Bill Harvey and Fadi Karam
Brand growth is the mantra of marketers. Today the number of new tools available to help marketers achieve brand growth is multiplying annually.
The new focus is on large scale databases, data science, artificial intelligence, biometrics, and the beginnings of a true marketing science. However, all of this is competing for attention within the context of established marketing processes, advertiser-agency relationship structures, and higher order degrees of complexity of communication. The result is that “good enough” often substitutes for adoption of proven innovations.
This is a case study not only of high relevance to all marketers for the importance of its discoveries, but also as exemplary of the openness to new ideas and the rigor of proofs that the best companies are employing to ensure brand growth against a horde of competitors who often have more marketing dollars. We thank the CPG advertiser for its leadership in giving permission to publish this anonymized case study for the benefit of all marketers.
II. Brand Background
Brand X is a CPG brand that competes against brands within the portfolios of some of the largest CPG companies in the world. By comparison with the spend levels of many of the competitors, Brand X has a modest advertising budget. In order to grow the brand under these circumstances, the company sought to understand the alternative innovations available for advertising optimization, and to select an approach that had already achieved credible third-party validation.
III. Creative and Media Silos Not Integrated
The content of an ad is generally not considered as relevant in deciding which media to buy. The agency generally optimizes reach against a specified target which has nothing to do with the ad; the target either is a demographic group or a group defined based on claimed, inferred, or actual past buying behavior.
What about the motivational appeals in the specific ad, though? Is an ad is motivationally appealing to affectionate people, for example, would it not perform better in a program with affection as one of the program’s central themes?
After a study of over 300 studies on the subject the Advertising Research Foundation (ARF) has concluded that these context effects are as significant as the lifts provided by the best forms of targeting. ARF now considers best practices to include consideration of both targeting and context as equally important.1
And yet there is reason to question whether these general practices are actually best practices or not. Ogilvy Vice-Chairman Rory Sutherland, writing in Campaign, says “Why try to identify your target audience in advance when the right creative gets them to identify themselves? In mass media, the targeting takes place in the mind.”2 In other words, the content of the ad should be considered in the targeting of the media. One ought to seek to reach people most likely to be responsive to the specific ad.
Practitioners have jumped to the conclusion that the most likely ad responders are people who buy competing brands, i.e. they already buy the category. This is definitely a relevant partial predictor of ad response, although it excludes those who have not yet been tracked buying the category because they don’t use frequent shopper cards and, in a growing category, it excludes what might be the largest source of ad responders, those who are considering the category.
What about the content of the ad? Would that not also be a partial predictor of likely ad responders? If an ad depicts a certain lifestyle, for example, isn’t it a reasonable assumption that people who aspire to that lifestyle would be more likely to respond to that ad?
IV. DriverTags and Need States
A group of researchers led by Bill Harvey rigorously developed a system called DriverTags by which it is possible to discern which of 265 psychological motivators are found in a specific ad. The same 265 motivators can be found in specific people, making it possible to target the ad to people most likely to be motivated by a specific ad.
The same 265 motivators can be found in the content of television programs so as to select the contexts for the specific ad most likely to enhance advertising effectiveness for that specific ad.
The 265 motivators (“DriverTags”) were derived in two stages. In the first stage, attitude scaling and factor analysis of over 10,000 words extracted by exhaustive and systematic means from the Oxford Unabridged Dictionary yielded 1562 motivators. These were then attributed to television programs by two out of three coder agreement, and used in a personalized program recommender tied to set top box measurement, that reduced the list to 265 which accounted for almost all of the variance in the conversion to loyal viewership of programs recommended.
Semantic clustering of the 265 motivators shows that they fall into 86 Need States, which help explain why specific motivators work. The 86 Need States telescope into 15 Values Domains about half of which were predicted by the work of Abraham Maslow3. A comprehensive description of this development work is provided in an ARF white paper4. How the system would be used in a next generation ROI cross platform media/creative/context optimization system is described in another ARF white paper.5 Further theoretical structure is provided in a pair of articles.6
V. Validation of DriverTags
Prior to this study there had been two other independent third-party validations of DriverTags. The first was presented by ARF EVP Horst Stipp at the March, 2017 ARF Annual Conference, and was a study sponsored by Turner, for which Turner chose to use Nielsen Catalina. The study showed that when ads and programs carrying those ads had DriverTags similar to one another, the sales effect of the advertising increased by double digits.7
Ad Context Matters – Sales Effect Lift
Nielsen Catalina study for Turner showed DriverTag Resonance is worth +36% Sales Lift.
