Putting Engagement To Work
Soon, measures of Time Spent were recast as engagement metrics. The ARF posited a definition of engagement: "Turning on a prospect to a brand idea." The definition proved to be a lightning rod, with many observing it rendered engagement unmeasurable. And to this day, conceptualizing engagement continues to remind of the story of the 5 blind men describing the elephant.
Erwin Ephron offered an interesting working definition of engagement: "The sum of all measurable variables that significantly affect the probability of viewer response to the ad message." Erwin's construct introduced two important concepts: the requirement that engagement variables be measurable; and, the idea that engagement is a sort of grand metric that is comprised of the sum of component parts. In a sense, it reminded me of the way the NFL ranks quarterbacks. Erwin's engagement definition represented a significant step toward measurability and relevance.
For the past two years, Eric Peterson of Web Analytics Demystified has been refining and vetting a working definition of engagement in papers and on his blog. In Peterson's definition, engagement is indeed the summation of an array of observable phenomena describing the visitor's interaction with a Web site. A couple of weeks ago, Peterson wrote on his blog about some work that comScore and Web Analytics Demystified have done to apply the calculation to comScore's audience data.
Previously, publishers had been able to adapt this calculation to their own internal data -- but that was of limited utility, save for providing the ability to trend one's own performance over time. But by running a modified version of Peterson's formula against comScore's person-centric panel, we are able to place a site's engagement score into context; we can let publishers track their performance compared to their content category, their peer group, their competitive set.
So what is engagement? In the work we've done, we streamlined Peterson's calculation somewhat in order to operationalize it against a syndicated data base. The four variables we settled on are: click depth; duration; recency, or "visit velocity"; and, loyalty (expressed as share of visits in-category.)
Suppose I told you that cnn.com had an overall engagement score of 25.9%, and that 22.5% of its audience is "highly engaged." What might you make of that? And of course, it's a rhetorical question; without context, there isn't much that can be made of it.
But suppose now I told you that the average engagement score for the news/information category was 17.5%, and that the average "highly engaged" visitor score for the category was 14.5%. With this context as backdrop, now we can make inferences from the engagement scoring. CNN's audience is almost 50% more engaged than the average news/information Web site audience; its concentration of highly engaged audience is about 55% higher than average.
Next, suppose I were an automotive marketer. Would it be interesting to me to know how the different ad-supported automotive Web sites compare to one another with respect to the percent of their audiences that are highly engaged? Another rhetorical question: one would expect to find a high correlation between the level of audience engagement to an automotive Web site, and the effectiveness of automotive advertising placed on the site.
Indeed, this might well prove a fertile ground for further exploration, since we can empirically study the effectiveness of advertising. Empirical, objective engagement scoring can allow us to quantify the relationship between audience engagement and ad effectiveness, enabling publishers and advertisers to more fully operationalize the concept of engagement into the buying and selling of advertising. Which was Stengel's original point at the 4As in 2005.
What do you think? I'm not going to ask if this is the precisely "right" way to conceptualize engagement, because I doubt a group as diverse and as well-informed as the MediaPost readership could ever reach consensus on so nuanced a topic. So let me ask instead: is this a useful construct? Let me know, either with a comment, or via email.