Get Better Results with Google Ads Automation

Get Better Results with Google Ads Automation

If you’re looking to get started with Google Ads to help grow your business, you’re in the right place.

In this blog, Mike Rhodes shares how to Use Google’s AI to Do the Heavy Lifting in Your Accounts!

There are 6 parts to this blog post – please see the Outline of this doc to jump to the relevant sections

Part 1

What Really Is Artificial Intelligence

Artificial intelligence. Machine learning. Deep Learning. GANs & CNNs. GPT-3!

Odds are, you’ve been hearing these terms more and more. Because in recent years the big giants of the online space—Amazon, Facebook, and especially Google—have been doing some amazing things with artificial intelligence (AI).

And if you’re a marketer or a company that depends on Google Ads to generate sales, you absolutely NEED to understand and leverage Google’s AI. Because if you don’t, it won’t be long before you’re left in the dust by more forward-thinking competitors.

Google’s CEO, Sundar Pichai, famously said:

“We are moving from a mobile-first world to an AI-first world.”

AI is becoming more & more important, more & more powerful, and more & more capable of generating better results inside of Google Ads.

But before we dive into how you can leverage AI inside your Google Ads account, let’s make sure we’re on the same page as far what AI actually is.

What Is AI?

Exact definitions aren’t needed here… which is good, because the world’s smartest geeks can’t all agree on an exact definition. All we need to know for now is that…

AI is the science of making things smart.

What we’re more interested in is a corner of the AI world called Machine Learning, or ML.

ML is about creating machines that learn without being explicitly ‘taught’ (or programmed).

In other words, ML doesn’t need to be told exactly what to do. Instead, it can learn and update on its own—which means it can (in many situations) learn much better and faster than we can program it manually.

And when you combine that rapid learning with a mountain of data (which Google happens to have), then AI & ML can do some really impressive things.

* Note. To make this article easier to read, I’ll use the term ‘AI’ to cover all the many flavours of AL & ML & Deep Learning (a form of Machine Learning). This isn’t technically perfectly correct, but makes reading easier.

How Powerful Is AI Today?

I’d like to give a little context on just how immensely powerful AI has become. This is important to understand if you’re going to trust Google’s AI with your (or your clients’) hard-earned money.

So first, a little story.

Back in the 1950s, computers started playing games for the first time. The programmers’ game of choice: tic-tac-toe (aka noughts & crosses). The computer took up the size of a large room, and it was really slow, but it was a rudimentary form of AI.

Fast-forward to the 1990s, when an IBM computer famously defeated Kasparov, then the best chess player in the world.

Then in 2014, the best AI minds on the planet believed they had found a near-impossible problem to crack: using Deep Learning to create a computer that would learn how to play the game “Go.”

Go is the most complicated game in the world. Chess, for example, is complicated but can still be brute-forced. (Meaning, at any point in a game the computer can run through almost every possible outcome to find the best move.)

But “Go” has more possible board positions than there are atoms in the universe. You can’t brute-force it; instead, it requires intuition. And many AI experts believed in 2014 that was an impossible task for a computer.

Well, just two years later DeepMind (now owned by Google’s parent company Alphabet) had built an AI that defeated the best human Go player in the world. It beat him 4 games to 1.

The following year, in 2017, they built an even better AI that beat the new best player in the world 3-0.

Pretty impressive, right? But they were just getting started. Here’s where the story gets really interesting.

They created a brand-new AI called AlphaGo Zero. But this time, they didn’t give the machine any examples of how humans play the game… they only gave it the basic rules of the game. Which consist of just 9 sentences.

After just 3 hours of simulation, the machine was as good as an amateur human. After 20 hours, it was as good as professional human player. After 3 days, it was as good as the machine from 2016.

After 21 days, it was as good as the machine from 2017 (the one that beat the world’s best player 3-0).

And after 40 days, it was virtually unbeatable. It played against that AI from 2017 and beat it 100-0!



