Book a call

Marketing data warehouse: what it does and when it pays

Analytics & measurement
Selective

By Wiesław Tkacz, Data Analyst and StrategistUpdated 3 November 2026

Where it sits among the other tools is on the ecommerce marketing technology stack page.

What does a marketing data warehouse do?

It stores raw data and lets you ask your own questions of it. Reports inside GA4 or an ad platform only answer the questions someone built into them.

For GA4 users, the starting point is the GA4 BigQuery export: a copy of your raw event data in Google's BigQuery warehouse. If you are new to GA4 itself, read Google Analytics 4 basics first.

A warehouse can also take in:

  • orders and refunds from your shop
  • ad spend from each platform
  • email and customer data, which is your first-party data

Once these sit in one place, you can join them. That is the real point.

Why do e-commerce brands use it?

Example (illustrative): Meta says 120 orders, GA4 says 90, the shop says 150. Nobody can say which customers came back, because each tool only sees its own slice.

Three reasons to consider one:

  • Thresholds. GA4 withholds some data, such as demographics, from reports and explorations so nobody can identify individual users, and you cannot switch this off1. Google lists BigQuery export among the things that make thresholds less likely1.
  • Cohorts. Cohort analysis needs order-level data. A warehouse gives you that without exporting spreadsheets by hand.
  • Your own rules. You can build your own attribution models from raw events instead of accepting each platform's.

How do we implement it?

1

Name the question

Write down the question that needs joined, order-level data. If you cannot name one, wait.

2

Check the source data

Fix broken tracking first, because a warehouse stores bad data faithfully. Agree one definition of revenue.

3

Build in your accounts

Keep the warehouse, access and data in your name. Then connect the shop, ad spend and the GA4 export.

4

Name an owner

Someone owns the data, the access and the definitions. Expect totals to differ from GA4 reports.

When should you not use it?

Skip it, for now, when:

If you are not sure, you can start with a free tracking audit. Our ecommerce tracking audit article lists the checks that change a budget decision.

What does it cost in effort?

We give no figures. The cost is mostly people and attention, not software:

  • An owner. Someone must own the data, the access and the definitions.
  • Build. Connecting sources, naming fields, and agreeing what "revenue" means.
  • Upkeep. Platforms change their data. Connections break quietly.
  • Questions. Raw data is useless until someone writes the query.

Our view: ask any supplier for all four in writing, and keep the accounts and the data in your name, so you can change supplier without starting again.

Pitfalls

  • Numbers that do not match GA4

    Google's documentation on behavioural modelling says modelled data is not in the BigQuery export2. If you use Consent Mode, your warehouse totals can differ from your GA4 reports. That is expected.

  • Sampling

    Google's help page says GA4 uses a default sampling limit of 10 million events on standard properties, across reports, explorations and the Data API3. For unsampled results without Analytics 360, it points to BigQuery export or a shorter date range3.

  • Not everything is exported

    Google says data from Google signals is not exported to BigQuery, and event counts per user can differ between GA4 and BigQuery1.

  • Three definitions of revenue

    If every table uses its own, you have rebuilt the disagreement, only more expensively.

  • Personal data

    Customer data raises consent and privacy questions. Your counsel decides what you may store and for how long.

  • No owner

    An unattended pipeline stops being right without telling anyone.

Our take: Selective

Our view: Selective. This is our rating, not a market fact or an industry scale.

Google documents the export and the threshold workaround1. It does not say your brand needs one. That depends on whether your questions need joined, order-level data.

Our view: a warehouse is a means to a question. If you cannot name the question, you do not need the warehouse yet.

FAQ

A rider in a red cape and bicorne hat on a rearing yellow plush horse, after David's Napoleon Crossing the Alps

What are GA4 data thresholds?

GA4 withholds some rows, such as demographic or search query data, when counts are too low to protect user identity1. You cannot adjust them; Google suggests a wider date range or BigQuery export to make them less likely1.

What does the GA4 BigQuery export give you?

Raw event data in your own BigQuery project, outside GA4's reports. Google points to it for unsampled results on standard properties3, but data from Google signals is not included1.

Does GA4 sample data in explorations?

It can. Google's help page gives a default sampling limit of 10 million events for standard properties and 100 million for Analytics 3603.

Will BigQuery numbers match my GA4 reports?

Not always. Modelled data is not included in the export2, so totals may differ where Consent Mode modelling applies.

Do I need a data engineer?

Not always, but you need someone who can write queries and keep the connections working. Without that person, the warehouse sits unused.

Is a warehouse the same as a dashboard?

No. A warehouse stores data. A dashboard shows a chosen slice of it.

Who should own it?

Our view: you should. Keep the cloud account, the access and the data in your name.

Sources

  1. [GA4] About data thresholds (Analytics Help)Primary source
  2. [GA4] Behavioral modeling for consent mode (Analytics Help)Primary source
  3. How data is stored and displayed (Analytics Help)Primary source

Let’s find what’s actually working.

Next step. Want a view on whether a warehouse is worth it on your account? Talk to us about your budget decision. Or read about our Analytics & tracking service.

Talk to us about your budget decision