Lana Cubric (Head of Marketing at Two Desperados) and Matej Lancaric (User Acquisition & Marketing Expert at Lancaric.me) answer questions about UA optimization and creative testing following their GameCamp webinar presentations.
You’ll always see different IPM vs. ROAS numbers, and often different trends, but Matej follows ROAS as this is what matters in the end. Matej considers IPM as a secondary metric and mostly takes IPM into consideration when thinking about creative iterations and testing.
Having D3/D7 ROAS benchmarks from past creatives helps identify creative winners quickly. Otherwise, it’s fine to follow IPMs if you know that for your app it follows the same trend as the ROAS.
Two Desperados used to try to spend as little as possible for creative testing and therefore had different campaign settings vs. Business As Usual campaigns. But now they have reverted back to the same campaign settings for testing, just a wider audience, and it works better for them (test campaigns even bring profit).
When testing new ad networks based on machine learning it’s better not to stop your test campaign because it might go against you: it’s probably better to iterate slightly and give more time to the campaign.
For other networks like Facebook or ironSource, Two Desperados usually run a 7-day ongoing test campaign. Before starting, they always define the average CPIs they’re expecting to know how much money to invest.
Two Desperados is not too rigid with their benchmarks anymore: it is a learning curve so if they’re hitting 70% of their benchmarks they continue to optimize and test.
No matter which event you’re optimizing for (e.g. purchase), you need to have checkpoints in your funnel (i.e. health metrics) to understand how your funnel is doing.
You can use memes to draw/attract the Lookalikes from other IPs. However be careful about using the exact same images as original memes. Try to use images first and move to video if you see results.
Usually, ads for games that are not monetized through IAPs perform better on the UAC inventory that has lower CPIs (AdMob), and the best creative format there tends to be portrait. Games monetizing primarily though IAPs tend to perform better on inventory with higher CPIs (YouTube), and the best creative format there tends to be landscape.
To find out what your audience likes, look at the Facebook interests of your audience (Audience Insights). Think about why they might like this and try to incorporate it in your creatives. Example: they like Family Guy -> they find this type of humor funny -> use this kind of humor in the ads.
Always look at the comments section on YouTube or Facebook to understand better what potential players actually like.
New creative people need to embrace failure and understand that most people won’t like their creatives. However what you need to be clear on is what you’re testing for: color palette, gameplay dynamics, characters vs. no characters, challenge the player or relax the player. Don’t mix too many hypotheses at the same time.
You’ll always see different IPM vs. ROAS numbers, and often different trends, but Matej follows ROAS as this is what matters in the end. Matej considers IPM as a secondary metric and mostly takes IPM into consideration when thinking about creative iterations and testing.
Having D3/D7 ROAS benchmarks from past creatives helps identify creative winners quickly. Otherwise, it’s fine to follow IPMs if you know that for your app it follows the same trend as the ROAS.
Two Desperados used to try to spend as little as possible for creative testing and therefore had different campaign settings vs. Business As Usual campaigns. But now they have reverted back to the same campaign settings for testing, just a wider audience, and it works better for them (test campaigns even bring profit).
When testing new ad networks based on machine learning it’s better not to stop your test campaign because it might go against you: it’s probably better to iterate slightly and give more time to the campaign.
For other networks like Facebook or ironSource, Two Desperados usually run a 7-day ongoing test campaign. Before starting, they always define the average CPIs they’re expecting to know how much money to invest.
Two Desperados is not too rigid with their benchmarks anymore: it is a learning curve so if they’re hitting 70% of their benchmarks they continue to optimize and test.
No matter which event you’re optimizing for (e.g. purchase), you need to have checkpoints in your funnel (i.e. health metrics) to understand how your funnel is doing.
You can use memes to draw/attract the Lookalikes from other IPs. However be careful about using the exact same images as original memes. Try to use images first and move to video if you see results.
Usually, ads for games that are not monetized through IAPs perform better on the UAC inventory that has lower CPIs (AdMob), and the best creative format there tends to be portrait. Games monetizing primarily though IAPs tend to perform better on inventory with higher CPIs (YouTube), and the best creative format there tends to be landscape.
To find out what your audience likes, look at the Facebook interests of your audience (Audience Insights). Think about why they might like this and try to incorporate it in your creatives. Example: they like Family Guy -> they find this type of humor funny -> use this kind of humor in the ads.
Always look at the comments section on YouTube or Facebook to understand better what potential players actually like.
New creative people need to embrace failure and understand that most people won’t like their creatives. However what you need to be clear on is what you’re testing for: color palette, gameplay dynamics, characters vs. no characters, challenge the player or relax the player. Don’t mix too many hypotheses at the same time.