Then in 2018, DeepMind took the learnings from that project & built AlphaStar. A machine designed to beat the best players in the world at the computer game StarCraft II.

This time it took a little longer… 200 years of simulation & a few months of some of the best geeks on the planet tweaking the dials. But they did it.

In late 2018, AlphaStar beat two top players in a game that:

  • Has a huge “game space”, which means there’s a LOT happening
  • Gives players imperfect information, meaning you have to guess what your opponent is doing
  • Requires a huge number of strategic decisions to be made each minute
  • And critically, the outcome of those decisions may not be discovered for some time, making it harder to know which were good & which bad

Now that sounds a lot like the average day for a marketer! Lots happening, not knowing what competitors are doing, many decisions to make & outcomes that aren’t immediately known.

And this is happening TODAY. Google own this technology. Some of it powers their various ‘smart bidding’ algorithms.

So why do I share this story? Because I want to drive home the point that AI is getting smarter & faster.

And there’s no company in the world with more AI expertise than Google.

Part 2

Why You Cannot Ignore AI In Google Ads Any Longer

When Google came out with their first automated bidding options, they were intriguing but the results they generated were inconsistent. Many advertisers who tested them came to the conclusion that manual bidding was still the best option.

But over time, that AI bidding has continued to get better and better.

So much you can’t ignore it for much longer—it’s just becoming too good, and too important. Before long, the advertisers who ignore these automated options will find themselves left in the dust.

Here’s why:

Google has access to an immense amount of data. For one thing, they have perfect data on every Google Ads account. They know every single search term ever searched, every single keyword ever triggered, every single ad ever shown, every single click and conversion—everything.

They also have really, really good data on us – the people who use Google every day! In fact, Google probably knows more about you than you realize.

Keep in mind, Google owns…

This is a MASSIVE amount of data that Google has on us. And this is all data that Google’s AI can leverage.

For example, imagine 2 people searching Google for the same exact keyword: “buy miele washing machine.”

Most advertisers would probably be willing to bid a lot on a bottom-of-funnel keyword like this. But they wouldn’t really know anything about these people beyond the terms they just searched for.

Google, on the other hand, knows a LOT.

  • Is this search activity usual for this person, or unusual?
  • Is this the first time they’ve performed this particular search? Or similar searches?
  • Have they watched any related YouTube videos recently?
  • What other websites have they visited? And what was their behaviour there?
  • What browser, OS & device are they using? How do they typically act on that device?
  • Are they in a typical location, or travelling & therefore likely to act differently?

For example, Google would know if one of these searchers has recently started searching for washing machine-related terms, or watching videos on “how to fix a broken washing machine.” Google would also know if one of these searchers previously visited the Miele website, which product pages they viewed, and how much time they spent on each of those pages.

In short, Google has much, much more context on the people performing these searches than you will ever have. And it can leverage this data to make more informed advertising decisions than any human is capable of.

And while Google’s AI isn’t perfect, it’s very good—and it’s getting better, every day..

We’ve hit that point in the curve where the AI has enough data that its improvements are growing more and more significant. (Just like the AlphaGo Zero, where 1 year was enough to beat the previous year’s AI 100-0.)

It’s a tipping point.

And if you ignore this AI today, then there’s a good chance that the next wave of AI innovation will leave you far behind.

So my recommendation is to stop trying to fight the robots. Instead, lean how to leverage AI. Let it take over some of the work in your campaigns, and use the extra time that frees up to work on more strategic marketing efforts.

If you do that, the end result for you & your clients will be more sales, revenue, and profit, at a better ROI.

Sound good?

Great. But let me say, right off the bat, that I am NOT recommending you simply hand over the keys to your Google Ads account and let Google do all the driving.

Not at all. Instead, I’m recommending that you TEST IT.

Always test before you trust.

And fortunately, there’s a way to do just that inside of Google Ads. Using a feature called Drafts & Experiments, you can set up an A/B split test to compare the performance of 2 different campaigns. We’ll dig into that in part 3.

