How to analyze your marketing data in BigQuery with Claude

Connect the BigQuery MCP

To explore the dataset with Claude, you copy the dataset name and hand it over. You need the BigQuery MCP connected: here’s the Detrics guide explaining the setup.

Claude, check via the BigQuery MCP whether you can see this dataset.

Claude runs a couple of SQLs, finds the dataset among the ones it can access and confirms it sees the tables, with their rows and sizes. While it’s at it, it runs a first analysis: accumulated spend on Meta and TikTok, detecting the currency situation and reporting in dollars.

The whole pipeline, visible end to end

An important detail: with the Detrics MCP also connected, Claude sees both ends: the transfer (the source) and BigQuery (the destination holding the data). It spots, for instance, the failed TikTok chunk the system repaired, and explains why the TikTok data starts in 2022: the platform only stores the last few years, and whatever wasn’t loaded in time is gone. One more reason to set up the transfers as soon as possible.

Asking your warehouse questions in plain language

Tell me, in total dollars, how much spend we’re tracking with these tables, what time period they cover, and how many unique ads we have.

Claude builds the SQL on its own. It realizes most of the spend is in pesos, looks up the dollar rate and converts: roughly 13 million dollars estimated across the history, with data starting in 2022 and a 3-year overlap between Meta and TikTok, plus a spend summary per year.

For more precision, you can load a table into BigQuery with the historical peso-dollar conversion and ask it to use that. These SQL queries used to be written one by one, by hand; now Claude does them, and does them very well.