Advertisers' Own Ad-Server-of-Record -- Why and How?

This study by Cornell University and TVision Insights published a few days ago found that "nearly a third of TV ads play to empty rooms." We kind of knew that already. But did you know, coincidentally, a third of digital ads were paid for but never served. You don't have to believe me -- i.e. you may see different things in your own campaigns. But how many of you have compared your DSP reports on bids won, by domain, with your ad server reports, by domain, like I have? Let me know if you have and let me know what you found. For the rest of you who haven't done this, this previous article Did You Get What You Paid For in Programmatic? shows that between 1/3 and 3/4 of the ads were not even served. The lower the CPMs, the higher the risk of encountering fraudulent sites where this discrepancy is high. The hypothesis is that bots quickly move on to spoofing the next ad, without waiting for the previous one to finish loading. Why wait, when the site already gets paid upon bids won? The bots can generate more fake impressions the quicker they move on.

A corollary question is "can you trust others' measurements" which determine how much you pay for digital ads? For example, Google tells you how many ads were served; and you pay based on that. Are those numbers accurate? Do you have a way to check? You'd have to assume those measurements were accurate and truthful, right? Well, don't assume. I've been a digital marketer for more than a quarter century and I've seen stuff. Hint: "don't trust, always verify."

"That's not a molehill-sized pile of crap; it's Everest."

Should you trust others' measurements? Yes, but to a point.

Let me start with some context. Over the last 25 years, I've been continuously pissed at metrics, analytics, or measurements that are woefully inept and inaccurate but yet still widely accepted by advertisers, including metrics that determine how much they have to pay. Some metrics are so wrong, that advertisers are in effect paying for something they didn't get (that's a nice way of saying they got ripped off). Think: "paid for bids won, but ads were not served" or "paid for ads assuming that they were shown to humans, but they were shown to bots instead." While some of the examples were due to difficulties in corroborating the measurements, most of the time, I've witnessed it was because advertisers didn't know to question the accuracy of the data or have any desire to investigate further. That's of course fine by me; after all it's their money being wasted. But for those advertisers who do want to know more, stay with me. I'll show you how some clients and I have been able to independently verify not only counts like impressions served, but also other metrics like viewability, fraud, in-geo accuracy, brand safety, and much more.

What comes next amounts to basic counting, not rocket science. But these counts matter a lot, because it determines what you got and what you have to pay for. While I don't think there was any malicious intent, the following cases regarding the two largest sellers of digital ads -- Google and Facebook -- illustrate how dependent advertisers are on these companies metrics and how their counting methods may not be your counting methods, but you still have to pay based on their numbers anyway. In fact, recent data from Ebiquity shows that Google, Meta (formerly Facebook) and Amazon accounted for more than $7 in $10 (74%) of global digital ad spending last year, and nearly half of all money spent on advertising." In the two examples below, if these companies that control the vast majority of digital ad dollars didn't voluntarily make the changes below, would any advertisers even have known they were paying based on inaccurate metrics?

Case 1: How many of you remember when DoubleClick quietly added a new metric called "downloaded impressions" in 2017 (screen shot below)? Why did they do that? Well, there were too many instances where their previous measurement called "impressions served" was so wildly discrepant from the counts of other measurement companies that something had to be done.

Source: https://www.pmg.com/blog/doubleclick-impression-count-methodology-update-2017-ready/

This was important because these counts impacted what advertisers had to pay to Google/DoubleClick for ad serving. "This new method will count based on download vs. served. Count on download means that DCM will count an impression when the creative [ad] download begins. Up until now, DCM has counted impressions when DCM receives a request for content." To put this in non-technical terms, ad impressions were counted as "ads served" when the initial request for the ad was made, not when the download of the ad creative had even started. That meant that a large portion of the ads were not actually served but yet the advertiser got billed for "ads served" anyway.