Part 3

How To Do Manual vs Automated Settings In Google Ads

Google Ads has a feature called “Drafts & Experiments” that you can use to test all sorts of things that you might not have realized were testable. For example, you can test different website URLs. Different account structures (which comes in handy when you’re taking over an existing account that you’d like to restructure). And different campaign settings.

I can’t overemphasize how powerful this is. You absolutely need to start using this feature in your accounts!

Here’s how to do it:

First, click “Drafts & experiments” on the left-hand side of your screen. Then click “+ Draft” at the top.

There are 2 steps to creating a draft.

First, you have to choose an original campaign to base your draft from. When you do this, the draft will automatically make an exact duplicate of that campaign so you can make changes to it without affecting the live campaign.

The second thing you have to do is decide on a name. I highly recommend putting the word “Draft:” at the beginning of all draft campaign names, to make it easier to tell the difference later between your live & draft campaigns.

Now your draft is created, and it’s an exact duplicate of the original campaign. The next step is to make changes to the draft that you want to test. For example, if you want to test different bid strategies, navigate to “Settings” in your draft and change the bid settings there.

When your draft is complete, click “Apply” at the top. You will see 2 options: you can apply the changes to the original campaign, or you can run an experiment.

You want to run an experiment, so make that selection and click “Apply.”

Google Ads will allow you to choose start & end dates, and specify how much of your traffic you want to include in this test. For faster results, use a 50/50 traffic split. If you want to be more conservative, you can choose to send a smaller amount of traffic to the new draft campaign.

You can view your experiment at any time by navigating to “Campaign Experiments.”

And by clicking on the experiment, you can view the data to see which variation is performing better. Google Ads will even do the math and tell you which results are statistically significant, and which are not.

All you have to do now is watch the experiment until it has enough data to tell you which variation performs better. Once you know that, you have only 2 choices:

  • If the original campaign outperforms the draft, simply stop the experiment.
  • If the draft outperforms the original campaign, then click “Apply” at the top of the screen to replace the original campaign with the draft.

This is a really powerful feature that only a small percentage of the top Google Ads agencies are using. It’s the perfect way to test Google’s AI, so I definitely recommend testing this in your own campaigns ASAP.

So now that you know how to run a split-test inside of Google Ads, let’s dive into which AI features you should test, and in what sequence.

Part 4

Level 1 of 3 AI In Google Ads

What sort of things can you use AI to automate inside of Google Ads?

To answer that, let’s take a step back and ask a really basic question: What is advertising?

For what purpose are we applying this AI stuff?

“Advertising is showing the right message to the right person at the right time, profitably.”

If we break down that definition, we get 3 distinct pieces that we can directly correlate to different aspects of your Google Ads campaigns:

  • Showing the right message—this is your ad.
  • To the right person, at the right time—this is your targeting.
  • Profitably—this is your bidding.

And when we look at that definition through the AI framework, we can organize those elements into a pyramid—with bidding at the bottom, keywords & audiences in the middle, and ads at the top.

These are the 3 big levels of AI that you can test inside Google Ads. Now let’s go through each one in detail, starting with the foundation—bidding and budgets.

Level 1: Use Google’s AI to Optimize Your Bidding

If you’re just getting started with AI testing, the first thing you should look at is your bidding. This is the thing that Google has had the most success with automating, and that’s why it makes up the foundational level of our pyramid.

And when it comes to testing automated bidding, there are really 2 main strategies you’ll want to look at:
target Cost Per Action & target Return On Ad Spend (tCPA & tROAS for short).

If you’re a lead-gen company, I recommend testing target CPA if you get at least 50 leads/month from Google Ads.

If you’re an e-commerce company, I recommend testing target ROAS if you get at least 100 sales/month from Google Ads.

If you don’t meet these minimum numbers, you can still test—just keep in mind that Google Ads will have less data to work with, and as a result your results will be less consistent.

