Advertising on the Web: is Tiiere Response before Ciicii

Advertising on the Web: is Tiiere Response

before Ciicii-Througii?

A study of Web banner advertising that measured attitudes and behavior found

important attitudinal shifts even without click-through. By using Millward Brown's

BrandDynamics^'^ system, along with other copytesting measures, the authors have

documented increases in advertising awareness and in brand perceptions to Web

banner ads for apparel as well as technology goods.

REX BRIGGS

Miliward Brown

Interactive

NIGEL HOLLIS

Miilward Brown

internationai

This b fhe full report t^tbe re-

search conducted jointly betieeen

HotWired and Millward Braum

International to assess the effec-

tiveness of advertising on fhe

World Wide Web. It xoas first

presented December 2, J996 to

an audience of over 100 repre-

sentatives of on-line ad agencies,

web sites, and the press at the

Jupiter Online confneticc in New

York. A summary version cfthe

report was issued at Ouit time, but

this reprrsents the full report which

includes minr details cfall the find-

ings touched on in the presentation.

SINCE ADVERTISING ON THE WORLD WIDE WEB beganin 1994, marketers have asked the same questionthey ask of all advertising in any media: Does itwork? More specifically, do banner ads (thosesmall, hyperlinked pixel displays popping up onpublic Web sites) actually provide a vehicle for ef-fective commercial communication? As marketersare projected to spend billions of dollars on Webadvertising in the next few years, this questionbecomes increasingly important.

Until now, the only available answer has beenpartial at best. The accepted wisdom suggests that,yes, ad banners do work as direct marketing ve-hicles—but only when viewers click on them fortransport to the advertiser's own Web site, wherea wide range of customized marketing processesbegins.

The problem is that only a fraction of all viewersclick on the banners they see. As a consequence, afew marketers have elected to pay only for provenclick-throughs, while the rest of the marketingcommunity, which pays for ad placements accord-ing to CPM (cost per 1,000 impressions), is left towonder whether the millions of impressions itsbanner ads generate without click-through aresimply wasted.

Are advertisers throwing away money on byte-sized electronic billboards which go unnoticedand un-noted in an environment unfriendly toadvertising? Or do Web banners, even withoutthe benefit of click-through, stimulate brandawareness, brand affinity, and purchase inter-

est as effectively as more traditional advertising

does?

To answer these questions, two research teams

collaborated in an on-line experiment. The re-

searchers came from the research department of

HotWired, Inc., the Internet publisher that inno-

vated the ad banner, and Millward Brown Inter-

national, a recognized leader in advertising effec-

tiveness research. The experiment was the first sig-

nificant research study on Web advertising

effectiveness, a study which dealt successfully

with the unique research challenges posed by the

Internet environment. This article describes the

methodology of this important study and details

its findings.

WHAT TO MEASURE

Many people have argued that the best measure ofadvertising response on the Web is the click-through rate. The advantages of this metric arethat it is a behavioral response and easy to ob-serve, and that it indicates an inxmediate interest inthe advertised brand. But many other factors arealso likely to influence the click-through response,and these factors may have more to do with theoriginal predisposition of the audience than withthe advertising itself. Thus the practice of evaluat-ing Web advertising on the basis of click-throughis like evaluating television ads for automobiles onthe basis of how many people visit a showroomthe next day. A showroom visit is an ideal re-

M a r c h . Apr i l 1 9 9 7 JDURntiL OF HDUERTISinG RESEHRCH 3 3

RESPONSE BEFORE CLICK-THROUGH?

sponse, but hardly the most likely one,

since relatively few people will be in the

market for a new car on a particular day.

Because most advertising does not

evoke an immediate behavioral response,

we designed our study to observe and

measure both the attitudinal and the be-

havioral responses to a Web banner. In

particular, our objective was to measure

whether or not banners themselves have

the potential to build brands, by creating

awareness and image. We divided the

population of our experiment into two

cells: an exposed cell, which saw a banner

for a tested brand on HotWired's

homepage, and a control cell, which did

not. Using these two groups, we sought to

assess tbe impact on an advertised brand

of one single incremental banner expo-

sure.

HOW WE MEASURED THE RESPONSE

The question we then faced was how todefine and measure the "impact" of oneWeb banner exposure. Traditionally, re-searchers have relied upon measures ofawareness, recall, and reported responseto indicate whether or not advertising hashad an impact. These measures allow usto assess tbe degree to wbich an ad basbeen noticed (a necessary precursor to anymore fundamental effect) and the degreeto which tbe advertisement is likely tolead to a purchase. We adapted MillwardBrown's proprietary measurement sys-tems to tbe Web environment to assess theawareness of, and reaction to, the ad ban-ners tbemselves. However, we sought togo beyond measures of awareness, recall,and reported purchase probability. Wesougbt to determine the contribution tbead banner makes to buiiding tbe brand. Todo tbis, we adapted Millward Brown'sBrand Dynamics'^" System to work in tbecontext of an online interview (Dyson,Farr, and Hollis, 1996).

The BrandDynamics™ System

Tbe BrandDynamics^" System is com-

posed of two modules: tbe Consumer

Value Model and tbe Brand Dynamics^"

Pyramid.

