+3 votes
We usually use IDs from the payers only and make lookalikes on them 3-5% based on the app market or country.

Which other option do you consider to use, test and try?
by (320 points)

4 Answers

+2 votes
We usually do certain engagement of a player in the game (except payers). Take a look at these examples, maybe some of them will help you:

PAYERS:
- top 250 payers (in certain month)
- 2+ special offers bought (in last XY days/months)
- XY soft currency bought (in last XY days/months)
- whales
- engaged payers - players with 10/20/30+ logins in last XY days & made a purchase
- top 10% purchasers (in last XY days/months)
- single pay over xy UDS (in last XY days/months)  - single payment over 20, 30, xy USD

ENGAGED:
- XY+ quests achieved in the game (in last XY days/months)
- XY PVP matches played
- engag with high LVL - e.g. all players with level 100
- XYZ+ logins in 1 day - (in last XY days/months)

Also, we use 1, 3, 5 or sometimes even 10% lookalikes. Definitely worth trying, since different lookalikes % tends to work for me in different games.
by (320 points)
+1 vote

Here are the top non-obvious lookalike seed audiences that we've used that come to mind:

  • Repeat purchasers, whales(top 10% of purchasers) - and other purchase-based lookalikes;
  • Most active users;
  • Engaged users (reached level X);
  • Consumed content relevant to theme X (eg - we're working with an esports app that lets users engage with a ton of games within this app - and we target lookalikes of users who have engaged with specific games or made purchases after playing these games. ie COD lookalikes, Clash of Clans lookalikes etc.);
  • Completed other pre-purchase event(for a lifestyle app we work with, users have to make a booking before they made a transaction -> so we built lookalikes off of 'made booking'. likewise for subscription apps this could be 'signed up for free trial');
  • Video watchers(we found this especially effective when we targeted rewarded video placement only in a campaign).
by (1.5k points)
+1 vote
Lookalikes that are more unique to your 1P data but FB can still pick up the signals clearly, such as attractiveness of users.

Also like to use VBLALs off of intent, such as inverse time to conversion.
by (160 points)
0 votes

If you think about Lookalike audiences as a manual implementation of AEO, the advantage of LaLs becomes pretty obvious: you get far more flexibility in defining what gets optimized for with LaLs than with AEO, because AEO events are pre-determined by FB (and custom events tend to not perform nearly as well for optimization as the AEO catalogue events do).

The reason LaLs are powerful when implemented with the same intention as AEO (ie. finding users likely to trigger some event) is that you can use increasingly down-funnel events as you scale your app and collect purchasers. The "cold start" problem with AEO is that without any historical data on purchasers -- ie. when the app is new and hasn't seen many purchasers yet -- Facebook's algorithm doesn't know what to look for in finding purchasers and very broadly experiments with ad exposure. If you have some down-funnel events that you know proxy well for making a purchase, you can pre-empt AEO by running MAI campaigns against engagement-based LaL lists that correspond to better quality traffic.

And because LaL list definitions are completely determined by the advertiser, these "events" can be much more flexible than with AEO (and don't need to be discrete events at all). An advertiser might pull a list of users who:

  • retained on Day 1 AND 2;
  • finished the tutorial in under X minutes;
  • joined a guild within X days of installing the game;
  • did some action X number of times within 3 days of installing the app;
  • etc.

Obviously, the disadvantage of LaL workflow relative to AEO is that it is manual and requires uploading a new custom audience for each new seed list, creating a LaL against that custom audience, and updating an existing campaign or creating a new one. This can be done via API but when done manually, it's tedious.

by (8.2k points)