Market Incentives That Drive Fraud: The Truth Behind Reach vs. Frequency

Last Click Attribution is an Incentive for Mobile Ad Fraud

As the digital advertising ecosystem becomes educated on mobile ad fraud, it’s important to understand that fraud comes from a variety of different angles. The majority of fraud is actually a side effect of the way attribution is performed. It is, interestingly enough, directly correlated to the predominance of mobile app install ads being priced on a CPI basis.

Roughly three-fourths of the fraud detected by Kochava is characterized as attribution fraud — where the install is legitimate, but fraudsters attempt to get credit for either organic traffic or installs driven by another network partner. Tactics to game the attribution system include click spamming, click stuffing, ad stacking, and many other techniques.

I once led a team of measurement analysts focused on campaign performance at one of the world’s largest software companies. We expressed performance in terms of incremental lift. This meant asking the question: Does touching a household with media have a measurable effect vs. an identical household that was not touched by media? Or: In the absence of an ad, how many people would have taken the action anyway?

First impressions matter

One of the more interesting artifacts from incremental research was how much lift was generated by impression. With enough data, it was possible to detail the incremental effect of the first, second, third impression, etc. Invariably, the first impression did the most ‘work’ in influencing behavior. This makes sense when the customer has no prior engagement with the brand.

Additional marketing touches are influential; however, while the first impression is the most important, it often takes multiple cumulative impressions to get someone to ‘pull the trigger’ and convert.

In the first graph, the overall lift for this campaign was $20. This was a cumulative lift based on impressions 1 through 5. In the second graph, when we calculate the delta between each impression, we see that the first impression had the most lift.

What does incremental lift by impression have to do with direct response marketing with the intent of driving app downloads? And, how does this relate to the amount of fraud we’re observing?

The short answer: Networks are incentivized to have the last click, not to reach the most prospects. While they should be maximizing their reach in making those valuable first impressions, networks instead focus on frequency to win the last click.

Where no lift was observed, we often found that the ads weren’t viewable. Also, partners like Moat Analytics or Integral Ad Science (IAS) provided clues as to whether the impression was seen by a human. In desktop/web marketing, that was as far as we’d go for fraud detection.

Viewability of mobile ads is important but difficult to implement, particularly in-app. Also, the mobile app world as a whole does not account for impressions, so ingesting impression data is the exception to the rule.

Last click may not be the most influential

Direct response attribution has been borne from the demand for nearly instant feedback loops happening in real time. To have instant feedback, an install must be immediately adorned and posted back to the ‘winning’ network. Thus, we have a last click attribution model.

With upper-funnel user acquisition, however, the marketer is most benefited by maximized reach, i.e., to serve as many first impressions as possible to the largest population of prospects. But the network incentive is to have the last click, so instead of maximizing reach, the dollars lie in maximizing frequency. The results are click spamming and the falsely gained attribution everyone is witnessing.

Industry perceptions of attribution must change

Determining attribution based on the last click casts a blind eye to the valuable touchpoints preceding it. That said, the attribution model should be improved to better reflect the most influential touchpoints prior to the install.

Opportunities for improvement include:

  1. Visualization of impressions and clicks
  2. Ability to ingest ad types, sizes, and consumption metrics
  3. Adoption of multi-touch (fractional) attribution
  4. Measurement of incremental lift generated from ad units

There is value in collecting all touchpoints leading to an install. Collecting the data is the hard part. While the measurement framework may not be ideal, the necessary elements are in place to improve. If attribution is about finding the channels that drove an install, then we need to consider more touchpoints outside of the last click.

About the Author

Grant Simmons is the Director of Client Analytics at Kochava and leads the team in analyzing campaign performance and business value assessments. He is the former head of Retail Analytics at Oracle Data Cloud where he worked with over 1,500 retail directors, VPs, CMOs and agencies to develop individualized test-and-learn strategies.