Testing and Optimizing Ad Creatives in Google UACs
Alessandra Miuccio is a Performance Marketing Lead at Glovo, optimizing campaign performance and developing user acquisition strategies. Before Glovo, Alessandra managed lead generation, install and conversion campaigns at JustEat, BurgerKing and CornerJob. Fascinated in understanding mobile device usage and the thrilling challenge of acquiring new app users, Alessandra currently steers user growth at the fast-growing delivery app in Europe.
Learn more about Mobile Hero Alessandra.
Nowadays, most app marketing channels apply machine learning to optimize performance campaigns. Machine learning helps automate the necessary but lengthy process of analyzing user data and setting the best target and bids for your campaign.
Google’s Universal App Campaign (UAC) enables marketers to do just that—but with few elements to control such as locations, device language and campaign event optimization.
However, with UAC’s limitations, it can be tough navigating ad testing. We cannot target different audiences, evenly split them, or implement proper A/B testing to define whether a new creative outperforms an existing one. So, I pose the question, “how can you master ad testing in UAC?” Creatives in UAC can be text ads (headlines with a max of 30 characters and descriptions 90), images, videos and HTML5.
As an app marketer, remember to listen to your users and apply their insights in your ads strategy. This process should not be static but happening continuously; it is a cycle we all know called testing and learning. When done correctly, it guarantees success. Know how the audience interacts with your ads to help decrease user acquisition costs.
After several years of creating and optimizing in UACs, I always advise marketers to dig into ad creatives to drive results and user growth. In this blog, I share four recommendations and tips for testing ad creatives in Google UACs.
1. Establish KPIs and benchmarks
Before we start testing, it is important to set quantifiable goals to help us track and measure an ad’s success. These are key performance indicators (KPIs) and benchmarks—the point of reference against which results may be compared.
For comparing performance among different creatives, use the conversion rate between impressions to installs or impressions to in-app events. A helpful second KPI is the absolute number of the event (installs, purchases) that each ad brings.
To have reliable results, make sure you reach a minimum number of conversions and that enough time passes to accumulate significant volume. I recommend waiting at least fourteen days to consider the weekly seasonality and ease out from the ads’ learning phase.
Here are the formulas and conversions benchmarks:
- Mobile App Install (MAI) campaigns:
Conversion Rate (CVR) = (Installs/Impressions) * 100
The benchmark must reach a minimum of 150 installs in each variant. - App Event Optimization (AEO) campaigns:
CVR = (Number of in-app event/Number of impressions) * 1000If your goal is an in-app event, multiply for 1000 instead of 100 to make the results legible. In most impression to in-app event conversion rates, the CVR is below 1, which makes it difficult to present or debate on results.The benchmark will depend on your type of app, monetization model and campaign budget.
2. Use ad groups to test messaging
Let’s say there is a booking app, and you have three different selling propositions to tryout: Free 24-Hour Cancellation / Best Price Guarantee / Stay for 3, Pay for 2. In this scenario, ad groups would be the perfect solution. As you input text in the ads, this improves the UAC algorithm’s targeting options and helps segment your audience.
Create an ad group for each proposition in your current campaign. In each ad group, develop the selling proposition in different text lengths and complete the five headlines and the descriptions available.
Since UACs perform better when all assets are complete, develop one template for images, video and HTML5 ads. The composition must be the same for all three ad groups, and the only thing changing is the text accompanying it. Be sure to include the best-performing creative dimensions: for instance, 1200×628 for banners and 1920×1080 for video ads.
Now you are ready to launch three ad groups on the same date. After reaching your benchmark in 1-2 weeks, define the winning message by identifying which ad group had the higher impression-to-conversion rate and brought the highest number of conversions.
3. Use market-based campaigns to test visuals
Imagine you want to rebrand a sub-product inside your app. For example, a language learning app where you want to change the program’s visual for kids: it’s always been blue and now you want to test a red background.
