffic from the organic link to the paid link, thus increasing cannibalization and raising the cost of the focal brand’s advertising. * We thank Christian Perez, Matthew Goldman and Giorgos Zervas for helpful comments. 1

Transcript of Competition and Cannibalization of Brand Keywords

Page 1: Competition and Cannibalization of Brand Keywords

Competition and Cannibalization of Brand Keywords∗

Andrey SimonovUniversity of Chicago (Booth)

Chris NoskoUniversity of Chicago (Booth)

Justin M. RaoMicrosoft Research

September 4, 2015

We describe and quantify the effects of competition on search advertising brand key-words. In this market, firms compete over advertising on their own keywords as well asthose of their competitors. Traffic often has a direct substitute: a similar organic link thatappears directly below the paid advertisements. Given this potential crowd out, real andperceived (non-incremental) metrics of ad performance will diverge, complicating the op-timization problem for firms. We utilize a large-scale, fully randomized experiment on theBing platform where paid search advertisements were selectively removed for subsets of users.The experiment gives us good data coverage for over 2,500 firms. We find that there is sub-stantial heterogeneity in the causal effect of one’s own brand keyword advertisements, whichis driven in part by how well known the brand is. Competitors greatly impact the market.First, if the focal brand is not present in the top slot, they can steal a substantial fraction ofclicks. Second, when the focal brand is in the top slot, the firms can steal a modest amountof clicks, but the primary effect is to dramatically shift the focal brand’s traffic from theorganic link to the paid link, thus increasing cannibalization and raising the cost of the focalbrand’s advertising.

∗We thank Christian Perez, Matthew Goldman and Giorgos Zervas for helpful comments.

1

Page 2: Competition and Cannibalization of Brand Keywords

1 Introduction

Advertising on brand keywords, defined as bidding on queries related to a company’s trade-

marked name, is a substantial fraction of advertising expenditure on sponsored search plat-

forms. At first glance, this practice seems to make a lot of sense: Consumers who search for a

firm’s brand name are signaling a high degree of product awareness and click-through-rates

(CTR) for the focal brand on brand keywords are consequently considerably higher than

similar non-brand keywords. Further, since bidding on another firm’s trademarked term

is legal, competing firms can step in and siphon off traffic from the focal brand. Bidding

aggressively can ensure the focal brand occupies the top slot in an effort to minimize “traffic

stealing.” On the other hand, the focal brand also occupies the first “organic” result shown

just below the paid link, creating the possibility that the paid link can crowd-out free clicks.

Indeed a recent paper, Blake et al. (2015), find almost complete crowd-out studying a sin-

gle, well-known brand, eBay. Using a controlled experiment, they document that when eBay

stopped bidding on its own keyword, 99.5% of traffic was retained via the organic link.

We are thus left with a bit of a mystery. Advertising on brand keywords remains big

business, yet published evidence on the topic indicates that it is practically worthless for the

focal brand. Moreover, the paper had a compelling story: users who intentionally seek out a

particular brand do not need to be continued to be advertised to. Competing explanations

of why firms bid on brand advertisements have fundamentally different implications for the

marketplace and our understanding of firm behavior. The first class of explanations posits

that advertisers focus on “nominal measures” of ad performance, such as the ad’s CTR, rather

than the causal impact on customer visits, which leads to an irrational interest in keywords

that appear attractive through this nominal lens. The second class of explanations points to

eBay being unrepresentative in some way. For example, eBay, as the second largest e-retailer

with a very established brand, may be immune to competitive effects among customers using

“eBay” in a search query. Under this explanation, other firms very well could be making

rational choices.

In this paper we arbitrate between these disparate explanations with a large-scale, fully

randomized experiment run on Bing.com. On both Bing and Google, the maximum allowable

number of ads placed above the organic search results is 4. In the experiment this cap was

exogenously reduced to 0, 1, 2 or 3 (cap 0 through 3). For example, for a query where one

2

Page 3: Competition and Cannibalization of Brand Keywords

ad cleared the auction reserve price, comparing conditions “Cap 0” and “Cap 1” effectively

replicates the eBay experiment. We have such experiments for tens of thousands of brands,

but focus our analysis on the top 2,500 brands, which account for 98.7% of brand searches.

Even within this group, firms vary widely in terms of size and other brand quality metrics.

We start by measuring the extent that organic search results serve as substitutes for paid

search links. We find that the results in Blake et al. (2015) are an outlier. In our sample of

firms, sponsored links on branded search queries drives significant incremental traffic—total

clicks to a firm’s website increases by 2-3 percentage points, or about 5–10% baseline CTR

of the organic link. This effect is significantly larger for lesser known brands, while the

strongest brands in our sample show effects closest to that of eBay.

Although incremental traffic is non-zero, “cannibalization” or “crowd out” of free (organic

link) clicks is generally an order of magnitude larger than the casual impact of the ad. Given

the large crowd out, the nominal “cost per click” (CPC), the standard pricing metric reported

by online advertising platforms, and effective cost per incremental click (CPIC) strongly

diverge. When the firm does not enjoy a high position in the organic ranking, which is true

of competing firms on brand queries and most firms on generic commercial queries, these two

quantities are roughly equal. In contrast, all the brand firms in our study occupied the top

organic position. We measure the wedge between CPIC and CPC for a broad cross-section

of firms advertising on brand keywords in the absence of competing bidders that clear the

reserve price. In this case, the vast majority of the time the focal brand is the only ad shown,

making is easy to form the counterfactual of what would happen if they stopped advertising

(at least in the short-run).

For firms that show a significant ad effect, CPIC is on average 11 times larger than the

CPC. For all firms, we establish lower bound of 22 times CPC. While these multipliers are

large, it is not immediately clear what conclusions we should draw since the pricing rule

of the generalized second price (GSP) auction rewards link relevancy and clickability with

lower prices—the focal brand almost always pays a much lower CPC than competitors. To

put our CPIC estimates in context we compare them to the CPCs focal brands pay on

keywords when they do not occupy a high organic position (meaning CPC represents the

real price) and their bids for brand keywords. CPIC exceeds CPC on non-brand keywords

in 90% of cases and their bid in the brand keyword auction in 73% of cases. Since we cannot

3

Page 4: Competition and Cannibalization of Brand Keywords

observe the potential differences in the value of an ad click versus an organic link click,1 nor

the overall value of a click (given bid shading in the GSP Gomes and Sweeney (2014)), we

cannot definitely conclude that firms do not fully understand average versus marginal CPC,

but the evidence does strongly point in that direction.

All of the results presented so far have focused on brands that do not face aggressively

bidding competing firms. Bidding on a competitor’s brand keyword may be an attractive

strategy to “steal” traffic from the focal brand and it is quite common in our data. We

find when the focal brand continues to occupy the top slot, competitors can steal 3-5% of

the focal brand’s traffic by bidding into the second through fourth slots. Weaker brands are

more susceptible to competitors and indeed face competition more frequently.

