Suppose you are a data-driven search marketer and want to maintain significant control of your main advertising assets such as keywords and ad copy. In that case, you must be disoriented by the trajectory that search advertising platforms, particularly Google, are taking. Starting on June 30, 2022, Google no longer allows the creation or editing of expanded text ads, leaving only responsive search ads as an option.
Last year in 2021, a similar move was made around broad match modified keywords. Everything indicates that the industry is going towards more machine learning-driven optimization, with fewer possibilities of manual control. The introduction of performance max campaigns demonstrates it as well. This presents new opportunities and challenges to face. How do you ensure that your ads are as efficient as possible when Google has more and more flexibility on what to deliver?
Understanding Machine Learning Optimizations
I think about an algorithm as a huge calculator that can process a lot of data in a matter of seconds. The key here is “a lot of data”. The easy trap to fall into is not to give your algorithm enough data to work with. In the instance of Google Ads, we can look at it at the ad group level to determine whether enough data is being collected. Your responsive search ad as well as your keywords function at that level, which is the reason why we look at ad groups.
You want to be careful about over-segmenting your ad groups, which could result in your responsive search ad not receiving enough impressions and clicks to optimize efficiently. When you have 15 headlines and 4 descriptions, that represents a lot of possibility for testing and Google needs to reach statistical significance for the ads to work well. To illustrate the problem with limited data, you can always refer to what happens when you create a new account.
In one of our accounts advertising apartments, we tested a full responsive search ad approach along with smart bidding and broad keywords. We realized very quickly that Google started delivering ads for location keywords completely irrelevant to where our apartment complex was. We took a rather extreme approach in giving Google full control and found out correlatedly that our account was for sure not ready for that. That said, when your account has enough data, things can go in the opposite direction, and that’s where you will fall in love with smart optimization.
If You Give it Enough Data, It Will Work
On the other hand, we did a similar test with an account with a history of recorded conversions. We were more diligent with the approach as we created a campaign where only a few broad match keywords were implemented within a single ad group with one responsive search ad. In essence, the campaign setup did leverage full machine learning-driven optimization but it had less to optimize at once and operated within an account that had historical data to work with. The results below were recorded during the first 3 months and were fantastic.
| Campaign | Cost | Conversions | Cost/click | Cost/conversion |
| Control Campaign(11 ad groups, 50+ keywords, audience targeting) | $2,8k | 21 | $2.88 | $133 |
| Test Campaign(One Ad Group, 2 broad keywords, audience targeting) | $1,6k | 21 | $2.46 | $77 |
The cost/conversion was reduced by ~70%, changing nothing else than the campaign structure. This tells you the importance of structuring your campaign appropriately as we transition to smart bidding, responsive search ads, and more utilization of broad keywords.
What if You Still Need To Segment Your Campaigns?
The question is then, do you really need to? Or did you get used to a methodology that you struggle getting away from? If the answer is that you need some ad copy control for example, which is fair and not rare, then you want to make sure that each of your ad groups have enough data to work with. This is true for impressions, clicks, and conversions – but you can give yourself an arbitrary number to commit to such as 10k impressions per month in any ad group. So you still can have multiple ad groups, but just keep in mind that they will need food (data) to survive and stay healthy.
Conclusion: Automated Optimizations are Here to Stay
Computers are getting faster and better at calculating, so it is natural that search advertising platforms will continue to implement machine learning-driven campaign types and optimizations. Privacy has become critical for consumers in recent years and that implies that advertisers will see less data as well. Understanding how to navigate the new search advertising era with fewer data and more leverage for computers will be key success drivers.
