With advancing technology, automation is increasingly having an impact on our decisions and our day to day lives. For marketers, Google Ads is no exception. Over the course of the past several years, advertisers have been forced to give up more and more control as Google’s machine learning algorithms have improved and automation has taken over much of what previously had to be done manually.
While some of these changes have caused marketers to lament the reduction in control or insight into their advertising accounts, overall the changes have been positive as they level the playing field and remove much of the optimization work that previously went into managing accounts. This allows marketers to spend more time focusing on strategy and testing new ideas rather than the manual manipulation of the many variables that go into optimization.
Google Ads Before Machine Learning and Automation
Before we dive into the details, humor me a little and take a ride in my time machine back to a time before machine learning and automation were pervasive within Google Ads (or Google AdWords as it was then called).
When you bid for a keyword with a $5.00 max CPC (Cost Per Click), you were entering the auction firmly with $5.00 as your maximum bid for a click on that keyword. You were not stating that you would be willing to pay more than $5.00 for a click if Google determined that the click was more valuable.
You then had to do historical analysis to determine which dimensions required adjustments to your baseline bid against the various dimensions. Weekdays and desktop devices may have required a higher bid due to stronger performance – and you would then apply a bid adjustment against these dimensions to compensate for the difference in performance.
This was the workflow as paid search marketers knew it and it required a lot of manual analysis and bid adjustment. In complex accounts this could take hours upon hours to accomplish at a granular level.
To advertise for mobile apps, there were various placements and targeting methods. There was the typical Search campaign setup with all of the targeting and functionality that Google Search campaigns have. There were also Display Mobile App Install campaigns, which you could optimize based on app placement or utilize interests and remarketing.
The ad formats for both were similar to typical Search and Display campaigns, where you could customize the text or image that you wanted to run per ad group. You could download and analyze the data similar to how you would in a typical Search or Display campaign. You could exclude or add placements and keywords that performed well. The world was your oyster. Most marketers I believe would agree that this level of optionality and flexibility provided for the best experience and results.
For standard Search campaigns focused on web traffic, when you added a keyword, you knew precisely which search terms could be triggered by that keyword’s match type and you structured your account accordingly. This often meant very granular campaign structures that were segmented by each major match type to maximize control and performance. This structure prevented keyword match type overlap and thus intra-account competition. It was logical, clean, and very human-friendly.
So what has happened since? Well…let’s now take a look at how machine learning and automation have impacted these three key areas of Google Ads described above – bidding, mobile app campaigns, and search keyword match types.
Target CPA Bidding – An Improving Gift from the Marketing Gods
Target CPA bidding was introduced to save the time and effort it took to do granular analysis, manual bid optimization, and dimension adjustment to dial in on the optimal bid.
Target CPA bidding places the burden of bidding at the optimal level on Google’s algorithm, as Target CPA bidding uses “historical information about your campaign and evaluating the contextual signals present at auction-time” and “automatically finds an optimal CPC bid for your ad each time it’s eligible to appear,” according to Google.
When Target CPA bidding was introduced in 2007, the algorithm was not nearly as refined as it is today. As a result, you had to give up control over performance by foregoing manual CPC bidding in hopes that Google would make the best bidding decisions.
It turned out that wasn’t very often the case (based on discussions and experience amongst our team). As a result, manual CPC bidding was still the bid optimization method of choice for many marketers seeking optimal performance.
Recently, Target CPA bidding and other Smart Bidding options that leverage Google’s machine learning algorithms have shown much more promise, with performance often meeting and sometimes even exceeding what is possible with manual CPC bid optimization.
At Growth Pilots, we keep a very open mind and constantly challenge our assumptions in pursuit of seeking optimal performance. In line with this philosophy, last year we decided to conduct a multi-client test of the impact Target CPA bidding has on performance – and we saw surprising results across the board.
In one extreme case, one of our clients saw CPA drop by 42% and conversion volume increase by 100% with Target CPA bidding compared to manual CPC bidding. This type of performance lift is not typical with Target CPA bidding, but it does go to show the value that machine learning can bring to the table in the right scenario.
Google gives you the option, but not the requirement, to use Target CPA bidding settings. This is important because there is still control and optionality that can be exercised at the discretion of marketers. Google still allows you to keep a tight control on your bids through manual CPC bidding if you choose to.
In fact, we still manually bid (manually meaning leveraging manual CPC bid settings within Google, but using technology to actually choose the ideal bid and executing the bid adjustment) a significant number of our clients as we have seen better performance by doing so.
Having the ability to test and compare different options is key to maximizing performance and Google makes this possible when it comes to bidding for inventory within its Search, Display, and YouTube networks. We are optimistic on the future of Target CPA bidding and we continue to test it against manual bid CPC optimization in pursuit of the best performance possible.