The Consumer Value Model. Tbe Con-

sumer Value Model allows us to identify

tbe probability tbat an individual will

choose a particular brand for tbeir next

purchase. We call this measure Consumer

Loyaliy. The prediction of Consumer Loy-

alty is based on:

• consumers' claimed brand consider-

ation

• a measure of brand size

• the price of tbe brand relative to others

in the category

The Consumer Value Model has been vali-dated against bebavioral data and marketshare in many different categories in bothNortb America and Europe. A higher av-erage Consumer Loyalty in the expwjsedcell than the control would suggest thatthe banner exposure had positively af-fected the likelihood that people wouldbuy the advertised brand.

The BrandDynamics^" Pyramid. Basedon a consistent set of sun. ey measures thatcan be applied across different brands,categories, and countries, the BrandDy-namics^" Pyramid allows us to explainthe variation we observe in respondents'Consumer Loyalty scores. Eacb level ofthe Pyramid is a composite of awarenessand imagery measures. We use tbese lev-els as building blocks to define an indi-vidual's relationship with a brand, asshown in Figure 1.

As we move up the levels of the Pyra-mid we describe a deepening attitudinalinvolvement with the brand. The brandrelationship moves from consciousness ofthe brand (Presence), through acquain-

tance and examination {Relevance and Per-

formance), to experience (Advantage), and

then, finally, to the point where the con-

sumer finds it difficult to consider alter-

native choices {Bonded). By comparing the

percentage of people in the control and

exposed cells who attain each level of

the BrandDynamics^" Pyramid, we ex-

plain any obser\'ed changes in Consumer

Loyalty.

Immediate vs. delayed effects

An important point to note is that thefindings reported in this document relateto consumers' reactions a day or so afterexposure. Tracking advertising responsein traditional media suggests that the ob-served increases would erode over timewithout the reinforcement of further ex-posure. A positive response to advertis-ing, while obviously desirable, is of littlebenefit to a brand unless it positively in-fluences the consumer's choice on the nextpurchase occasion. Therefore, a key func-tion of advertising is not only to establishbut to maintain key brand associations inlong-term memory so tbat they come tomind the next time the person encountersthe brand. Unfortunately, it is beyond tbescope of this project to evaluate the adver-tising in this respect. The objective of ourstudy was to identify the immediate effectof one additional banner exposure. How-ever, since exposure is rarely limited toone occasion, we assume tbere would beadditional effect with repeated exposure(and the results of this study suggest asmuch).

THE STUDY DESIGN

We designed our research to address theparticular concerns of botb the advertiserand the researcher. Advertisers need toaccurately measure the impact of adver-tising. Therefore, we measured the com-munication value of the advertising ban-ner (i.e., its impact on consumers' atti-

3 4 JOURIIflL or HDUERTISIIJG flESEHRCH M a r c h 1997


and September 16,1996, a random sample

was solicited for participation in a short

research study, "To Help HotWired Better

Understand Their Audience." Thirty-

eight percent of those solicited partici-

pated. Respondents completed a seven-

minute survey covering demographics

and webographics. The survey avoided

any indication of the true purpose of the

research. Upon submission of the survey,

respondents were returned to HotWired's

homepage. Respondents were randomly

assigned to Control or Exposed (test) cells.

One of three test ads, or a control ad, was

randomly placed in the designated adver-

tising space on HotWired's homepage.

Second day. Respondents were sent an

email thanking them for their participa-

tion and inviting them to participate in a

second survey (with the added incentive

of a chance to win $100), Respondents

were provided a URL (Web address) in

the email and directed to link to the Web

page in order to participate in the survey.

Sixty-one percent of the original respon-

dents participated in the second survey.

Respondents from the test cells were

tudes), as weU as its ability to eUcit an the Web site behavior of both Responders served one of three category-specific sur-

immediate behavioral response {the click- and Nonresponders enabled us to look for ^'^ys based on their exposure to tbe adver-

through). We controlled exposure care- significant differences between Respond- tisement on the first day. Respondents

fully to ensure that observed differences e rs /Nonresponders and Control /Test ^""^ ^^^ control ceU were randomly di-

were due to the advertising and not extra- CeUs. No significant differences were ob- ^^^t^d to one of the three exposure sur-

neous variables. And we tested ads which served. veys. The resulting final base sizes for

had varying levels of preexisting advertis- ^^^^ '̂ eU are shown in Table 1.

ing "weight" in order to gauge diminish- F*^* '^^y- ^^°^ *^^ universe of users who There was no special selection processine returns accessed HotWired between September 9 for the creative used in the experiment.

