Thresholding, '(other)', and a Demographic Remarketing Audience
Northline Supply runs a Shopify storefront on a single GA4 web data stream. Monthly traffic is about 62,000 active users, and the property collects ecommerce events. Two months ago the team enabled Google Signals and granular location/device data so it could build Google Ads remarketing lists. Since then the Age and Gender reports show a large '(other)' row and many demographic rows are blank; a revenue-by-age exploration returns rows only for a few age brackets. The growth lead wants a 30-day audience of 'women 25–34 who viewed a product but did not purchase' and expects several thousand users. The published audience size in Google Ads is only a few hundred. Meanwhile, purchase revenue in the ecommerce report is close to the Shopify back office, so the team is unsure whether tracking is broken. They propose turning off Google Signals to 'unhide' the missing demographic data.
1. Why are the demographic rows blank or grouped into '(other)', and why is the published Ads audience so much smaller than the growth lead expects? Separate the explanations and state what is not evidence of broken ecommerce tracking.
Read the worked answer
Two different mechanisms are at work.
- Thresholding applies because Google Signals and demographics are enabled. GA4 withholds rows when a cohort is small enough that an individual could be inferred. Blank age/gender cells and suppressed revenue-by-age rows are expected; they are a privacy control, not a tag failure.
- '(other)' is usually cardinality: the report has a row limit and low-volume values are rolled into a row. It can coexist with thresholding, but it is not the same thing.
- Audience size is small because the definition is narrow (women 25–34, product view, no purchase, 30 days) and because demographic audiences depend on Signals-eligible users; the published size in Ads can also reflect eligibility and privacy checks.
Ecommerce tracking is not disproven by missing demographics. The check is whether purchase events and revenue reconcile with Shopify after deduplicating transaction_id; demographic suppression says nothing about whether purchase events fire.
2. The team proposes disabling Google Signals so the demographic data becomes visible. Evaluate that trade-off and recommend a path if they still need the remarketing audience.
Read the worked answer
Disabling Google Signals does not 'unhide' the same data. It stops the collection and use of Signals data, so Age/Gender/Interest reports and Signals-based remarketing audiences lose their input. You may see fewer suppressed cells because there are fewer demographic breakdowns to suppress, but you have not recovered the suppressed cohorts; you have removed the capability. You also lose cross-device Signals data and any modeled conversions or behavioral modeling that depend on Signals eligibility, depending on property settings.
If the demographic audience is a hard requirement, keep Signals on and accept that small cohorts are withheld. Broaden the audience definition (for example, all product viewers 25–34 with a longer membership) so it clears privacy and volume thresholds, and let Ads optimize on the larger pool. Do not promise 'several thousand' users from a narrow, suppressed demographic segment. If the business can use first-party declared data (newsletter signup with age), that is a separate consented dataset, not a workaround for GA4 thresholding.
3. What would you check to confirm that purchase tracking is healthy, and what would you tell the growth lead about using BigQuery export to recover the suppressed demographic rows?
Read the worked answer
Confirm purchase health with event-level checks: compare purchase event count and revenue to Shopify orders for the same date range and timezone; deduplicate by transaction_id; verify currency; check that the purchase event fires once per order and that items are populated. Use DebugView for a test order and Realtime for a live order.
BigQuery export is not a demographic workaround. It is unsampled and unthresholded for event data, but Google Signals demographics are not exported as user-level age/gender fields you can join back to purchase. Do not imply you can reconstruct suppressed cohorts in BigQuery. If the lead needs age-based revenue, the options are a consented first-party survey or declared data, a broader modeled audience, or accepting aggregated suppression - not a query that re-identifies users.