Sponsored Search: A Brief History
`
`by Daniel C. Fain and Jan 0. Pedersen
`
`-
`
`Daniel C. Fain is a
`director of research
`at Yahoo! Inc.
`Jan 0. Pedersen is
`chief scientist for
`the search and
`marketplace division
`at Yahoo! Inc. They
`can be reached at
`701 First Ave.,
`Sunnyvale CA
`94089-0703; faind
`at yahoo-inc.com
`and jpederse at
`yahoo-inc. corn,
`respectively.
`
`W igating the Internet. It has become an essen-
`
`eb search is a fundamental technology for nav-
`
`tial part of the online experience - 98.8% of Internet
`users utilize search (Prospect, 2004) - yet, it is pro-
`vided free to consumers. Sponsored search, the
`delivery of relevance-targeted text advertisements
`as part of the search experience, makes this possible.
`In the week ending August 21,2005, search engines
`displayed 13 billion sponsored results (Nielsen/
`NetRatings, 2005). Sponsored search has evolved
`to satisfy users’ need for relevant search results and
`advertisers’ desire for qualified traffic to their web-
`sites, and it is now considered to be among the most
`-
`effective marketing vehicles available.
`The basic elements of sponsored search are
`Advertiser-provided content: a set of adver-
`tiser hyperlinks annotated with keyword tags,
`titles and descriptions
`Advertiser-provided bids that value traffic on
`specified concepts or keywords
`Review process combining manual and auto-
`mated methods to ensure that advertiser con-
`tent is in fact relevant to the target keyword
`Matching of advertiser content to user queries
`as they are received by a search engine
`Display of advertiser content in some rank order
`in some placement alongside other algorithmic
`(that is, non-sponsored) search engine content
`Processes that gather data, meter clicks and
`charge advertisers based on consumer clicks
`on their displayed content
`In 1998. GoTo. later Overture Services. was the
`first to combine these elements. Yahoo! acquired
`Overture in 2003, re-branding it to Yahoo! Search
`Marketing in 2005. BeFirst - now MIVA - followed
`with a similar product in 1999. Google adopted the
`model and modified it to incorporate click feedback
`in 2002. In 2005, Ask Jeeves adopted it, and MSN
`Search extended it to support behavioral targeting.
`Sponsored search engines display results not only
`on their own sites, but also in space rented on other
`sites. For example, when a user searches at MSN
`Search, the search query is passed toYahoo!’s spon-
`sored search engine, which returns results to the
`MSN server, which in turn renders the page that the
`end user sees. Similarly, Google rents space on AOL.
`
`In the early days of Web search, sites used anno-
`tation via <META> tags that contained keywords
`supposedly describing the site’s content. Some sites
`used these tags to outrank other sites that may have
`been more popular or relevant. Even worse, some
`sites - in particular, pornographic ones - manipu-
`lated <META> tags and appeared in the results for
`essentially irrelevant queries. To address this prob-
`lem, GoTo focused on roughly the top 1 ,OOO queries,
`employing editors to create handcrafted result pages
`with images, product and service offers and infor-
`mational links. When GoTo introduced sponsored
`search, each listing was associated with a keyword.
`The keyword set was open ended, and advertisers
`could request to add new ones.
`In other words, GoTo’s editors ensured the rel-
`evance of the advertiser content to the keyword.
`Given the rapid growth of the sponsored search
`industry, this process is now partially automated.
`The majority of yahoo!'^ sponsored search listings
`are initially evaluated by software.
`In the beginning, Internet advertising mimic-
`ked offline advertising. Viewers of print ads are
`more likely to remember ones that are more graph-
`ical and less text-oriented (Du Plessis, 2005, p.
`128). HotWired introduced graphical banner adver-
`tisements to the Web in 1994. Search engines relied
`on banner advertising before the adoption of spon-
`sored search, so they faced a dilemma - keep users
`on the site as long as possible to view more ban-
`ners or send the users promptly to the sites appear-
`ing in the search results.
`Paid search reconciled this ,dilemma by tying
`the search engine’s revenue to the act of transferring
`the user to an advertiser’s site. In 1996, the search
`engine Open Text briefly offered preferred listings,
`in which sites would pay to be inserted into the
`search result set for particular keywords.
`From the start, advertisers were concerned with
`how to measure effectiveness, and publishers were
`concerned with how to set prices. Some publishers
`charged a fixed fee to appear on a site, while others
`such as Netscape and Infoseek in 2005 used the cost
`per mille (CPM) - the cost to display an advertise-
`ment a thousand times - just as for offline advertis-
`ing. A very basic question was whether the banner
`
`Bulletin of the American Society for Information Science and Technology-Decernberlanuary
`
`2006
`
`AOL Ex. 1029
`Page 1 of 2
`
`

`

`______
`
`~
`
`advertisements were actually causing users to visit the site. Some
`advertisers were practicing direct response marketing, which
`seeks a measurable and attributable response. The easiest
`response to measure was the click on a hyperlink. In 1996 Yahoo!
`agreed to charge Procter & Gamble only when users clicked
`from Yahoo! to a Procter & Gamble Web page. The two com-
`panies used the cost per click (CPC) as a common currency,
`translating the advertiser goal into a basis for pricing. For search
`advertising, LinkStar introduced CPC pricing in 1996.
`Even without a common currency, tools became available
`that converted between CPM and an advertiser’s expected/
`effective cost per click (eCPC). A natural extension of CPC
`account management or pricing is cost per action (CPA).