The research objective was to execute a

rigorous and carefully controlled study

which avoided the bias sometimes en- T A B L E 1

countered in on-line research. We repre-

sented the "real world" as precisely as .T°*?.'. .̂̂ P.?*.®*! 9.?.

possible, by ensuring that the true nature Total sample 1,232 910 293

of the research was not immediately ap- ^^g^'s apparei brand 397 301 82p a r e n t t o p a r t i c i p a n t s . O u r r e c r u i t m e n t '•••-••" • •• •••••• ••'"•• ^ ' ] ^

Teiecommunication company ISP 422 312 95methodology guaranteed us a true cross-

t . , .,-4̂ . A Web browser 413 297 116section of users, and our ability to record

Figure 1 BrandDynamics™ Pyramid

M a r c h . April 1 9 9 7 JOURdRL OF HDUERTISIHG RESEHRCH 3 5


We believe it is possible for advertisers tocreate banners that compare favorably totbe ones tested in tbis study. Wbat wassignificant about tbe ad banners used wastbat each one represented a brand at a dif-ferent stage of developing a "presence"on the Web.

Men's apparel brand. For reasons of cli-ent confidentiality we bave used de-branded data in this paper. At the time ofthis study, the men's apparel brand waspreparing to launch their Web site. In fact,tbe very first Web advertising impressionstbat were seen for this brand occurred aspart of tbis study. Thus we expected thebanner exposure to increase Presence (con-sciousness of the brand) and Relevance(perception that the brand might suit theirclothing needs). We expected the advertis-ing to sway individuals who identifiedwith the Web toward purcbasing tbebrand that had recently become "wired."Seeing the advertising might also causepeople in this target audience who hadpreviously rejected the brand to reevalu-ate tbeir opinion, perbaps even to per-ceive an Advantage and then Bond with it.This could result in an increase in theirlikelihood to purchase the brand, as mea-sured by tbe difference in Consumer Loy-alty scores between the exposed and con-trol cells.

immediate behavioral response (a click-

through), we wondered whether we

would observe an attitudinal response at

all.

Web browser. The latest version of thisWeb browser was released witb a signifi-cant Web advertising campaign. Tbebrowser was a new product but witb per-vasive presence and a well-known parentcompany; therefore, we were prepared fosee little response af all to just one morebanner exposure. If, bowever, users didshow a measurable response to one addi-tional exposure to the ad banner, then wewould conclude that Web advertising hasthe potential to impact consumers even athigher levels of frequency.

DID THE ADVERTISING

AFFECT ATTITUDES?

To answer this question, we will start byconsidering the overall impact of an addi-tional ad banner exposure on tbe variousbrands tested using the Consumer Loyaltyscore. We will then explore the factors thatmay have caused the observed differenceson a brand-by-brand basis.

Consumer Loyalty

Tbe Consumer Loyalty score, wbich mea-

sures tbe likelibood tbat the consumer

will purchase tbe product on the next pur-chase occasion, has been shown to bebigbly predictive of consumer purchasebebavior. In each of our three test cells, theConsumer Loyalty score is higher thanthat for the control cell, suggesting tbattbe extra exposure did have an effect.(When considering the implications ofthese scores, it is important to bear inmind how they are derived. The basiccomponent of the Consumer Loyalty scoreis a consideration scale. Another key com-ponent is the relative price of fhe differentbrands witbin their category. For themen's apparel and ISP brands we calcu-lated a perceived relative price from thebrand image data. Tbe men's apparelbrand indexes 10 percent bigher tban tbeaverage khaki pants, and the ISP 30 per-cent higher than the average Internet ser-vice provider. To tbe extent that this per-ception-based measure differs from tbeactual relative prices in the market, the ab-solute Consumer Loyalty will be biasedup or down. However, any bias will beconsistent between the two cells. In thecase of the Web browser, the price index isset to nearly zero, given tbat if could bedownloaded for free.) Tbe difference ismost significant in the case of the men'sapparel brand, reflecting tbe fact that theexposure was tbe first time the brand had

A telecommunication company's inter-net service provider brand. While thetelecommunication company helped pio-neer Web advertising, the ISP brand hadnot been advertised on the Web prior toour study. Given tbe telecommunicationcompany's strong over-arcbing brandname, we expected tbat tbe ad bannermigbt reinforce the parent company's ge-neric presence rather than generate sig-nificant awareness for the ISP brand spe-cifically. And because tbe telecommunica-tion company's objective was to create an

Web browser

ISP

Apparel brand 13%

10% 20%

• Exposed

Ei Control

20%

30% 40% 50%

Figure 2 Effect on Consumer Loyalty

3 6 JDURORL OF HDU[RTISinG RESERRCH M a r c h . A p m 1 9 9 7


100% —90% —80% —70% —60% —50% —40% —30% —20% —10% —

0% i"

18

14

13

15

Exposed Control

• Not a type youwould consider

a Possit)le, txrt notone you wouidusually consider

• One of severaltypes you wouldconsider

• One of only 2or 3 types youwould cxjnsider

• The only type youwould consider

Figure 3 Consideration of Apparel Brand

ever advertised on the Web. The impact ofa single Web banner exposure on the Con-sumer Loyalty score ranged from a 5 per-cent increase for the Web browser fo anincrease of over 50 percent for the men'sapparel brand as shown in Figure 2.

