I used to be a vocal sceptic of all the AI & Machine Learning settings in Google Ads, and firmly said no to our Google team who regularly suggested we switch campaigns over.
Why? Quite simply we’d tested these settings multiple times for clients over the past 6 or 7 years and they didn’t work! We would switch them on for a client who had consistent (and improving) results, and start to see the results either declining, going haywire or ramping up the spend to Google’s benefit, but without many additional conversions.
Luckily, things are different now, so i’ll tell you what has changed, the best practices that you need to follow before you start and a full breakdown on all of the Google Ads AI & Machine Learning Optimisation options available for you.
So What’s Changed?
Google Ads campaigns have become more and more complex, and given the dynamic nature of search auctions, the “right” bid is a moving target that is now hard to reach due to the complexity. With more and more ways to adjust your cost-per-click bids based on signals that a user is likely to convert (ie. location, device, time, audience segment, interests etc.), it is now near impossible to adjust bids manually based on each user’s unique combination of signals.
Even if you are more of a sceptic than I was, the sheer time it takes to manually adjust bids on all of these metrics is an opportunity cost in itself – with things such as increasing the user experience of your website a far better and more impactful use of your time.
Along with the complexity of bidding manually, Google have also improved its AI & Machine Learning and now go a step beyond traditional signal analysis by recognising and adjusting for meaningful interactions between combinations of signals, and constantly considering new ones. These changes are making a positive impact to campaigns and as we re-tested these bidding settings slowly over time, they actually started to work. Just check out how much work they have done on these bidding options in the last few years:
These three factors piqued our interest so we roped in an AI expert from Google to come in and present to our team who showcased the three features that he thought were the key reasons advertisers should all be using AI bidding:
- True Auction-Time Bidding – Google Ads automated bidding offers true auction-time optimisation that sets bids for each individual auction as it happens rather than at an ad group or keyword level with updates every now and then.
- Adaptive Learning At The Query Level – Machine learning algorithms rely on conversion data to build bidding models that predict performance at different bid levels. This means the AI bidding is testing a range of different bidding options at once for every search and picking the level likely to get the best result.
- Richer User Signals & Cross Signals Analysis – Signals like time of day, presence on a re-marketing list, or a user’s device and location are key dimensions that AI bidding considers when determining optimal bids.
Since our meeting with Google, we’ve worked out the best way to use these settings and have been testing AI across all eligible campaign, with some pretty promising results.
Before You Get Started…
Before I take you through the options, there are a few things you need to do to get the best results:
1. Check Your Conversion Tracking
Double check all conversions are being tracked correctly and that each one is a valuable conversion for your business that will have an impact on your growth. Google will optimise based on the data you give them so you need to make sure you’re only counting things that matter.
2. Create Best Practice Campaigns
Ensure you have built a best practice campaign which has segmented ad groups, has awesome ads and offers, relevant keywords selected with the right match type, extensions in place AND all the other common elements in high converting campaigns.
3. Utilise Audiences
Audiences are a key signal for AI bidding and should be applied at either the campaign or ad-group level on all campaigns with the bid only option selected. Make sure to upload customer and prospects’ email addresses to create Customer Match audiences, create a re-marketing tag for your website to create segmented audiences from, and utilise Similar Audiences of each to apply to your campaign.
4. Use A Non Last-Click Attribution Model
Although I use last-click attribution in Google Analytics to compare channels against each other, as it de-duplicates conversions there is no reason you can’t turn on a different attribution model in Google Ads itself. Doing so will give the machine learning model a lot more data to work from, helping it set optimal bids at every keyword auction – even those earlier in the conversion path. I personally use time-decay for my accounts.
5. Wait Until You Have Enough Data
Google needs a lot of data to learn how to set its bids, so the more you have the faster you’ll speed up the algorithms’ learning period and the better the results will be. You should wait until your campaigns have been running for at least two months and are generating over 30 conversions per month before you test these bidding settings.
This table from bgtheory.com breaks down exactly how many conversions you need and the learning period for two of the main bidding types:
Google AI & Machine Learning Options
If you meet those three requirements and are ready to move your campaigns over to Google’s AI bidding, there are a few options you can test – which one you should try depends on the type of campaign you’re running.
The most effective for performance-focused campaigns are:
- Enhanced Cost Per Click (eCPC): Adjusts your bids up and down depending if a conversion is likely or unlikely. A successful eCPC experiment will increase the total number of conversions at a stable CPA.
- Maximise Conversions: Sets your bids to get as many conversions as possible within your campaigns budget, without caring about the cost per conversion. A successful ‘Maximise Conversions’ experiment will increase the total number of conversions compared to the control period while spending the campaign budget.
- Target Cost Per Acquisition (tCPA): Sets bids to get as many conversions as possible within a CPA goal – but with no target on volume of conversions. A successful Target CPA experiment will increase the total number of conversions while maintaining a stable CPA.
- Target ROAS (tROAS): Sets bids to generate as much eCommerce revenue as possible within a return on ad spend goal. A successful Target ROAS experiment will increase the total conversion value while achieving the set Target ROAS value.
Each of those options will give you a recommended bid, be it a bid per conversion or a return on ad spend amount. I would start with Google’s recommendation as if you bid too low or too high your campaigns will not work effectively.
There are a few other automated bidding options that I would be a bit careful with and would only apply for very specific campaigns if you know what you’re doing. They are:
- Target Search Page Location: Automatically targets your ad to the top of search results or onto the first page.
- Target Outranking Share: Lets you automatically outbid and appear above a selected competitor’s ads.
- Maximise Clicks: Automatically gets you the most clicks possible out of a certain budget.
Whatever AI bidding option you select for your campaigns, make sure you’re patient, which.. I know, is tough. The campaign will need time to learn and for the AI to test new bids so I would give it at least two months before you change anything as every bid or budget change you make will reset the learning period.
If you have a campaign that’s working and are scared to turn this on completely you should take a look into bidding experiments where you can test a new bidding strategy on just 50% of the campaign – it will take longer to get results though.
Since I am still a recovering sceptic I will point out one concern I have before I finish up – the power that Google is going to have if everyone has automation turned on. With a flick of a switch they could increase every cost per click on every keyword reaping billions without too many advertisers even noticing. That’s a lot of power to wield and I hope they use it wisely. Despite this, these bidding options are working so advertisers should still move ahead with testing are implementation.
If there are any other sceptics or converts out there – would love to hear how these settings are going for you in the comments.