There is a lot of buzz around artificial intelligence and machine learning. A recent study by global consultancy, Accenture, determined that AI has the potential to increase corporate profitability by an average of 38% by 2035.
Google and Facebook are both investing heavily in AI and machine learning. Some of this research and development has led to tools and services that are available to us today.
In this post we will look at how Google and Facebook artificial intelligence can help us find new customers.
Facebook Lookalike Audiences
If you provide Facebook with a sample of your existing customers, you can ask it to use its machine learning algorithms to identify other people who it believes look like your customers. This is based on mining the demographic, interests, activity, usage and behaviour data of its 2.33 billion monthly active users.
You can then create Facebook ads to target the lookalike audience – which you can further refine using standard Facebook ad targeting such as geographic, age, gender etc.
There are 3 sources of existing customer data you can ask Facebook to use.
- People who like your company Facebook Page
- People who have visited your website – the data is collected by adding the Facebook pixel to your website
- Uploaded customer list – Facebook attempts to match your contact list to its users based on email addresses and phone numbers)
You can create multiple audiences for people who visit your website. For example, people who visited a specific page, spent a certain amount of time on the site or completed an action – such as adding to basket but didn’t checkout.
One of the great benefits of lookalike audiences is that Facebook will identify many more “potential new customers” than the number of existing customers you provide.
In the example above Facebook is proposing to build 3 lookalike audiences based on the the fans of a company Facebook page (which has approximately 8,500 likes). In this case the 1% lookalike list (the “top 1% of lookalikes”) contains over 400,000 potential customers from an initial sample list of 8,500. The 2% – 5% contains over 1.2 million potential customers.
But how does the Facebook Lookalike Audience AI work?
Unfortunately, Facebook doesn’t explain exactly how the AI / machine learning determines lookalikes. It is a black box. However, we do know that is geared towards demographics and interests.
Here at Peak Demand, we’ve had success for our clients with using the 1% lookalike audience for website visitors, the 1% lookalike audience for people who have liked the Facebook page and the 1% lookalike audience for an uploaded customer list.
Recommended approach for Facebook Lookalike Audiences
Test, test, test. If you have a decent number of fans for your Facebook page then start with creating a lookalike audience based on this as it is the easiest way.
If you’ve already configured the Facebook pixel and carrying out Facebook retargeting then it is easy to extend this to lookalike audiences. Unless you are constrained by budget, we always recommend creating a separate Ad Set to track lookalike audiences so that you can see the performance. If you merge a lookalike audience with a retargeting audience you will lose the accuracy of analysis.
Google Similar Audiences
Lookalike audience on Google Ads are called “Similar Audiences”. If you have created an audience in Google Ads (or Analytics) then Google’s AI / machine learning will build a similar audience automatically for you. Unlike Facebook, Google doesn’t allow you to create audiences which have a specified level of similarity – it simply creates a Similar Audience for every natural audience you’ve created.
For Google you can create “natural audiences” through Google Analytics (most popular), Google Ads tag, YouTube, Google Play (for apps) and App Analytics, e.g. via Firebase.
As with Facebook you can create Google audiences using broad criteria – such as general website remarketing – or custom audiences based on specific actions or segments.
Google Similar Audiences can be used for targeting users with ads across the Google Display Network (GDN), for targeting YouTube users, App users, Gmail users and as per the example above for users carrying out Google searches.
In this case, you apply a similar audience to an ad group in the same way you would for a Remarketing List for Search Ads (RLSA) – either for dedicated targeting or observation only. As with RLSA you can use observation to adjust bids for people within the Similar Audience or targeting to show ads only to people on the list. A similar list must have at least 1,000 members to be eligible.
Note that Google Ads does not allow you to use a similar audience list for exclusion in search results.
How do Google Similar Audiences AI work?
For retargeting Google states:
“To find similar audiences, Google Ads looks across the millions of apps and sites on the Display Network. As your remarketing lists change, your similar audience will change as well. Similar audience lists are created using machine learning and are built to adapt to changing markets and trends dynamically. These models find high-performing users who are similar to existing customers.”. Source: https://support.google.com/google-ads/answer/2676774?hl=en-GB
We can assume Similar Audiences based on other sources work in a similar way.
Are Facebook Lookalike Audiences & Google Similar Audiences static?
No! They are dynamic. As your source data is updated so is your lookalike / similar audience. For example, if you’ve built audiences based on your retargeting website visitors, each day Facebook and Google will be refreshing the list of lookalike / similar audiences – finding new people and removing others. It is important to note that any user who is in your retargeting audience will not be in your lookalike / similar audience.
If you are are actively targeting your lookalike / similar audiences then the expectation is that users will move from your lookalike / similar audience and into your Facebook / Google retargeting audience.
Can I see who is in a Facebook Lookalike / Google Similar Audience
No. You simply see an audience number.
You have to trust Facebook and Google’s AI and machine learning algorithms and test / check the results from running ads to these audiences.
Do Facebook Lookalike Audiences & Google Similar Audiences work in the real-world?
Yes, for many – but not for all. As with every tactic in digital marketing the key is to test, test and test again.
Our clients see Facebook Lookalike audiences and Google Similar Audiences as a way to use the power of Facebook & Google AI / machine learning to extend their advertising reach to new people with similar characteristics and browsing experience to current fans / customers.
Recommended next steps
- Create a basic Facebook Lookalike Audience based on fans of your page, Pixel retargeting or Customer Match. Run a small budget ad campaign targeting the audience alongside an ad campaign targeting your natural audience and compare results.
- Apply an existing Google Similar Audience to a search campaign in observation only mode for a month. Compare clicks and CTR (click through rate) for the similar audience alongside a default audience.
Contact us for details of our Facebook campaign management and Google Ads management services.