This competitive siphoning, while statistically significant, might be considered relatively

modest, especially for stronger brands. However, these ads have a much larger impact

on cannibalization rates. When facing no competitors (in our experimental condition that

eliminated them), 50% of clicks to the focal brand’s website go through the paid link and

the remaining half are free clicks on the organic link. Holding the set of firms fixed, the

fraction of paid clicks increases to 61% when one competitor is present; 73% when two

competitors are present and 80% when 3 competitors are advertising. So while competitors

might steal relatively few clicks, they cause the brand to pay for substantially more. A

natural explanation is that more competitors creates a larger ad space pushing the brand’s

organic link down the page into a less visually prominent position.

Given that the focal brand has to pay for a large fraction of clicks, many of which are

non-incremental, why would it continue to advertise? In our final set of results, we look

at cases where the focal brand is not present but competitors are. These firms come in

two varieties. The first are firms that produce a product and sell it themselves, but other

retailers sell it as well, such as a laptop computer. These firms typically only get about 30%

of total clicks on brand searches. The second variety is the exclusive seller, such as a travel

website or car insurance company. These firms typically get upwards of 80% of total clicks.

It turns out that, in both cases, competing advertisers garner 20% of total clicks—far higher

than the 3-5% we observed earlier when the focal brand was present in the top slot. For

the first type of firms, only about one-third of this 20% comes from substitution from the

1We note that more than 90% of firms have the same “landing page” (where a user goes after clicking onboth.)

4

Page 5: Competition and Cannibalization of Brand Keywords

focal brand’s links. The rest come from clicks that would have gone to other firms present in

the organic links or from searches that would have resulted in no clicks. In contrast, for the

second type of firm, much like in the case of firms who bid on their own keywords, nearly all

the competing ad clicks come from the focal brand.

These results indicate that the returns to advertising for the focal brand depends critically

on the presence of competitors. In the absence of competitors, CPIC typically exceeds the

nominal CPC a focal brand pays on other keywords. Computing CPIC for firms that occupy

the top slot in the presence of competitors is more difficult because it requires us to estimate

a counterfactual we cannot directly observe at the firm level. Nonetheless, using the average

estimates from other firms predicts that firms of the second variety would lose a substantial

fraction of traffic—roughly one-fourth of total clicks—if they discontinued their ad. The

effects on firms of the first variety are more muted.2

The evidence brings us to something of a resolution to the question posed at the outset.

When competing bidders are present, occupying the top slot can effectively, but not com-

pletely, fend off competition. For these brands, advertising is primarily defensive. As more

competitors enter, the cost of this defense rises because cannibalization increases, giving

rise to an interesting tradeoff firms must navigate. When no competitors are present, the

paid ad does causally increase total clicks, but effect sizes are modest and decrease with a

firm’s brand capital. In this case, roughly speaking, for every 8 clicks a firm pays for, only

one represents a causal increase. We saw that the effective real cost typically exceeds other

quantities, such as the bid and the firm’s CPCs on non-brand keywords, that we would not

expect it to if firms behave rationally, although this is admittedly a highly tentative con-

clusion. Without the defensive value induced by competition, the real cost of these clicks

appears quite high and the overall returns are, at the very least, questionable.

2 Background and Context

Early experimental evidence, Reiley et al. (2010), showed that organic links and ads are

substitutes for each other. This substitution pattern is overwhelmingly present in our ex-

2Indeed the returns to these firms is more complex, since “competiting firms” are advertising the productthe focal brand produce, although they almost always sell other brands on their website.

5

Page 6: Competition and Cannibalization of Brand Keywords

perimental data as well, and has also been found in structural work (Jeziorski and Segal,

2014).3 Reiley et al. (2010) further show that more ads can increase total CTR on the top

slot because organic links act as (slightly better) substitutes for ads in their study.

We define brand keywords as queries that consists of a trademarked term and where the

trademark holder occupies the top organic slot. Competitors using a trademarked term to

guide their bidding is a contentious practice as the focal brand dislikes the fact that their

competitors can target a user that has expressed a particular interest in them. A firm may

use other forms of advertising to generate searches for their products, as shown in Lewis

and Nguyen (2014), and not want this traffic prone to competition. Indeed, firms have

sued Google, claiming trademark infringement, but their core claims have consistently been

rejected by the courts.4 In 2009, Google began allowing resellers to use trademarked terms

in their ads. Chiou and Tucker (2012) study this change and find that this change did not

damage the focal brand because it made the competing resellers less distinct.

2.1 Experiment and Data Description

The data in our study come from a fully randomized experiment on the Bing search engine.

On Bing, the sponsored listings that appear at the top of the page, above the organic listings,

are known as the “mainline.” A maximum of four mainline ads could be shown on a given

query – the same practice employed by Google. Clearly, absent an experiment, the number

of ads and their composition is endogenously determined by firms’ bids. A cross sectional

regression that looked at differences in the number of advertisements by keyword would

conflate true effectiveness of advertisements with differing environments across keywords.

We use our experimental variation to control for these confounding factors.

The experiment, conducted on a fraction of U.S.-located users over nine days in January

of 2014, randomized searches into one of five conditions. Four of these conditions had some

treatment, and corresponded to limiting the maximum number of mainline ads at 0, 1, 2,

3One paper, Yang and Ghose (2010), sharply diverges from all other papers we are aware and uses astructural model to assert that there is a positive interdependence between organic ranking and searchclick-through rate.

4Competing ads cannot use trademarked terms in their ad text, as would reasonably confuse a consumer.For example, while travel website Expedia is free to bid on “priceline,” it cannot include “priceline” in theirad text. This rule does not apply to licensed resellers, however.

6

Page 7: Competition and Cannibalization of Brand Keywords

and 3. The fifth condition was a control group, and corresponded to the maximum of 4

mainline ads. The control group corresponded to the situation that would have occurred in

the absence of the experiment. We note that just because a treatment group limited the

number of ads that could be shown, that doesn’t necessarily correspond to that number

of ads that actually appeared. For instance, in the treatment group that limited mainline

ads to a maximum of 3 (cap 3 to employ the terminology we will use throughout), if there

were not enough bidders that met the reserve price to fill the 3 slots, then fewer than 3 ads

were shown. We carefully control for this issue by selecting only queries that matched into

bidding data where an ad would have been shown in the absence of the experiment. See the

Appendix A for more detail on this process.

To identify brands, we extracted 87,000 retailer and brand names from the Open Direc-

tory Project.5 Search is characterized as brand search if and only if (1) the query is in this

list, meaning it is a verified firm brand, and (2) the query matches the domain name in the

first organic position. We focus only on brands that are in the first organic link because this

selects true brand keywords. Queries that generate brands that are not in the first organic

position might be searches of a different nature, perhaps not meant to get directly to the

brand page, but to a broader set of sites. This restriction assures us that we are getting true

brand search behavior and represents an underestimate of the true set of brand searches that

may be going on.

Figure 1 provides an example of a brand search query. The queries are simplified using

standard techniques, e.g. we treat macys.com, macys, www.macys.com and macy’s as the

same query. We focus on searches with 0 or 1 clicks on the page, ignoring queries with 2 or

more clicks.6

The majority of brands have very few exposures in the dataset. Table 1 presents the

number of brands binned by the number of observations for those brands. E.g., 64.7% of

all brands in the control group have less than 10 exposures but represent only 0.19% of all

traffic.

