Ad Fraud: The Enemy of Analytics!
Analytics depends on good data. Bad data in often means bad data out. For analysts we typically depend on a good accounting of the actual number of impressions seen by consumers. Because of the methods that are in use, we derive the result of those impressions based on a number of approaches.
Digital Media and Marketing Analytic Methods
Last Touch Attribution – This method measures clickthru’s and conversions as the primary measure of success
Statistical Modeling – This method uses a more indirect approach looking at variations in consumer drivers (such as impressions) and tries to correlate them (using regression modeling) to peaks and values in sales volume
Agent-based Modeling – This method uses a consumer-centric approach to model how consumers respond to media by modeling, among others, how advertising builds awareness, purchase intent and brand relevance
If the media data is incorrect, the three methodologies can give varying answers as to the effectiveness of a particular media channel.
Determining marketing effectiveness in the face of digital media fraud
If the fraud is the same, proportionately, for all periods, each modeling method should deliver accurate answers to the net effectiveness of the fraudulent channel. The accuracy of the methods is more dependent on when there are varying levels of fraud, period by period. If one period has more fraud then all other periods, this will have varying impact on the accuracy of the three methods as follows:
1. Last Touch Attribution
Last Touch attribution can generally determine the exact number of click thru’s generated through a specific media channel, period by period. The period with low fraud will have higher effectiveness. The period with high fraud will have lower effectiveness. If the media buy leads to higher fraud, then generally, the number of clickthru’s will be lower and the media will be measured as less successful. If the fraud is generated through bot-based clickthru’s then the clickthru’s may be higher, but the final conversions will still be low. Digital media analytics professionals must take care to measure the full impact of the media, not just the initial clickthru.
Digital media measurement using last touch attribution ignores cross-channel impacts from other media channels. Because Last Touch Attribution ignores the impact of fraud and can measure the impact directly fraud will have no effect on the measurement of success of other non-digital channels.
2. Statistical Regression Modeling (aka, Marketing Mix Modeling or Econometrics)
With statistical regression modeling the impact of fraud can lead to inaccuracies. Statistical modeling bases its calculation on the number of impressions purchased. When the level of fraud changes from period to period, it will lead to an overall underweighting of the entire media channel. The more fraudulent period will cause all periods to be underweighted, since it is looking at the average response across all time periods for that media channel. Generally, a statistical model treats all impressions (whether fraudulent or valid) as equal.
3. Agent-Based Modeling
Agent-based modeling may also lead to a false reading on the effectiveness of the media channel as well but to a lesser extent than with a statistical model. Because it models at a lower level of granularity the impact on the specific fraudulent media channel with the varying levels of fraud will be more isolated. Secondly, because of the greater flexibility and granularity available in agent-based modeling the impacts from last touch attribution can be used to inform the model in order to potentially compensate for an apparent higher level of fraud in a particular period.
Overall if the fraud is changing period to period and is not measured in any way, then statistical modeling will provide some level of error. Agent-based modeling will also provide some level of error, but it may be able to be mitigated when combined with insights from last touch attribution.
If you’d like to see how this is affecting the digital media market I hope you’ll join us at our upcoming webinar Tailwagging: Is your digital marketing a tail wagging the dog? Look forward to discussing this with you then.