Electronic Commerce Research and Applications 9 (2010) 488–494

Contents lists available at ScienceDirect

Electronic Commerce Research and Applications

journal homepage: www.elsevier .com/locate /ecra

An analysis of the importance of the long tail in search engine marketing

Bernd Skiera a,*, Jochen Eckert b, Oliver Hinz c

a Department of Marketing, Chair of Electronic Commerce, Goethe-University of Frankfurt, Grueneburgplatz 1, 60323 Frankfurt am Main, Germanyb Faculty of Business, University of Technology Sydney, PO Box 123 Broadway, Sydney, Australiac Department of Marketing, Assistant Professor for E-Finance & Electronic Markets, Goethe-University of Frankfurt, Grueneburgplatz 1, 60323 Frankfurt am Main, Germany

a r t i c l e i n f o

Article history:Received 1 September 2009Received in revised form 30 April 2010Accepted 1 May 2010Available online 12 May 2010

Keywords:Long tailSearch engine marketingOnline marketing

1567-4223/$ - see front matter � 2010 Elsevier B.V. Adoi:10.1016/j.elerap.2010.05.001

* Corresponding author. Tel.: +49 69 798 34649; faE-mail address: skiera@skiera.de (B. Skiera).

a b s t r a c t

Search engine marketing is currently the most popular form of online advertising. Many advertisingagencies and bloggers claim that the success of search engine marketing is driven by the ‘‘long tail”,defined in this research as the many less popular keywords employed by users to search the Internet,and suggest that search engine marketing campaigns need to have hundreds or thousands of keywordsto accommodate this phenomenon. We doubt this claim. Our data from three search engine marketingcampaigns in two countries, which report the success of a total of 4908 keywords over 36 weeks, covering10,104,015 searches and 492,735 clicks, show that the top 20% of all keywords attract on average 98.16%of all searches and generate 97.21% of all clicks. The use of the top 100 keywords in each campaignattracts on average 88.57% of all searches and 81.40% of all clicks. These results are fairly stable acrossa varying total number of keywords in use and suggest that the success of search engine marketing is dri-ven by relatively few keywords. However, we also show that the set of the top 100 keywords changesover time. Hence, advertisers need to concentrate on finding the top 100 keywords, but they do not needto bother too much about the performance of keywords in the long tail.

� 2010 Elsevier B.V. All rights reserved.

1. Introduction in the quality of the ads, see Edelman et al. (2006) or Varian (2006).

The market for search engine marketing continues to grow stea-dily throughout the world. Expenditures on search engine market-ing in 2009 were $10.7 billion in the US alone and thus account for7% of the total online advertising spending in the US (IAB 2010).Hence, search engine marketing is currently the most popular formof online advertising. It allows advertisers to place text ads that de-pend on the keywords entered in a search engine. These ads arelinked to advertisers’ websites, which hold further information re-lated to the entered search keywords.

In Fig. 1, for example, the user entered the search keywords‘‘two-way radios” into Google’s search engine. The user receivesthen two different results from the search engine provider: thelower, left-hand part of the screen shows the unsponsored, organicsearch results, whose ranking is driven by the relevance that thesearch algorithm assigns to these results. The other parts on thetop and the right side of the list present ads known as sponsoredsearch results. The display of the unsponsored search results is freeof charge, whereas advertisers pay for each click on their ads in thesponsored search results. The ranking of these ads is the result ofkeyword auctions for which the advertisers need to submit bidsfor the price per click that they are willing to pay. For more detailson keyword auctions, which also uses weights to reflect differences

ll rights reserved.

x: +49 69 798 35001.

In the example of Fig. 1, ads ranked 1 and 2 are above the unspon-sored search results, and the ads ranked 3–7 are on the right side ofthe search results. Generally, the leading ranks draw the mostattention of the users and are therefore most preferable, and thusthese ranks are sold to the bidder with the highest (weighted)bid. Hence, higher bids lead to higher and thus more attractiveranks, more awareness, more clicks and hence very likely to a high-er number of acquired customers (Hotchkiss et al. 2005). However,the prices per click on those ranks are also higher, which leads tohigher acquisition costs per customer.

The number of keywords that advertisers might use is ratherunlimited, and it is tempting to presume that the long tail, definedin this research as the many keywords that are available for searchon the Internet, is also fairly important in search engine marketing.Anderson (2006) coined the phrase to describe the phenomenonthat niche products can gain a significant share in total sales. Inline with this idea, online advertising agencies and bloggers nowalso claim that search engine marketing campaigns need to havehundreds or thousands of keywords because the long tail, that is,the many less popular keywords, drives success in search enginemarketing.