`Example actions are purchasing or signing up for a new
`account. By 1997 Doubleclick was already offering CPA pric-
`ing. Advertisers could configure their websites to relate a par-
`ticular action to a unique search. For example, a travel site
`may h o w that a passenger who bought an airline ticket arrived
`at the site by searching on “travel” on a particular sponsored
`search engine. In 2003, Overture, Google and FindWhat intro-
`duced automated tools to enable sponsored search advertis-
`ers to measure CPA and set appropriate CPC bids. SNAP
`adopted CPA as the standard pricing model for its search
`engine in 2004. It is even possible to determine whether tele-
`phone calls were prompted by a Web search by using specific
`telephone numbers. Ingenio started providing this service in
`1999, and both MIVA (previously FindWhat) and MSN inte-
`grated it as a pricing option for sponsored search in 2005.
`It may be easier for an advertiser to express a goal as CPA,
`CPC or CPM, depending on the circumstances. The ratio
`between CPC and CPA is known as the conversion rute. Click-
`through rate (CTR) is the ratio between cost per impression
`and CPC. An impression is one display of an advertisement,
`so the cost per impression is CPM/1000. If the goal is brand
`awareness, it is sometimes sufficient to see an advertisement
`without clicking, so there is some dispute over the importance
`of CTR (Briggs, 1997; Hoffman, 2000).
`Both conversion rate and CTR are conditional probabili-
`ties: of an action given a click and of a click given an impres-
`sion. In principle, with enough data, one can directly estimate
`these probabilities. A sponsored search campaign will usu-
`ally encompass a huge number of unique queries, and for most
`of those queries, there will be insufficient data. Somehow,
`data must be combined across contexts - for example, by
`grouping queries together or comparing the current campaign
`to others. Yet, additional variables such as the time of day or
`day of week can be introduced that predict variations in con-
`version rate and CTR. This variation is one of the reasons that
`sponsored search bids fluctuate throughout the day.
`In 1997 FlyCast and Narrowline introduced marketplaces
`that enabled advertisers to manage campaigns across multi-
`ple publisher sites, each of which placed a bid for advertise-
`ments. In 2003, Yahoo! was the first major sponsored search
`provider to offer a tool that enabled campaign management
`across multiple online advertising types, including banners,
`
`Briggs, R & Hollis, N. (1997). Advertising on the Web: Is there r
`esponse before click-through? Journal of Advertising Research,
`37(2), 33 et seq.
`Du Plessis, E. (2005). The AdveftisedMind. Sterling, VA: Kogan Page.
`Google. (2005). How are ads ranked? Available October 14, 2005,
`at https://adwords.google.com/support/bin/answer.py?answer
`=6111 &topic= 11 5.
`Hoffman, D and Novak, T. (2000). How to acquire customers on
`the Web. Harvard Business Review, 78(3), 179-1 88.
`iProspect. (2004). Search engine user attitudes survey April-May
`2004. Available October 14, 2005 at www.iprospect.com/
`prerniumPDFs/iProspectSurveyComplete. pdf
`Nielsen//NetRatings. (2005, September 7). Sponsored search link
`placements led online by EBay, Shopping.com Inc., and
`InterActiveCorp, according to NeilseWNetRatings. [Press release].
`Nielsen//NetRatings. Available October 14, 2005, at http://direct.
`www.netratings.com/pr/pr~050907.pdf.
`Varian, H. (1999). Economics and search. SlGlR Forum, 33, (I),
`- hal/papers. html.
`1-5. Available October 14, 2005, at www.sims.berkeley.edu/
`Weber, T.A., & Zheng, Z. (2004). A model of search intermediaries
`and paid referrals. Current version of a paper presented at the
`Workshop on Information Systems and Economics, Barcelona
`Spain, December 14-1 5, 2002. (Wharton School OPlM Opinion
`Paper No. 02-1 2-01 .) Paper available October 14,2005, through
`the Social Science Research Network at http://papers.
`ssrn.com/sol3/papers.cfm?abstract~id = 601 903.
`
`sponsored search and email campaigns. Both Yahoo! and
`Google now run continuous auctions, in which a competitor
`willing to pay more can displace an advertiser at any time.
`Yahoo! sorts search results first by the preciseness of the
`match, then by the CPC bid selected by the advertiser. Direct
`Hit ranked search results using CTR. When Google re-
`launched AdWords in 2002, they modified Overture’s rdnk-
`ing rules to use CTR*CPC instead of just CPC. Google gen-
`eralized the CTR prediction, incorporating an analysis of the
`listing text (Google, 2005). Similarly, Yahoo! calculates a
`Click Index score by comparing a listing’s observed CTR to a
`standard. This score is displayed in the advertiser’s interface,
`but not used when ranking search results. When a listing’s
`score is very low, it is flagged for possible removal.
`Yahoo! continues to transform the original sponsored
`search engine. Increasingly, there is academic interest in spon-
`sored search. Weber and Zheng (2004) proposed ranking
`search results by combining merchant ratings and bids. Varian
`(1 999) presented an argument for sometimes placing a more
`obscure interpretation of a query above a more likely inter-
`pretation based on the average user’s search cost in the eco-
`nomic sense. Yahoo! helped organize a workshop at the 2005
`E-Commerce Conference (http://research.yahoo.com/work-
`shops/ssa2005/sched.html) to bring together the fields of user
`experience, economics and information retrieval. Sponsored
`search continues to improve advertisers’ effectiveness in tar-
`geting their campaigns and to provide highly relevant results
`for searches with commercial intent.
`
`Decernber/Jonuory 2006-Bulletin of the American Society for Information Science and Technology
`
`AOL Ex. 1029
`Page 2 of 2
`
`