Criteo’s detection and mitigation of ad fraud is almost twice as bad as its peers in the AdTech sector, according to a paper by private due diligence investment group Gotham City Research.

Following on from its analysis of the Method Media Intelligence study last week, GCR has released a more detailed outline of its study and included some new observations, including:

  • Criteo appears closely linked to publishers that promote the spread of known malware;
  • Criteo links to and places ads on publisher sites that have millions of non-human visits.

Asked to comment on this latest report, a Criteo spokesperson this morning told Which-50, “The existing statement in relation to the topic stands.”

GCR also reaffirmed its view that over 50 per cent of web sites using Criteo are of suspect quality, and that the company links to and displays clients’ ads on publisher sites that contain content that appear to violate its own advertising guidelines.

The report also asks “Is Criteo a beneficiary of ad fraud, perpetrator, or both? Criteo exhibits qualities of a repeat offender.”

It quotes remarks from Ben Edelman in April 2015, saying “Criteo systematically buys adware injection inventory on the very sites where the advertisers already have reached the users on the advertisers’ own sites – I have caught this dozens of times for clients each time Criteo apologises – but each time they do it again.” Edelman is an associate professor at the Harvard Business School where he studies and teaches about the economics of online markets.

Edelman declined to be interviewed for the story. However, in an email he told Which-50, “I believe the Gotham and Method reports are broadly on target in the sense that I share their ultimate conclusion — that Criteo is not as good a deal for advertisers as the company leads them to believe. I agree at most partially with the Gotham and Method diagnosis of what’s going wrong — why it is that advertisers misunderstand the effectiveness of what they’re buying from Criteo; how so many advertisers end up overpaying; what specifically Criteo is doing wrong.”

However, he cautioned, “More generally, I would have analysed some of the facts differently, with different methodologies to support my conclusions. Indeed, I know of at least one serious and indisputable error in the Method report.”

He declined to identify the error.

When we spoke to Gotham City Research founder Daniel Yu this morning, he confirmed — in keeping with Criteo’s practice in the past — that it has not responded to any of the criticism in the report.

See also:

According to the study, “Criteo is worse than its peers based on our study of its web sites.”

The authors of the paper sought to build on the MMI research by addressing the question of materiality: what it describes as the percentage of Criteo’s domains/clicks/revenue that are derived from suspect web sites

“In other words, what percentage of Criteo web sites would its clients not want their ads displayed on (recall, Criteo does not disclose where its ads are displayed to its clients)? This is the central question, as Criteo charges its clients on a per-click basis, without regard to the quality of the web site or its traffic.”

To do this, Gotham City Research says it evaluated several different service providers that collected and stored web sites using Criteo (and their metadata).

“We found a database with information on nearly one million web sites using Criteo and/or its peers. According to our analysis, over 54 per cent of the web sites using Criteo are of suspect quality.”

“If over 50 per cent of Criteo web sites are of suspect quality, then its clients are not getting what they thought they were paying for. This means that they can reduce their spending on Criteo by over 50 per cent without impacting their sales. Furthermore, we believe that clients could reduce their total spend to Criteo by well over 50 per cent if they were to only pay for the publishers/sites of their choosing.”

Suspect web sites defined

The study defines a suspect web site as one with an extremely high likelihood or certainty of fake (non-human) traffic.

“After manually looking through many Criteo web sites, we determined that fake web sites tend to have little to no information about themselves.”

The researchers created what they describe as a simple 17-factor model that only showed Criteo web sites with exactly zero out of the 17 factors, from the following categories:

  • Social Media Presence;
  • Contact Information;
  • People;
  • Multimedia;
  • Geographic information.

The researchers say they applied the same assumptions and methodology to Criteo as they did to its peers and used all available data for Criteo and its peers “… for example six years of Criteo data.”

Furthermore, they assumed that the percentage of suspect clicks was the same as the percentage of suspect web sites, but added the following caveat: “This may be conservative, as there is evidence to suggest that click clustering, browser hijacking, etc. lead to nonlinear effects.”

And they suggest the same assumptions could be applied to Criteo’s revenue.

GCR also suggests that Criteo is targeting mid-tier clients who are less sophisticated in their understanding of digital ad fraud. As evidence, it points to a decline in average revenue per user since 2014. It also includes references from earnings calls where the company explicitly states it wants to grow its share of mid-tier customers.

Of course, in fairness to Criteo, there is also a benign interpretation of this strategy. There are more clients and volume in the mid-tier.

However, the study also outlines clear examples of how enterprise clients have been able to massively reduce online ad spending with no impact on sales by taking a more aggressive approach to ensuring the quality of the inventory they buy. (Note that these examples are not specific to Criteo.)

For instance, Proctor and Gamble’s digital ad spend dropped $US140 million — or 41 per cent — year-over-year, and yet it still grew sales by two per cent over that same period, beating both analysts’ predictions and P&G’s competition.

In this example, according to GCR, “P&G ran ads on 1,459 sites between January and May 2016, that number dropped to 978 sites in the same period in 2017 — a decline of 33 per cent in sites featuring P&G ads year-over-year.”

GCR also notes that Unilever’s digital ad spend dropped 59 per cent year-over-year between January and May of 2017, and that it is advertising on 11 per cent fewer sites this year for a total of 540. “For the sites where it ran ads during both time frames, it reduced spend by 57 per cent on 155.”

Perhaps most tellingly, “Chase Bank had ads on 400,000 web sites, and then reduced the number to just 5,000 sites. No impact on sales.”

Which-50 has sent questions to Criteo’s investor relations department in New York and is awaiting a response.

Previous post

Consumers resist facial recognition security

Next post

Baidu puts $1.5 billion behind driverless car start-ups

Join the digital transformation discussion and sign up for the Which-50 Irregular Insights newsletter.