Should you optimize Google Ads for machine learning?

I started doing Google Ads more than ten years ago and in that time, the game has changed dramatically.

Back in the day, only a few customers had conversion tracking installed. Many campaigns didn’t separate search and display. We would regularly tweak our bids by hand: a couple of cents up or down. 

If someone had started a 10-year sabbatical in 2010 and would like to resume running Google Ads today, they’d probably need to learn it from scratch.

Google Ads automation: the pros and cons

One of the biggest changes we’ve seen happening over the years is an increase in automation. The question is: Why would anyone manually tweak their bids if Google’s algorithm can do it better and more frequently?

Google has also been adding more and more automated campaign options that promise to improve your campaign performance and help you get better results. Google’s machine learning algorithm considers multiple factors, including the user’s location and their previous search behavior. You definitely wouldn’t be able to do all that manually.

On the other hand, some advertisers are wary of giving Google too much control over their campaigns. They’ve argued (and often for a reason) that Google can’t be trusted. After all, its objectives often differ from those of the advertiser.

At Supermetrics, we’ve been actively testing some of Google’s automation options. For example, we’ve used Google’s CPA bidding and portfolio bidding for several years now, and we’ve seen that they work well for non-branded search. We’ve found them to be very efficient in optimizing for target CPA. And even when we haven’t reached the target, Google has been automatically reducing the campaign/ad group spend.

How to optimize your Google Ads account for machine learning

Let’s say you’ve decided to take the plunge and start optimizing your Google Ads campaigns for machine learning. How exactly would you go about doing that?

The interesting thing here is that some of the recent recommendations are contradictory to established best practices. 

And that’s why I put together a list of the most notable new best practices:

1. Simplify your account structure

A granular account structure used to be the recommended best practice. We would split campaigns and keywords into small groups based on themes. In fact, it wasn’t uncommon to have SKAGs, aka single-keyword ad groups.

Currently, however, Google recommends that you simplify your account structure and combine keywords into larger ad groups and campaigns. The main reason for this is that the machine learning algorithm needs data to work with. And the more data it has, the better results it can get.

As Google analyzes some of the data on the ad group and campaign level, ad groups and campaigns with low impressions, clicks, and conversions will probably not be optimized efficiently. That’s why Google recommends having at least 3,000 impressions per week for an ad group and 30 conversions per month for a campaign.

2. Remove unnecessary splits

Similarly to the last point, many advertisers would also split campaigns and ad groups according to different principles. For example, they would separate ad groups or even campaigns for different match types (broad, phrase, exact), separate mobile and desktop campaigns, have individual campaigns for different countries, and so forth.

Many typical splits are now considered unnecessary, as the machine learning algorithm should automatically pick up on the best performing ads and optimize accordingly.

3. Don’t shy away from broad match keywords

Until recently, the best practice was to avoid broad match keywords whenever possible or use a broad match modifier.

In fact, many advertisers have wanted to only target the exact search queries they’ve chosen. These people have complained that Google has been removing more precise keyword options or that the exact match is no longer exact. This is partially true. Besides, using broad match keywords can easily get out of hand — especially in English language campaigns. 

On the other hand, according to Google, in 2017, around 15% of Google searches done by users had never been seen before. Even perfectly researched exact match keywords wouldn’t be able to target these new search terms.

4. Try dynamic and responsive ads

Some advertisers are still not comfortable with giving up control of their ad copy to Google, which is understandable. Personally, I’m also somewhat cautious about this.

Dynamic ads have been around for a while but they haven’t been working all that well for some advertisers. For example, in some cases, dynamic ads would simply cannibalize traffic from other campaigns.

Responsive search ads, on the other hand, are relatively new and it seems that Google might make them the default ad format in the future. It looks like they’ll require a slightly different mindset than regular search ads, which is why we’re starting to look into how to use them more efficiently. For example, traditional A/B testing of ads will no longer be possible, since Google will serve audiences ad copy that matches their search queries.

Should you optimize your Google Ads account for machine learning? 

Knowing all this, should we all just jump into the machine learning bandwagon and restructure our Google Ads accounts? 

The problem with answering this question is that you won’t know if it works unless you try. That’s why the best approach would be to take Google’s suggestions with a pinch of salt and gradually test different things.

But before you start experimenting, there are two important considerations that you should be aware of:

1. Firstly, any machine learning initiative is as good as the data you feed it. If your conversion tracking is poor, the algorithm won’t know it and it will optimize for whatever you’ve defined as a conversion. That’s why you’ll choose which conversion events you’re optimizing for and make sure your conversion tracking is airtight. However, this can be especially challenging for businesses with fewer conversions and longer sales cycles, such as many B2B companies. In these cases, you might want to focus on micro conversions that have a meaningful correlation with the actual conversion.
2. Secondly, you’ll want to answer these questions: What are your business goals? What do you want to achieve with paid search? Are there several goals or just one? What’s the value of a conversion? These answers will help you determine if machine learning is indeed the right way forward for you.

According to Google, campaigns should be structured by business goals, while ad groups should be structured by landing pages. This means that if you have multiple campaigns/ad groups that promote the same landing page and have the same business goal, you can consolidate them. 

On the other hand, if you have campaigns with different CPA targets, or if the conversion value from different regions is different, you should keep them separate. This way, the campaign split by products, locations, or target audiences is still valid, while as mentioned before, other types of campaign splits might not be necessary anymore.

Campaign testing at Supermetrics

For our own paid search campaigns, we started with auditing our Google ads account structure. We quickly found out that we have a lot of small non-brand campaigns and ad groups that have a similar structure and can be consolidated. 

Instead of having separate dynamic ad campaigns, we also decided to test adding dynamic ad groups to regular campaigns to expand these campaigns’ reach. Also, we started looking into ways to better use responsive ads.

We consolidated our smaller ad groups and campaigns into larger ones based on either business goals or landing pages. As our conversion path is relatively long, we still need some time to understand the impact on conversions.

However, our preliminary results show that simplifying account structure and adding dynamic ad groups drove additional traffic immediately. The largest consolidation we made, where we consolidated 41 (yes, forty-one) similar campaigns into one, has already resulted in 25.8% increase in weekly clicks. And if the additional traffic doesn’t convert well, CPA bidding will help us dial down the traffic and spend.

Conclusions

Using machine learning to optimize digital marketing campaigns is a growing trend, and Google ads are no exception. Google also has one of the largest data volumes and most powerful technologies on this planet. 

That’s why using Google’s campaign automation tools efficiently is important for success. In fact, I believe that those who don’t start automating their campaign optimization are likely to be left behind. 

But as new approaches can be quite different from the established best practices, it’s good to test them step-by-step to figure out what works for your business.

Editor’s note: If you’re looking for a quick way to analyze the performance of your Google Ads campaigns before and after machine learning, check out Supermetrics. You can now start a free 14-day trial of any Supermetrics product.