How to load 3 years of ad history into BigQuery with transfers

Create the transfers

Let’s create the transfer: load the entire history and set up deduplication.

Claude asks whether the transfer should include every account or just the sampled ones. The video picks the ones shown earlier, but you can pick all: if you do, every time a new account appears in the Meta or TikTok account you connected, the system picks it up and adds it to the transfer automatically. That keeps both platforms synced to your warehouse forever, for existing and future accounts.

On frequency: daily is the recommendation. Syncing more often can saturate the Meta and TikTok APIs, and it’s unnecessary because the data is usually read on closed days.

The historical load runs in chunks

Once created, both transfers start downloading. In the web app’s Transfers section you can see the data being cut into chunks: it’s a huge amount of information, and requesting it whole tends to make the APIs fail. Cut into small pieces, the download is manageable; if a chunk fails, the system takes care of recovering that piece later.

The historical load can take up to days, but it runs once. When it finishes, an email arrives confirming the history is complete: in the video, 2 million rows for Meta and 200,900 for TikTok.

From the initial load to steady state

Back in Detrics, you can see the first run (with one failed chunk the system later repaired) and, below it, the subsequent syncs: much smaller and faster, focused only on the refresh window configured in the table group (August 8 to 12, for example).

That’s the steady state: short, efficient, stable transfers that keep BigQuery always current. For each account you can drill into the loaded periods, when they ran and how many rows they brought.