In recent years, using cookie data in behavioral targeting strategies has been heralded as a game-changer by online marketers. A large part of this attention is focused on data management platforms (DMPs) that enable behavioral targeting through the use of cookie data to categorize a user’s website activity and optimize ad buying.
Although this has led to noted improvements in ad effectiveness for national campaigns, we posit that locally-targeted ad campaigns with strict geographic limits do not see as much of a benefit as from other targeting techniques, such as page-level semantic analysis, or contextual targeting.
This is an important insight for online marketers to be aware of, as the rapid implementation of DMPs is occurring at a time when retailers’ online marketing strategies are becoming increasingly focused on local markets.
Our examination of the effectiveness of behavioral targeting for retailers – large and small – casts doubt on the effectiveness of such targeting in locally-focused online display campaigns. Our case study suggests contextual targeting is superior at the local level in four ways:
- Stronger engagement in the form of higher click-through rates (CTR)
- Increased ad relevance to site placements
- Lower cost per thousand views (CPM)
- Lower cost per acquisition (CPA)
In this study, we compare one behavioral targeting strategy with a corresponding contextual targeting strategy at the local level to evaluate their respective CTR, relevance of site placement, CPM and CPA.
We challenge the hypothesis that big data analytics and behavioral targeting are the next frontier for locally-focused marketers by presenting data from 28 display advertising campaigns which employed carefully controlled and optimized behavioral and contextual targeting strategies. We discuss the findings of our case study in light of a recent shift by national and local retailers to focus on micro-campaigns with narrow geographic targeting parameters.
According to the advertising tracking firm, BIA/Kelsey, retail spending on local advertising will exceed $26.8 billion in the United States in 2013, with $4.2 billion being spent on digital media alone. This includes national advertisers that want to tailor their marketing to local markets as well as local advertisers that only operate within specific regions.
With this sharpened focus on local digital advertising by retailers, behavioral targeting strategies have been brought to the forefront as a potential solution for locally-focused campaigns. While the merits of behavioral targeting to large-scale advertisers are undeniable, it is unclear if such merits are equally valuable to locally-focused advertisers.
The below research represents one case study of how behavioral targeting strategies at the local level compare to identically designed and controlled contextual targeting strategies.
From August to November 2012, our company launched 28 display ad campaigns across 14 locations in a southern state on behalf of a national veterinary clinic franchise. This veterinary franchise relies predominantly on business from customers that reside within a five-mile radius of their individual units.
At the client’s request, we designed two test campaigns for each of the 14 target geographies. Each geographic target had one contextual and one behavioral targeting campaign. The data used in the behavioral targeting campaign was provided by Blue Kai. The contextual targeting campaign was powered by Peer39.
We applied additional attributes for results optimization, such as time-of-day, day-of-week, properties of the webpages, ad placement and size of the ad. We used the same creative design, text and call-to-action for each market and campaign, with the only exceptions being the city of the clinic as indicated on the ﬁnal frame of the display ad, the embedded map feature (which indicated the location of the unit on an interactive map), and the destination landing page.
Because our methodology only tracked click-through as an indicator of engagement and not acquisition/conversion, we cannot exclude the possibility that a few of the lower-performing campaigns may have resulted in higher ROI. However, in this scrutiny of 28 display ad campaigns across 14 cities, a small number of such inaccuracies would not materially affect our broad conclusions.
A total of 3.8 million impressions were served evenly across 14 cities within the same southern state from the period of 15 August 2012 to 10 November 2012; impressions were divided equally between two targeting strategies for each unit. The geographic targets for each location consisted of zip-codes within a ﬁve-mile radius. The table below summarizes the 14 units and CTR per campaign over the 87-day period.
Per above, in every unit, the CTR of contextual targeting exceeded that of behavioral targeting.
While the bar plot demonstrates a higher CTR across the individual units, the box plot shows that, regardless of location, contextual targeting outperformed behavioral. Additionally, the notable separation between contextual and behavioral CTR in the box plot buttresses the undeniably superior performance of contextual targeting in this study.
Taken as a null hypothesis that contextual and behavioral targeting are equally effective on average at generating clicks, it would be expected that the number of units where CTR for behavioral targeting exceeds CTR for contextual targeting would be about equal to the number of units where the reverse is true.
With a p-stat dramatically lower than 0.00001%, we must reject our null hypothesis as it is a virtual certainty that the observed difference in CTR did not occur by chance. Overall, contextual targeting generated clicks more than twice as effectively as behavioral targeting.
For large-scale campaigns at the national level, it is popular to believe that behavioral targeting optimizes to give a superior conversion rate. However, in the above local test, a significant handicap for behavioral targeting with tight geographic controls appears to be the limited data set available within the geographic parameters, which, in the case of the veterinary clinic, were paramount.
The contextual targeting campaigns – not being reliant on the data set within the geo-target – were able to identify relevant, page-level ad placements that boosted the CTR to approximately twice that of the cookie-based targeting strategy.
Our experience in this case, as well as with other locally-minded retailers, indicates that the effectiveness of behavioral targeting relies not only on the number of impressions served, but also on the size of the geographic target and population within. Only when these two variables are ample will there be adequate ad buying opportunities to be selective. In practice, behavioral targeting is outperformed by contextual targeting at the local level.
Brand Safety Results
The table below shows the top 15 sites per targeting strategy during the August to November market test.
As shown in the table above, another practical disadvantage of behavioral targeting is site list relevance. For example, a pet-owner cookie resulted in the veterinary clinic ad being shown on sites such as meetme.com and wowhead.com.
However, using contextual targeting, the same ad naturally experienced higher click rates on sites such as petﬁnder.com and kittyﬂix.com. While the site lists for contextual targeting are also unpredictable, the table above indicates a seemingly greater degree of brand safety afforded the advertiser, given their target.
An analysis of the CPM of each targeting strategy revealed that, on average, behavioral targeting was 1.94 times more expensive than contextual.
The above charts demonstrate the ratio of behavioral CPM to contextual CPM. Of significance in the box plot is the concentrated grouping of contextual targeting, compared to the more widely distributed CPM for behavioral targeting. For media planning purposes, stability and predictability of CPM for locally-minded campaigns are critical in budgeting both spend and impressions.
With a p-stat dramatically lower than 0.00001%, we must reject our null hypothesis as it is a virtual certainty that the observed difference in CPM did not occur by chance. This ratio ranged from 1.37 to 2.67.
Since CPA is proportional to CPM/CTR, the ratios of CPA for contextual versus behavioral targeting are larger than those for CPM, ranging from 2.32 to 7.36.
The above calculated values of CPA assume a ﬁxed conversion rate consistent with typical conversion rates for demonstration purposes. Though the conversion rate of contextual and behavioral targeting are equal for all units, we make no assumption about the numerical value of this conversion rate.
As the difference between behavioral and contextual CPA shows a consistent direction (behavioral targeting is more expensive per acquisition than contextual targeting), the p-value of the null hypothesis is again 6.104 x 10^(-5) by a sign test. The higher CPM and lower CTR for behavioral targeting results in a cost-per-acquisition that is 3.18 times more than that of contextual targeting.
Big data solutions such as behavioral targeting rely on just that, big data. In the absence of a sufficient data set – which is the case for many local advertisers, especially those outside of urban settings – behavioral targeting does not have a sufficient number of ad buying opportunities to be selective.
Retailers are better served in terms of engagement (CTR), brand safety and CPM/CPA to employ targeting at the local level that enables real-time identification of inventory that is relevant to the advertiser’s message.
Opinions expressed in the article are those of the guest author and not necessarily Search Engine Land.