The primary motivation of this research is that we doubt theimportance of the long tail in search engine marketing. Thus, webelieve that the set of keywords required to successfully run searchengine marketing campaigns can be fairly small so that the man-agement of search engine marketing campaigns is easier than

Sponsored Search Results

Unsponsored Search Results

Keyword Rank

Fig. 1. Unsponsored search results and sponsored search results.

B. Skiera et al. / Electronic Commerce Research and Applications 9 (2010) 488–494 489

many online advertising agencies claim. Therefore, the aim of thispaper is to examine the importance of the long tail in search enginemarketing. Stated differently, we would like to study whether theuse of many different keywords drives the success in search enginemarketing or whether this success is driven by a few but importantkeywords. To accomplish this, the remainder of this paper is orga-nized as follows. Section 2 provides an overview about search en-gine marketing and Section 3 discusses the idea of the long tailphenomenon and its relevance for search engine marketing. Sec-tion 4 describes the set-up of the empirical studies and Section 5presents their results. We conclude in Section 6 with a summaryof our findings.

2. Search engine marketing: an overview

Search engine marketing is by far the largest source of revenuefor Google (Edelman et al. 2006), which is the market leader ofsearch engine providers in most Western countries, usually clearlyahead of Yahoo and Microsoft (Ghose and Yang 2009). The displayof the unsponsored (organic) search results is free of charge,whereas advertisers pay for each click on their ads that appearamong the sponsored (paid) search results. Typically, the term‘‘search engine optimization” labels efforts of firms that aim to im-prove the ranking of the ad in the unsponsored search results andthe term ‘‘search engine marketing” describes efforts towards opti-mizing the placement of ads in the sponsored search results. Theseefforts together account nowadays for approximately 50% of all on-line advertising expenditures (IAB 2010). The main reason for thisvery high share is that search engines have become the main toolconsumers use to locate information (Rangaswamy et al. 2009).Despite this huge importance for firms, research on search enginemarketing is still limited, which is not too surprising as this kind ofadvertising is rather new. It was introduced in 1997 by GoTo,which was acquired by Overture and is now part of Yahoo (Edel-man et al. 2006). Surprisingly, Yahoo allowed Google to use theirproprietary pricing mechanism, which is patented under US law.

Basically two streams of research on search engine marketingexist. One stream deals with the optimal design of keyword auc-tions and possible improvements (Chen et al. 2009, Edelman

et al. 2006, Feng 2008, Liu et al. in press, Varian 2006) or examinesbidding behavior in keyword auctions (Edelman and Ostrovsky2007). Other studies analyze the effect of click fraud on the searchengine provider’s revenue (Wilbur and Zhu 2009, Midha 2008).Mahdian and Tomak (2008) comment on the challenges of design-ing mechanisms for pay-per-action instead of pay-per-click mod-els. These studies have great relevance, though primarily forsearch engine providers.

Another research stream analyzes key questions from theadvertisers’ perspective, including forecasts of the success of singlekeywords by modeling the relationships between bid and rank aswell as rank and click-through rate (Kitts and LeBlanc 2004, Fenget al. 2007, Ghose and Yang 2009), the interdependencies amongkeywords (Rutz and Bucklin in press) or between organic andsponsored search (Yang and Ghose in press). He and Chen (2006)investigate the impact of ad ranks on the quality signal perceivedby uninformed consumers. These studies primarily provide recom-mendation for advertisers to better bid in keyword auctions.

3. Long tail and search engine marketing

The phrase ‘‘long tail” became popular with the best-sellingbook of Anderson (2006) and, more generally, describes the phe-nomenon that the distribution of demand across products hasshifted away from blockbuster products to niche products. Thisidea builds upon work by Brynjolfsson et al. (2003), who show thata large proportion of book sales come from books that are not fre-quently purchased. There is an under-exploited spectrum of cus-tomers’ tastes to which pre-Internet retailers could not cost-effectively cater (Tucker and Zhang 2007). The existence of thelong tail phenomenon is also supported by studies of Brynjolfssonet al. (2007) and Elberse and Obergolzer-Gee (2006). Although El-berse (2008) and Fleder and Hosanagar (2009) did not observesuch a shift, the popularity of long tail-thinking among practicingmanagers is very high (see, e.g., Enge’s comments at searchengine-watch.com/3635039).