The large majority of traffic is generated by a relatively small subset of the total brands:

Almost 96% of traffic comes from the 1045 brands that have 1000 or more exposures. In

5dmoz.org, the project uses volunteer annotators to “classify the web.”6In these rather rare occurrences, the searcher often visits all advertisers, making it less interesting to

study. Further, search engines often refund clicks from such patterns.

7

Page 8: Competition and Cannibalization of Brand Keywords

Figure 1: Brand search example

This example has two mainline ads: own brand ad in mainline 1 and competitor’s ad in mainline2.

8

Page 9: Competition and Cannibalization of Brand Keywords

Table 1: Majority of brands have few exposures and less own ads in mainline 1(based on Control condition)

Number of Number of Percentage Percentage Percentage Percentageexposures brands of brands of traffic of own ads of competitor’s adsin Control in Control (%) (%) in ML1 (%) in ML1 (%)

1 4869 23.1 0.02 3 30.62 2773 13.1 0.02 4.1 32.43 1686 8 0.02 6.3 30.8

4 - 10 4315 20.5 0.12 10.2 34.511 - 100 4200 19.9 0.64 19.8 34.6

101 - 1000 2202 10.4 3.6 42.64 28.5> 1000 1045 5 95.6 43.8 13.6Total 21090 100 100 14.4 31.4

Percentage of ads is computed across companies. For example, companies with 4 exposures andcompanies with 10 exposures are given the same weight in group 4-10. Total frequency is also

computed across companies, unweighted.

particular, the starting point for our sample selection is the 2517 companies with over 350

exposures.7 These firms offer us relatively precise estimates and cover 98.7% of the market

activity.

In Figure 12 (Appendix A) we show that for this sample, 32.7% of companies advertise on

their own brand keywords more than 90% of the time. 39% of brands do not advertise on their

own keyword at all. The remaining brands advertise selectively, turning brand advertising

on and off, or (more likely) advertising in some geographic regions but not others.

3 Results: Causal Effect of Ads

In this section we look at how effective advertising on one’s own brand keyword is in driving

traffic to a brand’s website.

7With this selection rule we are balancing the number of firms against the inclusion of brands that don’tprovide meaningful information because they are so small. We have done substantial robustness around thisthreshold and very little is affected.

9

Page 10: Competition and Cannibalization of Brand Keywords

3.1 In the absence of competing ads

We begin by looking at the case of firms that choose to advertise on their own keyword

over 90% of the time. For the 2517 high exposure companies, this case corresponds to

32.7% of companies.8 For 42.7% of this set (352 companies), competitor is occupying the

second paid slot less than 20% of the time. This represents the most extreme form of

potential substitution between paid and organic search links because the link directly below

the mainline advertisement goes to the brand’s website and there is no competitor bidding to

disrupt this substitution. To estimate effects for this case, we compare our treatment group

that limited advertising to one mainline ad (cap 1) and this ad went to the brand website

that was bidding on the paid link to our treatment condition where no advertisements (cap

0) were shown for the same set of queries. Effectively we forcibly and experimentally remove

the paid search advertisement despite the fact that the firm was still bidding on it. Our

primary measure of interest is the probability that an individual arrives at the website of

the searched brand. We estimate this probability based on a user’s behavior on the search

page – clicks to any own organic or paid link on the first search page indicates arrival at that

brand’s website. No click after the search or clicks on other organic links indicates that the

user did not arrive to the brand’s website.

Figure 2 plots the probability that a user arrived at a brand’s website either from organic

or paid search links by the cap 0 vs. cap 1 groups. As shown, advertising on one’s one

keyword drives an incremental 2.27% of traffic to a brand’s website. This average effect is

statistically significant, although perhaps economically small.

Next, we look at whether this effect varies by observables in the data. We focus on

a subsample of companies with sufficient amount of traffic to get reliable company-specific

estimates of the probability to get a click9. This provides us with a sample of 493 companies.

Figure 3 presents an estimate of the density of the estimates. The average effect is 0.0214,

and the average standard deviation of the estimate is 0.0488. We plot the corresponding

normal density over the distribution of effects. The distribution of effects has heavier tails,

which suggests there is heterogeneity. We test it formally and reject the hypothesis that

8If we focus on companies that advertise on their own keyword at least once, these companies correspondto 53.6% of companies and 19.6% of traffic

9We keep companies with more than 80 exposures in each condition.

10

Page 11: Competition and Cannibalization of Brand Keywords

Figure 2: The Effect of Advertising on One’s Own Keyword

there is no heterogeneity10.

What is the source of this heterogeneity? We first ask whether the effect is more incre-

mental for firms with brand queries in more competitive environments. In many cases, firms

bid on their own keywords but competitors also bid on them. We might expect that in this

environment the incremental effect of having one’s own advertisement in mainline 1 would

be more incremental than compared to queries where competitors don’t bid. Importantly,

regardless of the number of bidders, we are comparing a condition where only at most one’s

own ad appears in mainline 1 – we’ve experimentally eliminated the competitors ads for the

purposes of this comparison, and are simply comparing the cap 0 to the cap 1 condition

for firms that would have had competitors show up on their keyword in the absence of the

experiment. Below, we systematically explore the effect of adding the competitors’ ads back

in. Here we are simply looking at the selection effects across different types of firms.

Table 2 divides the sample into groups depending on the probability (across queries and

time) that a competitor would have ended up in mainline 2 in the absence of the experiment.

The table shows that there is clearly significant heterogeneity in the incremental effect of

10We perform a series of standard tests for normality, including Shapiro-Wilk, Jarque-Bera, DAgostinoand other tests. All of them reject normality of the distribution.

11

Page 12: Competition and Cannibalization of Brand Keywords

Figure 3: There is heterogeneity in the effect of own ad in mainline 1

advertising on one’s own keyword. For firms whose competitors never bid on their keyword,

the incremental effect is around 1.4%, whereas for companies who almost always (above 90%)

have competitors bid on their keyword, the effect is around 4.9%.

We can also cut the data by the number of other competitive bidders for a given query.

Table 3 shows the estimates binned by the number of bidders on a brand’s keyword. The

similar pattern emerge: effect of one’s own brand ad in mainline 1 increases from 1.4% to

2.7% as the number of bidders increases.

Table 4 provides regression results of competition on the effect of own brand ad in mainline

1. There is a significant correlation between the frequency of a competitor’s ad in mainline

2 and the magnitude of the effect. The number of bidders does not correlate significantly

with the effect of own brand advertisement.

A second observable descriptor is the amount of brand capital that a firm has. Blake et

al. (2015) showed that bidding on one’s own brand ad does not have much of an effect for

eBay, a company with substantial brand capital. With our broader sample we can explore

how this effect differs with the level of brand capital of the firm.

