To paraphrase the tagline used in commercials by Ally Bank: "Even children know it takes more than a single ad to convert a prospect."
When optimizing online campaigns, the overwhelming majority of advertisers still credit only the last ad event that occurred prior to the conversion. That last event is typically in the form of exposure to a display banner or a click on a paid search link. This is usually referred to as the ‘last ad’ model. Some advertisers will only attribute credit to an ad if it was clicked – the ‘last click’ model.
It’s no wonder, then, that every third post on this blog seems to be about "attribution."
For those who have spent the last several years living under a rock, when media practitioners say "attribution" or "multi-touch attribution" they are usually apportioning credit for each conversion measured across all advertising a consumer was exposed to prior to converting. That seems like the obvious way things should be done, but it is a giant leap forward from the traditional approach still commonly used today.
Indeed, in recent years we have seen a sharp increase in usage of attribution management as advertisers look to extract maximum value from their digital budget. Multiple players and methodologies compete for the best way to attribute conversion to the ad/media that drove it. The field is data and method heavy with companies usually boasting about the superiority of their statistical techniques or the visual allure of their dashboards as key service differentiators. In some instances, advertisers conduct their own analysis with campaign data available from third party servers such as Atlas and Doubleclick.
There is, however, a major challenge to the explanatory value of attribution analyses as currently performed. And it is not in using logistic regression instead of Bayesian networks.
Most attribution management solutions today can only credit the ad events and touchpoints that were tagged before the campaign even started. Obviously, what you don't tag you won't see - and very often you simply cannot put a tag on the proverbial elephant in the room.
As a result, attribution platforms tend to over-credit Paid digital media - which is usually properly tagged - while ignoring other digital touchpoints, e.g., earned media and competitive marketing. The latter, in theory, should be given negative credit if it takes consumers away from conversion on your site - not doing so masks the positive impact of native advertising in balancing/neutralizing the competitive messages (an ad seems to be ineffective in driving the conversions, but things would be much worse without it). Not all of the factors out there that affect ad performance can be influenced by an advertiser - yet they have to be accounted for as they have the potential to skew the results of the campaign.
Another problem is dependence on cookies. A proper attribution analysis can only be done on the respondent level, across that unique person's purchase life cycle. Cookie deletion, when its rate is high, makes a single respondent appear as two, three or 14 different individuals, inevitably eroding the model. Moreover, the access from multiple devices ( work and home PCs, tablets and smartphones) by the same users brings additional noise into the system.
The most crucial omission of the current solutions, however, is their limited ability to answer the "why" questions. Why was that ad so influential? Why was that particular path to purchase so popular? Is that because of ad message, ad placement or the effects of targeting that served the ad to the right demographic/behavioral segment? What was the state of target audience before the exposure to the campaign? The crucial demographic and behavioral variables are rarely available - the only variables to explain the performance of sites and ads remain, well, the sites and the ads.
In order to overcome this, attribution solutions should go hybrid, combining tagging/cookie pool data with the data from online behavioral panels. The former will ensure that the entire span of the paid media campaign is captured by the analysis; the latter will correct for cookie deletion while enriching the data sets with a wealth of competitive information as well as explanatory insights on campaign audience - the demographic, psychographic and behavioral variables that cannot be gleaned from cookies and tags. The panel data can also be used to account for the impact of the Earned media touchpoints that cannot be tagged - and assign due credit to them as well.
Onward to Attribution 2.0!