In line with these ideas, online advertising agencies and blog-gers postulate that search engine marketing campaigns need tohave hundreds or thousands of keywords because a large number

Table 1Description of three search engine marketing campaigns.

Travel 1 Industrialgoods

Travel 2

Number of keywords 2018 1206 1684Average rank 1.36 1.12 3.71Total number of searches 1,826,534 4,281,462 3,996,019Average number of searches per

keyword905.12 3550.13 2372.93

490 B. Skiera et al. / Electronic Commerce Research and Applications 9 (2010) 488–494

of rather less popular keywords drives the success in search en-gine marketing (see, e.g., www.davechaffey.com/blog/seo/new-keyword-research-tool/; www.searchengineguide.com/matt-bai-ley/keyword-strategies-the-long-tail.php; and www.seomoz.org/blog/illustrating-the-long-tail). The implication of the potentialimportance of the long tail in search engine marketing is thatthe management of search engine marketing campaigns is rathercumbersome because advertisers need to administer many key-words and optimize their prices and texts. A likely consequenceis that advertisers are reluctant to run search engine marketingcampaigns themselves and instead hire online advertising agen-cies to do so. Advertisers might thus perceive management of asearch engine marketing campaign as rather complicated. Addi-tionally, it might be an argument to extend search engine market-ing campaigns even if ‘‘short tail” search engine marketingcampaigns are not profitable, which might also be driven by therather high prices that have to be paid at search engines suchas Google. Lilienthal and Skiera (2010), for example, documentan average price per click on an ad of rank 1 of $2.44 and outlinethat clicks on top ranks in the finance and insurance industry cancost up to $16.37 in the United States and $6.30 in Germany. Suchprices, in combination with low conversion rates, make searchengine marketing quickly unprofitable for advertisers.

We doubt that the long tail, based on the large number of ratherless popular keywords, drives the success of search engine market-ing campaigns. We therefore empirically analyze the followingthree propositions:

� Proposition 1 (the concentration proposition): The distributionof searches, clicks and conversions across keywords isconcentrated.� Proposition 2 (the increasing keywords proposition): An increase

in the number of keywords has a rather small effect on thenumber of searches, clicks and conversions.� Proposition 3 (the most popular keyword proposition): The use of

the most popular keywords at one point in time is sufficient tocapture the vast majority of searches, clicks and conversions infuture periods.

Such knowledge is important because support for these propo-sitions could encourage advertisers to run search engine marketingcampaigns by themselves or to concentrate their supervision to themost popular keywords. Additionally, it would be a clear signal fortop management that running and managing such campaigns isnot too complicated and good results might be achieved by focus-ing on the most popular keywords. The optimal number of key-words certainly depends on the costs that are associated withthe administration of each keyword. These costs might vary sub-stantially across advertisers and depend on the size of the admin-istration costs (which might be reduced if bid managementsoftware is used), the structure of the administration costs (fixedor variable) and the opportunity to allocate the administrationcosts to the number of clicks of keywords. Higher and more vari-able administration costs certainly lead to situations in which itis more attractive to focus on rather few keywords while the opti-mal number of managed keywords is certainly higher if variablecosts are low. A detailed analysis requires detailed informationabout administration costs, which is beyond the aim of this studybut seems to be an interesting avenue for further research.

Total number of clicks 89,846 262,530 140,359Average number of clicks per

keyword44.52 217.69 83.35

Average click-through-rate (CTR)(%)

4.91 6.13 3.51

Average conversion rate (CR) (%) 7.10 0.574 2.97Total number of conversions 6379 1507 4168Average number of conversions 3.16 1.25 2.48

4. Description of empirical studies

We used data of search engine marketing campaigns fromthree firms of two European countries (Germany and Spain) thatcover 36 weeks from August 2007 to April 2008 in two different

industries. The first campaign, ‘‘Travel 1”, was conducted in Ger-many and aimed at selling boating vacations. The second cam-paign, ‘‘Industrial Goods”, was run in Spain and used ads todraw interest for industrial goods such as industrial weights. Athird campaign, ‘‘Travel 2”, was carried out in Germany andwas intended to sell cruises. As other studies have (Ghose andYang 2009, Rutz and Bucklin in press, Yang and Ghose in press),we rely on data that were provided by these firms. We recognizethat they might not be representative of the search engine mar-keting campaigns of other firms. Yet, in contrast to many otherstudies, we provide results for more than one firm and more thanone country. Table 1 provides descriptive information about thesethree search engine marketing campaigns.

