Really, Bob? Really? Another Supply Chain Transparency Study?

Remember when I tweeted that "when I get pissed off, I write?" Notice the number of articles and newsletters coming from me lately? Don't worry, I'm not really that pissed off that much. But this AdAge article from minutes ago led to the thought "Really, Bob? Really?" Hence this post.

Not only have there been FOUR previous studies on "programmatic supply chain transparency" they also all found pretty much the same thing -- that 50% or less of every dollar spent by advertisers goes to publishers for showing ads. The rest goes to middlemen and fraudsters, all of whom are maximizing their profits by feasting on these ad budgets.

And it's almost like Bob's been ignoring me since 2014 when I started reporting on the "productivity of digital ads." It wasn't all about fraud. In the chart below, you can see that even back then, other problems like "not in view, not in geo, not on target," etc. were all sapping the potential productivity of digital ads. The fact that unit prices for ads (CPMs) were cheaper still could not outweigh the waste and percentage of every dollar going to adtech middlemen and fraudsters. You'd have to spend 2X the amount on digital ads, if only 50% went towards showing ads. You'd have to spend 10X the budget on ads if only 10% of every dollar went to showing ads. That's exactly what we see below. Using just industry averages published by others, the productivity continued its predictable decline from 13% possible productive in 2014, to 10% in 2017, and then 1% in 2019. It's actually far worse than these averages would suggest.

But why does "digital" appear to be performing so well, and fraud appear to be so low? Well, I wrote a whole article on this a few hours ago, called Why Does Digital Marketing Appear to Perform So Well and Fraud Appear So Low?

This article will focus on a few of the underlying data points that show why digital ads are not productive, or at least not as productive as what marketers assume. The entire deck that I published in June 2021 is embedded below. I will refer to a few of the slides in the next section.

You can't depict that in a waterfall chart

The ANA's "investigation" started with an RFP in May of 2021. Immediately I could see that they were not even asking the right question -- they asked for a waterfall chart to depict ad summarize "where all those dollars really go." Unfortunately, the construct of a waterfall chart is too simplistic and will fail to account for the vast ranges of each of the observable phenomena, like ad fraud, viewability issues, ad blocking, etc. It will also fail to depict the forms of fraud and waste that are not observable or easily quantified -- like undisclosed arbitrage done by media agencies. The equivalent supply chain transparency study published in May 2020 by the ISBA [PDF] (British Advertisers) found that despite working with 15 of the largest advertisers and 12 of the most mainstream publishers in the UK, only 12% of the 267 million ad impressions could be "successfully matched" from one end of the supply chain to the other (from the advertiser spending the dollar to the publisher receiving the money and showing the ad). "The rest could not be mapped due to low data quality." This was in a "well lit neighborhood;" imagine what this would look like in open programmatic. Further, among the 12% of impressions that were "matched" they found that 15% of the dollars could not be accounted for --"an ‘unknown delta’ of unattributable costs."

This is a perfect example that illustrates the waste and fraud that is not "observable" or "quantifiable" even with complete supply chain log-level data. A media agency silently keeping an unspent $4 million out of a $10 million budget, and then cooking the books by altering the effective CPM to make it appear the entire $10 million was spent, does not show up in log level data. But it does show up if financial records are subpoenaed.

The "most realistic" scenario

You can look through the rest of the slides above yourself. I will use the following summary slide to illustrate what I consider a "more realistic" or "most realistic" scenario for the productivity of digital ads purchased through programmatic channels.

On the left side of the chart, we have the "known costs" -- these are the costs that can be quantified if the parties involved disclosed them truthfully. You can imagine most ad tech companies don't want you to know what margins they make from you, so we are likely to get ranges, not absolute numbers. If you assume these are truthful, you can see these numbers depicted in the orange bars -- 1) agency fee - 10%, 2) exchange fee - 20%, 3) data and verification fees - 20%. These are the most likely round numbers. Note the orange bar extends above and below these numbers. That represents the range that I have observed in the wild, outliers excluded. Yeah, there are some outliers that are hard to believe so I will spare you the shock. If you just took the left side of the chart, the readily observable and quantifiable costs, you are already at the "50% of a dollar" goes to working media.

On the right side of the chart, in red and green, you can see various other forms of fraud, waste, and suboptimal spend. The same problem exists with these. In order to plot them on a waterfall chart, you have to take a number, likely an average. And averages hide the true problem because the range is left out. Take ad fraud for instance. Look at the IVT/NHT red line -- it ranges from 1% to 100%. What's a rational average to use in the waterfall chart? If you assume a 42% rate of fraud, the other 58% is "fine" (marked in green). If you do the same for the other categories, you will see 45%, 85%, 58%, 75%, 20%, 34%, reading from left to right. These are just a few illustrative categories and there are more that are not shown and some that are not known even by me. If you multiply all these green percentages you get a very small number -- that's the 1% "likely productive ads" shown in dark blue on the far right. This is the most realistic scenario. There are far worse outcomes if the rate of fraud were higher, the amount of ad blocking were higher, the amount of viewable ads were lower, etc. compared to the averages that had to be assumed above to draw the waterfall chart.

This is why I have said before that another supply chain transparency study is pretty much futile -- it won't yield any new information compared to the previous four, if the same methodology were used. But if the same methodology were used, it would not properly account for the ranges, because by necessity waterfall charts can only be drawn using averages, and that requires assumptions to be made. These hide the harsh truth. A scenario, depicted above as 1% productive, is likely not something Bob has the balls to publish, given that he has been leading the ANA for the last 10 years and telling all his members that digital appeared to be performing well and fraud was low. See: https://www.dhirubhai.net/pulse/why-does-digital-marketing-appear-perform-so-well-low/

Prove me wrong, Bob.

(This "study" is dead before it starts because "hope" is not enough to get fraudsters, or anyone else benefitting from the flow of dollars, to hand over log level data that is the damning evidence that exposes the fraud they've been perpetrating under your watch for the last decade.)