You’ll always see different IPM vs. ROAS numbers, and often different trends, but Matej follows ROAS as this is what matters in the end. Matej considers IPM as a secondary metric and mostly takes IPM into consideration when thinking about creative iterations and testing.
Having D3/D7 ROAS benchmarks from past creatives helps identify creative winners quickly. Otherwise, it’s fine to follow IPMs if you know that for your app it follows the same trend as the ROAS.
Two Desperados used to try to spend as little as possible for creative testing and therefore had different campaign settings vs. Business As Usual campaigns. But now they have reverted back to the same campaign settings for testing, just a wider audience, and it works better for them (test campaigns even bring profit).
When testing new ad networks based on machine learning it’s better not to stop your test campaign because it might go against you: it’s probably better to iterate slightly and give more time to the campaign.
For other networks like Facebook or ironSource, Two Desperados usually run a 7-day ongoing test campaign. Before starting, they always define the average CPIs they’re expecting to know how much money to invest.
Two Desperados is not too rigid with their benchmarks anymore: it is a learning curve so if they’re hitting 70% of their benchmarks they continue to optimize and test.
No matter which event you’re optimizing for (e.g. purchase), you need to have checkpoints in your funnel (i.e. health metrics) to understand how your funnel is doing.
You can use memes to draw/attract the Lookalikes from other IPs. However be careful about using the exact same images as original memes. Try to use images first and move to video if you see results.
Usually, ads for games that are not monetized through IAPs perform better on the UAC inventory that has lower CPIs (AdMob), and the best creative format there tends to be portrait. Games monetizing primarily though IAPs tend to perform better on inventory with higher CPIs (YouTube), and the best creative format there tends to be landscape.
To find out what your audience likes, look at the Facebook interests of your audience (Audience Insights). Think about why they might like this and try to incorporate it in your creatives. Example: they like Family Guy -> they find this type of humor funny -> use this kind of humor in the ads.
Always look at the comments section on YouTube or Facebook to understand better what potential players actually like.
New creative people need to embrace failure and understand that most people won’t like their creatives. However what you need to be clear on is what you’re testing for: color palette, gameplay dynamics, characters vs. no characters, challenge the player or relax the player. Don’t mix too many hypotheses at the same time.
Notes for this resource are currently being transferred and will be available soon.
[💎@1:15:40] You’ll always see different IPM vs. ROAS numbers, and often different trends, but Matej follows ROAS as this is what matters in the end. Matej considers IPM as a secondary metric and mostly takes IPM into consideration when thinking about creative iterations and testing.
Fake creatives can drive IPM up and CPIs down but then the retention sucks.
Matej
[💎@1:17:14] Having D3/D7 ROAS benchmarks from past creatives helps identify creative winners quickly. Otherwise, it’s fine to follow IPMs if you know that for your app it follows the same trend as the ROAS.
Lana
Two Desperado’s lifecycle is quite long but they also have benchmarks that show early indicators that allow them to find out how a creative performs within the first 3-7 days.
[💎@1:18:15] Two Desperados used to try to spend as little as possible for creative testing and therefore had different campaign settings vs. Business As Usual campaigns. But now they have reverted back to the same campaign settings for testing, just a wider audience, and it works better for them (test campaigns even bring profit).
Matej
Spending $400 is typically enough for their game to get the installs needed to understand ROAS. If you don’t have that, then yes you look at proxy metrics.
Lana
All channels are different so there isn't one single approach.
[💎@1:21:25] When testing new ad networks based on machine learning it’s better not to stop your test campaign because it might go against you: it’s probably better to iterate slightly and give more time to the campaign.
[💎@1:21:54] For other networks like Facebook or ironSource, Two Desperados usually run a 7-day ongoing test campaign. Before starting, they always define the average CPIs they’re expecting to know how much money to invest.
With new networks they usually stop the campaigns to check how the cohort’s LTV are developing.
[💎@1:22:32] Two Desperados is not too rigid with their benchmarks anymore: it is a learning curve so if they’re hitting 70% of their benchmarks they continue to optimize and test.
To select a network, ask your account managers what kind of players and inventory/placements they should expect to make sure that it’s a good fit.
Give yourself some time to understand what’s successful or not.
Matej
Always tries to get to a ROAS campaign on ad networks, as soon as he can. There is a learning phase where you’re optimizing on CPI but it usually doesn’t work very well.
For ROAS campaigns he tries to look at the early results, even though it’s always going to evolve. If there is a trend of increasing ROAS he spends more to get more data.