All campaigns used a large number of different keywords,which varied between 1206 keywords for ‘‘Industrial Goods” and2018 for the ‘‘Travel 1” campaign. On average, the keywords oftwo campaigns were displayed on fairly well-ranked positions(1.36 and 1.12), whereas those of the third campaign were shownon less attractive ranks (3.71). The total number of searches forcampaign ‘‘Travel 1” was 1826,536. Divided by the number of key-words (2018), this yields an average number of searches per key-word of 905.12. The division of the total number of clicks(89,846) by the number of keywords (2018) leads to an averagenumber of clicks per keyword of 44.52. The click-through rate(CTR) is the share of clicks at the number of searches, thus the totalnumber of clicks (89,846) divided by the total number of searches(1,826,534). When searchers click on one of these ads, they are di-rected to the advertisers’ landing page, which usually provides thepossibility to act. In this context, the general term ‘‘action” com-prises either a lead, to which registrations and downloads are as-signed, or an order. The term ‘‘conversion” is frequently used toindicate either leads or orders and in our three empirical studiesall conversions reflect orders and thus actual purchase transac-tions. The conversion rate (CR) describes the percentage of clickson an ad that finally leads to a conversion, which is in case of cam-paign ‘‘Travel 1” 7.10% (6379 conversions divided by 89,846 clicks).

5. Results of empirical studies

5.1. Distribution of searches and clicks across keywords

In line with previous research on the long tail (e.g., Brynjolfssonet al. 2003, 2007, Ghose and Gu 2006, Hervas-Drane 2007), we usedifferent measures to assess our Proposition 1, based on the con-centration of searches and clicks across keywords. In particular,we calculate the share of the top 20% and top 100 keywords inthe total number of searches, the total number of clicks, and theGini coefficient. Table 2 provides the results for each of the three

Table 2Concentration of searches and clicks across keywords.

Campaign Top 20 (%) Top 100 (%) Ginicoefficient

Concentration of searchesTravel 1 98.57

(91.32a)86.24(68.36a)

0.9521

Industrial goods 97.05(92.83a)

88.71(73.86a)

0.9325

Travel 2 98.85(98.82a)

90.75(81.29a)

0.9641

Average over allcampaigns

98.16(94.32a)

88.57(79.45a)

0.9496

Concentration of clicksTravel 1 96.91

(95.84a)75.68(79.65a)

0.9181

Industrial goods 97.11(94.69a)

87.58(79.03a)

0.9307

Travel 2 97.60(99.21a)

80.95(88.75a)

0.9309

Average over allcampaigns

97.21(96.58a)

81.40(82.48a)

0.9266

a Corresponding share of conversions.

B. Skiera et al. / Electronic Commerce Research and Applications 9 (2010) 488–494 491

campaigns. We also report the corresponding share of conversionsfor the top 20% and top 100 keywords.

The top 20% keywords on average represent 98.16% of allsearches and generate 94.32% of all conversions. The top 100 key-words cover 88.57% of all searches and 79.45% of all conversions.The Gini coefficient can range from 0 to 1, and values close to 1indicate a high concentration. The average value of 0.9496 forthe Gini coefficient also points out that most searches concentrateon rather few keywords.

Travel 1

70%

80%

90%

100%

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35Weeks

Rel

. con

trib

utio

n to

sea

rche

s/cl

icks

0100200300400500600700800

No.

of k

eyw

ords

Top 100 keywords (searches) Top 100 keywords (clicks)No. of keywords

Trav

70%

80%

90%

100%

1 3 5 7 9 11 13 15 1W

Rel

. con

trib

utio

n to

sea

rche

s/cl

icks

Top 100 keywords (searches)No. of keywords

Fig. 2. Development of the weekly share of the top 100 keywords i

Analysis of the concentration of clicks yields similar results: thetop 20% keywords generate on average 97.21% of all clicks and96.58% of all conversions. The top 100 keywords cover 81.40% ofall searches and 82.48% of all conversions. The average Gini coeffi-cient of 0.9266 also indicates that clicks are concentrated in a fewkeywords. Interestingly, from comparing the concentration mea-sures for searches with those for clicks, all measures indicate alower concentration for clicks. Still, we find strong support forour Proposition 1.

