Finding Bobby Orr

In 1972, Phil Esposito of the Boston Bruins set the record for goals by an NHL player, with 76 (shattering the previous record of 58.)  Earlier in his career, Espo had been known as a "garbage man;" a guy who would camp out by the goal and reap the benefits of his teammates' hard-fought play, finishing up an offensive foray by bashing the puck past the sprawled goalie.

That same season, Espo's teammate Bobby Orr obliterated the all-time record for assists, a far less glamorous stat, notching 102.  In the spring of '72, I watched that Bruins team eliminate my beloved Rangers from the playoffs, and I will tell you now -- and I'm sure anyone who saw that team play would agree -- that Orr, the set-up guy, was by far the better player.  He was poetry in motion, the most engaging player of his era.

(Both records have since been broken by Wayne Gretzky.  Gretzky was sort of like the Google of hockey.)

Last week I moderated the "Metrics Maelstrom" panel at Ad:Tech.  A couple of key themes emerged from panelists Jon Gibs (Nielsen Online), Young-Bean Song (Atlas), and Beth Uyenco (Microsoft, and new co-chair of the IAB Research Council.)  One of these themes was the need to understand the relative roles and impacts of the broad plethora of marketing communications to which a consumer is exposed. 

Like hockey statisticians, we advertising researchers have a tendency to attribute undue weighting (or credit) to the things we can easily measure and see, especially to the extent that these things are close to the goal (which, for marketers, is a sale.)  Young presented some of the work Atlas has recently done on Engagement Mapping (that's hyperlinked for a reason, and I heartily recommend you go and check it out.)  The premise of Atlas' Engagement mapping is to allow advertisers and publishers to "evaluate every touch point, not just the last."  Young showed a great illustration of the concept using the Corona beer brand, which I won't repeat because I don't want to spoil it for him, since I'm sure it kills at all the conferences.

David Smith called this effect the "Multiple Attribution Model" in this very space; others have called this the "assist model," because it endeavors to give credit to the touches that proceed the activity associated with the desired, elicited consumer behavior (hence my hockey metaphor.)  The general idea is that as digital platforms enable advertising and commerce to exist in the same platform, we are accruing data that tells us, for example, what percentage of exposures to an ad lead to a click, and then to a sale.  But understandably, we've fixated thus far on the last ad, tacitly allocating all the credit to that exposure, despite the fact that we know that isn't the way advertising works.

Consider the consumer who goes to Google and types in "buy iPod." (Sorry, I've got iPods on the brain; mine just crashed.)  Three ads pop up at the top, and eight down the side, all promising an iPod retail experience a click away. (The first organic result, happily, is the online Apple store.)

But what brought the consumer to Google in the first place?  Obviously the fact that the search was for a brand -- as opposed to a generic (" buy mp3 player")-- shows that there were other factors at play, other marketing communications, as the consumer made her way through the funnel.

At my company, we've done much work on the latent effects of search, and on the offline impact of online advertising.  With respect to the former, in one study of consumer electronics searchers, we found that 25% of all searchers ultimately made a purchase in the category.  However, of these purchasers, 92% of them made that purchase offline; and, of the 8% who made their purchase online, only 15% made the purchase in the same user session; we observed the other 85% making their purchase in a subsequent online session.  The value of those search clicks would be vastly understated if we failed to take into account the offline and subsequent-session behaviors, and the "assist" construct allows us to do just that.

Once upon a time, it was easy to conceptualize the affects of advertising.  You spent a lot of money on TV, and your sales went up, and the increase covered the cost of the TV advertising, or it didn't.  And if it didn't, your accountant told you that advertising built the bookkeeping asset of "Good Will," and so you shouldn't think about that money as a short term measurable expense anyway.  So, heck, next quarter you increased the TV budget 10% and everyone was happy.

Today, though, it feels like we all have just enough data to be dangerous.  We have what looks like causal data on the immediate impact of ad exposure on sales; but as Young and others have wisely pointed out, we need to look past this sort of obvious data if we are to disentangle the true effects of various channels, vehicles, exposures and executions on sales (not to mention brand equity, recall, awareness, engagement, and all that other soft squishy good marketing stuff.)  It seems the more data we have, the more we realize we need.  Ultimately, we will develop complex models that aggregate the experiences of individual consumers into metrics for evaluating advertising performance at different places in the funnel.  This will require a consumer-centric approach to measurement; we'll need to follow individuals throughout their media and commerce lives, and build up holistic predictive and evaluative models from the most granular (and rich) data set: the customers themselves.

Tags: metrics
Recommend (1)