The Billboard Effect - Cornell SHA

Exhibit 1Domain visitation (60 days prior to reservation)BookingChannelReservationsSite Visitation Prior to ReservationOTAsOTA2,776Direct2,317Hotel SitesWeb SearchTripAdvisorOther Meta48%68%39%33%66%34%21%65%Note: Sample OTAs include Expedia.com, Hotels.com, and Booking.com. Sample Hotel sites include Hilton.com, Marriott.com, and IHG.com. Searches include searches atGoogle, Yahoo, and Bing. Sample Meta sites include Kayak.com, Trivago. Com, and GoSeek.com.Exhibit 2Average number of visits per reservation (60 days prior to reservation)BookingChannelReservationsSite Visitation Prior to ReservationOTAsHotel SitesWeb SearchesTripAdvisorOther 2.3A primaryreasonforchangethis changeis consolidationA primaryreasonfor thisis consolidationandand innovationamongthe onlinefirms. Expediainnovationamongthe onlinetravel travelfirms. Expediahashas acquired both Travelocity and Orbitz, while Pricelineacquired both Travelocity and Orbitz, while Pricelineacquired Kayak and Expedia and also took a major equityacquiredKayakand Expediatookactivitya majorhasequityposition inTrivago.Much of andthis tivityhasbeenallowed (from a competition standpoint) by the movesof GoogleandaTripAdvisorto become metaOTAsites,ofasallowed(fromcompetition standpoint)by iontoGoogle and TripAdvisor to become meta OTA sites, as arebecoming full fledged OTAs that offer facilitated directKayak and Trivago, and their continued evolution tobooking. There has also been an upsurge in hotel-OTAbecoming full fledged OTAs that offer facilitated directinteractions, with several large hotel brands launchingbooking.There campaigns.has also beenin hotel-OTAdirect bookingAllanof upsurgethis activityhas chingaged research findings that imply that the billboard effectis dead.This conclusionthe rapiddirectbookingcampaigns.stemsAll offromthis activityhasgrowth ouraged research findings that imply that thebillboardtake a larger share of online transactions (and transactionseffect is dead. This conclusion stems from the rapid growthin general). To understand changes in consumer onlineof the two major OTAs (Expedia and Priceline), which nowbehavior we revisit aspects of the billboard effect throughtakea largershareof onlinetransactionsuse ofpubliclyavailabledatasources. (and transactionsin general). To understand changes in consumer onlinePre-PurchaseWebSearch,OTAbehaviorwe revisitaspectsof theSocial,billboardandeffectthroughVisitationuseof publicly available data sources.In this study we use a randomly selected sample ofmore than 50,000 consumers from a panel of over twothe sample members’ 2015 online behavior.3 In our analysis we focus on some 13,000milliononline consumers maintained by comScore, whichtravel-related reservations (including air, rental car, and hotel). A total of 5,093 hotel3tracks allofmadethe bysamplemembers’behavior.reservationswerethe sample:54.5 percent 2015(2,776) onlineof these reservationswereIn ouranalysiswe focussome13,000travel-relatedmadeat OTAsand the remaining2,317on(45.5percent)were madedirectly at hotelIn this study we use a randomly selected sample of more than 50,000 consumers from apanel of over two million online consumers maintained by comScore, which tracks all ofwebsites.4 Using domain level information for each website visited prior to the hotelreservation,3 Thewe comScorefocus on travelrelatedbehavior60 days priorto ,desktopcomScoreonly provides domain level information (e.g., Hilton.com), we have informationpanelists.on which domains consumers visited (and how often), but we don’t necessarily knowreservations (including air, rental car, and hotel). A totalof 5,093 hotel reservations were made by the sample:54.5 percent (2,776) of these reservations were made atOTAs and the remaining 2,317 (45.5 percent) were madedirectly at hotel websites.4 Using domain level information for each website visited prior to the hotel reservation,we focus on travel related behavior for 60 days prior topurchase. Because comScore only provides domain levelinformation (e.g., Hilton.com), we have informationon which domains consumers visited (and how often),but we don’t necessarily know which pages or contentconsumers focused on.5 We do know whether they visitedweb search related sites (Google, Yahoo, or Bing), butwe don’t know which keywords they searched. For websearch related visits, we do know which site they wentto after visiting the search engine. If this next site was atravel related domain, we can infer that this was a travelrelated search. Exhibit 1 summarizes the percentages