12

Page 13: Competition and Cannibalization of Brand Keywords

Table 2: Effect is higher for brands with more competitors on their keywordsTotal Competitors’ ads frequency in mainline 2

0-0.1 0.1 - 0.25 0.25 - 0.75 0.75 - 0.9 0.9 - 1

Ncomp 493 192 99 125 34 43

pnoad 0.779 0.811 0.790 0.756 0.739 0.708(0.001) (0.002) (0.003) (0.003) (0.005) (0.005)

pad 0.800 0.825 0.808 0.776 0.783 0.757(0.001) (0.001) (0.002) (0.002) (0.004) (0.003)

pad − pnoad 0.021 0.014 0.018 0.020 0.044 0.049(0.002) (0.002) (0.003) (0.004) (0.007) (0.006)

whereNcomp - number of companiespnoad - probability to get to brand’s website when there are no adspad - probability to get to brand’s website when there is own ad in mainline 1 and no competitor’sads

To investigate the relationship between brand capital and the effect of brand advertising,

we examine the relationship between the own brand effect and measures of brand capital

collected from Alexa.com. In particular, we collect US and global rankings, bounce rate,

the fraction of traffic from search, pages viewed per day and time spend per day. Table 10

(Appendix B) provides a summary of brand capital measures.

Table 5 presents the results of four regressions of own brand ad effect on brand capital

measures. Specification (1) shows that companies with higher rankings tend to have a lower

incremental effect of one’s own ad in mainline 1. On average, the effect of an own brand

ad for a well-known company like amazon.com or alibaba.com (log(US) ranking around 2.5)

is 3 percentage points lower than for an unknown brand (log(US) ranking around 11). As

we add more controls for our brand capital measure in specification (2), the coefficient on

log ranking is still significant. In specification (3) we add controls for how much space the

brand links have on the webpage: the number of “deep links” (sub links below the main

link), the number of “dcards” (detailed informational panels including things like maps and

ratings), and the number of organic links of one’s own brand on the first page11. Finally, in

specification (4) we also control for the level of competition.

We conclude that brands with higher brand capital gain less from advertising on their

11For a description and examples of deeplinks and dcard please see Appendix C

13

Page 14: Competition and Cannibalization of Brand Keywords

Table 3: Effect is higher for brands with more bidders on their keywordsTotal Quantiles of number of bidders

0-0.1 0.1 - 0.25 0.25 - 0.75 0.75 - 0.9 0.9 - 1

Ncomp 493 48 74 246 74 51# of bidders 11.55 3.7 6.22 10.88 16.72 22.42

pnoad 0.779 0.819 0.807 0.780 0.743 0.749(0.001) (0.003) (0.003) (0.002) (0.004) (0.005)

pad 0.800 0.833 0.825 0.802 0.766 0.776(0.001) (0.002) (0.002) (0.001) (0.003) (0.003)

pad − pnoad 0.021 0.014 0.018 0.022 0.023 0.027(0.001) (0.004) (0.004) (0.002) (0.005) (0.006)

Table 4: Companies with higher competition have higher own ad effect

Dependent variable:

Effect of own ad in ML1

(1) (2) (3)

Frequency of competitor in ML2 0.032∗∗∗ 0.034∗∗∗

(0.007) (0.007)Number of bidders 0.001 −0.0002

(0.0004) (0.0004)Constant 0.012∗∗∗ 0.014∗∗∗ 0.013∗∗∗

(0.003) (0.005) (0.005)

Observations 493 493 493R2 0.043 0.005 0.044Adjusted R2 0.041 0.003 0.040

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

own ad in mainline 1. This could be either because consumers are less likely to substitute

away from going to a firm’s website when that firm has higher brand capital or because the

organic link of these firms take up more room on the page, thus making the organic link

more of a substitute for the paid link. In either case, the implication for managers remains

the same – firms with higher brand capital benefit less from bidding on their own brand

keywords.

14

Page 15: Competition and Cannibalization of Brand Keywords

Table 5: Companies with higher brand capital have lower own ad effect

Dependent variable:

Effect of own ad in ML1

(1) (2) (3) (4)

log(us) 0.004∗∗ 0.009∗ 0.006 0.006(0.001) (0.005) (0.005) (0.005)

Deeplinks −0.002∗ −0.002(0.001) (0.001)

Dcard −0.002∗∗∗ −0.001∗∗

(0.001) (0.001)Number of own organic links −0.001 −0.0002

(0.002) (0.002)Frequency of competitor in ML2 0.017∗∗

(0.008)Constant −0.008 0.024 0.062∗∗∗ 0.052∗∗

(0.012) (0.020) (0.024) (0.024)Other brand capital controls No Yes Yes Yes

Observations 492 492 492 492R2 0.013 0.038 0.079 0.089Adjusted R2 0.011 0.026 0.062 0.070

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

15

Page 16: Competition and Cannibalization of Brand Keywords

3.2 In the Presence of Competition

First, we look at situations where a competitor is the only bidder that clears the reserve

price and they therefore end up in the mainline. This is a selected sample because only firms

that do not bid on their own keyword are included in it.12 However, for this set of firms we

can use the experiment to look at what happens when a competitor’s ad appears in mainline

1 but using the experimental variation that removes that competitor’s ad. Second, we look

at the situation where, conditional on competitors bidding and ending up in mainline slots

2-4, we experimentally removed those ads.

3.2.1 Competitors in the Top Position

We select the set of firms who do not bid on their own keywords but where a competitor

does and explore the effect that the competitive advertisement has on traffic both to the

competitor’s website and to the firm’s own website (via the unpaid organic link). We note

that this is clearly a selected group of firms. And in particular that firms that choose to not

“protect” their own brand keywords by bidding on them are fundamentally of a different

type. For instance, consider the case of brand keywords for upstream firms that do not

have their own web retail presence. In this case, the mainline slots for these keywords might

be bid on by downstream retailers that can actually sell the product. Consumers in this

particular case might not be interested in going to the brand’s own website. Rather, they

are interested in shopping for the product. A firm advertising on its own keywords wouldn’t

make any sense.

One marker of this form of heterogeneity is the probability, in the absence of any adver-

tisements (a case created by the cap 0 condition in our experiment), that consumers click on

the firm’s organic unpaid link. In the case where consumers don’t click on the organic link

for the brand, even in the absence of advertisements, we might think that they are typing

in the brand keyword in an attempt to get to a different type of website. In the case where

we can match the probability of clicking on the organic link between firms that advertise on

their own keywords and those that don’t, we might think that we are somewhat effectively

controlling for some of the heterogeneity.

12Recall that one of the downsides of the experiment is that we were not able to place ads in situationswhere they wouldn’t have existed otherwise – we were only able to subtract them.

16

Page 17: Competition and Cannibalization of Brand Keywords

Figure 4 displays the probability of a consumer arriving (through an unpaid link) on a

brand’s website for the 49 firms that fit the criteria described in the data description, who do

not bid on their own keywords, and who have a competitor in mainline 1 above 90% of the

time in the control group for our experiment. It’s clear that 1) The probability of arriving

at a firm’s website through an organic link is lower for these firms on average than for firms

that advertise on their own keyword and 2) That there is substantially more heterogeneity

in this effect relative to those firms that advertise on their own brand keyword13.

Figure 4: Heterogeneity in firms that do not bid on their own brand keywords

We now explore the causal effect of a competitor appearing in the mainline for this set

of firms. Based on the data in figure 4, we divide the sample into firms that receive a low

percentage of traffic on their own keywords (to the left of the first red line) and those that

receive a high percentage (those to the right of the second red line). We surmise that the

second of these groups is more like firms that do advertise on their own keywords although

note that there is still selection into this sample.