Why does the Consumer Loyalty scoreincrease so strongly for the men's apparelbrand? Do those who are exposed to theadvertising banner suddenly say that thisbrand is the only brand they will buy?Though the impact of the advertising isnot this dramatic. Figure 3 shows how theadvertising makes some people a bit moreiikely to consider this brand compared toother brands. The differences noted arestatistically significant. (Statistically sig-nificant differences have been marked asfollows; 95 percent confidence level de-noted with **, 90 percent confidence leveldenoted with *.)

The first appearance of a banner for themen's apparel brand seemed to causesome people to move the brand higher upin their consideration hierarchy, Acomplementary question, designed to de-termine if the brand might fulfill a specificrole in relation to other brands peoplemight consider, demonstrates that expo-sure to tbe men's apparel brand ad bannerincreased the likelihood that the brandwill be bought for a purpose that other

competing brands don't fulfill {exposed:14 percent versus control: 5 percent).

Tlie ISP and Web browser brands showa different pattern of response, perhapsindicating a degree of diminishing returnsassociated with increased advertisingweight on this particular measurement.Neither brand shows statistically signifi-cant increases in the consideration hierar-chy. However, those exposed to the ISPbanner clearly demonstrate an increasedpropensity to "buy" the service for a pur-pose that other competing brands don'tfulfill (exposed: 26 percent versus control:14 percent).

It is important to note that the differ-ences reported refer to future purchaseconsideration. There is no difference be-

fween the exposed and control cells inferms of prior brand experience or own-ership for any of the three brands, con-firming that exposure to a single addi-tional advertising banner has the potentialto increase purchases,

Name recognition

Given that a basic function of all advertis-ing is to increase the presence of the ad-vertised brand in the marketplace, andthat the men's apparel brand had neveradvertised in this medium before, it is notsurprising that awareness of the men's ap-parel brand is higher in the exposed cellthan the control, as shown in Figure 4.This result is also observed with the Webbrowser, despite the fact that the level ofawareness of that brand is already ex-tremely high.

If appears that, in addition to increasingthe Consumer Loyalty scores, the adver-tising increases passive name recognitionas well. So, how does the advertising ap-pear to affect the levels of the BrandDy-

Pyramid for each brand?

THE BRANDDYNAMICS™ PYRAMID:

MEN'S APPAREL BRAND

All levels of the pyramid increased fromthe Control to the Exposed cell, with thelargest effect observed at the Presence level(see Figure 5). The score at each of the

nWeb browser

\

ISP

Apparel

C

_

)% 20%

• • • 1

40%

• • • 1

• • 6 1 %161%

60%

|73%

80%

^•96%**

i i

| 84%**

100%

Exposed

Control

Figure 4 Aided Brand Awareness

March . April 1997 M M V Of ilDUEflTISIIlG RESeflRCH 3 7


Bonding

Advantage

Performance 1

1Relevance I

Presence 1

Control ExpoBad

Figure 5 BrandDynamics^'^ Pyramid—Men's Apparel

Pyramid levels is a composite of aware-ness and image measures. By looking atthe individual measures which compriseeach level, we can further diagnose theeffects of the advertising. This increase inPresence is driven entirely by an increasein jusl one of the measures used to definePresence—endorsement of the statement,"Comes readily to mind when you thinkabout (clothes)." (Fifty percent of the ex-posed agree with this statement, com-pared lo thirty-two percent of those in thecontrol cell.) The increase of the Presencedimension then fuels the majority of theincrease at higher levels of the Pyramid,suggesting that many people held latentgood opinions of the men's apparel brandwhich emerge upon exposure to the ad-vertising banner. There are also somemodest increases observed for attributesthat affect the higher levels of the Pyra-mid. These are shown in Figure 6.

As can be seen from the chart, the larg-est difference is in the attribute "just thesame as other brands." While not ex-pressly used to build the Pyramid, this at-tribute contrasts "offers something differ-ent," which is used, in part, to derive Ad-vantage and Bonding. This finding suggeststhat the advertising has a positive affect

on perceptions as well as Presence, by cre-ating a distinctiveness for this brandamong the Web audience.

The hypothesis that Web advertisingcan positively impact perceptions is con-firmed by the fact that other diagnosticmeasures also exhibit a difference be-tween the exposed and control cells. Oulof 18 adjectives which might describe themen's apparel brand with which peoplecan agree or disagree, 5 are higher in theexposed cell compared lo the control (seeFigure 7).

For this brand, the findings appear con-clusive. The effect of its first exposure onthe Web has a strong immediate effect onthe brand. Consumer Loyalty and brandPresence increase, perceptions of thebrand's personality are positively im-pacted, and the advertising helped to dif-ferentiate the brand from its competition.We attribute these effects both to the con-text in which exposure takes place and thepower of the banner on the target audi-ence. The Web and the HotWired site

have strong positive images for these

people; HotWired delivers conlent they

find relevant through a medium—the

Web—which they find exciting. Advertis-

ing on the Web transferred some of this

excitement and relevance on to the men's

apparel brand itself.

THE BRANDDYNAMICS^'^ PYRAMID: ISP

As might be expected from the ConsumerLoyalty results, we do not see quite thesame degree of impact resulting from thefirst exposure to the ISP banner (see Fig-ure 8).