5.2. Effects of an increase in the number of keywords

Our results so far have shown that the top 100 keywords of thethree campaigns, with a total of 10,104,015 searches, 492,735clicks and 4908 keywords, generate on average 88.57% of allsearches and 81.40% of all clicks as well as the vast majority ofall conversions. While those results show that rather few keywordsare responsible for a very large share of all searches, clicks and con-versions, they do not reveal the effects of an increasing number ofkeywords (Proposition 2). Therefore, we next analyze the effect ofan increasing number of keywords on the share of the top 100 key-words in the total number of searches and clicks. We thus rank allkeywords according to their weekly volume of searches and clicks.We then select the top 100 keywords (i.e., the keywords that rankbetween 1 and 100 with regard to the number of searches andclicks) and calculate their share of the weekly number of allsearches and clicks. We also determine the number of keywordsused for every week. Fig. 2 illustrates the development of theweekly share of the top 100 keywords in the total number ofsearches, clicks and keywords over time. Fig. 2 also illustrates thatthe number of keywords varies between 340 and 750 keywords perweek.

Industrial good

70%

80%

90%

100%

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35Weeks

Rel

. con

trib

utio

n to

sea

rche

s/cl

icks

0

100

200

300

400

500

600

700

No.

of k

eyw

ords

Top 100 keywords (searches) Top 100 keywords (clicks)No. of keywords

el 2

7 19 21 23 25 27 29 31 33 35eeks

0100200300400500600700800

No.

of k

eyw

ords

Top 100 keywords (clicks)

n the total number of searches, clicks and keywords over time.

Table 3Results of the estimation of the share of top 100 keywords in the total number of searches (clicks).

Campaign N R2 Model sign. bc,0 bc,1 bc,2

Travel 1 36 0.875 (0.878) 0.000 (0.000) 6.203115** (5.218205**) �0.011133** (�0.010321**) 0.000008** (0.000007**)Industrial goods 1 36 0.207 (0.286) 0.008 (0.001) 7.150672** (10.283465**) �0.016985** (�0.028159**) 0.000015** (0.000024**)Travel 2 36 0.586 (0.887) 0.000 (0000) 9.655388** (10.337504**) �0.021403** (�0.025167**) 0.000016** (0.000018**)

Note: **p < 0.05; ( ) = if share of top 100 keywords is based on clicks.

492 B. Skiera et al. / Electronic Commerce Research and Applications 9 (2010) 488–494

Fig. 2 indicates a negative correlation between the number ofkeywords and the share of the top 100 keywords for all campaigns(with the exception of the weekly share of top 100 keywordsamong the total number of searches for the campaign ‘‘IndustrialGoods”). This result is in line with our expectation that the shareof the top 100 keywords would decrease with the number ofkeywords.

We estimate an aggregate logit model to further determine theinfluence of the number of keywords on the weekly share of thetop 100 keywords in the number of searches and clicks. Eq. (1)illustrates the general structure of our model, while Eq. (2) showsthe model specification.

s100;c;t ¼evc;t

ðevc;t þ 1Þ ð1Þ

with s100,c,t is the share of the top 100 keywords among the totalnumber of searches and clicks for campaign c in week t; and vc,t

the function parameter of the share of the top 100 keywords insearches (clicks) for campaign c in week t:

vc;t ¼ bc;0 þ bc;1 � nc;t þ bc;2 � n2c;t þ ec;t ð2Þ

with bc,0, bc,1, bc,2 are the function parameters for campaign c; nc,t,total number of keywords for campaign c in week t; ec,t, residual

Travel 1

75%

79%

83%

87%

91%

95%

99%

337 382 427 472 517 562 607 652 697 742

No. of keywords

Rel

. con

trib

utio

n to

se

arch

es/c

licks

Top 100 keywords (searches) Top 100 keywords (clicks)

Trav

70%

75%

80%

85%

90%

95%

100%

376 408 443 478 5No.

Rel

. con

trib

utio

n to

se

arch

es/c

licks

Top 100 keywords (searches

Fig. 3. Influence of the number of keywords on the weekly share o

in week t for campaign c; t, index set of weeks; and c, the indexset of campaigns.We estimate our model via ordinary least squaresregression by taking the logs of our model according to Eq. (3):

lnS100;c;t

1� S100;c;t

� �¼ vc;t ¼ bc;0 þ bc;1 � nc;t þ bc;2 � n2

c;t þ ec;t ð3Þ

Table 3 reports the results of the estimation, which indicate thatall parameters are highly significant for all campaigns. The R2

shows that the explained variance is substantially higher for thetwo travel campaigns compared to the campaign for industrialgoods. As already pointed out in Fig. 2 there is some unexpectedpositive correlation between the weekly share of top 100 keywordsand the total number of searches and we expect some exogenouseffects that are not captured in the available data and that accountfor the low explained variance.