Figure 5 graphs the difference in the number of clicks on the brand’s own organic link and

the competitor’s ads across the cap 0 (no advertisements) and cap 1 (only the competitor

appears in the mainline area).

13Figure 14 in Appendix E shows a similar histogram for 493 companies which advertise on their keywordmore than 90% of the time

17

Page 18: Competition and Cannibalization of Brand Keywords

Figure 5: The incremental effect of bidding on one’s competitors’ keywords

In the case of low organic traffic (the left-hand panel), on average, 30% of clicks in the

cap 0 condition (when there are no ads) end up at the brand’s website. In contrast, in the

high organic traffic case (the right-hand panel), on average, around 80% of clicks are on

the brand’s website – a number very similar to the case when firms do advertise on their

own keyword discussed above. When we experimentally add the competitor’s advertisement

(the cap 1 case), two things happen. First, the percentage of traffic to the brand’s website

drops. For the low organic companies it drops by 8 percentage points. For the high organic

case by about 17 percentage points. This is traffic that a competitor is “stealing” from the

branded website. Second, the competitor gains a large number of clicks – around 20% in

either case. In both cases the number of clicks that the competitor gains is larger than the

amount that is lost by the brand website indicating that there is an overall increase in clicks

to the combination of mainline advertisements (in this case 1) and the top organic link (the

brand’s own website). Some traffic was diverted from links further down on the page or from

clicks that wouldn’t have happened in the absence of the competitive advertisement.

18

Page 19: Competition and Cannibalization of Brand Keywords

3.2.2 Effect when competitors occupy positions 2–4

Conditional on a the focal firm winning the brand keyword auction, what is the effect of

competitors in mainline slots 2, 3, and 4? Do competitors capture any traffic and does that

decrease the probability of clicks on the brand’s paid or organic links? Again, we deal with

a bit of a selected sample – keywords where both the brand and competitors are already

bidding. But conditional on this selection, we use the experimental variation to remove each

of the competitive advertisements in the mainline slots.

Figure 6 displays the causal effect of competitive bidding in each of the mainline slots.

We narrow our sample down to firms where, in the absence of the experiment, 4 mainline

ads would show up. We then systematically remove each one of those advertisements and

ask what happens to traffic to the brand’s website. Looking at the left panel, the first point

in the graph displays the total traffic that goes through both the organic and paid link to

the brand’s website. This is the experimental condition where all other advertisements were

removed. The second through fourth points display the traffic to the brand’s website adding

in competitors into the mainline slots 2 through 4, respectively. Going from no competitive

advertisements to 3 (slots 2-4) reduces the traffic to a brand’s own website by around 4

percentage points. In the right panel we separate advertisers by a median split of website

traffic. The lower, blue line represents firms below median size and the red above. We can

see the smaller firms have a lower baseline CTR and a slightly greater loss of clicks when

facing 3 or 4 competitors.

Figure 6 demonstrates that competitive advertising can affect traffic to a brand’s own

website, but does this benefit the competitors who are bidding? Figure 7 shows the traffic

that these competitive advertisements generate to the competitors’ websites. The top (red)

line shows the amount of traffic that is siphoned off by the competitor in the 2nd mainline slot.

The first point is for the situation where we place only 1 competitive advertisement (and the

brand’s own advertisement in the first mainline slot). The second point is the traffic that goes

to the competitor in the mainline 2 slot when there are 2 competitive advertisements, and the

3rd point for when there are three. Interestingly, adding more competitive advertisements

does not decrease the traffic that goes to the competitive advertiser in the 2nd mainline slot.

It stays static at around 2.5%. The 2nd (blue) line does the same thing for the advertiser in

mainline slot 3, and the green dot for the competitive advertiser in mainline slot 4. All told,

19

Page 20: Competition and Cannibalization of Brand Keywords

Figure 6: The effect of competitive ads in mainline slots 2-4

(a) Average effect (b) Upper and Lower 50% by Brand Capital (USRanking)

competitors in all slots are able to causally siphon off around 4.3% of traffic from the focal

brand by bidding on its keyword.

In this section we’ve documented some of the effects of competition on brand advertising.

For the sample of firms that we observe that do not bid on their own keyword and have a

competitor in mainline 1, this competitor receives around 20% of clicks for those keywords.

When we remove these advertisements, on average, clicks on the brand’s own organic link

increases by between 8 and 17 percentage points. These numbers provide some justification

for the common belief that firms should bid on their own keyword in order to protect their

brand keywords from competitors. For the cases in the data where competitors have this

opportunity, they gain a significant number of clicks at the expense of traffic to the brand’s

website. In the case where a brand does bid on its own keyword and therefore ends up in the

mainline 1 slot, we examine the effect that competitors have in slots 2-4. Here the evidence

is much more muted. Competitors gain around 5% percent of clicks, split between the 3

advertising competitors. Most traffic still goes to the brand’s website, either through the

paid link or the organic link.

20

Page 21: Competition and Cannibalization of Brand Keywords

Figure 7: The effect of competitive ads in mainline slots 2-4

4 Results: Cannibalization, CPIC, and Bidding

Up until now we have focused on the overall probability that a searcher ended up at either the

brand’s own website or a competitor’s site, combining the paid and organic links together.

But, of course, whether or not this traffic goes through the organic or paid links matters a

lot to the firms. If the traffic goes through the paid link, they must pay the search engine for

each click; if it goes through the organic link, it is free. Bidding on one’s own keyword not

only affects the overall probability that searches end up on the brand’s website (as described

above), it also affects the percentage of traffic that goes through each of these links on

the search results page. Furthermore, competitors not only steal clicks, they also push the

organic links further down on the page, forcing more traffic through the brand paid link.

This section documents both of these effects and argues that the right measure for thinking

about the success of one’s campaign is not ROI computed with the cost per click metric

(CPC), rather firms should be computing the cost per incremental click (CPIC). We define

and compute this metric and measure the difference between the CPIC and the CPC for a

wide variety of firms.

21

Page 22: Competition and Cannibalization of Brand Keywords

4.1 Cannibalization without Competitors

We earlier documented that without competitors brand advertising does generate incremen-

tal lift—on average around 2.27%, but that comes at the expense of driving a significant

portion of traffic through the paid link instead of the organic link. Figure 8 shows the

percentage of traffic going to one’s own brand website through the paid link in mainline

1 for the sample of always advertising on their keyword. On average, 46.3% of clicks go

through the paid link, with up to 80% for some companies. This corresponds to an average

cannibalization rate of 44.6%.

Figure 8: Histogram of the fractions of total clicks going through paid links

Figure 9 uses the experimental variation and compares the cap 0 (no advertising) con-

dition with the cap 1 (brand advertising on one’s own keyword) condition, emphasizing the

cannibalization that occurs when advertising on one’s own brand keyword. The difference

in the overall height of the bars between the cap 0 and cap 1 conditions represents the incre-

mental clicks to a brand’s website as discussed above (average effect of 2.27%). The drop in

the pink bar shows the amount that paid search cannibalizes organic clicks for our sample.

Half of the traffic now goes through the paid search and almost all of it would have ended

22

Page 23: Competition and Cannibalization of Brand Keywords

up at the brand’s website in the absence of brand keyword advertising.