The results for this brand may suggestthat increasing brand awareness for newproducts and services (especially whenbranded under a strong parent name) maytake more than a single banner impressionto have a dramatic impact on consumers'perceptions and purchase intentions.While there is little change in the overallPyramid, there are two individual attrib-utes that exhibited a significant difference.

One measure, "is growing more popu-lar," helps define perceptions of Advan-tage and Bonding (38 percent of the ex-posed cell agreed with this statementcompared to 26 percent of the control).The other attribute, "is a good Internetservice provider," helps to define the Per-formance level of the pyramid but onlyamong those who have first achieved thelevel of Releimnce. Overall, 43 percent ofthe exposed cell agrees that this ISP wouldbe a good Internet service provider versus31 percent among the control. The in-crease we obser\'ed in these two measuresoccurred among f>eople who had failed toqualify for either the Presence or Relevancelevels of the Pyramid.

The effect of its first exposure on the Web has a strong

immediate effect on the brand.

3 8 JOUrniHL OF RDUEHTISinG RCSEHRCH M a r c h • Apr i l 1 9 9 7


Arejust the sameas other brands of

(clothes}

You have a hi^ieropinion of than other

brands

Appeal to you morethan other

brands

Are better thanothers

21%32%"

• Exposed

B Control

10% 20% 30% 40%

Figure 6 Apparel Brand Attributes

THE BRANDDYNAMICS™ PYRAMID: THE

WEB BROWSER

The overall Pyramid and its underlyingattributes provide intriguing insight intoconsumer perception of the parent brand,as well as the advertising for this particu-lar product (see Figure 9).

Even with ubiquitous advertising lev-els, there is a modest positive increase inPresence as a result of one more exposureto the Web browser's ad banner onHotWired's homepage. We also see somemarginal increase at the levels of Perfor-mance and Advantage. The Bonding score,

however, declined. This decline may havebeen caused by the negative press sur-rounding the brand during the time of theadverlising effectiveness test.

To a similar point, the Presence levelcomprises not just an active awareness ofthe brand, but an active awareness of thebrand promise (defined by an acceptanceor rejection of the notion that the brand isrelevant to the individual's needs). In thecase of the Web browser brand, the aware-ness component for the exposed cell ishigher than the control cell (47 percentversus 41 percent). However, active rejec-tion is also 5 percentage points higher (12percent in the exposed cell versus 7 per-cent in the control cell). Similarly, whenwe turn to the higher levels of the Pyra-mid, "have a higher opinion of than other

Web browsers" is lower among the ex-posed cell (6 percent versus 12 percentamong the control). This increase in activerejection, in part, accounts for the reducedscore at the Bonding level.

Overall, it would appear that the singleadditional advertising exposure increasedthe awareness and Presence of the brandamong some viewers; however, it also re-minded a few of their negative views ofthe parent company, which may havebeen generated by the intense scrutiny fo-cused on the security provisions withinthe product.

These findings suggest that exposure toan ad banner on HotWired's homepage canhave an immediate impact on consumerperceptions of the advertised brand. Wenow turn to the advertising-specific datato better understand these reactions.

REMEMBERED EXPOSURE:

BRAND-LINKED AD AWARENESS

We know that all those in the exposedcell had one more opportunity to see thead banner than those in the control cell.This additional exposure should result inthose exposed having both a higher ievelof brand awareness, in general, and ahigher levei of ad banner awareness spe-cifically.

We will now assess the results for spe-cific brand-linked ad awareness, a mea-sure that has been found to have a strongpredictive relationship with sales in tradi-tional media {Hollis, 1994). Figure 10 con-firms that the proportion of people whoclaimed to remember seeing the testedbrand advertised on the World Wide Webin the past seven days was higher in allcases. The chart shows that the increase inad awareness varies by brand. This vari-ance is a function of the overall power ofthe banners, diminishing returns relatedto the individual brands' Web adver-tising weights, and the original level, the"base," from which ad awareness hadbeen increased.

To properly understand the last factor,we need to consider the fact that it mustbe incrementally more difficult to raiseawareness from a base level of 73 percentthan from 7 percent. Millward Brown'sFORCE score (the First Opportunity to see

Response Created by the Execution) {Dyson

spirited

AdventunDus

Confident

Real

Traditional

15%"

• Exposed

• Control

140%"

•i0% 10% 20%

..... 135%

I I30% 40%

148%'

50%

Figure 7 Apparel Brand Personality

M a r c h . April 1 9 9 7 JOUROHL OF RDUCRTISIflfi ReSEHRCH 3 9

RESPONSE BEFORE CLiCK-THROUGH?

Control Exposed

Bonding

Advantage

Performance

Relevance

Presence

1%

Figure 8 BrandDynamics™ Pyramid—ISP Brand

and Farr, 1995) accounts for the effects ofdiminishing returns and consequently al-lows a like-for-like comparison of thethree banners ability to create brand-tinked ad awareness.