ofhotel bookers who visited travel related sites. Exhibit1 is separated into two rows: the first row representsconsumers who book hotels at OTAs and the second row4 We focus only on hotels that have at least 30 days pre-purchaseinformation (reservations made in February onwards) and those whichthere was a gap of at least 30 days followi ng any prior travel relatedreservations.5 One methodology to address this issue is eye tracking. See: Bref-fni Noone and Stephani K.A. Robson, “Using Eye Tracking to Obtaina Deeper Understading of What Drives Online Hotel Choice,” CornellHospitality Report, Vol. 14, No. 8 (2014), Cornell Center for HospitalityResearch.which pages or content consumers focused on.5 We do know whether they visited websearch related sites (Google, Yahoo, or Bing), but we don’t know which keywords they4searched. For web search related visits, we do know which site they went to after visitingthe search engine. If this next site was a travel related domain, we can infer that this wasa travel related search. Exhibit 1 summarizes the percentages of hotel bookers whovisited travel related sites. Exhibit 1 is separated into two rows: the first row representsThe Center for Hospitality Research Cornell University

Exhibit 3Distribution of visits to hotel websites prior tobooking via a hotel site vs. an OTArepresents consumers booking directly at hotel websites.The exhibit summarizes the percentage of these consumers who visit OTAs, hotel websites and search engines,as well as sites such as TripAdvisor and meta sites (e.g.,Kayak, Trivago, GoSeek) within 60 days prior to making ahotel reservation.The percentages in Exhibit 1 are reasonably consistentwith those from our 2011 study.6 At that time, about 75percent of consumers who booked directly with a hotelonline visited an OTA prior to purchase (compared to65 percent in this study), while 83 percent of consumersperformed a web search in the earlier study (compared to66 percent in this study).As shown in Exhibit 2, the average number of visitsper reservation is not radically different for those whobooked on the OTA, compared to those who bookedwith the hotel brand directly. In terms of web visits, theonline research behavior is consistent between OTAbookers and hotel direct bookers, but hotel direct bookersvisited TripAdvisor about 33 percent more often thanOTA consumers. On average, hotel direct bookers makeabout twice as many visits to hotel websites (6.5) as OTAbookers (3.4). However, the distribution of these visitsversus just the average (see Exhibit 3) shows that thosetwo groups’ behavior is fairly consistent. That is, thoseOTA bookers who visit hotel websites tend to visit aboutthe same number as those who book direct. The averageshown in the exhibit is smaller because only about half ofthe OTA bookers visit hotel websites prior to booking atthe OTA.Exhibits Exhibits4, 5, and 6 showthe x-axis)the distributionsthe numberof daysbefore4, 5,(onand6 show(on theofx-axis)thedistribubooking that consumers perform web searches, visit TripAdvisor, or go to an OTA. (Thetions of the number of days before booking that consumsix y-axes, showing relative frequency, are on the same scale, allowing a comparison ofers perform web searches, visit TripAdvisor, or go to andirect bookers with those using OTAs.) The figures are noteworthy as they indicate thatweb search activity (Exhibit 5) is happening fairly consistently during the entire 60-dayresearch6phase (although it gradually picks up just before the booking), whereas visits toThe sample in 2011 included hotel direct bookings for July andTripAdvisor (Exhibit 5) and OTAs (Exhibit 6) tend to be intensive just prior to the booking.August of 2008, 2009, and 2010, with data provided by comScore.For OTA visitation prior to OTA booking (left panel of Exhibit 4), we exclude the OTA visitduring which the transaction occurred. The intensity of TripAdvisor and OTA visitationprior to booking indicates that these travel sites may be greatly influencing the purchaseOTA. (The six y-axes, showing relative frequency, are onthe same scale, allowing a comparison of direct bookers with those using OTAs.) The figures are noteworthyas they indicate that web search activity (Exhibit 5) ishappening fairly consistently during the entire 60-dayresearch phase (although it gradually picks up just beforethe booking), whereas visits to TripAdvisor (Exhibit 5)and OTAs (Exhibit 6) tend to be intensive just prior tothe booking. For OTA visitation prior to OTA booking(left panel of Exhibit 4), we exclude the OTA visit