This new metric called "downloaded impressions" was meant to solve the discrepancy. But, if you read that new definition more closely you realize that "when the creative download begins" STILL does NOT mean the ad creative was finished downloading -- i.e. arrived in the device and rendered on-screen. "Downloaded impressions" that you now pay for may still never have actually arrived in the device and rendered on screen. DCM (CM360) counted it, and you paid for it? With a FouAnalytics in-ad tag in the ad itself, we count how many ads finished downloading into the device and rendered on screen, because the FouAnalytics tag is set to "asynchronous" and does not fire until the DOM of the ad becomes interactive. In non-technical terms, the FouAnaytics tag measures ads when they finish rendering on-screen. By comparing this, to the number of ads served, as reported by the ad server, and to the number of bids won, you get a very complete sense of whether you got what you paid for.

Case 2: Then, there's the example of advertisers suing Facebook for allegedly "failing to disclose [a] key metric error" related to video ad viewership. Source: WSJ, October 16, 2018 -- "Facebook told some advertisers that it likely overestimated average time spent watching videos by 60% to 80%. The plaintiffs alleged that the error was much larger and that the average viewership metrics had been inflated by some 150% to 900%. Many brands already were skeptical of the practice by Facebook and other tech giants to closely guard their internal ad data—one top executive likened it to “grading their own homework.” The incident fueled renewed calls for Facebook to allow independent measurement and auditing."

An Independent Standards Body or Auditor to Check

With all these different counting methods flying around, there came a collective call for an industry trade body to set some standards so everyone can count impressions in the same way, especially because the counts determined what advertisers had to pay. It was great that the MRC (Media Ratings Council) created some standards for impression counting, viewability, invalid traffic (IVT) and other digital metrics. But everyone knows that while standards are great, fraudsters and criminals don't play by the rules and may even take advantage of the standards to do more fraud.

Case in point, since the MRC definition for a viewable display ad was 50% of the pixels for 1 second, fraudsters refresh display ad slots at exactly 1 second to load as many ads as possible. Every ad would be marked as "viewable" and thus advertisers were billed for it. Similarly for video ads, the standard says "50% of the pixels for 2 seconds." So what do you think the criminals do? Right, refresh the video ad slots exactly at 2 seconds. All the video ads were marked as viewable and advertisers paid for it. My point is that standards are good, but they don't prevent fraud, any of it.

As you can imagine, fraudsters will cheat in every way possible; even when they are caught, they find workarounds. Having studied ad fraud for the last ten years, I have witnessed many examples. For example in 2013, Mike Shields and I documented many examples of sites cheating on their comScore traffic counts. They did this by taking the same comScore beacon, that was meant for one site, and sticking it on thousands of "satellite sites" to count all of the traffic as if it were from the main site, thus drastically inflating the traffic numbers and the ads sold to advertisers. Other ad tech companies did the same to trick Quantcast's traffic counts. Eventually Quantcast forced sites to break out the primary (or "parent") site from satellite sites so advertisers could see where their ads actually ran -- e.g. was it actually on heavy[.]com or complex[.]com? Finally, remember the case where Newsweek was caught using code to alter viewability measurements so advertisers paid for ads thinking they were all viewable, when they were not. Fraudsters also write false data into Google Analytics, do click flooding to claim credit for app installs that happened already, do cookie stuffing to earn affiliate revenue shares under false pretenses, and many many other forms of fraud and metrics manipulation. See: Why Does Digital Marketing Appear to Perform So Well and Fraud Appear So Low?

Advertisers depend on the measurements by ad tech companies; those measurements tell them how much they have to pay. If those measurements were not accurate, advertisers would be paying for something they didn't actually get. Industry trade bodies and standards-setting bodies might be helpful. But, who audits the auditors; and do the auditors even do a good job? Well, industry trade bodies that give out "certifications" for completed paperwork and payment of annual dues are useless in preventing advertisers from getting ripped off by criminals, ad tech companies, and fraudsters. TAG, for example, doesn't audit any of their "certified" companies during the course of the year to see if they are actually doing what they claimed in the self-attested paperwork; so that certainly doesn't help. Who audits these auditors that are clearly not doing their damned job? Can you rely on the accuracy of any ad tech company's measurement, if the measurements are self reported, and they stand to make more money from you when the counts are higher? The simple answer is no.

