TikTok iOS - Which Conversion Metric to Trust?

[randomguy19]

Hey everyone,

We recently launched our first TikTok iOS mobile app campaign and are trying to figure out which conversion metric to trust for performance evaluation. We are using Branch as our MMP with SAN integration and PAM enabled.

For the same period we are seeing four different numbers:

- Branch (MMP): 34 trials
- TikTok Conversions: 18 trials
- TikTok Conversions (SKAN): 23 trials

From what we understand:
- TikTok Conversions uses click and view-through attribution for ATT opted-in users only
- TikTok Conversions (SKAN) comes from Apple's SKAdNetwork framework, covering all iOS users, including opted-out
- Branch captures the broadest set via probabilistic modelling and PAM but may include some over-attribution

A few questions for the group:

1. Which of these metrics do you use as your primary benchmark for evaluating TikTok iOS campaign performance and making budget decisions?

2. Do you trust TikTok Conversions (SKAN) as a reliable deterministic number?

3. How do you reconcile the gap between your MMP and TikTok's reported conversions? Do you apply an adjustment factor or use a blended approach?

Any insight from those who have been running TikTok iOS campaigns with an MMP would be really appreciated.

Thanks

 

[davamosk]

My two cents:
first - don't run campaigns with TikTok based on SKAN. Go with the probabilistic attribution.
second - try to confirm that the campaigns have some incrementality, either by looking at the blended or by looking at the app referral traffic on your app store connect insights

regarding your Qs:
1 - I check TikTok numbers, and then I check the incrementality in the baseline plus the traffic in App Store Connect.
2 - no
3 - just answered above

 

[randomguy19]

Thanks for the response.

When you say "Don't run campaigns based on SKAN. Go with probabilistic attribution." Does that mean this current setup needs to have the checkbox unchecked for "SKAN Attribution"? (see image attached)

Will it automatically use probabilistic attribution if we uncheck it?

Side note: We've confirmed with Branch that our setup is correct for PAM.

 

[davamosk]

yes exactly, the checkbox needs to be off and for that, you need to run SKAN first for a couple of days.
When you run with SKAN (checkbox ON), you dont use probabilistic so you'll deal with signal delays and so on (SKAN downsides basically)