duringwhich the transaction occurred. The intensity of TripAdvisor and OTA visitation prior to booking indicates thatthese travel sites may be greatly influencing the purchasedecision.This observational data indicates that consumersremain actively engaged in researching their hotel stay.Review sites and OTAs are critical components of thepurchase decision, although consumers rely less on searchengines compared to our 2011 report, probably as a resultof OTA consolidation and increased familiarly with theinternet.Exhibits3, 4,5and5 illustratethetravelrole travelWhileWhileExhibits3, 4, andillustratethe rolesites playsitesplayinonlineresearch,inExhibit7wein online research, in Exhibit 7 we focus on thefocusstart onof thatthe start of that travel research. Exhibit 7 lists the traveltravel research. Exhibit 7 lists the travel related sites whererelated sites where the consumers’ research phase wastheconsumers’researchphaseinitiated inItadvanceinitiatedin advanceof thehotelwasreservation.summa- rstrizes the percentages of first visits occurring at metaofsites,Trip Advisor,sites,OTAs,web searchesacrossvisitsoccurringhotelat metasites,TripandAdvisor,hotel rectOTAs, and web searches across all consumers as wellasbooking channels. It indicates that web search and Tripseparated into OTA versus direct booking channels. ItAdvisor share similar percentages as the initial site forindicates that web search and Trip Advisor share similarboth OTA and direct bookers, while OTA bookers havepercentagestheinitial siteofformetabothandOTAandvisitationdirectalmost ealmosttwicethedirect bookers.frequency of meta and OTA visitation as direct bookers.Implications for the Billboard EffectResearch conducted since our 2009 report reflectsnew opinions regarding the billboard effect. Estis Greenthan indicated earlier.7 They summarize work done by P.K. Kannan at the University ofand Lomanno contend that the effect is considerablyMaryland describing online consumer behavior using comScore data from 2012 and 2014less prevalent than indicated earlier.7 They summarize(the same data used here, but from different years). For ease of discussion we show awork doneby P.K.Kannanat theUniversityof Marylandreproductionof resultsfrom thisstudy in Exhibit8. Thekey insights fromthis exhibit arethelow probabilitiesof consumersmoving froman OTA (labeledas comScorean intermediary)datato (9.3andpercentfor 20127.0 percent2014) versushigh dhere,thebutof consumers moving from OTA to OTA (90.7 percent in 2012 and 93.0 percent in 2014).different years). For ease of discussion we show a reproThis indicates that it is unlikely that awareness is created at an OTA with consumers thenduction of results from this study in Exhibit 8. The keyswitching sites and booking with hotels directly (as suggested by the billboard effect).insightsthistheGreenlow andprobabilitiesconOnedetail thatfromreceiveslessexhibitattention inarethe EstisLomanno report ofis thatsumersmovingfromanconsecutiveOTA (labeledas(froman intermediary)theseswitchingprobabilitiesare forwebsite visitst-1 to t in Exhibit 8)andfor the consumer’sprocess.As summarizedin Exhibit2,to nota hotelwebsiteentire(9.3researchpercentfor 2012and 7.0percentforResearch conducted since our 2009 report reflects new opinions regarding the billboardeffect. Estis Green and Lomanno contend that the effect is considerably less prevalentconsumers who visit OTAs, prior to booking direct with hotels do so 7.2 times on average,not once. We can create an approximation of the 9.3 and 7.0 figures from Exhibits 1 and7 Estis6. Exhibit1 showsthat 65Cpercentof consumersbookingwith the hotel thevisited anGreen,and MVLomanno.2016.directly“DemystifyingOTAprior Marketplace:to booking direct,Spotlightand Exhibiton6 (right-handpanel) showsthat about18 percentDigitalthe HospitalityIndustry,”HSMAI(ofthis 65 percent) visit an OTA on the day of the booking (day 0 on x-axis of Exhibit 6),Foundation.the product of these two being 11.7 percent. That figure is higher than the 7.0 or 9.3percent as it ignores other (non-OTA) travel site visits on the same day of the bookingthat might have occurred between the OTA visit and the hotel direct booking. ThisCornell Hospitality Report April 2017 www.chr.cornell.edu Vol. 17, No. 11decision.estimate of the OTA impact (like any click-to-click switching probability) ignores all theother OTA visits in Exhibit 6 (those not on the same day as the booking) and provides aconservative estimate of the effect.5