Why should you care, and what can you do about this?

If you can't rely on the accuracy of measurements, especially from companies that have the self-interest to make more money, and you can't even rely on trade bodies and auditors, what can you, the advertiser, do to protect your own digital investments? Of course there are many things you can do, but let me focus on just one here -- your OWN ad server of record. You heard that right -- you, the advertiser, serve your own ads, complete with detailed, auditable analytics.

This idea is not new. I have served ads for selected clients, so we have been seeing all the data for years. The slide above from 2015. By serving your own ads, instead of using DCM (now called "CM360") you have your own detailed telemetry and first hand data. You won't have to let Google spy on all your campaigns; think about it, Google has analytics on your site and on many publishers' sites, Google operates the "stock market for ads" and plays both the buyer and seller, and Google determines what you pay for ads and reports the ad impressions and clicks you got. There may not be anything dishonest about that, but that is certainly a heavy concentration of power.

By serving your own ads, you pay lower costs; and most importantly, YOU determine what you will pay for. This solves all of the data INaccuracy problems above, because you have the single source of truth. As the advertiser spending the money, you can dictate that the numbers coming from your ad server is the single source of truth -- i.e. the "official counts" of ads that you will pay on. Everyone else is a service provider; if they want your money, they have to play by your rules. You're the official time-keeper for this race.

You're the official time-keeper for this race.

Is it a "heavy lift" to start serving your own ads?

Nope. In fact, there's no lift at all. You don't have to deploy your own tech, or hire engineers. Even the processes you have in place will remain the same. You, the advertiser, create the ads. You send the creatives to the agency and they upload into the FouAnalytics ad server, instead of to DCM/CM360. They then give a spreadsheet of invocation codes to the DSP so it can call the ad once you win the bid. It's all the same. The FouAnalytics ad server is a PaaS (platform as a service) -- you pay as you go like a cloud service, there's no tech to set up, and all the data is your data; I have no other use for it other than to help you do better digital marketing. This has been the case since the beginning of FouAnalytics. I am an independent researcher, not an ad tech company.

By having your own server, and the "official count," you will no longer be dependent on any third party ad tech company whether for the ad serving or the ancillary measurements. It eliminates any self-interest of any third party vendors "getting in the way of accurate measurement" and it no longer matters whether a standards-setting body deems the metrics to be accurate or not. The only thing that matters is the counts and measurements from your own server-of-record. That is what you will pay for, when you are satisfied with the completeness and accuracy of the data.

Furthermore, some advertisers have also experienced the pain of trying to bring together log level data from different sources to analyze. Cleaning the data and getting into the proper format so comparisons can be made often takes an unknown amount of time; so most advertisers don't even embark on this endeavor. By having your own ad server of record, not only will you have all your ad impression counts in one place, you will also have viewability, IVT, in-geo, and brand safety measurements fully integrated together in the same platform. See: Viewability, IVT, Out-of-Geo, Brand Safety in FouAnalytics It's like having the log level data that DoubleVerify, IAS, Moat don't want to give you, or will charge you a lot of money to get, and not needing to do big data processing. handle log level data. And by the way, all of these measurements are included in the FouAnalytics ad server of record, so you can save the millions of dollars you pay to these vendors for questionable measurements. See: Why Black Box Fraud Detection Tech Sucks

The counts from your own ad server of record are what you can completely validate yourself, from ads rendered (not just served or downloaded), viewable, non-fraudulent, human, in-geo, and brand safe. These ads are the valuable ones. Isn't this what you should be paying for? (the only thing you should be paying for?).?You control your own ad-server-of-record. It is the single source of truth and the "official counts." You own all the data and all the related measurements are integrated in one place. This is the foundation for good digital campaigns. Some would call this the holy grail.?

If you want to experiment with your own ad server of record, please let me know. We can set up a campaign quickly, so you can see what we can see, how much more accurate the data can be, and how much cost savings can be achieved.

P.S. please also let me know if you think this is a dumb idea. I'd welcome the push-back and feedback.