Youtube Serving Issues: Human Error or Algorithm?

The news of Adalytics study on Google serving as much as 80% of video ad inventory on low quality placements that would qualify for refunds has been making the rounds and the reaction is much what folks would expect:

"Google is evil"

"We've come to expect this"

"Google doesn't care about us anymore"

Google's response attempts to address this, however it's understandable how it might have come across as missing the point. It addressed the facts of the claims, however didn't address the "feeling" that has been growing for the past few quarters that Google is trading its "don't be evil" mantra for "algorithmic cash cow". Note, I don't actually believe Google is evil. Many ethical and dear industry friends work for Google and have been champions for us advertisers, agencies, and tools. However I understand why some might feel this way.

I'm not going to defend Google or the efficacy of this study. These are my feelings based on data on how to take the news as well as what actions you can take to ensure your account is firmly on the path to profit and victory. I encourage everyone to read both documents in depth (the actual Adalytics study and Google's response) so you can come to your own conclusion on the facts and potential implications for your business.

How Much Does Google UI Encourage Human Error?

Candidly, the first thing I thought of when I read the SEJ post on this was: looks like Search Partners with Display expansion has come to Youtube. For those of you who don't know what that is, search advertisers by default would be opted into search partners (third party sites that allow ads on their in-site search) as well as display spots. The issue with this was:

• Most advertisers didn't take the time to create image assets (I'm ignoring the bad times before this was possible and text ads would be put in display spots).
• Search Partners often had drastically cheaper average cost per clicks (cpc) and display absolutely would. So advertisers would get false positives that they could afford transactional search, when a third or more of their budget was going to these other placements with cheaper clicks, lower click through rates (ctr), and usually worse conversion rates.
• Making the choice to opt into or out of Search Partners/Display Expansion was a campaign level choice (Microsoft lets you make the choice at the ad group level), so you'd have to make the calculation on whether to say yes to it for your entire campaign (even if some ad groups were suffering for it).

The reason this came to mind on the Youtube question is it feels like history repeating itself: campaign settings leading advertisers astray because they don't understand what they're saying yes to. To quote the study:

"To summarize, it appears that TrueView media buyers can choose to opt-out of buying on the GVP network for brand awareness and reach campaigns. Media buyers cannot opt out for any other campaign type including TrueView conversion campaigns. It appears the Google Video Partner network is mandatory for TrueView campaigns optimized to a performance goal." - Adalytics 2023 youtube study

And they're right - if you choose a stand alone video campaign opted into conversion focus settings, you have to "eat" the display network:

Conversely, if you choose non-skippable (i.e. you'll be charged for every view), you can opt out of display:

If you want conversions as the focus, you have to eat the work of proactively putting protections in place. These protections include device/operating system:

Exclusions (done after campaign creation) for either topics, specific urls, or display/video keywords. I grant this is a "hidden" feature and it's easy to miss that you can change the exclusion types. Shameless plug for Optmyzr's Smart Exclusion tool.

So...yes Google does make you go through some steps to protect your account from understandably frustrating placements. However saying that they force advertisers to serve on them over the advertised Youtube placements is just not true. What might be happening (and this is something I'm seeing with Performance Max or PMax campaigns), is that the budget or bids aren't aligned with the targeting goals of the campaign. Google is ultimately an auction and if the bids are not high enough to earn placements on premium spots, others will be found.

A good way to audit against this is to see if your campaign is taking longer than normal (2-3 business days) to start serving, or if it seems like you're getting an above average view rate with really cheap CPMs/CPCs. Just like seeing a high ratio of visual ad channels in PMax can be taken as a sign you're not budgeting enough for transactional search, so to seeing a cheap but high view rate percentage can be take as a warning sign.

If Google takes anything away from this episode, hopefully they will take away that their UI invites confusion. They've gotten better at sharing the specifics of their algorithm and I hope this extends to easier navigation.

How Many Accounts Need To Be Impacted For This To Be Statistically Significant

I was incredibly impressed with the technical breakdown of how Adalytics structured their study and processed their findings. There's one little problem: the amount of accounts is tiny.

1000 accounts is nothing in the grand scheme of all things Google and we don't know the ins and outs of how each account was set up (i.e. how much was human error vs a valid bug or algorithmic oversight the cause of the bad placements). I was hoping as I read their study to find some context on this, as well as get insight on the amount of accounts they included. However, because they focused on brands, and only shared data based on single account performance, I'm left struggling to fully buy into the claim.

Data sets typically need to revolve around thousands (usually 10K+) of accounts to capture a meaningful sample size of the Google advertiser market. We don't bring claims forward unless there is enough statistical significance to substantiate it. Yes, that might mean a case study gets scrapped. Yes it might mean that the outcome I was hoping to find would come up short (this happened when I was looking at bidding strategies).

I think there are some absolutely valid claims to be made here and I think the light should be shown on Google for how easy they make it to opt into incorrect settings. However making the claim that they violate their own terms of service feels more like an attention grab than anything else.

The study itself provides a masterful understanding into the innerworkings of how ads get served and which should be charged. Again, everyone should read it for themselves and come to their own conclusions.

What Should I Do With All This Information

Candidly - advertise as normal, just be proactive about which settings you opt into.

It's not lost on many of us who have been in the industry for a while that PMax placements often lean into youtube and display. You can build exclusion lists to protect yourself from bad placements on both stand alone video and PMax campaigns. You also can advertise elsewhere.

Microsoft's Audience Network has been getting a lot of love lately and they've always made it much easier to see and add problem placements. If you're burned out on Google, consider testing some budget there (especially with the news on predictive targeting).

So...am I saying Google is innocent? No. However, they're not guilty either.