Exhibit 4Time before booking of web searchesOTA BookersDirect Bookers2014) versus the high probabilities of consumers movingfrom OTA to OTA (90.7 percent in 2012 and 93.0 percentin 2014). This indicates that it is unlikely that awarenessis created at an OTA with consumers then switching sitesand booking with hotels directly (as suggested by thebillboard effect). One detail that receives less attention inthe Estis Green and Lomanno report is that these switchingprobabilities are for consecutive website visits (from t-1 tot in Exhibit 8) and not for the consumer’s entire researchprocess. As summarized in Exhibit 2, consumers who visitOTAs, prior to booking direct with hotels do so 7.2 times onaverage, not once. We can create an approximation of the9.3 and 7.0 figures from Exhibits 1 and 6. Exhibit 1 showsthat 65 percent of consumers booking directly with thehotel visited an OTA prior to booking direct, and Exhibit6 (right-hand panel) shows that about 18 percent (of this65 percent) visit an OTA on the day of the booking (day 0on x-axis of Exhibit 6), the product of these two being 11.7percent. That figure is higher than the 7.0 or 9.3 percent asit ignores other (non-OTA) travel site visits on the sameday of the booking that might have occurred between theOTA visit and the hotel direct booking. This estimate of theOTA impact (like any click-to-click switching probability)ignores all the other OTA visits in Exhibit 6 (those not onthe same day as the booking) and provides a conservativeestimate of the effect.To illustratepotential impactof the 7.2 (on average)a consumeratTo theillustratethe potentialimpactOTAofvisitstheconsider7.2 vestotwowebsites).Usingthe2014OTA visits consider a consumer at an OTA who only makesresults from Exhibit 8 there is a 7 percent chance that she moves to a hotel website and 93two transitions (i.e., moves to two websites). Using the 2014percent chance she moves to or remains at an OTA. Following that click she could alsoresultsfrom Exhibit 8 there is a 7 percent chance that shemove to a hotel or an OTA. Exhibit 9 shows all the possible outcomes of a consumer makingmovestoa hotelwebsitefromand93(H)percentshetwo transitions, addingthe transitionshotelwebsite to chancehotel website(60 movespercentto or ance)well as hotelOTAtransitions(40 percent).thatBecauseof thesetransitionsaconsumerat aorOTAa 10.71Exhibitpercent (0.042 0.0651)chanceending up atmove whoto astartedhotelanhasOTA.9 shows alltheofpossiblethe hotel website, up from the original 7 percent. So the probability that a consumer ends upoutcomes of a consumer making two transitions, addingthe transitions from hotel (H) website to hotel website(60 percent chance) as well as hotel to OTA transitions(40 percent). Because of these two transitions a consumerwho started at a OTA has a 10.71 percent (0.042 0.0651)chance of ending up at the hotel website, up from theoriginal 7 percent. So the probability that a consumerends up booking directly at a hotel, given she was at anOTA earlier, depends upon how many of these websiteto-website transitions are made. This probability converges at about 15 percent after about four transitions.The 15 percent (and the 7 percent) are path independenttransition probabilities, which means the chance of aconsumer moving from Expedia.com to Hilton.com isthe same whether she is starting her travel research or isalmost finished and knows where she wants to stay.We can’t read too much into these transition probabilities as they are simply click-to-click behavior anddon’t include the entire search process. In fact, as notedby Estis Green and Lomanno there is a stronger effectof consumers moving to OTAs from hotel direct sitesversus the opposite, with a single click probability of 40percent of consumers clicking over to OTAs from hoteldirect liAnotherAnotherway to examinetheseprobabilitiesto consider themin aggregateacrossthetoentireresearch processfrom click-to-clickactions.In oursample ofties isconsiderthemversusin aggregateacrosstheentire5,093 hotel reservations, 4,273 of these consumers visited OTAs, with 2,776 booking atresearch process versus from click-to-click actions. InOTAs and the remaining 1,497 booking direct with hotels (see Exhibit 10). This indicatesour sample of 5,093 hotel reservations, 4,273 of these35 percent of hotel room purchasers who visited OTAs eventually booked direct. OurconsumersOTAs,2,776OTAssamplealso showsvisited232 customerswho withvisited OTAsbutbookingbooked directatwithhotelsand theremainingbookingwith ofhotelswithoutvisitinghotel websites1,497prior to thepurchase. directThis 5.5 percentthe samplerepresentsunique shoppers,as