Figure 9: Organic click cannibalization compared to incremental clicks

High cannibalization rates mean that companies are paying more than just the cost

per click (CPC) for traffic. Instead, to figure out the actual costs that firms are incurring

we need to compute the ratio of the incremental clicks that occur because of paid search

advertising to the overall amount that the firms spend on paid search. To compute the costs

per incremental click we need to form the counter-factual: what would happen if the firm

did not advertise. For firms that typically do not face competing ads, this can be computed

by a simple comparison of the Cap 0 to Cap 1 conditions. As such we restrict our analysis

to ad listings that face competition less than 20% of the time (the results are not sensitive

to this threshold). In this case, we can compute CPIC as

CPIC =xp1

(p1 − p0)CPC

where x is the fraction of clicks going through an ad, p1 is the probability of getting to one’s

own website with the own brand ad in mainline 1, and p0 is the probability of getting to

one’s own website without the own brand ad in mainline 1.

23

Page 24: Competition and Cannibalization of Brand Keywords

We compute the CPIC/CPC ratio based on estimates of x, p0 and p1 from the analysis

above. Notice that it is only possible to compute CPIC if there is a positive effect of own

brand advertisement on traffic to company’s website. In our sample, effect of advertisement

was positive and significant for only 16% of companies. This restricts us from computing the

average CPIC/CPC ratio across all companies. Instead, we compute CPIC/CPC ratio for a

company with average effect size. For this subset of companies, average effect of own brand

advertisement is 1.6%14, and average amount of traffic going to own brand website through

the paid link is 36%. The resulting estimate of CPIC to CPC ratio is 22.5. We notice that

this estimate is a lower bound on the expected CPIC/CPC ratio, given that the CPIC/CPC

function is convex in the advertisement effect, p1 − p015.

Additionally, we compute the CPIC/CPC ratio for companies which had significant effect

of own brand advertisement in mainline 1. This corresponds to a 43 companies. Figure 10

shows the histogram of the CPIC/CPC ratio estimates for these companies.

Figure 10: Histogram of CPIC/CPC estimates for firms that do not typically face competingads

14Effects for these companies are presented in columns 1 and 2 of Table 215We provide more details on this in the Appendix D

24

Page 25: Competition and Cannibalization of Brand Keywords

The median CPIC for the 43 companies with a significant impact of the ad is 9.1, and

the mean is 11.07. This means that for each marginal click, brands on average pay 9.1 times

their CPC. For companies with an insignificant effect of their own brand ad in mainline 1,

the CPIC/CPC ratio is either bigger (if we do not have enough power to show significance),

or does not make sense (if there is no effect).

Table 6: Summaries of CPC, CPIC and bids

Unweighted average Weighted averageAll brands Significant All brands Significant

Ncomp 268 43 268 43¯CPCown ($) 0.15 0.15 0.09 0.06

¯CPCcompet ($) 0.86 0.78 0.61 0.47¯CPCown−other ($) 1.36 1.43 1.14 0.92

¯bidown ($) 11.48 6.45 4.8 1.9¯bidcompet ($) 1.8 1.97 1.15 0.87

¯bidown−other ($) 2.42 2.26 1.96 1.49¯CPICown ($) 1.12 1.42

CPICown > CPCcompet (%) 51.16 92.73CPICown > CPCown−other (%) 55.88 90.62

CPICown > bidown (%) 34.88 73.72The weights are based on the number of exposures

¯CPCown−other is computed for about 85% of companies in the sample

Table 6 present the summary of bids and CPC. A number of things are apparent from

this table. First, the per click price on the focal brand’s own keyword is relatively low, the

average unweighted CPC is 15 cents (the average weighted CPC is 9 cents). But given the

amount of traffic that goes through these links, the amount of money involved is far from

trivial. Second, the average CPIC for the focal brand on one’s own keyword is much higher

– almost $1.42 per click. This is substantially higher than the $0.47 that competitors are

paying per click. Third, for 93% (51% unweighted) of the companies in the sample, the

CPIC that they are paying is higher than the CPC that their competitors are paying on

those same keywords.

All of this strikes us as being rather curious. Firms are spending a substantial amount of

money on brand keyword advertising when the CPIC calculations indicate that they would

actually pay less per incremental click for non-brand keywords. Similarly, firms are paying

more to attract these customers than their competitors are. Why would a firm pay so much

25

Page 26: Competition and Cannibalization of Brand Keywords

relative to the competition to maintain a customer’s click that is presumably already very

loyal? One explanation is that these customers are more valuable to the firm and that

firms are doing the right thing in spending large amounts on marketing to these customers.

Another explanation is that firms do not fully understand the difference between CPC and

CPIC. When they bid, they use CPC metrics given to them by the search engine.16 While

none of these numbers conclusively show that firms do not understand the difference between

CPC and CPIC, it starts to paint a picture that tips in that direction. We leave the unpacking

the underlying causes, such as principle agent problems, for future work.

4.2 Cannibalization with competitors

One effect of competition on brand keywords is that a competitor may steal some traffic

from the brand owner. Another is that paid search competition shifts the organic links

further down the page. This pushes more traffic through the paid search link for the brand,

increasing cannibalization. To the best of our knowledge this paper is the first one to point

this out and provides evidence for this additional effect that competition can have on its

rivals. For a competitor, simply bidding on a brand’s keyword can cause them harm through

this increased cannibalization channel, even if no customers click on the competitive link.

In fact, from the competitor’s perspective, it works better if no customers click on the ad

because then the competitor doesn’t pay anything.

To examine the magnitude of this effect, we go back to the set of keywords that, in the

absence of the experiment, would have generated 4 different mainline advertisements. We

then experimentally put ads in slot 1 (the brand advertisement) and slots 2-4 (advertisements

from competitors). This compares the experimental cap 0 through cap 4 conditions. In each

case we compare the total traffic that goes to the brand’s own website and the fraction of

that traffic that goes through the organic versus the paid link. Figure 11 plots these effects.

The height of the overall bar indicates the overall traffic that makes it to the brand’s

16We wouldn’t argue that firms don’t intuitively understand that some of this cannibalization is happening.The question is to what extent firms measure and adjust their bids based on it. Our sense, based on workingwith large firms on their keyword bidding strategies, is that many of them train machine learning models tomaximize ROI using the CPC and not the CPIC metric. If the difference between these two metrics is large,as we are arguing in this paper, then these machine learning algorithms will over-bid on brand keywordsrelative to non-brand (for a fixed marketing budget).

26

Page 27: Competition and Cannibalization of Brand Keywords

Figure 11: The effect of competition in increasing cannibalization

website. Even in the face of 3 competitive advertisements, consumers tend to make it to

the brand’s website. However, there is a dramatic difference in the percentage of that traffic

that goes through the paid link.

