ARF Insights: Understanding Earned Media: ARF Interview with Seth Duncan, Research Director at Beyond

While word-of-mouth marketing was once mainly the purview of PR agencies, social media data has given market researchers a way to quantify earned media like never before. Here at the ARF, we spend a lot of time helping our members figure out how to translate social media data to business intelligence. We caught up with an expert on the subject to get some insight. Below, Seth Duncan, Research Director at Beyond shares his thoughts about how to optimize earned media.

ARF: Briefly describe the challenges facing advertisers when trying to get insights from "uncontrolled" social media data.

Seth Duncan: Consumers can easily share campaign content through social media channels in a way that wasn't possible a decade ago. Whether you want them to or not, consumers often take content from commercials, print ads, digital ads, and social media campaigns and share them on their own blogs or on social networking sites such as Facebook or Youtube. Just 10 years ago, if someone saw a commercial that they did not like, they might make a snarky comment to the other 2 or 3 people in the living room. Now, if someone does not like a commercial, they can share a snarky comment with hundreds or even thousands of people through Twitter.

This creates two huge measurement challenges for advertisers. First, media buyers can no longer report exactly where your campaign appeared, how many impressions it received, which demographics and psychographic profiles it reached, etc. In most cases, the main audience for a campaign will still be the people who are reading a specific magazine or website where the advertisement appeared, or are watching television. Planned or not, there is the occasional campaign that "goes viral" and the audience is much more difficult to track and measure in these cases. This is especially true for small to mid-size campaigns where the audience is too small to be measured by existing online panels. The immediate impressions from a known campaign placement have to be supplemented by additional data sources, including data from social media listening software, channel-specific analytics data such as Facebook insights, and brands' own web analytics data. Social media measurement is a nascent practice, so there is no single sophisticated solution for measuring "viral" campaign reach. Figuring out exactly who saw your campaign can be an intensive and time-consuming process.

The Second challenge is that advertisers no longer have complete control over the messaging around the campaign. Most of the time, when consumers share advertising campaign content with their friends or family online, they share the same messages that the brand intended when the campaign was created (e.g., the thousands of people that write "the real thing" or "a coke and a smile" on Coca Cola's Facebook wall). There are other times, however, when campaign content is shared in a decidedly off-message manner. In some cases, being off-message is relatively innocuous (e.g., consumer-generated parodies of MasterCard's' priceless commercials). In some cases, though, consumers can add very negative commentary to advertising campaigns. For example, many people saw negative Tweets discussing how Groupon's Super Bowl commercials were insensitive, even if they never saw the actual commercials. When an advertising campaign receives a lot of attention in social media, not only do you have to figure out where those conversations are happening, but you now have to figure out whether or not the intended campaign messages are being disseminated as well.

ARF: What challenges do social media data pose to advertising researchers?

Seth Duncan:Ultimately, gaining insights into how advertising campaigns reached online audiences is much more exploratory than advertising research has been over the past few decades. There has always been a discovery-component to how campaigns resonate with target audiences, but now the discovery reaches all the way back to what audiences actually saw the campaign and which campaign-related messages they saw. In this respect, measurement for social media marketing looks very similar to the media measurement methodologies that Public Relations professionals rely upon to understand how reputations are shaped by influential peoples: searching through hundreds or thousands of written pieces of content (whether newspaper and magazine articles, or blog posts and other digital forms), and performing content analysis to understand the topics, messages, and sentiment being discussed around a brand. Social media presents an unusual case where advertising research can learn something from public relations research now that consumers, not just industry experts and journalists, act as influencers in the media.

ARF: How can advertisers and researchers tell which social media channels are being used to share content?

Seth Duncan:To understand the interplay between paid and earned media, brands must integrate data from a wide-variety of sources including their media buyer, the social media channels that they curate (Twitter, Facebook, YouTube), earned media channels such as blogs, and digital properties that they created themselves. At the moment, there is no single searchable database that aggregates all social media conversations (or, at least not one that is exhaustive), so this process involves searching multiple databases and doing a little forensic research. Below is a list of data sources that brands should leverage to understand how their content is being shared:

a) Social media monitoring databases such as Radian6, Sysomos, or SM2. These databases are useful for collecting posts on blogs, message boards, Twitter, and the public sections of Facebook. Like any database, you have to find content using specific keywords or tags.

b) Web analytics for campaign websites.If consumers are using unexpected language to describe your campaign, you might not be able to discover their content through these databases. Web analytics referring site reports for campaign websites often reveal conversations around the campaign that were missed by social media monitoring tools.

c) Channel-specific analytics and APIs. Many social media channels only offer analytics and site-referral data through their own UI or API. For YouTube videos, for example, you have to download an analytics report directly from you own channel to see the top referring sites—sites that might not be picked-up by social media monitoring tools. Similarly, for campaigns that might have a location-based social media component, conversations on Foursquare (i.e., "tips") must be collected directly from Foursquare.com.

d) Mainstream media databases. The comment sections of online traditional media sources often contain discussions of advertising campaigns. The leading social media monitoring tools only cover a limited number of traditional publications, so you have to rely on media databases that license premium content (e.g., Wall Street Journal, New York Times, etc.), such as Dow Jones Factiva or LexisNexis.

ARF: What role does statistical modeling play in gaining social insights?

Seth Duncan: Statistical modeling touches nearly every part of digital media measurement now. From the logistic regression models that Google Website Optimizer uses to help brands with their campaign landing pages, to the Bayesian regression used in everything from online sentiment analysis to advertising effectiveness control groups (e.g., comScore's AdEffx control groups), statistical modeling is inescapable in digital marketing research today.

Applying statistical models to understand the effectiveness of advertising campaigns' secondary, viral effects in social media is relatively rare today, but it is necessary because the attributes of social media conversations are unpredictable and cannot be measured using experimental design. You can use experiments to test which types of online ads, for example, produce better recall, or drive higher conversion rates than others because you can manipulate which aspects of the ads differ from each other, including the copy, the background color, size, etc and expose different variations to test and control groups. But, at the moment at least, web analytics tools are not very adept at connecting consumer conversations online with KPIs. Organic online conversations never occur in controlled laboratory conditions with test and control conditions that allow advertisers to make causal inferences about aspects of an ad campaign are effective. Consequently, just like any other type of naturalistic field research, we have to apply statistical modeling to make inferences about how various online conversations impact online business outcomes.

Want to hear more? Join Seth Duncan for a presentation on "Earned Media Optimization" the ARF Audience Measurement 6.0 Symposium. Registration is open.