The best way to rebrand is by running an A/B test at the user-level. Although this process is slow and expensive, it provides reliability and statistical significance in your results. However, as we described above, this isn’t an option in UACs.
When it is impossible to randomize at the individual consumer level, an alternative solution is to randomize the audience at the geographic market level. Split your audience into little units based on location—this assignment can be randomized (also known as geo-test or geo experiments, Vaver and Koehler 2011, Kerman et al., 2017) or carefully selected based on a combination of demographics, population density, and purchase behavior (also known as a matched market test, Gordon et al., 2016, Tim Au, 2018).
This option is only valid for:
- Advertisers promoting their app at a broader geo level, such as countries or regions. Those at postal code levels should not use this approach because of the risk of ad serving inaccuracy and the possibility of consumers commuting across geo boundaries.
- Upper funnel outcomes, like clicks or installs. Avoid lower-funnel conversion outcomes such as purchases. The lower we go, the results of the test are less reliable.
Once you have geo lists, create the control and test group campaigns. Following our language learning app example, the control group contains the current blue assets, and the test group has red assets.
Make sure to follow these best practices:
- Pick an event different from the current campaign to avoid audience overlap during campaign setup.
- Keep the same headlines and descriptions in both control and test campaigns. As explained, text ads are used to target different audiences. To maintain an equal baseline in both campaigns, we use the same copies.
- Maintain visual consistency across all assets in each campaign. In the control group, only use the blue background images and videos; in the test group, use only the red.
In the first 3 to 5 days, monitor the spend and target CPAs to ensure you reach the daily budget and that the investment is equal in each campaign. The winning campaign brings the highest absolute number of installs.
4. Upload ads to your live campaign / ad group and test video or image variants
Uploading ads in your current campaign is the most straightforward experiment to implement and probably the fastest to get quick and precise insights. You don’t segment the audience at all: simply make the new assets compete in the same campaign or ad group as the old ones. The risk here is audience overlap, but remember, there is no way to control the order or hierarchy of user exposure of an ad to another in UAC campaigns. That is why this option is considered the fairest since the test will happen under “business-as-usual”’ conditions.
There are two ways of measuring an ad’s success, depending on the number of creatives you want to test.
a. Having ≥ 2 new video or image variants
Let’s talk with more examples. You own a fashion e-commerce app, and it is sales season. You are looking to test the release of a “best deal” message in your video’s first three seconds versus showcasing the products and closing it with the promo message.
If you upload an image today and a variant one week later, you won’t be able to compare results against each other–one will have an additional week of historical data. When there are two or more variants, each variant must enter the learning phase simultaneously to compare results later. Define the winner with the absolute number of conversions.
b. Having only one new video or image
Now imagine you own a mindfulness app and that you have three different ad groups: one focusing on sleep, another for stress, and a generic one. After detecting missing content on the second ad group, you develop a brand new video showing users’ testimonials. So, how does this new video perform against the old ones?
The KPI for comparing a single asset against the existing videos is the impression to conversion rate. Compare the brand new asset’s performance against the ad group and campaign’s average rate during the same time frame in which the new creative was live.
Include both dimensions as benchmarks to detect opportunities or challenges. For instance, imagine the stress testimonial video reaches an impression to install rate of 12%, but the stress ad group is currently at 15% and the campaign 10%. This tells us that other videos bring greater results in the ad group, but overall the campaign performs better than the average. So your challenge: the testimonial video might not convey the best message to users dealing with stress. And your opportunity: the new asset stands out at the campaign level, so it might be a good idea to create a new ad group based on testimonials and test new variations.
In an environment where measurement efforts have proven challenging, a campaign’s success comes from having a clear and understandable process for testing and learning. Once you start implementing these tips, your UAC performance will improve. Your end goal is to find the most accurate way of measuring ad effectiveness for optimizing your marketing strategies in this network.
That wraps up Alessandra’s tips on creative testing on Google UAC. For more from our Mobile Heroes, check out our latest blogs or join the Mobile Heroes Slack Community and chat with over 2,000+ mobile marketers.