Millward Brown utilizes the FORCEscore in traditional media such as televi-sion and print. As the FORCE score mod-els out the effects of time, exposureweight, diminishing returns, and baselevel, FORCE scores can be directly com-pared across media types. While time-series modeling is normally employed tocalculate the FORCE measure, in this caseit was not necessary. The aspects of time,exposure weight, and base level were allcontrolled for by the test design and asimple calculation can be used to allow forthe diminishing returns effect.

As the median FORCE score for televi-sion advertisements is only 10 percent thescores reported in Figure 11 {an averagescore of 20) suggest that the Web bannerswe tested compare very favorably to most

TV ads in terms of creating brand-linkedad awareness.

Though this finding may seem surpris-ing at first glance, previous work done byMillward Brown in the United Kingdomsuggests that print also compares very fa-

vorably to TV in its ability to create brand-linked awareness of advertising (Farr,1994). The median FORCE score for printads is 18 percent, as can be seen in Fig-ure 12.

While TV has the advantage of beingmore intrusive (moving visuals, sound,etc.), it is nonetheless a passive medium.The content is displayed to the viewer ir-respective of whether they are actually at-tending to it. Conversely, Web- and print-based media both have the advantage ofactive reader involvement. The consumeractively engages with the content as theyperuse it and search for items of interest.This engaged state, that the nature of theWeb encourages, seems to result in higherinitial recall of advertising than might oth-erwise be expected.

We have now observed that, overall.Web advertising can create strong in-creases in brand-linked ad awareness, thenecessary precursor to the changes inbrand-related measures. But how do theindividual banners compare?

As the men's apparel brand banner hadnot run anywhere else on the Web prior tothe study, we are able to see a true pictureof the effect of one exposure. Seven per-

Control TMt

Bonding

Advantage

Performance

Relevance

Presence

Figure 9 BrandDynamics^"^ Pyramid—Web Browser

4 0 JflURnm flf HflUERTISIiiG RESEflRCH March • Apri l 1 9 9 7

RESPONSE BEFORE CLiCK-THROUGH?

Apparel

ISP

Web browser

21%*

40%'• Exposed• Control

31%

82%'

73%

- r I I I I I I I 1 I0% 10% 20% 30% 40% 50% 60% 70% 80% 90%

Figure 10 Ad Awareness

cent of the respondents in the control cellclaim to have seen an ad for the brand inthe past seven days. While we know thisto be contrary to fact, this level of baseawareness is not unusual; in fact, it is to beexpected from an established and rela-tively well-known brand like the onetested. The increase of 200 percent for theexposed cell is considerable, even by thestandards of television and print advertis-ing. The resulting FORCE score, however,is only marginally higher than the ob-served increase in ad awareness. This isdue to the fact that the effects of dimin-ishing returns and exposure weight(which negatively impact the absolute in-crease observed for the other brands) areminimal in the case of the men's apparelbrand, since it started with a low level ofawareness.

The telecommunication company's ban-ner generated a 29 percent increase in gen-eral awareness of advertising for thebrand on the Web. The higher base leveland the increased impact of diminishingreturns account for the difference in thesize of this increase relative to the one ob-served for the men's apparel brand.

The results for the Web browser'sFORCE score stand out from the others.Though the base awareness level in thecontroi cell is extremely high in compari-son to any point-in-time measure for tra-ditional media, it is clear that an extra ex-

posure on HotWired's homepage is ca-pable of improving even on that high levelof awareness. The FORCE score indicatesthat the browser's banner created a highlevel of brand-linked memorabilitv—a

THE EFFECT OF DIMINISHING RETURNS

What is the incremental value of the ad-ditional ad banner exposure? Analysis ofWeb server-log files by one Web mediacompany suggests that diminishing re-turns on the behavioral click-through be-gin immediately and that, at a frequencyof three exposures, little to no additionalclick-through benefit is garnered. If thissuggestion is correct, does the brand com-munication value generated from the ex-posure of the banner exhibit a similar pat-tern?

To fully analyze the diminishing re-turns function requires a tightly con-trolled experiment which measures the ef-fect at multiple levels of exposure for eachbanner ad. While fhis type of measure-

. . . the Web banners we tested compare very favorably

to most TV ads in terms of creating brand-imi(ed ad

awareness.

level far higher, in fact, than the averagesfor television and print. The publicity andother marketing activity surrounding thebrowser's introduction may well haveplayed a part in the increased attentionthat people paid to this ad, but the score isdramatic nonetheless.

ment is beyond the scope of this study, thefollowing analysis does provide some in-sight into the incremental benefit onemight expect to find at different levels offrequency. The analysis is based on a com-parison of the single exposure for thethree different ads tested. On average,

Apparel

ISP

Web browser

•••

-r0%

15%

•1

110%

^ H 15%

13%

• •1

15%

•1

20%1

25%1

30%

l33%

135%

Figure 11 FORCE Score

M a r c h . Aprii 1 9 9 7 JOOROHL OF HOyEeTISlOG RESERRCH 4 1


Television

Print

Web"

-r0%

15%

^ ^

110%

9 Average

i15%

FORCE score

B%

^ ^ 20%

120%

Figure 12 Comparison of FORCE Score by Media Type

these brands have different levels ofweight ranging from the men's apparelbrand (no previous exposure) to the Webbrowser (significant exposure on aver-age).

