Rampant gaming of view-through conversions

"What if a user doesn't interact with my ad, but then later converts on my website?"

"View-through conversions" were created to give display ads more credit for their contribution to conversions. After all, someone may have SEEN the ad, but didn't CLICK on it. But later, the same user came directly to the site -- in the subsequent, 30, 60, or 90 days -- and bought something -- i.e. "converted." Shouldn't the display ad get credit for the conversion? This is yet another adtech thing that was a nice idea in theory, but made it super easy for bad guys to commit rampant amounts of fraud, by simply gaming this form of attribution -- i.e. by claiming credit for sales that the ad didn't actually cause.

Let's start with Google's definition of "view through conversion."

About view-through conversions

Source: https://support.google.com/google-ads/answer/6270625?sjid=13627188442445612270-NA

"Your “View-through conversions” column tells you when customers see, but don’t interact with your ad, and then later complete a conversion on your site.

View-through conversions are a helpful way to track the value of your display or video ad campaigns. For display campaigns, for example, they measure the conversions where a customer saw—but didn't click—an ad before completing a conversion.

For Display Network ads, the last viewable impression will get credit for the view-through conversion. With Google's Active View technology, an impression of a display ad is considered viewable when at least 50% of the ad is onscreen for at least 1 second.

View-through conversions automatically exclude conversions from people who have also interacted with any of your other ads. View-through conversions are not included in the “Conversions” column, only in the “View-through conversions” and "All conversions" columns.

View-through conversions from browsers that don’t allow cross-site cookies cannot be reported."

View-through conversion window: Definition

Source: https://support.google.com/google-ads/answer/7320922?hl=en

"The period of time after an impression during which a view-through conversion will be recorded. View-through conversions occur after an ad impression,?if the user doesn’t interact with the ad, then later converts.

You can set your view-through conversion window when you create or edit conversion actions. For example, if you pick 30 days for a conversion action, then view-through conversions that happen within 30 days after an impression are tracked. A longer view-through conversion window (like 60 days or 90 days) will usually increase the number of view-through conversions Google Ads records."

In other words, Google attributes any sale in the 60 to 90 day period after ad exposure as a conversion for that ad. So the ad appears to be performing really well and driving a very high ROAS, when those conversions and sales could have or would have happened anyway. The same problem exists with retargeting and remarketing. A user who knows a particular online merchant will be very likely to buy from them again. View-through conversion reporting always OVER-attributes the sales to ad impressions, when those sales would have happened anyway.

How bad guys easily game view-through conversions

Now, can you see how easily anyone can game view-through conversions? If not, let me lay this out for you. If a user were shown an ad, they are considered an "exposed" user. Google knows this user because they set a first party cookie on them on ad exposure (cookie is set by a Google-owned domain). If that user goes to a website and completes a purchase ("conversion") within 30, 60, 90 days after the ad exposure, that conversion is attributed to that ad impression. Google also has conversion tracking on the website and can read cookies set by a Google-owned domain; so in DCM/CM360 reporting that conversion is attributed to that ad exposure as a "view-through conversion." This makes sense, on the surface, right? But how can this be gamed or used for fraudulent purposes?

Here's the tried and true technique used by fraudsters and adtech companies alike -- expose as many devices/users as possible so that whenever any of those exposed devices complete a purchase, they can claim credit for that conversion through view-through conversion reporting. This can be easily done by using the lowest cost display ads and brute force "exposing" every single device in the market. Sprinkle in some magic-dust algorithms and you can reduce your costs by exposing users/devices that are known to have purchased from an advertiser before (remarketing). Those users are likely to buy again, even without seeing any ads. But view-through conversion reporting makes it APPEAR that the ads drove those sales. View-through conversions make it easy for bad guys and adtech companies to claim credit for sales that would have occurred anyway 30, 60, or 90 days after ad exposure; the view-through conversion reporting in DCM/CM360 attributes conversions to the ad impression even if there was no actual causation or even correlation. Some adtech companies may actually run ads for their advertiser clients. But fraudsters can trick the view-through conversion reporting by running super-low-cost ads in the background to ensure that all 320 million smartphones in the U.S. are marked as "exposed."

I have written about this before (article below), citing examples of how mobile ad networks were claiming credit for "footfall" they didn't actually cause through their ads. By exposing every single mobile device in North America (using invisible ads in 1x1 windows, in the background, while users were playing mobile game, etc.), whenever any such device walks into a Taco Bell, Dunkin Donuts, or McDonald's, the mobile network can claim credit for that store visit, even if the user never saw the ad and walks into those stores anyway -- like their daily morning coffee at Dunkin. This is "taking credit" for the conversion under false pretenses.

There's a more sinister variation of this form of attribution fraud -- i.e. the adtech company is claiming credit for sales that had already occurred. This is done by observing which visits resulted in a purchase and then inserting a fake pageview at the beginning of just those visits to make it look like the visit came from a click on an ad. The perp in this example continues to perpetrate this form of fraud by writing the false data directly into their own customers' Google Analytics. In some cases, they ran no ads at all, because it was so simple to just write the false data into Google Analytics, there was no need to incur the cost of actually running ads and waiting for anyone to click on them.

So what?

If you've stuck with me thus far, hopefully you realize how easily view-through conversion reporting in DCM/CM360 can be gamed. Whether fraudulently or innocently, view-through conversions almost always over-attribute conversions to certain ads. This makes the advertiser believe they are getting very very good ROAS from those campaigns, even though those conversions/sales would have happened anyway. In other words, the advertiser is still over-spending on these ads because they APPEAR to have caused the conversions.

The way advertisers are using FouAnalytics to suss out this kind of mis-reporting or fraud is to measure the ads (to make sure ads actually ran) and to measure the clicks from those ads and the arrivals on the site. For example, we can see if large portions of ads are run in 0x0 or 1x1 pixel windows, in mobile apps, in overnight hours when humans are not awake and interacting with the app, etc. Remember, view-through conversions don't require clicks on the ads; the loading of the impression is sufficient to label that device as "exposed."

Finally, the best way to check for OVER-attribution of sales to certain types of ads is to pause the ads and see if the velocity of sales changes. If the rate of sales DOESN'T go down in the absence of the ads, you know instantly that those ads were given credit for the conversions when they didn't deserve such credit. You can then correct your ROAS calculations and re-allocate your budget to channels and types of ads that actually drive more incremental sales (sales that would not have occurred anyway), and away from ads that are just falsely claiming credit for sales due to how view-through conversion reporting works.

To help reduce the reliance on view-through conversions which can be gamed to make certain campaigns appear to be driving much better ROAS than they actually are, have a look at the following -- Cookieless and Clickless attribution with FouAnalytics. Let me know what you think.

For more case examples and screen shots from FouAnalytics, follow and subscribe here: https://www.dhirubhai.net/in/augustinefou/recent-activity/newsletter/