Lana
Have never included k factor into any reporting data for campaigns, but they do see a strong correlation between scaling UA and seeing organics grow. With iOS 14.5 they will have to take a different approach: look at the investment, look at the returns.
Matej
For his games Matej is looking at the blended ROAS with a payback expected on D30.
Lana
Not as aggressive with their expected lookback window as Matej, because they’re looking to grow/expand.
Matej
Always look at the funnel: onboarding flow, tutorial but also what happens afterwards (e.g. levels completion rate).
For non-gaming, Matej used the free trial event combined with an action in the onboarding (e.g. fitness test).
For gaming, there are predictive events in Firebase that he has tested but so far no success.
Lana
[💎@1:30:45] No matter which event you’re optimizing for (e.g. purchase), you need to have checkpoints in your funnel (i.e. health metrics) to understand how your funnel is doing.
They tried proxy events happening before purchases but it was too much effort and not enough results, so now optimizing for purchases. If launching a new game they optimize for retention (e.g. reaching level X).
It was however beneficial to research these proxy events beyond acquisition, but the challenge is that there isn’t necessarily a return on investment from that.
Matej
[💎@1:33:24] You can use memes to draw/attract the Lookalikes from other IPs. However be careful about using the exact same images as original memes. Try to use images first and move to video if you see results.
Mladen
[💎@1:35:15] Usually, ads for games that are not monetized through IAPs perform better on the UAC inventory that has lower CPIs (AdMob), and the best creative format there tends to be portrait. Games monetizing primarily though IAPs tend to perform better on inventory with higher CPIs (YouTube), and the best creative format there tends to be landscape.
Matej
Be mindful that the best practices you’re hearing might not work for everyone. You need to test it out.
Lana
Test, don’t assume anything because it might not make sense for your app.
Two Desperados mostly creates 30 second videos to get into AdMob’s inventory because YouTube is not working for them. So they’re always waiting for that display inventory to pick up.
Matej
On Facebook, Matej is using custom placements with different sizes. In the testing campaign he uses single creatives, in the BAU he uses the dynamic creative sets.
Lana
Tried dynamic testing but it did not improve the performance significantly. So now they test with single creatives and in square (largest inventory on Facebook). For ironSource and Google they also do it in portrait mode and adapt to the platform.
Matej
[💎@1:39:18] To find out what your audience likes, look at the Facebook interests of your audience (Audience Insights). Think about why they might like this and try to incorporate it in your creatives. Example: they like Family Guy -> they find this type of humor funny -> use this kind of humor in the ads.
[💎@1:40:45] Always look at the comments section on YouTube or Facebook to understand better what potential players actually like.
Lana
The community gives them a lot of ideas, especially about what kind of emotions the creatives and the game are triggering.
Matej
It depends how well you know your audience. The more you learn about your audience, the better your ratio will increase. Getting to 50/50 would already be an achievement.
Lana
[💎@1:43:30] New creative people need to embrace failure and understand that most people won’t like their creatives. However what you need to be clear on is what you’re testing for: color palette, gameplay dynamics, characters vs. no characters, challenge the player or relax the player. Don’t mix too many hypotheses at the same time.
[💎@1:15:40] You’ll always see different IPM vs. ROAS numbers, and often different trends, but Matej follows ROAS as this is what matters in the end. Matej considers IPM as a secondary metric and mostly takes IPM into consideration when thinking about creative iterations and testing.
Fake creatives can drive IPM up and CPIs down but then the retention sucks.
Matej
[💎@1:17:14] Having D3/D7 ROAS benchmarks from past creatives helps identify creative winners quickly. Otherwise, it’s fine to follow IPMs if you know that for your app it follows the same trend as the ROAS.
Lana
Two Desperado’s lifecycle is quite long but they also have benchmarks that show early indicators that allow them to find out how a creative performs within the first 3-7 days.
[💎@1:18:15] Two Desperados used to try to spend as little as possible for creative testing and therefore had different campaign settings vs. Business As Usual campaigns. But now they have reverted back to the same campaign settings for testing, just a wider audience, and it works better for them (test campaigns even bring profit).
Matej
Spending $400 is typically enough for their game to get the installs needed to understand ROAS. If you don’t have that, then yes you look at proxy metrics.
Lana
All channels are different so there isn't one single approach.
[💎@1:21:25] When testing new ad networks based on machine learning it’s better not to stop your test campaign because it might go against you: it’s probably better to iterate slightly and give more time to the campaign.
[💎@1:21:54] For other networks like Facebook or ironSource, Two Desperados usually run a 7-day ongoing test campaign. Before starting, they always define the average CPIs they’re expecting to know how much money to invest.