For instance, specific banner awareness(elicited by showing people the actual adbanner) displays a classic diminishing-returns function consistent with findingsfrom traditional media. Initial bannerawareness is highest for the Web Browser(see Figure 13), since many of these peopleare likely fo have been exposed to thisbanner, or one like it, elsewhere, buf theincremental benefit of one additional ex-posure is the lowest of the three (see Fig-ure 14). The converse can be seen for theapparel brand. Initial awareness is lowest,but one additional exposure results in fhestrongest increase. The suggestion fhatthere may be diminishing returns does notpreclude the fact fhat fhere is likely to be(as in the cases tested) measurable incre-mental effects at higher levels of fre-quency. Whether the incremental benefitwarrants the cost of exposure is a questionthat can only be answered on a brand-by-brand basis.

A key finding is that the benefits arelikely to be derived even after multipleexposures. In addition, findings for themen's apparel brand seemed fo indicatethat the communication power was par-

tially a function of the surrounding pro-

gram context

AD REACTION

The questions employed fo understandconsumer reaction to the advertising ban-ners themselves were adapted from Mill-ward Brown's Link copy test. The resultsreinforce fhe notion thaf the Web is aunique medium. If resembles traditional

print media in some respects but exhibitssome similarities with outdoor advertis-ing in fhe way people describe their reac-tion to the content.

The main finding in this area is the lackof variation in response to the ad banners.If tends to corifirm that part of the adawareness increase observed for the Webbrowser's banner could be due to synergywith other marketing activity and expo-sure. While the lack of variation in thisdiagnostic data might be a function of fhebanners tested, it does suggest fhat thehigh-involvement, prominent placementaspects of advertising on the Web domi-nate the differences in creative impact.

An example of this uniformity of re-sponse can be seen in how the bannersexhibit little difference in the degree towhich people thought they were "eye-catching," as shown in Figure 15, Thesimilar results are especially noteworthygiven that the men's apparel brand andISP were animated banner ads, while the

. . . ad banners are similar to direct maii envelopes, en-

ticing the recipient to open it up, browse further, and

taice action.

50% -

45% -

40% -

35% -

30% -

25% -

20% -

15% -

10% -

5% -

0% '

50%

18%13%

Apparel Web browser

Figure 13 Baseline Specific Brand-Linked Awareness

4 2 JoufinflL Of flESEHRCH March . April 1997


25% -

20% -

1 5 % -

10% -

5% -

0% +•

23%

16%

Apparel ISP Web browser

Figure 14 Incremental Effect on Specific Brand-LinkedAwareness

Web browser was not. The power of anadvertisement to hold a consumer's atten-tion and be remembered at a later datevaries widely in other forms of media,even given a constant level of exposure.The main causes of the variation differ bymedium. In print, category interest is key,since people are actively searching foritems of interest; the creative power of anad plays a supporting role. For TV adver-tising, creative power dominates andviewer involvement is key. (Viewer in-volvement is a function of enjoyabilityand the degree to which attention in-volves active processing rather than pas-sive) (Hollis, 1995). Our results from thisstudy suggest that the Web is far moresimilar to print than TV in this respect.

DETERMINING FACTORS

OF CLICK-THROUGH

A focus of the Web advertising debate hascentered on the value of click-through.Click-through on ad banners transportsreaders from content focused Web sites toadvertiser sites where direct marketingcan occur. From this point of view, adbanners are similar to direct mail enve-lopes, enticing the recipient to open it up,browse further, and take action. Using this

logic, many people have argued in favorof the click-through rate as the best mea-sure of advertising response on the Web.However, we believe that the click-through response is determined by fivebasic factors, three of which relate to thepredisposition of the audience, not the ad-vertising itself.

Audience-related factors are:

• innate tendency to click on advertise-

ments

• immediate relevance of the product tothe audience

• preexisting appeal of the brand or com-pany

Advertising-related factors are:

• immediate relevance of the message tothe audience

• involvement or intrigue created bythe ad

The results of this experiment indicatethat those who click through had, asmight be expected, actual recall of the adbanner (which they may have seen in lo-cations other than HotWired). More im-portantly the brand had immediate per-sonal relevance to the user. People whoremembered the ad and who said theproduct category was of interest to themwere three times more likely than the av-erage to claim to have clicked-through.The fact that we observe little differencebetween the reported click-through ratesfor any of the banners tested indicates thatclick-through in itself may not predictthe level of brand building likely to berealized.

Men's apparel

ISP

Web browser

60%

50%

55%

77%

72%

75%

-I- I I I I I I I I0% 10% 20% 30% 40% 50% 60% 70% 80%

Somewhat eye-catching Very eye-catching

Figure 15 Standout Power

M a r c h . Apr i l 1 9 9 7 JQUHIIRL DP ROUERTISinO RESEHHCIJ 4 3


. . . the use of click-through rates alone are likely to un-

dervalue the Web as an advertising medium.

