Facebook Machine Learning
The Facebook Machine Learning algorithm is comprised of over three million lines of code. It is also constantly being updated. There is one thing that Facebook has over Google when it comes to tracking their users. When you install the Facebook pixel on your website it will cookie those who visit. For example those who click on your ad. If those people have their Facebook open in their browser, the pixel will now be able to track their online behavior as well as their Facebook behavior. Facebook has officially announced that their Machine Learning algorithm is now performing just as well as ads that are manually managed, but there is a catch. You have to set it up properly. They are calling this the ‘Power Five’.
They have created a framework of five core tactics that drive the best CPA’s (Cost Per Action) possible. In other words you have to set it up in a particular way for this to work. In this article lets look at each of these more closely.
1 - Campaign Budget Optimization
Facebook has announced that it will start making everyone use this feature eventually. Right now you have the choice to turn it on or off. Marketing professionals have had mixed feelings about this. With this feature you are essentially allowing Facebook Machine Learning to make the decisions about which audience to spend your budget on. In other words, you are forced to trust Facebook. At the time of writing this article you still have a choice whether you want to use it or not.
Inevitable
At first Facebook was telling us that everyone would be forced to use this feature by February 2020 - but there was so much protest they have postponed when this is going to happen.
With CBO, you set one budget for all your ad sets. It will then automatically and continuously distribute your budget to the top performing ads. There are still budget constraints that you can apply. You can set a minimum or a maximum ad spend and FB will not exceed these limits, however, FB does not recommend that you use constraints...
“...we recommend using the least amount of spending constraints. Instead of using minimum and maximum spend limits with a CBO campaign, use cost and bid or return on ad spend (ROAS) controls to express ad value while maintaining spend flexibility.”
No Budget Constraints?
Of course having no budget constraints may not be the best option for those who have smaller budgets. It is important to note that FB suggests when using CBO - ‘if your active ad sets use spend limits, we recommend caution when changing your campaign budget.’
In general FB advises that you use caution when pausing or deleting ad sets when using CBO. So basically, you can easily screw this up.
“For example, a campaign with a $50 budget and 5 ad sets each with a maximum spend of $10 would make your total maximum spend limit $50. However, if you paused one of the ad sets, your total maximum spend would decrease to $40. Even though your overall campaign budget would remain $50, we would now spend up to $40. Or for example, imagine you have a campaign budget of $50 and a maximum ad set spend limit of $50. If you increase your campaign budget to $60 without also increasing the ad set spend limit, we can't use that additional $10. ”
Facebook explicitly discourages people from using both minimum and maximum ad spend controls at the same time when using CBO.
2 - Automatic Placements
Placements are all the places that Facebook can show your ads. Feeds, stories, In Stream, Search, Messenger, In-Article and apps and websites. By choosing automatic placements you let Facebook have flexibility as to where your ad is shown. Often, marketers prefer to not use this feature, and to just advertise in the news feed for example.
“ Automatic placements enable us to get the best results available from across all default placements. Because we can choose results from the widest range possible, automatic placements are typically the most efficient use of your budget and helps control costs. We recommend it for most advertisers. ”
When using this feature Facebook may not always distribute your ad to the lowest bid but will deliver the lowest cost overall. Not necessarily the lowest cost for each placement. The Facebook Machine Learning algorithm prefers this.
3- Dynamic Ads
This refers to both dynamic creative as well as dynamic product catalog ads. It tracks how your prospects interact with products online and in the news feed. It shows variations of offers. It will choose messaging or copy that it determines users will respond to the best. It is easy to set up, just toggle this button at the ad set level.
In the ad section when you choose ‘Single image or Video’. You now can choose up to ten images. The Facebook Machine Learning algorithm will now rotate and test in conjunction with different text, headline and description options.
Once ‘Dynamic Creative’ is toggled at the ad set level you will then notice in the ad level that it will now allow you add up to five variations of text, headline and description.
4- Auto Advanced Matching
This is something that is done in conjunction with pixel events. Auto matching is set up in the ads manager. Automatic Advanced Matching will tell your pixel to look for recognizable form fields and other sources on your website that contain information such as first name, last name and email address. This allows Facebook to increase the number of attributed conversions. To find this setting got the ads manager > pixels > settings > advanced matching.
It is worth noting that this wont work if your pixel is set up in a iFrame or if you are using an image pixel. This will also not work if you are doing business in a regulated vertical.
5- Account Simplification
Finally we have account simplification. Instead of splitting out your ad sets into a lot of different audiences, keep it simple. If there is any overlap, put the audiences together in the same ad set. Campaigns and Ad sets and Ads. It’s like a Russian doll. Campaigns are where you choose the objective. Ad sets is where you set the budget and the audience. Ads is where you build out the creative. When using the power five, keep the ad sets below four or five.
Conclusion
The Facebook Machine Learning Algorithm is bound to just get better and better over time. It seems that this method has a lot going for it and it is recommended by Facebook itself. Having said that, at this point one is still able to do all these things manually, but at some point that will no longer be possible.
Author John Victor