theynever visitedwebsitesofuntilthe purchase(see Exhibit10). Thisindicates35hotelpercenthotelroommoment. They are also active travel researchers, making an average of 14.1 visits topurchasers who visited OTAs eventually booked direct.travel related sites in the 60 days prior to booking. The 5.5 percent figure serves as theOur sample also shows 232 customers who visited OTAslow end of this switching behavior and the 35 percent figure serves as a high endbut booked direct with hotels without visiting hotelestimate, with the billboard effect falling somewhere in between. Exhibit 11 provides awebsitesprioreffectto thepurchase.5.5 percentofourthesummary of billboardestimates,comparingThisthe originalestimate from2009booking directly at a hotel, given she was at an OTA earlier, depends upon how many ofthese website-to-website transitions are made. This probability converges at about 15percent after about four transitions. The 15 percent (and the 7 percent) are path6independenttransition probabilities, which means the chance of a consumer moving fromExpedia.com to Hilton.com is the same whether she is starting her travel research or isalmost finished and knows where she wants to stay.report with the current estimate, as well as an estimate based on step-to-step transitionprobabilities and steady state transition probabilities from Estis Green and Lomanno.The Center for Hospitality Research Cornell University

Exhibit 5Time before booking of TripAdvisor visitationOTA BookersDirect BookersExhibit 6Time before booking of OTA visitationOTA BookersDirect Bookerssample represents unique shoppers, as they never visitedhotel websites until the purchase moment. They are alsoactive travel researchers, making an average of 14.1 visitsto travel related sites in the 60 days prior to booking. The5.5 percent figure serves as the low end of this switchingbehavior and the 35 percent figure serves as a high endestimate, with the billboard effect falling somewhere inbetween. Exhibit 11 provides a summary of billboardeffect estimates, comparing the original estimate fromour 2009 report with the current estimate, as well as anestimate based on step-to-step transition probabilities andsteady state transition probabilities from Estis Green andLomanno.One aspectof thecomScoredata iscomScorethe need to codeURLsappropriateOneaspectof thedataisintothetheneedto avelindustry.Duringthecodingofcode URLs into the appropriate travel categories, whichURLs into our specific categories of interest (namely, OTA, meta, TripAdvisor, hotelrequires understanding of the travel industry. During thedirect, and web search), we coded major brand sites (e.g., marriott.com, hilton.com) ascodingof URLs into our specific categories of interesthotel direct and also coded independent hotel websites and hotel specific sites as hotel(namely, OTA, meta, TripAdvisor, hotel direct, and websearch), we coded major brand sites (e.g., marriott.com,hilton.com) as hotel direct and also coded independenthotel websites and hotel specific sites as hotel direct. Distinct from the earlier study we also subdivided intermediaries into a series of categories (i.e., OTAs, web search,meta, TripAdvisor, airline direct). The result of this codingshows considerably different site share than that reportedin the study summarized by Estis Green and Lomanno. Ifwe focus on just hotel direct and OTAs we find 34.5 percent of these visits are to hotel direct and 65.5 percent toOTAs, compared to their 2014 numbers of 15.2 percent tohotel direct and 84.8 percent to intermediaries. Similarly,this coding shows 16 percent of consumers visited hoteldirect only and did not visit OTAs, versus the 7 percentreported in Estis Green and Lomanno, and 28.5 percentvisiting OTAs only (versus 64 percent) and 55.5 percentdirect. Distinct from the earlier study we also subdivided intermediaries into a series ofcategories (i.e., OTAs, web search, meta, TripAdvisor, airline direct). The result of thiscoding shows considerably different site share than that reported in the studyCornell Hospitality Report April 2017 www.chr.cornell.edu Vol. 17, No. 11summarized by Estis Green and Lomanno. If we focus on just hotel direct and OTAs wefind 34.5 per-cent of these visits are to hotel direct and 65.5 percent to OTAs, comparedto their 20

Cornell Hospitality Report April 2017 www.chr.cornell.edu Vol. 17, No. 11 . 1. The Billboard Effect: Still Alive and Well. A. s a follow-up on two earlier studies, this report confirms the so-called billboard effect on demand that occurs when online travel ag