Ad Networks Explained Part II: Best practices for user acquisition

Can Sölömbaz
Can Sölömbaz
Data Scientist
Ad Networks Explained Part II: Best practices for user acquisition

In this second post of the blog series Ad Networks Explained, we will give hints about how the ad networks work so that your user acquisition (UA) teams can get the most value from their marketing campaigns.

In our first post of the series, we clustered the ad networks in three groups to provide a high level understanding of how they work. However, each network has its own black-box machine learning (ML) models to target audiences and reach the set CPI objectives. They have different learning phases, targeting options and data needs to optimize the performance of a campaign. In this post, we will walk you through each group of ad networks to highlight the differences and provide best practices so that UA professionals can maximize the return on their efforts. 

Before diving deep into the ad networks, let’s define our decision variables and their characteristics.

Decision variables and parameters: LTV, Bid, Budget

Ultimately, user acquisition efforts have one main goal, which is acquiring profitable customers. When the lifetime value (LTV) of a customer is measured, the bids and budgets can be adjusted to reach the desired outcome. So, we can say that when you’re making a change on your user acquisition campaigns, there are three significant variables you need to consider: LTV, bid, and budget; with LTV being the leading variable.

LTV Prediction

Before deciding on increasing your campaign budget or changing your bid, the most critical parameter to look at is your users’ LTV. Accurately predicting the LTV will make your business decisions easier since you will know the range of profitability and sustainability. The LTV prediction is a vast topic, and to be honest, it needs a separate blog post to dig into its intricacies, but we can point out some critical aspects as a start.

1. Do not use fixed period ARPU as an LTV

Accepting a 7-day ARPU as your LTV in all ad groups will seriously prohibit the campaign performance. Users coming from different user acquisition networks will have different retention rates. Also, when you extend your retention horizon as your LTV horizon, you can significantly bid higher values and scale your campaigns.

2. Be careful about attribution 

The past revenue data is completed after 2-3 days regardless of the platform. For iOS 14.5+ campaigns, eliminating the last three days of attributed data is suggested. The elimination of missing data or completion by estimation is critical in LTV predictions.

3. Distribution of organic installs

Including organic installs in calculations increases the LTV and bids dramatically, which can be crucial for scaling your marketing campaigns. Adopting this strategy enabled one of our customers to reach #1 on the Top Chart.

LTV prediction requires top-notch machine learning know-how and automation capabilities that you can have by using UAhero.

Bid & budget variables

After predicting the LTV of an ad group, UA specialists need to decide on the bid and budget variables. Bids are the customer acquisition cost for a specific customer segment. But your actual cost per install (CPI) or cost per action (CPA) can be different from your bid. It is dependent on the learning phase and the estimation capabilities of ad networks’ black-box algorithms. Budget is also an important decision variable to keep your costs under control and eliminate unexpected results. The daily budget of your campaigns should be sufficient for your campaigns to show performance with the given bid amount. 

Network group-based best practices

Now, let’s get into the best practices for ad networks. In our previous post on the topic, we classified ad networks into three groups.

I – The ad networks which are not social networks: AppLovin, IronSource, Unity Ads, Vungle…

II – The ad networks which are also social networks: Facebook, Snapchat, TikTok…

III – The ad networks that can’t be specified as either: Google Ads, Apple Search Ads…

I - Non-social ad networks

The networks in this ad network group have similar characteristics regarding the learning phase, targeting options, and bidding strategy. The practices that we’ll share are proven to be effective on all of them.


Targeting options on these ad networks can be divided into country, publisher (source app), and device.

The first guidance we can provide on targeting is not grouping countries. Users of countries that have a similar economic level can have different LTVs and CPIs. Additionally, these networks have capabilities to perform in a fragmented campaign structure. Individually bidding for each country will increase your return on ad spend (ROAS).

Secondly, go granular on publisher and device level. The users coming from a puzzle game will have a significantly different retention horizon than a hyper-casual game. Moreover, an iPhone 13 user can be more valuable than an iPhone 7 user for your game. 

Keep in mind that having sufficient data is crucial to calculate granular level LTV accurately. A Bayesian approach helps in this situation. To explain it with an example; you can build a prior distribution for a country-publisher segment based on your other games, campaigns, or historical data of the campaign. Then you can update your prediction with fresh data day by day.

Lastly, a search algorithm should explore opportunities. The performance of a country or publisher can diminish in time. Increasing bids on your historically successful targets can elevate the performance since the market structure or competition changes over time.

This bit of information is also relevant for other ad networks, whether it’s a social network or non-social.

Learning phase

A commonly used tactic when launching a campaign is starting with a higher bid than your estimated LTV. For instance, in Unity Ads starting with two times of your LTV helps the algorithm to explore promising segments. After data collection, making new LTV predictions and adjusting bids accordingly in a couple of days is preferred. This practice can be applied to all non-social ad networks, but you might need to wait for 5-6 days for some networks such as ironSource to perform. 

