Machine Learning: The Future of Google Ads Bidding Strategies?

From June 2017, Google Ads has a new Smart (Automated) Bidding strategy called Maximize Conversions. Google Ads Smart Bidding is defined as A subset of automated bid strategies that use machine learning to optimize for conversions or conversion value in each and every auction—a feature known as “auction-time bidding”.

Experimenting with Google Ads Smart Bidding

Google Ads automated bidding offers three core capabilities:

True Auction-Time Bidding:

True auction-time optimization means that Google Ads sets bids for each individual auction, not just a few times a day. This gives advertisers a much more precise level of bid optimization and the ability to tailor bids to each user’s unique search context.  Google Ads bidding algorithms evaluate relevant contextual signals present at auction-time. These signals include the time of day, specific ad creative being shown, the user’s device, location, browser and operating system. Google Ads can detect the presence of the above-mentioned signals to more accurately predict conversion rate and set a more informed bid for every search query.

Adaptive Learning at the Query Level:

According to Google, Machine learning algorithms “rely on robust conversion data to build accurate bidding models that predict performance at different bid levels. While high-volume head terms often provide plenty of conversion rate data for modelling, accounts typically have some low-volume or new keywords with little performance history that must be taken into account. For these low-volume keywords, bidding solutions rely on models to set bids that reflect the best estimate of conversion rates at that time.”

Google Ads automated bidding uses query-level data across your account, giving bidding algorithms significantly more data to make decisions with. This reduces performance fluctuations when keyword-level conversion data is scarce (which is quite often the case with new keywords or Google Ads accounts).

With Query-Level learning, Google Ads algorithms don’t have to relearn performance data from scratch when an account manager adds new keywords or moves keywords to a different ad group.

Query-Level learning implies that the algorithm learns at the search query level rather than the keyword level. The greatest benefit of this is that if a search query has already been matching on other parts of your campaigns, the algorithms simply apply what they’ve learned about it across the Google Ads account to make smarter bidding decisions.

Richer Contextual Signals & Cross-Signal Analysis:

As Google states, “Every user search is different and bids for each query should reflect the unique contextual signals present at auction-time. Signals like time of day, presence on a remarketing list, or a user’s device and location are key dimensions to consider when determining optimal bids.”

Over and above the abovementioned signals, Google Ads automated bidding also includes additional signals like “a user’s operating system, web browser, language settings, and many more to optimize for performance differences across platforms and users.”

All of these signals helps capture meaningful context for every search, which allows the Google Ads algorithm to increase the accuracy at which it predicts the conversion likelihood of each auction and set the optimal bid.

Experimenting with “Maximize Conversions” automated bidding:

As impressive, promising and convincing as Google Ads’ Smart (Automated) bidding seems, with years of experience we have learned not to get over-excited and implement new features immediately on all of our Google Ads accounts.

Instead, we have been running experimental campaigns on specific Google Ads accounts and campaigns. Conducting these experimental campaigns within Google Ads allows us to test potential major changes to campaigns by applying these changes only to that specific campaign, using only a portion of the available daily budget from the existing campaign.

These experiments have shown that this bidding strategy has varying results – Not only from Account to Account, but also varying results between different campaigns within the same account.

Take Account A into consideration, for instance. Two experiments were set up, one for Campaign 1, and one for Campaign 2. These campaigns have different keywords and keywords themes, but target the same market and geographic regions.

Results from Campaign 1:

Original Campaign

Experimental Campaign 1










Average Cost per Click



Total Cost



Average Position






Cost per Conversion



Conversion Rate




Results from Campaign 2:

Original Campaign

Experimental Campaign 2










Average Cost per Click



Total Cost



Average Position






Cost per Conversion



Conversion Rate



The above data clearly indicates that these Smart Bidding strategies are by no means a “one-strategy-fits-all” type of solution. As you can see from the data, one of the campaigns outperformed the base campaign when using the automated bidding, while the other performed significantly worse.

Even within the same Google Ads account, different campaigns could show very different results when implementing the new Maximize Conversions Smart Bidding Strategy. It also needs to be mentioned that these experiments have only been running for about 14 days each. Google does state that their Machine Learning could take up to 30 days to generate enough data to base decisions on, so it’s possible that things could improve.

That said, with the “core capabilities” offered by Google Ads automated bidding, I would expect that the algorithm shouldn’t technically need quite as much time to learn and adapt – but hey, I’m only human, what do I know?

One thing is for sure – we will keep gradually experimenting with this new feature and apply it where it does improve performance on our accounts. Where Automated Bidding doesn’t improve a campaign’s’ performance, though, we’ll keep things old-school of now. 🙂