Table 7 extends the sample of firms to include those that would have only had a competi-

tive ad in the 2nd mainline slot, those that would have competitive advertisements in slots 2

and 3, and those that would have had ads in all 4 slots. The last column thus corresponds to

the numbers in figure 11. Having a competitor’s ad in mainline 2 increases the fraction of ads

going through the paid link of one’s own brand by 10 percentage points, having competitors’

ads in mainline slots 2 and 3 - by 9 percentage points more, and having competitors’ ad in

mainline slot 4 - by 6 percentage points more. The average fraction of paid clicks for one’s

own brand ad with 3 competitor’s in the mainlines is 0.843 which implies that over 4 out of

5 clicks on one’s own brand keyword are paid.

We conclude that competitors can hurt a brand not only by stealing clicks, but also by

increasing a brand’s costs. It might really make sense from a competitive standpoint to bid

on rivals’ keywords. If the competitor gets a lot of clicks, then their decision to advertise

27

Page 28: Competition and Cannibalization of Brand Keywords

Table 7: Competitors’ ads in mainlines 2, 3 and 4 increase cannibalization

Effect of competitors’ ads inML 2 ML 2-3 ML 2-4

Ncomp 784 682 564x1 0.539 0.566 0.595

(0.0029) (0.0042) (0.0133)x12 0.655 0.68 0.699

(0.0032) (0.0039) (0.013)x13 0.773 0.782

(0.0038) (0.0118)x14 0.843

(0.0091)

whereNcomp - number of companiesx∗ - fraction of clicks leading to own brand website through ad in mainline 1 with- x1 no competitors’ ads- x12 competitor’s ad in mainline 2- x13 competitors’ ads in mainlines 2 and 3- x14 competitors’ ads in mainlines 2, 3 and 4

will depend on the CPC and the quality of these clicks. However, even if competitors do not

get many clicks (imagine they get no clicks), they might want to advertise, as they would

increase the costs of the focal brand without paying anything.

5 Conclusion

In this paper, we’ve described and documented the competitive environment for brand key-

word advertising on Bing. Brand keyword advertising has an interesting property in that a

clear free substitute exists in close proximity to the paid advertisement. While this substi-

tute is more dramatic and obvious relative to other advertising situations, the general point

extends well beyond brand keywords. Consider, for example, the decision of a large firm

such as Amazon of whether or not to advertise on Bing. A consumer search process might

begin at a search engine but also include other websites that have Amazon advertisements

on them. Similarly, it might also include direct navigation to Amazon’s own home page.

In the absence of search advertisements, these other substitute channels might continue to

28

Page 29: Competition and Cannibalization of Brand Keywords

drive traffic, creating a wedge between the CPC of a search advertisement and the true

incrementality (CPIC) of those ads. Organic links provide a convenient lens to study this

wider phenomenon.

In the presence of substitute channels, questions about whether or not a firm should

advertise to either its own customers or try and advertise on the keywords of competitors

becomes more complex.

We demonstrated that advertising on one’s own keyword drives a significant amount

of traffic to one’s own website, but it comes at a cost – much of that traffic would have

flowed through the organic link in the absence of that advertisement. There is substantial

heterogeneity in the true incrementality of own brand keyword advertising. For large, known,

brands, this advertising drives little to no incremental traffic. For less well-known brands,

this number can be larger. In the face of this large amount of cannibalization, firms should

not look at the CPC of their advertisements, rather they should compute the cost per

incremental click. We demonstrated that the wedge between these metrics is quite large and

briefly discussed whether firms seem to understand this metric.

One reason for bidding on one’s own keyword is to prevent competitors from doing so

and appearing in the first mainline slot—“defensive advertising.” We demonstrated that

for a selected sample, when a competitor bids on its rival’s keywords, it can capture a

significant amount of traffic – around 20%. Most of that comes from traffic that would have

gone to the focal brand’s website through its organic link in the absence of the competitive

advertisement. Even when a focal firm is bidding and winning its own keyword, a competitor

can still significantly affect the search environment. By advertising on its rival’s keyword

and ending up in mainline slots 2-4, it can both steal some traffic (on average around 5%)

and, importantly, push the focal firm’s organic link further down on the page, driving a

much higher percentage of traffic through their paid link. This raises the costs of the focal

firm dramatically. This raising rival’s cost phenomena might apply to many other setting of

internet advertising and is a yet unexplored but we think interesting competitive element.

29

Page 30: Competition and Cannibalization of Brand Keywords

References

Chiou, L. and Tucker, C. (2012). How does the use of trademarks by third-party sellers

affect online search? Marketing Science, 31(5):819–837.

Gomes, R. and Sweeney, K. (2014). Bayes–nash equilibria of the generalized second-price

auction. Games and Economic Behavior, 86:421–437.

Jeziorski, P. and Segal, I. (2014). What makes them click: Empirical analysis of consumer

demand for search advertising. Technical report, Working papers, Haas School of Business.

Lewis, R. A. and Nguyen, D. T. (2014). A samsung ad for the ipad? dis-

play advertising’s competitive spillovers to search. Available at SSRN

http://dx.doi.org/10.2139/ssrn.2374414.

Reiley, D. H., Li, S.-M., and Lewis, R. A. (2010). Northern exposure: A field experiment

measuring externalities between search advertisements. In Proceedings of the 11th ACM

conference on Electronic commerce, pages 297–304. ACM.

Yang, S. and Ghose, A. (2010). Analyzing the relationship between organic and spon-

sored search advertising: Positive, negative, or zero interdependence? Marketing Science,

29(4):602–623.

30

Page 31: Competition and Cannibalization of Brand Keywords

6 Appendices

6.1 Appendix A: Matching with Auction Data

Experiment we use randomly restricts the number of possible paid links on top of the search

page. The control group corresponds do the default, which is a maximum of 4 advertisements

in the mainline (Cap 4). There are 4 experimental conditions: Cap 0, 1, 2 and 3. The idea

is similar to the control: e.g. Cap 0 does not allow any advertisements in the mainline, and

Cap 3 allows at most 3 advertisement in the mainline.

This design of the experiment restricts us to studying only the cases where advertisement

is eligible to be shown in the mainline, which means that in the absence of experiment

advertisement will be shown in the mainline. E.g. we cannot study the effect of advertisement

for a company that does not advertise on its own keyword: there will be no own brand

advertisements in both Cap 0 and Cap 1 condition.

Thus, we restrict our attention to cases where companies advertise. We are still facing a

challenge: if a company advertises only 50% of the time on a given query and selects search

traffic where the effect will be higher17, we cannot compare occasions with advertisement to

the treatment condition where the ad will be removed. E.g. if we would like to estimate

the effect of own brand advertisement in mainline 1 when in 50% of the cases company

advertises, and in 50% of the cases there is no advertisement, comparing occasions in Cap 1

conditions with a paid link shown to the entire Cap 0 conditions will bias the estimates.

To find the right treatment group, we need to allocate the occasions where ad was actually

removed from the mainline. In the example above, we would like to compare occasions with

own brand paid link in Cap 1 to occasions in Cap 0 when own brand paid link would have

been shown. To find such occasions, we collect the auction data for the search queries in the

experiment. The allocation of positions in mainline follows the standard GSP auction rules:

players submit the bids for price of a click, platform computes the ”rankscore” of a given

player, and players are allocated the positions in mainline based on their rankscores. Given

that the reservation level is cleared, company with the highest rankscore gets position 1,

company with the second highest rankscore gets position 2, etc. Rankscore is proportional

17E.g. using geo-targeting

31

Page 32: Competition and Cannibalization of Brand Keywords

to the bid and a probability of click on the ad as computed by the platform

RSj ∝ bjpclickαj (1)

where bj is a bid of company j, pclickj - probability of company j to get a click, and α is

the tuning parameter.

