Who you show your ads to is just as important – or more important – than what you show them. Every advertiser knows this and uses it. Platforms like Facebook and Google App Campaigns have given us incredible tools to slice and dice audiences so the most likely buyers rise to the top.
But just like with creative, those audiences can wear out.
Audience fatigue is why advertisers can’t just pick one lookalike audience and leave it at that. If the same people see the same creative over and over they’ll just start to ignore your ads.
But that’s not the only reason audience fatigue sets in. Some people will respond to your ads. But after they’ve downloaded your app, or bought something, or whatever action you’re optimizing for, they aren’t going to do it again. At a certain point, audience fatigue happens just because everybody who’s going to buy/downloads/etc have already done so. You’ve prompted all the actions these people are going to take. It’s time to find a new crowd.
This is why ad performance is best when users first see an ad. As the ad frequency increases, performance begins to decline. We also believe that Facebook shows ads to users most likely to convert first. So if you haven’t gotten the responses you wanted from those first ad exposures, as time goes on, you’re only less likely to get the response you want. From there, your ROAS will start to fall.
So what’s the solution? To continually test new audiences, but to do it in a systematic way so you can extend the life of your advertising creative as long as possible. Given how expensive it is to develop new creative – especially high-performing creative – the longer you can make those ads last, the better.
There are a number of ways to do this, but most of them start with a custom audience.
First, choose 7,500 customers (via device IDs, phone numbers, email addresses, etc) who took an action you want your prospects to take. This could be downloading an app, making an in-app purchase, or whatever action you want more of.
You can create a custom audience with as few as 3,000 people, but we find that 7,500 is the sweet spot. It’s possible to go up to 25,000, but again, we find that 7,500 is most efficient. And once you get below 7,500, the algorithm doesn’t perform as well.
That’s the standard way to create a basic lookalike audience. You’ve probably done it at least a hundred times.
Here’s how to make it more effective.
If we manipulate the LTV data we give Facebook’s algorithm, we can target high-value prospects more effectively. To do that, create a new custom audience with LTV (lifetime value) data. Here’s where you do that in Facebook’s ad manager:
Back to the Facebook ad manager setup: Before you upload the file, take the top 10% most valuable customers – 750 of the top payers in this example – and multiply their value by ten. For example, say the top 10% of payers had been worth 99 cents. Now we’ll tell Facebook they’re worth $9.99.
We’ll also take the bottom 10% of payers – the least valuable 750 people of this 7,500-member audience – and reduce their value by 90%. So we’ll take users that had been worth, say, 50 cents, and tell Facebook they’re now worth 5 cents. Here’s an example of how we break out audiences with this technique:
By stretching the data like this, Facebook will now be more likely to target the super-high value prospects.
This method doesn’t have to use just 90% and 10x either. It can be 75%, 5X – whatever makes sense for the audience you want to target. The idea is to manipulate the data in such a way that Facebook sees a specific set of users as super-valuable, and that it prioritizes finding more users like that.
The next thing to define is how many people we want the algorithm to find. With each new seed/custom audience, you’re introducing a new set of users into the lookalike audience. So you feed in a custom audience – 7,500 people – and you ask Facebook to give you a certain percent of the available audience. Like this:
If you choose only the 1% group, you’ll have a high-quality audience, but a smaller group of people. Having that quality is great, but it means this audience will fatigue quickly. If you want to make your creative last as long as possible, it’s worth testing the less-valuable audiences – reaching into 2%, 3%, or even 20% of the available universe of people.
If all the other settings of your campaign are lined up, and you’ve got some really excellent creative, you may be able to profitably show ads to these broader, less high-value audiences. Google App Campaigns will do this for you automatically, and Facebook will do it for you if your campaigns’ goals are set to either App Event Optimization or Value Optimization.
The algorithms will always go look for people who appear to be most appropriate for the action you’re looking for. So it’s possible that you don’t have to show ads only to the super-most-likely audiences. There could be affordable conversions to be found in broader audiences.
Using controls like this means that as you manipulate the data, you are telling Facebook or Google to look at different people based on different views of the data. It’s a really simple way to use your own data to find new audiences. Every new audience you can find means new people to show your ads to, so your expensive creative assets last longer.
But that’s only one way to manipulate the data. Here’s another technique to find new people.
Just like before, you’ll be creating custom audiences from the top 1%, 2%, 3% – up to 25% of the potential audience.
The difference here is you’ll be testing time frames. So target only the people who paid within the last day. Or people who paid within the last seven days. You can define custom audiences for people who have bought (or have taken whatever action you want more people to take) within the last seven, thirty or sixty days. Here’s how we break out audiences with this method:
The next table below shows the same recency-based audiences, but it also includes the manipulations of LTV values we talked about earlier. For these audiences, we’ve increased the value of the top buyers by 30X, and decreased the value of the bottom 10% of payers by 90%.
It’s also possible to “stack” or combine lookalike audiences. So, for instance, you move all your 2% people across multiple custom audiences into one larger custom audience group.
You can nest audiences, too. So you can pull, say, a 2% audience, then subtract your 1% people from it. That leaves you with just 2% of people. This can be useful if you want to isolate different tiers of prospects.
These lookalike audiences can also be split out by country. This often surfaces excellent prospects who might not have been identified by the algorithm before. You can use all the same techniques described here – stacking, nesting, defining by time since purchase – but at the country level.
Just be careful: For countries with 70 million people or less, don’t target a 1% audience. Instead, begin with a 3% audience. This helps you get a broader reach and helps Facebook maintain good data.
If this all seems a little complicated, you’re right. It’s complicated enough that we actually built an in-house tool called Ad Rules Audience Builder that makes this slicing and dicing of audiences much easier and more efficient.
Here’s how AdRules Audience Builder works:
1. We start by creating a custom audience.
2. Next, you specify the criteria for the audience you’d like to pull. For example, you can choose Top 1% of Purchasers, Top 5% of Purchasers, App Launches All Users, App Launches in Android or App Launches in iOS.
3. Then you choose your lookback window – or multiple lookback windows.
4. The audience creation tool will then take the number of app events/criteria and multiply that by the number of lookback windows. In this case, there were seven app events selected, multiplied by three lookback windows, resulting in 21 custom audiences.
5. Once the custom audiences are created, we can then create lookalike audiences off of those custom audiences. In the screenshot below, we’ve chosen one audience with three countries/country groups and four percentages. So we’ve now got twelve new lookalike audiences derived from that one original custom audience.
If you took those twenty-one custom audiences we created above, then added three additional countries/country groups and four percentages, you’d have 252 unique lookalike audiences. That would be a nightmare to manage manually, but with the AdRules Audience Builder, it’s all selected, labeled and managed for you.
That’s really the superpower of this tool: It lets you create a handful of strong custom audiences, then lets you turn them into hundreds, if not thousands, of lookalikes based off of those custom audiences. And it can all be done in just a few clicks, with no worries about human errors in setup.
Having an abundance of audiences like this means you can systematically move through fresh, well-targeted audiences. That, in turn, means you can sustain a piece of creative for considerably longer than if you were just doing basic lookalike audiences.