On the flipside, if you get 500 conversions/month or more (per campaign), then the machine will probably do a better job than you can do on your own.

Remember: the algorithm needs data to work its magic.

Now use Drafts & Experiments to test a new bid strategy against the old, manual way.

Part 5

Level 2 of 3 AI In Google Ads

The next level of Google Ads AI testing is your audience & keyword settings.

And there are a lot of ways to do this. Here are the ones you should consider testing in your campaigns:

Remarketing Lists for Search Ads (RLSA)

The RLSA feature in Google Ads allows you to ‘layer’ your own remarketing lists over top of your search campaigns, adjusting your targeting & bidding based on who has visited your website in the past. There are two ways you can use this feature:

  1. Targeting. This lets you narrow the reach of your campaigns, so that your ads will show ONLY to people who are on your remarketing list.
  2. Observation. This gives you the ability to use RLSA to keep showing ads to ALL people using your keywords, but adjust your bids for people also on your remarketing list — for example, you might be willing to bid more for a person who has visited your website in the past 7 days.

You can change this option in the “Audiences” tab on the left-hand side of your Google Ads interface:

My recommendation is to use the “Observation” setting, as using “Targeting” badly will greatly reduce the performance of your campaign. Also, generally speaking, you should probably be willing to bid more on a person who has visited your website in the past.

And you can get really specific with this, too! For example, most e-commerce advertisers would do well to dramatically increase their bids for shopping cart abandoners. In this example, we increased those bids by 50% but didn’t pay any extra and got a 37% DECREASE in cost/conversion:

My recommendation is to use the “Observation” setting, as using “Targeting” badly will greatly reduce the performance of your campaign. Also, generally speaking, you should probably be willing to bid more on a person who has visited your website in the past.

And you can get really specific with this, too! For example, most e-commerce advertisers would do well to dramatically increase their bids for shopping cart abandoners. In this example, we increased those bids by 50% but didn’t pay any extra and got a 37% DECREASE in cost/conversion:

Demographics for Search Ads (DFSA)

DFSA is similar to RLSA, but instead of using your remarketing lists it layers demographics data over top of your search campaigns.

This is a powerful setting that everybody should be testing.

Most companies have a certain demographic that their product or service appeals to. If you sell a premium product, for example, then you should probably bid higher on people in a high-income bracket. If you sell walking canes, then you only want to advertise to older people. And so on.

To find this setting, just click the “Demographics” tab on the left. Then at the top of the screen you can narrow down your results based on demographics information such as age, gender, and income.

Similar Audiences

This is Google’s version of Facebook’s Lookalike audiences. Google is not as good at this as Facebook is, yet, but Similar Audiences can still be useful—and they’re getting better all the time.

To test one of these audiences, start with a remarketing list and tell Google to go and find people that are similar to the people on that list.

Another place to start would be to use your current customer list. That way, Google will try to find people who are similar to your existing customers—which, theoretically, means that those people should be more likely to become new customers. (Note there are some restrictions to using the “Customer Match” feature. Get in touch if you get stuck).

Intent (In-Market) Audiences

Some of the most powerful AI settings inside of Google Ads are those that target people based on their intent. And this is much, much more powerful than targeting people based on demographics.

Using its plethora of data, Google can look back at a person’s recent search history and try to figure out what that person is trying to do.

When they search for “cars,” for example, are they actually interested in buying a new car?

These are called In-Market Audiences because they are people who are “in the market” for a certain product or service.Google currently has around 700 of these audiences created. For a list of all Google’s In-Market Audiences, head to and click the “InMarket” tab at the bottom:

And you’ll see a complete list of all 690+ In-Market Audiences:

Custom Audiences

In-Market Audiences are great, but some advertisers may be wondering:

What if there isn’t an In-Market category that fits my product or service?

In the past you would have been out of luck. But now you can create your very own custom audiences.

