Thanks To Deal-Based Ads, Last-Click Attribution Was More Accurate In Q4

visual-iq-last-click-april2013Online display advertising is still undervalued when it comes to attribution based on last click data, but during the holiday season 2012, last click attribution was slightly more accurate, according to marketing intelligence company Visual IQ.

The company analyzed cost per acquisitions for Visual IQ clients for the time period August to October 2012, and found that most display ads were undervalued. Visual IQ has its own TrueCPA metric, which it compares to the traditional last click CPA, creating a Cost Per Acquisition Skew, which determines how over- or under-valued the ad is.

"We use a set of algorithms to re-calculate clients' metrics so that it takes into account all cross-channel and cross-campaign influences so our clients are not basing their optimization strategies on last-click metrics, where 100% of the credit for the conversion is being given to the last click," explained Visual IQ CMO Bill Muller. "This conversion skew reflects the difference between those attributed metrics and last-click metrics."

According to Visual IQ's data, between August and October 2012, 94% of online display advertising spend went to placements where the last-click CPA was at least 70% higher than Visual IQ's calculated TrueCPA, showing how undervalued the online display advertising was with only last-click metrics. Yet looking at October to December 2012, only 52.4% of online display ad spend went to such ads. On a few occasions, online display ads were overvalued, with 0.6% of Visual IQ's clients' display ads having a CPA Skew of 70% or more between August and October, jumping to 6.6% of ads during the holiday season.

"With our clients, we look at the data the way they see it, which is generally last click, and we compared that to what we have, which is the data we're discussing now and how different those can be," said Brian Suh, director of analytics for Visual IQ. "Then we transition to, instead of using last as a reference point, they would use our metrics, which accounts for cross-channel and all the multiple touchpoints."

The data Visual IQ analyzed was, therefore, not a reflection of what clients were actually doing (which was generally using the Visual IQ TrueCPA), but was a look at what they would have seen if they had used traditional last-click metrics.

Comparing the August to October time frame to the October to December 2012 time frame, online display ads were slightly more accurately valued during the holiday season, as the deal- and discount-heavy display ads contributed more conversions.

"The trend that we noticed during the holiday season was that last click was getting slightly more accurate," said Suh. "I describe it as 'we are less incorrect than we were before,' but we're still pretty darn incorrect. From what we can tell, it was really driven by the deal-based nature. If you think about Black Friday or Cyber Monday, they are huge deal days and there are huge discounts being advertised there, which makes sense with the data."

As clients turn to the tools provided by Visual IQ, and other cross-channel marketing analytics companies, Muller and Suh hope attribution metrics can become more standardized and accurate industry-wide.

2 Comments

  1. Matthew Anthony

    The premise of this article is frankly ridiculous. If a performance metric's accuracy is a function of attributes of the things it's measuring, then the metric is flawed and incorrect. The viability of an estimator/metric should be fundamentally sound across all applicable elements to which it can be applied - see the idea behind BLUE/BUE estimators in statistics. Can we please stop peddling in bad-math and pseudoscience as if it were profound?

    Reply
  2. Thanks for the comment Matthew – you make a great point and we completely agree that accuracy should not be a function of what’s measured. Last-click methodology decreases inaccuracy during the holiday period, which is exactly why it is flawed. At any given time one has no idea how accurate or inaccurate the metric is, and that’s why making business decisions based on that optimization methodology can be so dangerous. Fortunately, Visual IQ’s TrueMetrics accuracy does not change as a function of what’s being measured. Our machine learning algorithm adapts to the nuances of any data set and consistently provides accurate attribution, which we validate through further testing and analysis.

    Reply

Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>