Bid and budget strategy

The usual bid strategy in the non-social networks is the target CPI. The target CPI is given to the algorithms to receive as the actual CPI. However, especially on iOS 14.5+ campaigns, which at the time of writing this post is more than %80 of Apple devices, there is a fluctuation on actual CPI due to loss of attribution and estimation capabilities. In this case, examining CPI for a longer period (weekly or even monthly) would be sensible. Even though black-box models cannot reach the target CPI daily, you can see a convergence in a monthly average CPI.

Since the target CPI in these networks control your costs, enter a high budget to make sure that the performance is not limited. 

II - The social networks

Social networks such as Facebook, Snapchat, and TikTok have similar targeting options, but their best practices may vary slightly.


When advertising in these ad networks, there are several optimization goals to consider for targeting. Users can be targeted mainly in 3 objective functions: Installs, event (purchase or retention), and value (target ROAS). 

UA teams simultaneously manage campaigns with different goals, which helps ad network algorithms to converge different user segments in their social graphs. 

In the campaign launching phase, it is crucial to start with a broad audience so that algorithms can try different user segments to understand which players are more valuable for your business. 

When the audience gets saturated, which inevitably happens over time, and performance decreases, you can try the lookalike audience (LAL). The lookalike audience, which is most likely to perform as your seed audience, basically narrows the targeted population in the ad network. Ideally, the seed audience would be the top 5000 existing users sorted by revenue. In the first step, you can narrow the population down to 10% LAL, which is a broader lookalike audience. 

To avoid the poor performance of audience saturation, you can try reducing the targeting to 1% of the audience, which is likely to perform as your seed audience. Also, when the campaigns are not performing adequately on detailed targeting, you can try “Automated App Ads” on Facebook or demographic segmentation in other ad networks. Using the Automated App Ads (AAA) might deliver better results, especially on iOS campaigns since its delivery models are improved for app install, app event, and value optimization.

Learning phase

It often takes more time for algorithms to explore and improve their estimations in the learning phase of social ad networks. 

For instance, it is expected for a Facebook campaign to take at least two weeks to reach its full performance. So, being patient is a crucial step in these ad networks.

Bid and budget strategy

For the install and event campaigns, using a lowest-cost bid strategy at the start is suggested. Lowest-cost strategy allows algorithms to attend more auctions and search for more users in the beginning. When performance deteriorates, using the bid cap option prevents the high CPI in user acquisition. For the value campaigns (target ROAS), starting with a 100% target ROAS and gradually increasing it usually brings the best results.

When using a bid cap, keeping the spending limit high is suggested to avoid limiting performance in these ad networks except for TikTok. Ads on TikTok perform better when the ad sets are duplicated. Therefore to increase or maintain the performance of your campaign on TikTok, it’s a better idea to duplicate the ad sets rather than increasing the spending limit while using a bid cap. 

III - Other big players

Google Ads

Google has similar bid and budget options as social networks however, targeting options are more limited. Google simplifies the process for UA professionals by only giving geo-targeting options and it optimizes the performance on its own. The best practice for Google is not changing bids and budgets frequently and more than %20. Otherwise, Google algorithms go back into the learning phase, which limits the performance.

Apple Search Ads

Apple Search Ads is a relatively new network, however, with the deprecation of LAT, it is gaining popularity and becoming a leading platform for mobile game advertising. Apple Search Ads combined with the Apple Store Optimization (ASO) help you gain users through the keywords you use for your campaigns. The relevancy of your keywords and the bid amount you selected are crucial factors in winning auctions apart from ASO. There are two modes for campaign creation on ASA; basic and advanced. To always be in control, using the advanced campaign setup is suggested. In order to have a successful Apple Search Ads campaign, the first thing to do is to optimize your Apple Store presence, as ASA and ASO work together and ASA success relies massively on ASO. In addition to ASO, you need to choose relevant keywords. Apple gives you a recommended bid amount and an indication of its strength. Apple's baseline suggestions can be used to start with and then adjusted according to the campaign performance.

In this part of the blog series we talked about LTV, bid, budget, and targeting methods in different ad networks. 

To sum up;

  • The first step of the user acquisition funnel is accurately predicting the LTV of your players. 
  • Secondly, going granular in all the ad networks is key to scaling performance.
  • Lastly, using the best practices in each ad network and continuously adjusting campaign bids and budgets are critical to reaching optimal.

We understand that maintaining several campaigns for each game is time-consuming. It can even affect the decisions you make negatively since it’s confusing to juggle between platforms to make meaningful inferences from the data. UAhero, the mobile marketing platform, is here to automate all the efforts and fine-tune decisions for each ad network. The AI-powered algorithms analyze your user acquisition campaigns and offer LTV predictions to save you precious time spent on calculations and leave you only the decision-making and application of the new variables through a few clicks on the platform. Our algorithms optimize all your campaigns across ad networks, while you still maintain control by deciding on which recommendations get implemented. To benefit from UAhero’s features and schedule a demo, contact us at or visit

Stay updated
Get the latest news on UAhero and the mobile marketing world.
Thank you! Check your inbox to confirm your submission.
Oops! Something went wrong.
Your privacy is important to us, see how we handle your data in our Privacy Policy.