The second study was presented by Simmons at ARF Conferences in 2017-2018, and showed that DriverTags explained variance in brand usage incrementally to demographics and incrementally to ~600 attitudinal questions, across all 3,830 brands tested. Where the brands were television programs, DriverTags explained far more variance than demographics or attitude questions.
A separate Simmons white paper reported that when used as fusion variables, DriverTags outperformed demographics and even outperformed specific program viewing, in accuracy and in self donation (identification of the best match between respondents in two overlapping samples, where only the DriverTag method assigned donor-recipient pairs which were actually the same person in both samples).8
VI. Marketing Decision Process
Primarily based on the Nielsen Catalina Solutions (NCS) validation, DriverTags were selected as a platform that would be used by Brand X in 2018. Because it was assumed that NCS had already proven DriverTags predictive of sales lifts, the company decided to measure branding effects instead of sales effects in its 2018 study of DriverTags.
Although most of the media buy had already been committed, the agency was able to include certain networks which would, given the specific two ads in the campaign, increase the incidence of higher resonance impressions in the database according to the DriverTag data, providing more of a spread in resonances and therefore a sounder basis for the analysis.
605 was engaged as the data analytics partner to conduct the brand attribution study and provide third-party verification of findings. Selection of 605 reflected the importance of national sample size of 12M+ households (given the low penetration of Brand X) in addition to 605’s unique single-source approach to TV attribution and brand lift measurement.
605 developed a custom brand survey that contained specific brand metrics (tracked by client) including “Unaided Awareness”, “Aided Awareness”, “Purchase Intent” and various “Brand Perception” attributes.
605 provided as-run logs to RMT containing details regarding each/every ad placement, ad position, time-stamp, etc. RMT scored each of the two creatives and placements within the 3,000+ pre-existing programs in the RMT database, in addition to tagging all new shows in the linear plan (to ensure measurement across each/every ad placement.)
RMT Resonance scoring is based on the DriverTags in the ad. One ad had 24 DriverTags, the other ad 26. If a program had only 1 DriverTag in common with the ad that had 24 DriverTags, the calculation for ad-program Resonance was 1 (DT in common) divided by 24 (# DTs in the ad). The Resonance score would then be 4.2%.
Using these Resonance scores, 605 appended those scores to the impression record for each household in the set top box (STB) measurement sample. For example, if the program whose Resonance with the first ad was 4.2%, was shown to 605 STB household 11,111,111 at 8:00PM Eastern Time on August 12, 2018, that impression Resonance score of 4.2% was added to that household’s record relative to that exposure. In this way, for the ~12,000,000 605 households, Resonance scores for ~216 million ad exposures were appended.
Ad exposures were matched onto the households by 605 using monitoring data which established the exact minute:second each ad started and ended, and the channel on which the ad was run. Any household with a STB tuned to that channel during that time span was counted as exposed.
Over 14,000 households anonymously completed the company’s online questionnaire. ~82% of these were exposed to the campaign. 605’s proprietary algorithm was used to control for differences between the exposed and unexposed groups based on ~5000 demographic and behavioral attributes leaving only the ad campaign to explain any differences in branding metrics between exposed and unexposed.
For analysis purposes, each household exposed to the campaign was coded based on the average Resonance of all campaign exposures received by that household, and then these averages supported the provision of a choice of three codes for each household: Low Resonance, if the average Resonance of all exposures received was under 10.0%; Moderate Resonance if 10.0%-14.9%; and High Resonance if 15.0%+. These three groups were of roughly equal size, hence, approximately tertiles. The average Resonance in the Low Resonance group was 7.9%; in the Moderate Resonance group, 12.7%; in the High Resonance group, 17.7%.
