
Background
Some attentive readers may recall reports that Uber sued their mobile agency Fetch in 2017. Uber had argued the agency was “running a wild west of online advertising fraud,” and claimed credit for app downloads that happened without customers clicking on ads. When the case was not assigned to a San Francisco state court, Uber dismissed the case in December 2017. In 2018, Fetch then sued Uber for the unpaid invoices.
Uber returns to the court
On 5 June 2019, Uber however re-filed their lawsuit. The official documents can be viewed here: https://webapps.sftc.org/ci/CaseInfo.dll?CaseNum=CGC19576493
This time the lawsuit alleges fraudulent concealment, negligence and unfair competition from the apparently 100 companies – mainly ad networks, exchanges and unknown publishers – that were engaged by Fetch.
There are some very interesting data insights presented here. For example, the court document shows a statistic on page 12 with more daily clicks than daily app users from one of the companies (e.g. 180,862 daily clicks versus 14,742 daily users). Of course, anyone will ask: why should every user click multiple times on an ad to download an Uber app?
Ad fraud – Yes, but whose responsibility?
Interestingly, according to statements linked to the first lawsuit, everyone seems to have agreed that there were issues with ad fraud. But one of they key questions appears to be who was in charge of mitigating the ad fraud risk. Fetch claimed that Uber had been suggested measures to combat ad fraud before, while Uber saw the responsibility with the agency who subcontracts the suppliers to deliver the ads. Of course, there are legal and there are ethical responsibilities. I leave this to everyone to reflect on this: even if a client may not potentially follow all advice [immediately] of their agency, does this mean an agency can look away when there is ad fraud? A careful reminder here that the most common industry standard is to be paid by volume – CPCs or CPMs.
Lessons for marketers
Independent of the outcome of this lawsuit, there are three very important lessons for marketers here:
- Ad fraud is real – always request log files for all media buys (even if only as backup).
- Marketers need to review their contracts – buyer beware!
- Don’t use last-click attribution – rely on proper experiments to examine directly measurable ad effects.
To support point 3, let me cite my favourite part from the court documents (page 17):
Just before Uber suspended the entire Uber Campaign in March 2017, which included payments to Defendants, Uber was spending millions of dollars per week on mobile inventory purportedly attributable to hundreds of thousands (even millions) of Uber App installs per week. Had the ads been legitimate, one would expect to see a substantial drop in installations when mobile advertising was suspended. Instead, when Uber suspended the Uber Campaign, there was no material drop in total installations. Rather, the number of installations supposedly attributable to mobile advertising (i.e., “paid signups”) decreased significantly, while the number of organic installations rose by a nearly equal amount. This indicated that a significant percentage of the installations believed to be attributable to advertising were in fact stolen organic installations. In other words, these installations would have occurred regardless of advertising.
In short: here we appear to have some evidence again that most attribution methods can be gamed. While ‘last touch’ or ‘last click’-attribution is one of the worst methods to investigate ad effectiveness, even algorithmic attribution models are likely to still give credit to ineffective media (just less credit than last touch). The most accurate and robust way to uncover the impact of your media buys are experiments – just compare the KPIs of interest for two different regions or groups where one is [additionally] exposed to the extra advertising we wish to investigate. No complex math needed. The possible impact should be seen in any basic dashboard and sales graphs (between the groups or regions).