The negative value for parameter bc,1 indicates that the share oftop 100 keywords in the total number of searches (clicks) declineswith an increase in the number of keywords but does so at adecreasing rate (see the positive value of parameter bc,2). Fig. 3illustrates these results graphically for different numbers of key-words for each campaign, which vary between the minimum andthe maximum number of keywords that are used in their respec-tive campaigns (see Fig. 2).

Industrial good

87%

88%

89%

90%

91%

92%

93%

94%

437 452 467 482 497 512 527 542 557 572 587No. of keywords

Rel

. con

trib

utio

n to

se

arch

es/c

licks

Top 100 keywords (searches) Top 100 keywords (clicks)

el 2

13 548 583 618 653 688 723 of keywords

) Top 100 keywords (clicks)

f top 100 keywords in the total number of searches and clicks.

Table 4Share of top 100 keywords, according to the number of searches in the first 10 weeks, in remaining weeks.

Campaign Week Share of top 100 keywords of week 10 at total. . . Top 100 keywords of week 10 New keywords in top 100

Searches (%) Clicks (%) Conv. (%) No. of keywords Still in use Still in top 100 New Former bottom

Travel 1 10 89.34 79.00 80.08 493 99 91 0 915 91.59 74.61 81.01 471 91 81 2 1720 86.66 65.34 71.10 482 91 77 5 1825 59.65 51.72 55.40 613 72 56 36 830 51.39 48.70 54.92 712 78 55 39 635 52.81 46.47 52.77 688 77 55 36 9

Industrial goods 1 10 88.59 87.09 71.11 535 98 92 0 815 86.90 84.43 64.71 576 96 88 4 820 86.30 80.06 77.78 554 96 89 4 725 69.29 71.41 72.35 601 96 80 15 530 73.01 75.52 61.91 634 96 78 15 735 73.12 74.77 80.50 619 95 79 13 8

Travel 2 10 94.05 86.62 92.55 489 100 90 0 1015 55.89 70.18 70.73 548 100 68 8 2420 77.83 67.33 66.00 544 100 63 9 2825 80.92 63.45 62.98 718 99 63 9 2830 81.23 65.04 60.15 697 98 58 15 2735 72.64 61.97 64.87 707 98 57 15 28

Table 5Share of top 100 keywords, according to the number of clicks in the first 10 weeks, in remaining weeks.

Campaign Week Share of top 100 keywords of week 10 at total. . . Top 100 keywords of week 10 New keywords in top 100

Searches (%) Clicks (%) Conv. (%) No. of keywords Still in use Still in top 100 New Former bottom

Travel 1 10 87.23 81.93 90.73 493 100 78 0 2215 89.80 76.43 90.00 471 90 73 3 2420 84.00 68.60 79.25 482 90 64 7 2925 58.17 53.50 67.89 613 74 54 32 1430 41.04 48.03 57.88 712 75 50 35 1535 48.38 47.18 60.90 688 74 55 32 13

Industrial goods 10 86.44 88.61 77.78 535 98 84 0 1615 84.81 86.84 82.35 576 97 87 4 920 82.95 82.29 80.56 554 97 82 4 1425 66.77 72.62 72.35 601 97 75 12 1330 71.23 76.89 66.67 634 97 77 14 935 71.60 76.36 87.81 619 97 78 12 10

Travel 2 10 92.91 88.38 92.55 489 100 81 0 1915 55.09 70.92 71.55 548 100 63 13 2420 77.03 68.65 66.00 544 100 62 14 2425 79.71 63.15 64.28 718 99 60 10 3030 79.91 63.75 60.15 697 98 52 15 3335 71.32 61.72 66.22 707 98 50 18 32

B. Skiera et al. / Electronic Commerce Research and Applications 9 (2010) 488–494 493

Fig. 3 illustrates that the share of the top 100 keywords in thetotal number of searches and clicks is fairly stable across the differ-ent numbers of keywords. Even though we detect a decreasingshare of top 100 keywords with an increasing number of keywords,this decrease is diminishing and converges to zero at about 650keywords in use for campaigns ‘‘Travel 1” and ‘‘Travel 2” and atabout 550 for the campaign ‘‘Industrial Goods”. The share of thetop 100 keywords in the total number of searches and clicks forthe campaign ‘‘Industrial Goods” stays approximately at 90%. Forthe campaigns ‘‘Travel 1” and ‘‘Travel 2”, the share of total clicksdecreases to about 80% for a large number of keywords, and theshare of total searches does not drop below 90%. Our findings im-ply that an increase in the number of keywords leads to onlyslightly more searches and clicks. Interestingly, for two of the threecampaigns, the share of the top 100 keywords in the total numberof searches decreases more strongly with the number of keywordsthan does their share of the total number of clicks.