This implies that knowing the rankscores of bidders and reservation level for a search

query allows to say which advertisement would be shown in the mainline in the absence of

the experiment. To get this information, we exploit auction data collected by the advertising

team. Experiment that we use was designed by removing the potential advertising slots from

mainline, but bidding data was still collected.

Table 8: Summary of matching of the experimental and auction data

Condition Searches Searches % of eligible ads inTotal Matched Matched, % ML 1 ML 2 ML 3 ML 4

Cap 0 3162615 1506827 47.6 30.6 9.2 4.6 2.4Cap 1 6342073 3568054 56.3 41.8 11.2 4.7 2.9Cap 2 6338914 3568918 56.3 41.8 10.9 6.1 3.2Cap 3 6348311 3577819 56.4 41.9 11 5.7 4.1

Control 22209220 12506083 56.3 41.9 11 5.7 3.5

Table 8 presents the summary of matching experimental data and collected auction data

for Cap 0-4. For Cap 1, 2, 3 and 4, around 56.3% of search queries in experimental data

where matched with the auction data. A search query will not be recored in the auction

data if no advertiser submitted a non-trivial bid18, so the unmatched data can correspond to

queries with no bidders.The majority of unmatched queries correspond to occasions where

no advertisements were shown, which supports this explanation.

Cap 0 condition has a higher percentage of unmatched search queries. This indicates

the problem with the matching, given that the experiment was constructed to be balanced

between the treatment and control groups. We further find that percent of advertisement

eligible for the mainline 1 position in Cap 0 is substantially different from the percent of

advertisements eligible for mainline 1 position in Cap 1, 2, 3 and 4.

18As defined by the platform

32

Page 33: Competition and Cannibalization of Brand Keywords

This creates a potential problem for using the occasions with eligible brand advertise-

ments for Cap 0 condition. Consider the case of estimating the effect of own brand ad-

vertisement in mainline 1. Using the matched data, we would like to compare occasions

with own brand advertisement from Cap 1 condition to occasions with eligible own brand

advertisement from Cap 0 conditions. We know that some occasions with eligible own brand

advertisement are missing from Cap 0. If this mismatch is correlated with the probability

of a click on own brand weblink, our estimate of advertisement effect will be biased.

To check if there is a selection problem in Cap 0 matching, we estimate the effect of own

brand advertisement in mainline 1 for companies which always advertise in mainline 1 on

their own keyword19. For these companies, comparison of Cap 0 to Cap 1 provides the casual

effect of own brand advertisement: we know that, if not the experiment, search results in

Cap 0 will have own brand advertisement in mainline 1. We also can estimate the effect

using only eligible own advertisement occasions in Cap 0 and Cap 1. If the estimates of the

effect based on two methods are different, we can confirm that the occasion in Cap 0 which

have eligible own brand advertisement in mainline 1 are correlated with the probability of

click on own brand website.

Table 9: Effect of own brand ad in mainline 1 is significantly underestimated when usingeligible ads

All queries When own ad is eligibleNcomp 391 391pown0 0.7867 0.8115

(0.0022) (0.0029)pown1 0.8035 0.8179

(0.0015) (0.0015)pown1 − pown0 0.0168 0.0063

(0.0027) (0.0033)pown0 is the probability of a click on own brand link in Cap 0

pown1 is a similar probability in Cap 1

Table 9 presents the estimation results. Advertisement effect estimate based on all traffic

is 1.68 percent points. Advertisement effect estimated based only on traffic with eligible

own advertisement is 0.63 percent points. The difference in two estimates is statistically

19Companies that have own brand advertisement in mainline 1 more than 99% of the time

33

Page 34: Competition and Cannibalization of Brand Keywords

significant20.

We thus confirm that the occasions of eligible own brand ads in mainline 1 in Cap 0 are

correlated with the probability to get a click on own brand weblink. This restricts us from

using the eligible advertisements occasions to compare Cap 0 and Cap 1. Instead, we focus

only on companies that have a paid link in mainline 1 more than 90% of the time. For these

companies, comparison of Cap 0 and Cap 1 gives casual effect of advertisement.

Figure 12 shows that around 50% of companies advertise at least 10% of the time, with

around 33% advertising more than 90% of the times. Restricting the analysis to the later

group gives us 824 companies which always advertise on their keyword.

20Difference in estimates is 0.0105, with standard error of the difference being 0.0043, which correspondsto a t-stat of 2.46

34

Page 35: Competition and Cannibalization of Brand Keywords

Figure 12: Frequency of ads in mainline 1 for 2517 most popular brand queries

(a) Frequency of own ads in ML1 (b) Frequency of competitor’s ads in ML1

(c) Frequency of ads in ML1

35

Page 36: Competition and Cannibalization of Brand Keywords

6.2 Appendix B: Brand Capital Measures

Table 10: Brand capital measures

Mean Standard deviationlog(expos) 7.59 1.41

log(Rankglobal) 10.52 2.14log(RankUS) 9.05 1.96

bounce rate, (%) 35.1 14.3time spent per day (minutes) 4.6 2.98

pages viewed per day 4.5 2.76search traffic (%) 19.85 9.19

36

Page 37: Competition and Cannibalization of Brand Keywords

6.3 Appendix C: Deeplinks and Dcard Example

Figure 13: Deeplinks example

37

Page 38: Competition and Cannibalization of Brand Keywords

6.4 Appendix D: Lower bound on CPIC/CPC

We compute the CPIC/CPC ratio for a company with average effect size. This estimate

gives a lower bound on the average CPIC/CPC ratio as CPIC/CPC is a convex function in

the effect size:

CPIC/CPC =xp1

p1 − p0Fix the xp1 and assume the effect size p1 − p0 > 0. Denote the LHS of the CPIC/CPC

equation as f(∆(p)), where ∆(p) = p1 − p0 Then, by Jensen’s inequality, E(f(∆(p))) >

f(E(∆(p))). Hence, computing CPIC/CPC in the average effect point E(∆(p)) gives a

lower bound on the expected value of CPIC/CPC.

38

Page 39: Competition and Cannibalization of Brand Keywords

6.5 Appendix E

Figure 14: Probability to get a click for firms that bid on their own brand keywords

(a) Cap 0 (b) Cap 0 versus Cap 1

39

Page 40: Competition and Cannibalization of Brand Keywords

6.6 Appendix F: Competition and Brand Capital Relationship

Table 11: Companies with higher brand capital have less competition

Dependent variable:

Frequency of competitor in ML2

(1) (2)

log(us) 0.052∗∗∗ 0.028∗∗∗

(0.009) (0.009)Deeplinks −0.023∗∗∗

(0.008)Dcard −0.022∗∗∗

(0.003)Number of own organic links −0.039∗∗∗

(0.012)Constant −0.126∗ 0.483∗∗∗

(0.075) (0.109)

Observations 492 492R2 0.065 0.216Adjusted R2 0.063 0.209

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

40