Question 1

What can a Databricks model send to TikTok Ads?

Accepted Answer

Custom audiences and offline conversion events. For audiences, each row the model returns becomes a member, matched on email or phone number. For conversions, you can report events that happened away from TikTok, which is how the platform learns about a sale that completed in your billing system. Both run off ordinary models, so one table can drive targeting and measurement.

Question 2

What stops a TikTok audience being usable?

Accepted Answer

Three things, all on TikTok's side. Audiences below 1,000 members are not usable in targeting, which TikTok documents and enforces in its interface even though the API has accepted smaller ones in testing. Membership takes up to 48 hours to process, so a freshly built audience is not immediately available. And an advertiser cannot hold more than 400 audiences, which matters if the plan was a segment per campaign.

Question 3

How should I schedule this against my SQL warehouse?

Accepted Answer

Group it with your other syncs rather than running it on its own timetable. Every run is a query against the warehouse you configured, and if that warehouse has auto stopped in between, the query wakes it and you pay the warm up. Given TikTok takes up to 48 hours to process membership anyway, a frequent schedule buys nothing while paying that cost repeatedly. Daily, aligned with other reverse ETL runs, is usually the sensible shape.

Question 4

Does membership reconcile against the model?

Accepted Answer

Not on TikTok. Add mode adds and remove mode removes, both available here, but there is no mirror mode reconciling the audience against your model automatically, which Microsoft and Meta both offer. A shrinking audience therefore means a second model describing who should come out, run as a remove sync beside the add one. Two syncs rather than one, so it is worth deciding before the audience is built rather than after.

Question 5

Is this better than exporting a list?

Accepted Answer

It is, once the definition of a good audience depends on anything computed. A manual upload is accurate the day it happens and drifts from then on, while a model reruns and reflects what changed. The bigger gain is that the seed list can come from scoring built on product and billing data, which is a far better input to a lookalike than a filter on a field the ad platform happens to hold.

Sync Databricks to TikTok Ads

Sync Databricks to TikTok Ads with Polytomic

Email and phone hashed

Real limits to plan for

Offline conversions too

Or sync TikTok Ads to Databricks

Trusted by businesses worldwide

Step-by-step integration guide

Connect Databricks in Polytomic

Connect TikTok Ads in Polytomic

Create a sync between Databricks and TikTok Ads

Map fields and enable the sync

Frequently asked questions

The sync and ETL platform that does it all

Powerful data sync

Advanced analytics

Enterprise security

Easy setup

Ready to connect Databricks with TikTok Ads?