With new networks they usually stop the campaigns to check how the cohort’s LTV are developing.
[💎@1:22:32] Two Desperados is not too rigid with their benchmarks anymore: it is a learning curve so if they’re hitting 70% of their benchmarks they continue to optimize and test.
To select a network, ask your account managers what kind of players and inventory/placements they should expect to make sure that it’s a good fit.
Give yourself some time to understand what’s successful or not.
Matej
Always tries to get to a ROAS campaign on ad networks, as soon as he can. There is a learning phase where you’re optimizing on CPI but it usually doesn’t work very well.
For ROAS campaigns he tries to look at the early results, even though it’s always going to evolve. If there is a trend of increasing ROAS he spends more to get more data.
Lana
Have never included k factor into any reporting data for campaigns, but they do see a strong correlation between scaling UA and seeing organics grow. With iOS 14.5 they will have to take a different approach: look at the investment, look at the returns.
Matej
For his games Matej is looking at the blended ROAS with a payback expected on D30.
Lana
Not as aggressive with their expected lookback window as Matej, because they’re looking to grow/expand.
Matej
Always look at the funnel: onboarding flow, tutorial but also what happens afterwards (e.g. levels completion rate).
For non-gaming, Matej used the free trial event combined with an action in the onboarding (e.g. fitness test).
For gaming, there are predictive events in Firebase that he has tested but so far no success.
Lana
[💎@1:30:45] No matter which event you’re optimizing for (e.g. purchase), you need to have checkpoints in your funnel (i.e. health metrics) to understand how your funnel is doing.
They tried proxy events happening before purchases but it was too much effort and not enough results, so now optimizing for purchases. If launching a new game they optimize for retention (e.g. reaching level X).
It was however beneficial to research these proxy events beyond acquisition, but the challenge is that there isn’t necessarily a return on investment from that.
Matej
[💎@1:33:24] You can use memes to draw/attract the Lookalikes from other IPs. However be careful about using the exact same images as original memes. Try to use images first and move to video if you see results.
Mladen
[💎@1:35:15] Usually, ads for games that are not monetized through IAPs perform better on the UAC inventory that has lower CPIs (AdMob), and the best creative format there tends to be portrait. Games monetizing primarily though IAPs tend to perform better on inventory with higher CPIs (YouTube), and the best creative format there tends to be landscape.
Matej
Be mindful that the best practices you’re hearing might not work for everyone. You need to test it out.
Lana
Test, don’t assume anything because it might not make sense for your app.
Two Desperados mostly creates 30 second videos to get into AdMob’s inventory because YouTube is not working for them. So they’re always waiting for that display inventory to pick up.
Matej
On Facebook, Matej is using custom placements with different sizes. In the testing campaign he uses single creatives, in the BAU he uses the dynamic creative sets.
Lana
Tried dynamic testing but it did not improve the performance significantly. So now they test with single creatives and in square (largest inventory on Facebook). For ironSource and Google they also do it in portrait mode and adapt to the platform.
Matej
[💎@1:39:18] To find out what your audience likes, look at the Facebook interests of your audience (Audience Insights). Think about why they might like this and try to incorporate it in your creatives. Example: they like Family Guy -> they find this type of humor funny -> use this kind of humor in the ads.
[💎@1:40:45] Always look at the comments section on YouTube or Facebook to understand better what potential players actually like.
Lana
The community gives them a lot of ideas, especially about what kind of emotions the creatives and the game are triggering.
Matej
It depends how well you know your audience. The more you learn about your audience, the better your ratio will increase. Getting to 50/50 would already be an achievement.
Lana
[💎@1:43:30] New creative people need to embrace failure and understand that most people won’t like their creatives. However what you need to be clear on is what you’re testing for: color palette, gameplay dynamics, characters vs. no characters, challenge the player or relax the player. Don’t mix too many hypotheses at the same time.
[💎@1:15:40] You’ll always see different IPM vs. ROAS numbers, and often different trends, but Matej follows ROAS as this is what matters in the end. Matej considers IPM as a secondary metric and mostly takes IPM into consideration when thinking about creative iterations and testing.
Fake creatives can drive IPM up and CPIs down but then the retention sucks.
Matej
[💎@1:17:14] Having D3/D7 ROAS benchmarks from past creatives helps identify creative winners quickly. Otherwise, it’s fine to follow IPMs if you know that for your app it follows the same trend as the ROAS.
Lana
Two Desperado’s lifecycle is quite long but they also have benchmarks that show early indicators that allow them to find out how a creative performs within the first 3-7 days.