And in fact, if you check your Google Ads account you’ll see that Google Ads has automatically created some of these for you! All you have to do is apply one of these audiences to a campaign and you can start running ads to it. (Note the name changed in mid 2019 from custom intent, to just custom audiences).

This is a game-changing shift, especially when it comes to the Google Display Network (GDN). Now, rather than displaying your ads based on the content of a particular website or the demographics of a user, you can display your ads based on what the user actually WANTS.

Because let’s face it: if you know that someone is in the market for a car, it doesn’t really matter what website they’re on. That is a person that you definitely want to show your ads to.

Custom Audiences are, in my opinion, the future of GDN. Eventually I think they will replace most other GDN targeting methods.

Part 6

Level 3 of 3 AI In Google Ads

The third and final level of Google Ads automation is to use the Google Ads AI to optimize your messaging.

Unlike levels 1 & 2, which in many cases are already really effective, level 3 automation is not quite there yet.

It’s getting better, of course, but generally speaking the automatically created ads in Google Ads are not very good. It’s going to take some time before we can trust Google Ads to create really effective ad copy on its own.

In the meantime, however, there are 2 narrow use cases where messaging AI is already working very well.

First, there’s a service called Lexio. This is a tool you can hook up to your Analytics and it will automatically create your Analytics reports for you.

Here at WebSavvy we’ve used this as an internal tool only, rather than a client-facing one. But it can still save a lot of time and provides some useful insights.

The second use case is a tool called Phrasee, which writes high-converting email subject lines. It doesn’t write the whole email—just the subject line—but what it does, it does well. This tool has made many millions of dollars for some of the companies smart enough to test it.

So depending on your business, you may want to give Lexio and/or Phrase a look. Other than that, you still need to create your own messaging manually.

At least…for now.

Also check out the examples online of GPT-3. This is writing by an AI .. nad some of the examples it’s generated are amazing. So machines will be able to write. Maybe sooner than you expect.

Where Should You Get Started?

OK, so we’ve covered the 3 main levels of Google Ads automation:

  • Bidding automation
  • Targeting automation
  • Messaging automation

So where should you get started? What’s the best place to begin your Google Ads AI testing?

Here are the most important things you need to start doing today:

Test & experiment more.

Use Drafts & Experiments inside of Google Ads. Test new settings. Anytime Google Ads comes out with a new feature, test it.

Start at the bottom of the pyramid and test automated bidding first (target CPA or target ROAS).

Layer audiences on top of Search & Shopping.

Then, start to test some of the different audience features like RLSA, DFSA & In-Market Audiences.

Layer those on top of your existing campaigns to improve the focus of your targeting.

Diversify your spend.

Don’t put all your eggs in one basket. Right now there are a lot of companies that would absolutely crumble if Facebook marketing stopped working. Don’t put your company in that kind of risk.

Instead, diversify your advertising portfolio. Get on the Google Display Network and test some of their In-Market & Custom Audiences.

Get great at writing persuasive copy.

Invest in skill that computers still can’t do very well, like copywriting. When you combine great copywriting & design with the AI features we’ve already talked about, your results can be really amazing.

Keep asking “Where do I add value?”

Because that’s the game. That’s what really matters in the end. The more value you provide, the more you will receive in return.

And remember that just because Google’s AI can be very good at optimizing for certain things, it still needs to be applied strategically. You can’t just hand over full control of your campaigns to Google Ads. Otherwise you’ll end up paying “the Google Stupidity Tax.”

Google can optimize for a lot of things, like CPA, CTR, ROAS, and so on.

But it doesn’t optimize for profit!

Optimizing for profit is hard, and it still requires high-level strategic thinking & testing to grow your company’s profits.

So even though the Google Ads AI is getting better and better at automating certain parts of the advertising process, you are still the most essential part of that process. You have to be the one to guide that AI.

Because in the new wave of Google Ads AI, the companies that come out ahead will be the ones that are the best at quickly testing new tactics and features, and using them strategically.