Also for analysis purposes, each household exposed to the campaign was provided a choice of two codes. If at least one campaign exposure was received at Resonance of 30.0% or higher, the household was coded “Peak”, or if none of the campaign exposures were received at that level, the household was coded as “No Peak”.
- All brand metrics showed higher lifts where average Resonance was higher. For example, the lift in Unaided Awareness attributed to campaign exposure was 62% higher in the High Resonance group than in the Low Resonance group.
- For Aided Awareness, the lift was 24% higher in the High Resonance group vs. the Low Resonance group.
- For Purchase Intent, the lift was 37% higher in the High Resonance group as compared with the Low Resonance group.
- Across the 8 Brand Perception statements, the lift produced by the campaign averaged 32% higher along the High Resonance vs. Low Resonance groups.
- As with all television advertising campaigns, about a third of all households reached received Low Frequency (known informally as “TV Quintile Effect”), which in this case was fewer than 7 exposures during a 15-week campaign. As to be expected, all brand effects metrics lifts were far below average in this group. However, within the Low Frequency group, some households received a single exposure with 30.0%+ Resonance (and a handful received more than one). Those Low Frequency households coded as “Peak” for having received at least one (typically just one) exposure at Resonance 30.0% or higher, showed far higher lifts in Unaided Awareness and in Purchase Intent – almost 6X higher in Unaided Awareness, and infinitely higher in Purchase Intent (as the Low Frequency/No Peak group showed a slight decline in Purchase Intent compared to the unexposed control).
- It was shown that the agency could have for the same budget have achieved more than double the average Resonance for the campaign, without sacrifice of efficiency, reach, or target density. The campaign actually averaged 10% Resonance and could have achieved 22% by making the same type of buy (all rotations) on a different set of network dayparts. Based on the findings of the 605 study, had the more optimally Resonant schedule of network dayparts been used, the average brand metric lifts would have all significantly increased.
IX. Implications and Recommendations
The study is further proof that context makes a large difference in the branding effects of advertising campaigns. The ARF brought forth a vast amount of similar evidence in March 2017 as relates to the importance context effects. The size of the upward effects are in line with similar size effects achieved by the use of purchaser targeting.9
And yet, while purchaser targeting is now widespread, context optimization is not. For brands seeking to grow at the expense of their competitors, the use of context optimization is strongly indicated, particularly before instituted by those competitors.
In television where the skew to heavier viewers (the 40% heaviest viewers receive 80% of the ad impressions) always results in about a third of the exposed group impoverished in terms of frequency, having more 30%+ Resonant exposures serves as a form of “low frequency insurance”. Having a low frequency group cannot be avoided, but its negative effects can be upgraded by Resonance.
Although in many cases the far higher Resonance levels which can be achieved by buying programs rather than rotations would justify twice the premiums being charged for buying programs, conservative media agencies can continue relying on rotation buys and still achieve significant increases in Resonance by knowing which network dayparts Resonate best with their ads.
1 Stipp, Horst, “Best Practice: How Context Can Make Advertising More Effective”, Journal of Advertising Research, June 2018
2 “Advertising is in crisis, but it’s not because it doesn’t work”, Campaign, January 24, 2019 https://www.campaignlive.co.uk/article/advertising-crisis-its-not-doesnt-work/1523689
3 Maslow, Abraham, Toward a Psychology of Being, 2015
4 Harvey, Bill, The Next Frontier: Content Analytics, Advertising Research Foundation, June, 2015
5 Harvey, Bill, Crossmedia ROI Optimization Must Include Creative, Advertising Research Foundation, June, 2016
6 Harvey, Bill, Engagement without Attraction is Not Enough, MediaVillage, January 23, 2019
Harvey, Bill, Discovery of the 86 Need States, MediaVillage, January 25, 2019
7 Harvey, Bill and Shimmel, Howard, Quantifying the ROI Impact of DriverTag Context Resonance, March 22, 2017
8 Pellegrini, Pat, Millman, Steven, Barber, Tamara, and Harvey, Bill, Using Measurement Science to Link Traditional and Emerging Research Methods with Calibration Panels, Simmons Research, October 29, 2017
9 Larguinat, Laurent and Harvey, Bill, The New Research is Bringing Transparency Between Marketing and Finance, ARF, AMS 2011