[💎@1:18:15] Two Desperados used to try to spend as little as possible for creative testing and therefore had different campaign settings vs. Business As Usual campaigns. But now they have reverted back to the same campaign settings for testing, just a wider audience, and it works better for them (test campaigns even bring profit).
Matej
Spending $400 is typically enough for their game to get the installs needed to understand ROAS. If you don’t have that, then yes you look at proxy metrics.
Lana
All channels are different so there isn't one single approach.
[💎@1:21:25] When testing new ad networks based on machine learning it’s better not to stop your test campaign because it might go against you: it’s probably better to iterate slightly and give more time to the campaign.
[💎@1:21:54] For other networks like Facebook or ironSource, Two Desperados usually run a 7-day ongoing test campaign. Before starting, they always define the average CPIs they’re expecting to know how much money to invest.
With new networks they usually stop the campaigns to check how the cohort’s LTV are developing.
[💎@1:22:32] Two Desperados is not too rigid with their benchmarks anymore: it is a learning curve so if they’re hitting 70% of their benchmarks they continue to optimize and test.
To select a network, ask your account managers what kind of players and inventory/placements they should expect to make sure that it’s a good fit.
Give yourself some time to understand what’s successful or not.
Matej
Always tries to get to a ROAS campaign on ad networks, as soon as he can. There is a learning phase where you’re optimizing on CPI but it usually doesn’t work very well.
For ROAS campaigns he tries to look at the early results, even though it’s always going to evolve. If there is a trend of increasing ROAS he spends more to get more data.
Lana
Have never included k factor into any reporting data for campaigns, but they do see a strong correlation between scaling UA and seeing organics grow. With iOS 14.5 they will have to take a different approach: look at the investment, look at the returns.
Matej
For his games Matej is looking at the blended ROAS with a payback expected on D30.
Lana
Not as aggressive with their expected lookback window as Matej, because they’re looking to grow/expand.
Matej
Always look at the funnel: onboarding flow, tutorial but also what happens afterwards (e.g. levels completion rate).
For non-gaming, Matej used the free trial event combined with an action in the onboarding (e.g. fitness test).
For gaming, there are predictive events in Firebase that he has tested but so far no success.
Lana
[💎@1:30:45] No matter which event you’re optimizing for (e.g. purchase), you need to have checkpoints in your funnel (i.e. health metrics) to understand how your funnel is doing.
They tried proxy events happening before purchases but it was too much effort and not enough results, so now optimizing for purchases. If launching a new game they optimize for retention (e.g. reaching level X).
It was however beneficial to research these proxy events beyond acquisition, but the challenge is that there isn’t necessarily a return on investment from that.
Matej
[💎@1:33:24] You can use memes to draw/attract the Lookalikes from other IPs. However be careful about using the exact same images as original memes. Try to use images first and move to video if you see results.
Mladen
[💎@1:35:15] Usually, ads for games that are not monetized through IAPs perform better on the UAC inventory that has lower CPIs (AdMob), and the best creative format there tends to be portrait. Games monetizing primarily though IAPs tend to perform better on inventory with higher CPIs (YouTube), and the best creative format there tends to be landscape.
Matej
Be mindful that the best practices you’re hearing might not work for everyone. You need to test it out.
Lana
Test, don’t assume anything because it might not make sense for your app.
Two Desperados mostly creates 30 second videos to get into AdMob’s inventory because YouTube is not working for them. So they’re always waiting for that display inventory to pick up.
Matej
On Facebook, Matej is using custom placements with different sizes. In the testing campaign he uses single creatives, in the BAU he uses the dynamic creative sets.
Lana
Tried dynamic testing but it did not improve the performance significantly. So now they test with single creatives and in square (largest inventory on Facebook). For ironSource and Google they also do it in portrait mode and adapt to the platform.
Matej
[💎@1:39:18] To find out what your audience likes, look at the Facebook interests of your audience (Audience Insights). Think about why they might like this and try to incorporate it in your creatives. Example: they like Family Guy -> they find this type of humor funny -> use this kind of humor in the ads.
[💎@1:40:45] Always look at the comments section on YouTube or Facebook to understand better what potential players actually like.
Lana
The community gives them a lot of ideas, especially about what kind of emotions the creatives and the game are triggering.
Matej
It depends how well you know your audience. The more you learn about your audience, the better your ratio will increase. Getting to 50/50 would already be an achievement.
Lana
[💎@1:43:30] New creative people need to embrace failure and understand that most people won’t like their creatives. However what you need to be clear on is what you’re testing for: color palette, gameplay dynamics, characters vs. no characters, challenge the player or relax the player. Don’t mix too many hypotheses at the same time.