Universal App Campaigns – A Black Box with Very Limited Options
On the other end of the extreme, Google’s solutions for mobile app advertising paint a stark contrast to the level of control advertisers are left with after automation comes into play.
Before Google migrated over to Universal App Campaigns in 2017, if you had an app that you wanted to advertise, there were plenty of options.
You could run app install search ads on their own, AdMob (Google’s mobile display network) app install ads on their own, or YouTube app install ads on their own. Within these campaign types, targeting options were plentiful and nearly identical to what you could achieve on web-focused campaigns with significant control and granularity. And further, you could pull reports segmented by any of the typical dimensions (keyword, creative, placement, etc).
Then Google launched Universal App Campaigns (UACs) and everything changed. With UACs, you are forced to relinquish most of the control over creative, targeting, and optimization to Google’s algorithm. And the level of reporting and insights that can be garnered from UACs is very shallow. So in short, Google is making most of the decisions in UACs and exposing very little to the advertiser.
So what does this mean for us control-obsessed performance marketers? It means we have to be very thoughtful about the limited things that we can control to ensure the Google algorithm has the best information possible to do its job.
The relationship between marketers and Google’s algorithm has evolved into a symbiotic one when it comes to Universal App Campaigns. You can still control the creative inputs like which text, image, and video to use, however Google determines the best combination of these components. You cannot control which audiences to target, but you can map app events within your app to Google in-app conversion events and select which event you want to optimize towards.
It seems very likely that Google will continue to take advertiser control away from other campaign types over time, and it’s not unimaginable for Google’s entire advertising platform to resemble Universal App Campaigns at some point in the future.
We spend a decent amount of time discussing and debating this internally at Growth Pilots. One thing we all agree on is this: if and when it happens, it will be critical to quickly adapt for your workflow to continue to drive optimal performance, so mentally preparing and discussing implications is important.
Keyword Match Types – Tomato, Toemotto
While Universal App Campaigns are on the extreme end of the spectrum when it comes to automation and machine learning, they also represent a small fraction of the advertising spend that flows through Google Ads so most advertisers haven’t had to use them.
If you look at Google’s core advertising network (Google Search Ads), machine learning automation plays a much smaller role, but it has certainly has been increasing over time. Let’s take a look at how keywords have been affected by automation and what that means for advertisers.
Previously, an exact match keyword meant you were matching exactly with the search term a user was searching. For example, the exact match keyword [banana stand] meant your ad for banana stand would only show if someone searched precisely for “banana stand” and not “banana stands”, “bananas stand” or any other deviation.
In 2012, the first change to how keyword match types was introduced to the exact match type – the inclusion of plurals and misspellings. That meant the keyword [banana stand] could now be triggered by search terms like “banana stands”, “bannanna stand”, “banana stnd”, etc. This was a reasonable and understandable change – and only mildly annoyed the most granular marketers who held that misspellings and plurals achieved slightly different performance.
In 2017, Google took things a step further by including or ignoring function keywords, such as binding words, conjunctions, prepositions, pronouns, quantifiers, modals and auxiliary/hedging verbs. For example, the keyword [LA flight to New Orleans] could be matched to the search term “New Orleans from LA flights” with a change in preposition and word order. While the meaning is similar, performance between those two keywords could conceivably differ.
This led to frustration as marketers has to develop negative keyword workarounds to keep their clean account structures intact.
Then most recently in October 2018, Google extended the meaning of exact match once again to now target keywords with “the same underlying meaning as the keyword.” Essentially, this means exact match now targets keywords with the same intent as your keyword. Taking our previous keyword example of [LA flights to New Orleans], this could potentially match to the search term “fly to New Orleans from Los Angeles”, as the underlying meaning is the same. This creates a much more difficult challenge for marketers looking to keep tight account structures with minimal overlap.
All of these keyword match type changes have the good intentions of expanding audience reach and making keyword list building simpler and quicker which are great. That being said, they come at the expense of much less control and messy search term reports that can overlap across campaigns. This means that more than ever, marketers need to be diligent about pulling Search Term reports and implementing negative keywords to accommodate performance and structural requirements.
Buckle Up and Get Ready for More Automation
Machine learning and automation are beginning to show true promise after years of product development and experimentation by Google. It’s clear that automation within the paid marketing platforms is the future and marketers would do well to determine what this means for their workflows and strategies. Automation is likely to accelerate now that it’s starting to outperform human workflows.
That being said, we’ve found the best approach is dipping your toes in slowly (assuming you’re not forced to dive right in – *cough* UAC) when it comes to new automation features. This allows you to evaluate performance relative to your existing workflow and options before drinking the Kool-Aid.