The results in Fig. 3 illustrate that the top-performing keywordsgenerate the majority of searches and clicks and that this result isfairly stable between different numbers of keywords. Our results

support Proposition 2 and contradict the assumed importance ofthe long tail in search engine marketing, which claims the neces-sity of having many less popular keywords in use for successfulsearch engine campaigns.

5.3. Prediction of the success of short-tailed strategies

Our results so far suggest that the use of few keywords is suffi-cient to capture the vast majority of searches, clicks and conver-sions. Yet, these are ex post results, meaning that we firstobserved the success of all keywords and then determined theshare of the best keywords in this success. In reality, firms needto first decide on the set of keywords they use; they are then ableto observe the success of these keywords. Therefore, we also testProposition 3 and conduct an ex-ante analysis in which we usethe first 10 weeks of our observation period to select the top 100keywords and then determine their share of searches, clicks andconversions during the following weeks. We also illustrate therespective effect of changes in the number of keywords and differ-ences in the composition of the set of the top 100 keywords.

494 B. Skiera et al. / Electronic Commerce Research and Applications 9 (2010) 488–494

Table 4 shows that the top 100 keywords of the first 10 weeksgenerate the vast majority of searches, clicks and conversions inthe following weeks but that they do so at a declining rate. Forexample, the top 100 keywords of the ‘‘Travel 1” campaign gener-ate 91.59% of searches, 74.61% of clicks and 81.01% of conversionsin Week 15, but their share declines to 52.81% of searches, 46.47%of clicks and 52.77% of conversions in Week 35. Even though theremaining two campaigns show a more stable share of the top100 keywords, this share is also declining. The column ‘‘Still inUse” in Table 4 indicates that the firm stopped using 23% of thetop 100 keywords in the campaign ‘‘Travel 1”, which might explainwhy the shares dropped more steeply in this campaign than in theother two campaigns. Table 4 also shows that there is not a cleartrend in how new keywords or former bottom keywords, whichwere not part of the top 100 keywords in the first 10 weeks, joinedthe top 100 keywords in later weeks. The share of new keywords ishigher for the first campaign but lower for the second and thirdcampaigns. The share of former bottom keywords is higher forthe first and the third campaigns but lower for the secondcampaign.

Table 5 shows the corresponding results for the top 100 key-words according to the number of clicks (in contrast to searchesin Table 4). Both results are fairly similar, indicating that the setof the top 100 keywords changes over time. Thus, we only findmoderate support for our Proposition 3.

6. Conclusion

Since Anderson (2006), the term ‘‘long tail” has enjoyed hugepopularity, and advertising agencies and bloggers claim that thesuccess in search engine marketing is driven by the long tail, whichwe defined in this research as the many less popular keywords en-tered by users to search the Internet. Advertising agencies havestrong incentives to argue that search engine marketing campaignsneed to have hundreds or thousands of keywords. The results ofour empirical studies, however, tell a different story: the top 100keywords—and not the very long tail composed of other keywords:generate the majority of searches, clicks and conversions. The top20% most-searched keywords are responsible for 98.16% of allsearches, 97.21% of all clicks and 94.32% of conversions, and thetop 100 keywords cover 88.57% of all searches, 81.40% of all clicksand 79.45% of all conversions.

We show that these results are fairly stable between changes inthe total number of keywords because an increase in the number ofkeywords beyond the top 100 keywords only slightly increases thenumber of searches, clicks and conversions. The implication of thisfinding is that rather few keywords are important for generatingthe vast majority of searches, clicks and conversions. However,our findings also show that the set of top 100 keywords varies overtime, and new keywords, as well as keywords that previously didnot perform very well, may replace some of the top 100 keywords.As a result, advertisers need to constantly analyze which of thekeywords perform best, but they can still focus on the top 100 key-words to further analyze the success of their search engine market-ing campaigns.

References

Anderson, C. The Long Tail: Why the Future of Business is Selling Less of More.Hyperion, New York, 2006.

Brynjolfsson, E., Hu, Y., and Simester, D. Goodbye pareto principle, hello long tail:The effect of search costs on the concentration of product sales. Working paper,Sloan School of Management, MIT, Cambridge, MA, 2007.

