2 min read
In brief
- Server events improve TikTok optimisation when browser coverage is genuinely poor — in-app browsers, iOS restrictions, ad blockers — not by default.
- The measurable gain is event volume recovered, so compare matched events per hundred orders before and after.
- As with Meta, mismatched event ids turn an improvement into double counting.
The conditions that make it worth building
- A large share of traffic arrives inside the TikTok in-app browser, where third-party cookie behaviour is least predictable.
- Purchase events are missing at a rate you can measure — count pixel purchases against orders for a week.
- You already run a server-side service for another platform, so the marginal work is a new endpoint rather than new infrastructure.
- Your audience skews toward regions or devices where browser tracking coverage is weakest.
If none of those apply, the pixel is probably reporting what it needs to. Building server-side infrastructure to fix a problem you have not measured is how tracking projects become permanent.
How to tell whether it changed anything
- 1
Establish the baseline
For one week, count TikTok-attributed purchases against store orders and compute matched events per hundred orders.
- 2
Ship the server stream for purchase only
One event, same id as the browser copy. Do not add five events at once; you will not know which one helped.
- 3
Re-measure for a week
Same metric, same traffic mix. An improvement shows as more matched purchases against unchanged order volume.
- 4
Check for duplication
If reported purchases exceed orders, you have an id mismatch, not an improvement.
What it will not fix
- Optimisation that is being fed a poor creative. Better measurement of a bad advert produces a more accurate account of the loss.
- Attribution honesty between platforms. TikTok will still claim what its window allows it to claim.
- Consent obligations. Sending events server-side does not remove the need for a lawful basis.
Build it when you can show purchases are being lost, ship one event first, and prove the gain against the same metric you used to justify the work. Otherwise the pixel is doing its job.
