In my previous article, I talked about how my client was making some mistakes with her Facebook ads. Since then, we’ve iterated the ads, they’ve been running for a month and here are the results: 

Purchase ROAS (Return on Ad Spend) increased by 2.18 times

33% increase in result rate (# of purchases / # of impressions)

43% decrease in cost per purchase

Given that ad performance has improved, I thought that I could share some of the changes we’ve made. As a disclaimer, I would like to highlight that there is no one single ‘model’ way to optimize ads. These were merely what I hypothesized would improve performance, which were then put into action! 

#1
Past: Multiple audiences in an ad set 
Now: One audience in an ad set

Previously a bunch of audiences were all under a single ad set, but now that we’ve split the audiences up and placed a single audience in one ad set each, we’re able to know exactly which are performing better when we view the results at an ad set level. Doing so allows us to identify and only include such audiences in new campaigns moving forward. 

Although one should ideally be running a split test to accurately determine which audience performs best, testing the audience as a variable is however often too costly for small businesses with a limited marketing budget. 

For example, a month-long campaign with a daily budget of $86 is required for the estimated test power (the likelihood of detecting a difference in your ads if there is one to detect) to hit the recommended 80%. This is also assuming that one is only looking to test 2 audiences, which might not be the case.

Certainly, the test can be ended early if a winning ad set is found, but running this test would require preparing to spend this amount, which can be more than one’s budget or severely reduce one’s budget for other ads for small businesses.

#2
Past: Horizontal videos that were 60 seconds long
Now: Vertical videos that are less than 15 seconds 

The video creatives used previously were framed horizontally, which are not mobile-friendly since mobiles are often used vertically. Now, the videos have been customized for each placement (4:5 for Feeds, and 9:16 for Stories for e.g.) so that the ad viewing experience each time is the best possible one. 

Given that my client is reaching out to potential customers who have not heard of the brand, it is even more crucial to catch their attention right from the first frame. This is why the video started right where the ‘action’ is - showcasing the fresh meals it offers as opposed to its brand logo. The video is also shorter so that the key message can be conveyed as soon as possible given that there’ll be view drop-offs.

#3
Past: Repeating audiences across campaigns
Now: Unique, non-overlapping audiences across campaigns

Previously when audiences were repeated across campaigns of the same objective (as shown below), it drove costs up as ads competed against one another during an ad auction.

Having cleaned up the audiences across campaigns and ensuring that there are no repeated audiences, this has likely contributed to the lower cost per purchase. 

Despite not repeating audiences across campaigns, there is still a risk of huge overlap amongst audiences which would similarly drive costs up too. Thus it is important to use Facebook’s audience overlap tool (example screenshot below) to ensure that there are no major overlaps amongst the audiences utilised.

Next steps

Now that we’ve made some changes and seen some improvements, there are still a few other things that can be done. In fact, reviewing and iterating should be done consistently, instead of leaving the ads to idle for months on end.

1. Creating an ad funnel

In my previous article one of the mistakes mentioned was not having an exclusive campaign for retargeting, and so changing that would mean creating an ad funnel. This doesn’t necessarily mean having to create a full one - it can be as simple as a partial, 2-step funnel where there’s a retargeting campaign to support existing campaign(s). 

2. Explore the different sizes of lookalike audiences (LLA)

One way to be even more precise in terms of identifying better-performing audiences would be exploring the different sizes of LLA and narrowing down on that.  

We’re currently using an audience size of 10% for all our LLAs, which means that Facebook would determine the top 10% of Singapore’s population that is most similar to the source. I have an inkling that this might be too broad for effective spend, but the only way to find out would be to test it out, starting with an audience size of 1 or 2% and then increasing the size if required.

Conclusion

I hope you’ve had a good read and gleaned some learning points from this! Here’s what I did, but I would love to hear how you would approach the iteration instead. As usual, there’s no right or wrong answers so don’t be afraid to chime in!