Brynjolfsson, E., Hu, Y. J., and Smith, M. D. Consumer surplus in the digital economy:estimating the value of increased product variety at online booksellers.Management Science, 49, 11, 2003, 1580–1596.

Chen, J., Liu, D., and Whinston, A. B. Auctioning keywords in online search. Journal ofMarketing, 73, 4, 2009, 125–141.

Edelman, B., and Ostrovsky, M. Strategic bidder behavior in sponsored searchauctions. Decision Support Systems, 43, 1, 2007, 192–198.

Edelman, B., Ostrovsky, M., and Schwarz, M. Internet advertising and thegeneralized second-price auction: selling billions of dollars worth ofkeywords. American Economic Review, 97, 1, 2006, 242–259.

Elberse, A. Should you invest in the long tail? Harvard Business Review, 86, 7/8, 2008,88–96.

Elberse, A., and Obergolzer-Gee, F. Superstars and underdogs: An examination of thelong tail phenomenon in video sales. Working paper no. 07-15, HarvardBusiness School, Boston, MA, 2006.

Feng, J. Optimal mechanism for selling a set of commonly ranked objects. MarketingScience, 27, 3, 2008, 501–512.

Feng, J., Bhargava, H. K., and Pennock, D. M. Implementing sponsored search in Websearch engines: computational evaluation of alternative mechanisms. INFORMSJournal on Computing, 19, 1, 2007, 137–148.

Fleder, D., and Hosanagar, K. Blockbuster culture’s next rise or fall: the impact ofrecommender systems on sales diversity. Management Science, 55, 5, 2009, 697–712.

Ghose, A., and Gu, B. Search costs, demand structure and long tail in electronicmarkets: Theory and evidence. Working paper no. 06-19, Net Institute, SternSchool of Business, New York University, New York, NY, 2006.

Ghose, A., and Yang, S. An empirical analysis of search engine advertising:sponsored search in electronic markets. Management Science, 55, 10, 2009,1605–1622.

He, C., and Chen, Y. Managing e-marketplaces: a strategic analysis of non-priceadvertising. Marketing Science, 25, 2, 2006, 175–187.

Hervas-Drane, A. Word of mouth and recommender systems: A theory of the longtail. Working paper no. 07-41, Net Institute, Stern School of Business, New YorkUniversity, New York, NY, 2007.

Hotchkiss, G., Alston, S., and Edwards, G., 2005. Google eye tracking report, 2005.Available at www.enquiroresearch.com/eyetracking-report.aspx.

IAB. IAB Internet Advertising Revenue Report: 2009 Full-Year Results.PriceWaterhouse Coopers, New York, 2010. Available at www.iab.net/media/file/IAB-Ad-Revenue-Full-Year-2009.pdf.

Kitts, B., and LeBlanc, B. Optimal bidding on keyword auctions. Electronic Markets,14, 3, 2004, 186–201.

Lilienthal, M., and Skiera, B. Decomposing the effect of increases in searchadvertising expenditures on prices-per-click and number of clicks. WorkingPaper, University of Frankfurt, Frankfurt, Germany, 2010.

Liu, D., Chen, J., and Whinston, A. B. Ex ante information and the design of keywordauctions. Information Systems Research, in press.

Mahdian, M., and Tomak, K. Pay-per-action model for on-line advertising.International Journal of Electronic Commerce, 13, 2, 2008, 113–128.

Midha, V. The glitch in on-line advertising: a study of click fraud in pay-per-clickadvertising programs. International Journal of Electronic Commerce, 13, 2, 2008,91–112.

Rangaswamy, A., Giles, C. L., and Seres, S. A strategic perspective on search engines:thought candies for practioners and researchers. Journal of Interactive Marketing,23, 1, 2009, 49–60.

Rutz, O. J., and Bucklin, R. E. From generic to branded: a model of spillover in paidsearch advertising. Journal of Marketing Research, in press.

Tucker, C., and Zhang, J. Long tail or steep tail? A field investigation into how onlinepopularity information affects the distribution customer choices. Workingpaper 4655-07, Sloan School of Management, MIT, Cambridge, MA, 2007.

Varian, H. R. Position auctions. International Journal of Industrial Organization, 25, 6,2006, 1163–1178.

Wilbur, K. C., and Zhu, Y. Click fraud. Marketing Science, 28, 2, 2009, 293–308.Yang, S., and Ghose, A. Analyzing the relationship between organic and sponsored

search advertising: positive, negative or zero interdependence? MarketingScience, in press.