While more extensive studies need to beconducted to fully understand other fac-tors that might drive click-through, thesefindings suggest that ad agencies and con-tent providers who agree to be remuner-ated solely on the basis of click-throughrates are tying their revenues to currentlevels of interest in the advertised brandsof their clients. Since our study has proventhat advertising on the Web has sizableeffects on brand loyalty and attitudeswhich are not reflected in click-through,the use of click-through rates alone arelikely to undervalue the Web as an adver-tising medium.

If this finding proves to be a generalone, then tactics designed to generateclick-through may also underutilize thepower of Web advertising. The practice ofrunning "unbranded" banners, which hasbeen reported to yield high click-throughrates, surely runs counter to the concept ofbrand building through ad banner expo-sure. In short, click-through certainly hasvalue (especially if the goal is to create animmediate behavioral response such asdownloading software), but the click-through rate itself is unlikely to be indica-tive of the overall value of a banner expo-sure.

CONCLUSIONS

So what do our findings tell us about Webbanner advertising? Our results tell us,simply and unequivocally, that it works.Moreover, it works with or without theadded benefit of click-through. One single

banner exposure generated increases inConsumer Loyalty scores ranging from 5percent for the Web Browser to 50 percentfor the men's apparel brand. Advertisingawareness increased from a low of 12 per-cent (for the brand with the highest baseof ad awareness) to a high of 200 percentfor the previously unadvertised men's ap-parel brand. The FORCE scores we calcu-lated for the three banner ads suggest thatWeb banners may rival or surpass TV andprint advertising in producing brand-linked ad awareness. We observed, as ex-pected, a diminishing returns factor asso-ciated with levels of advertising weight.This factor dampened but did not extin-guish the advertising effects.

In addition, our findings to date suggestthat the high-involvement nature of theWeb and the prominence of the advertis-ing seemed to play a larger role in defin-ing ad effectiveness than did consumer re-action to any particular execution wetested. In this respect advertising on theWeb appears to work similarly to printrather than TV. A combination of factorsappear to generate click-through, the pri-mary one being the nature of the audienceand the inherent interest the product cat-egory holds for them. The appeal of thecreative plays a secondary role at best.

The Web offers unique and undeniableadvantages over other media in terms oftargeting and direct marketing. One suchadvantage is the ability of advertisingbanners to serve as gateways to an adver-tiser's own Web site. But our results sug-

gest that the ad banner is a legitimate ad-

vertising vehicle in its own right. This ex-

periment shows that banner ads on the

HotWired Network have a significant im-

pact on their viewers, an impact that de-

monstrably builds the advertised brand.

Banner ads remind people of a brand's ex-

istence, stimulate latent or dormant brand

associations, and can cause people to

change their attitudes toward the brand,

thus increasing their likelihood to pur-

chase. The unique marketing power of the

on-line environment and the established

communication benefits of traditional ad-

vertising combine to make the Web a

powerful new advertising medium with

real potential for brand building.

REX Bmeos is vice president of MBinteractive, a

division of Millward Brown tnternaiionai. Formerly, Mr.

Bnggs was director of research for HotWired. Mr.

Briggs began analyzing the impact of the Ihternet on

marketing in 1993. San Francisco-based

MBinteractive is a fuii service research practice

dedicated to heiping clients measure and understand

interactive marketing.

The Web offers unique and undeniable advantages over

other media in terms of targeting and direct marketing.

4 4 JOURnni DF RBUEfUISIHG RESEHHCH M a r c h • Apr i l 1 9 9 7

. Hoius is the Group Research and

Deveiopment Director at Miiiward Brown International,

in his current role he has globai responsibility for ail

projects related to improving Millward Brown's curreni

services or developing new ones. Over the last two

years he has worked on projects reiated to brand

equity, and interactive and on-line research.

Mr. Hoilis started work in market research at

Cadbury Schweppes m the United Kingdom. After

gaining a broad experience of different research

methodologies at Cadbury. he joined Miiiward Brown

In 1983. There he had a key role m the developnwnt

of Miiiward Brown's successfui TV Link pre-test. In

1988 he transferred to the United States and. until

moving to his curreni role, worked on a variety of

client businesses, mostly related to the analysis of

tracking research.


He has had papers published in the Journal of uity." Journal of Advertising Research 36, 6 advertismg effectiveness, available from IPC

,̂ £^611/517?̂ Research, Admap, plunung und analyse. (1996): 9-21. Magazines or Miiiward Brown International.

and the Journal of the Market Researcfi Society.HOLLIS, N . "The Link Between TV Ad Aware-

, AND A. FARR. "Effective Frequency,ness and Sales." Journal of the Market Research

New Evidence on an Old Debate." In the Pro-Society, January 1994.

REFERENCES ceedings of the 1995 Market Research Society

Conference. . "Like It or Not, Liking Is Not

DYSON, P., A. FARR, AND N . HOLLIS. "Under- Enough." Journal of Advertising Research 35, 5

standing. Measuring, and Using Brand Eq- FARR, A. "Ad Track 94." A study of magazine (1995): 7-16.

March . April 1 9 9 7 JDDHiL QF flDUERTISinG RESERUCH 4 5

Documents

Advertising on the Web: is Tiiere Response before Ciicii