The Automation of Creative

5 years ago I was on a call with a team at Google. They were excited to discuss implementing an automated bidding system for a client. This was the advent of the transition to an automated age which I was highly skeptical about embracing. The idea of automating different elements of a digital advertising campaign is an intimidating prospect and in light of the recent updates and changes on Google and Facebook Ads, I want to address some concerns and shed some light on the situation.

“Machine learning” is a phrase you’ve probably heard mentioned recently. It sits at the apex of the changes that we are seeing in the digital marketing industry right now and there’s a lot of confusion as to what it really is. Machine learning is essentially a process by which a computer does an immense amount of calculations and computations on a data set to determine the best strategy. 

Say for instance you’re noticing that your advertisements are working but the cost of bringing in a new customer is too high. Normally, an advertiser would do some segmentation analysis on your audience and review if certain kinds of individuals are more expensive to convert to customers. For example you noticed that females are roughly 50% more expensive to bring in as new customers. So by bidding a bit less aggressively for the female audience you are able to lower your average cost per customer. Similarly, imagine that you looked at the age of your audiences and determined that people who are between 35 and 44 are the best performing age segment and everyone else was too expensive, so you were willing to bid less for anyone who is outside of this target age range. Once you apply these changes, you’re able to receive customers at a lower cost of acquisition.

Where does machine learning come in? Let’s take a closer look. Imagine our data shows that women aged 45 – 54 are actually very high value targets for our campaign. If they have children, however, they are actually more expensive to convert, but if they have children and they are in the top 10% income bracket, this is not the case. Now imagine this level of complexity exists across dozens of demographic, psychographic and interest-based segments. How are you going to find the time to discern and decipher all of these complicated audience targets, especially when you look at all the permutations of combining and segmenting these audience demographic categories? This is where machine learning excels and allows you to lean on real-time computation and analysis to show ads to your best performing audiences based on historical results in a way that no human could ever compete with. This is machine learning. This is automation.

There are many in the digital advertising space who are still struggling with the above concept as the industry continues to evolve. The procedure of automating the advertising bidding process is now a few years old and we are seeing the evolution of how this automation process can apply to the creative and display content that we show customers.

I’ve written extensively on the dynamic search ad system. It’s a very volatile yet very powerful system that I study closely and use consistently across a variety of advertising accounts. There’s some obvious work that needs to be done to refine the system but in general I’m comfortable with how it works. Believe it or not, this system can actually write your ads for you based on the kind of content you have on your website’s landing page in combination with what a potential customer is searching for. But I don’t want to go too deep into the wormhole that is dynamic search ads. Back to where we left off…

In the second quarter of 2021 many advertisers were speaking to their Google account teams about the fact that static written advertisements would no longer be available by the third quarter of the year. This kind of advertising format is referred to as the “expanded text ad” and allows  advertisers to write a specific combination of text fields that display to a potential customer when they searched for a term that they were targeting. The shift away from expanded text ads mark the beginning of a period where, instead of writing a very specific ad that will show exactly as advertisers prepare it, advertisers prepare a number of text fields that they are comfortable displaying to a potential customer. The system then decides which piece of creative it should use based on what the individual is searching for as well the historical performance of the ad text narrative.The same approach is now also being deployed across Facebook to a degree.

In simple terms, this means that customers and advertisers are no longer able to have the same type of control over their ad creative text on Google and Facebook. Some see this as a potential crisis but I genuinely can tell you that the system delivers better results and drives higher revenues for our clients when used correctly. There are no instances where an advertiser does not have reasonable control over what kind of text is prepared and used in their ad accounts.

To look at some specific examples of this you can see below an ad preview shows five potential headlines that the system can experiment with. The goal is to find the one that is most effective at getting users to click and then convert into a customer on the website. Each one of these was prepared specifically and written by our team.

The above five headlines were written and uploaded into the system to be tested. With the aim of driving the most number of users to a website, our goal was to optimize the ad creative around the ad text which produces the best performing click-through-rate.

After some time in a testing phase you can see the best performing combination by click-through-rate below.

Looking at this specific ad account in its entirety, when we use the automated creative system we receive an average click-through-rate of 6.75% compared to a 4.34% click-through-rate when we use the old expanded text ad system. The performance gains have been outstanding when using the new responsive search ad system.

If you have any further questions about how this process works, it would be our delight to speak with you further about it. You can reach us at linda@apparatus.com.