Beta Position Eleven is live and still being built.
Home › Writing › Analytics

Looker Studio will happily show you a wrong number

Looker Studio is genuinely useful and it is very good at rendering something plausible from data that does not support it. Nothing warns you.

Blending, which is not a join

Blending is a join — Looker Studio offers left outer, right outer, inner, full outer and cross, and defaults to left outer. The mistake is almost never the join type. It is the key: blending Google Ads and GA4 on date alone, then displaying cost from one and conversions from the other.

That produces a table where each row's numbers come from different systems with different attribution, aggregated on a shared date key. It looks like a unified view of performance. It is two unrelated series printed adjacently.

Blend on a genuine shared key — campaign identifier, not date. And know your join type: a left outer join silently keeps unmatched left-hand rows with blank right-hand values, which renders as zero on a chart and reads as a real zero rather than an absent match.

Date range mismatches

Every chart can carry its own date range, and it is easy to end up with a page where the scorecard shows last 28 days and the trend shows this month. Both are correct; together they are misleading.

Set the range at page level, and only override deliberately.

Sampling and quotas

The GA4 connector is subject to sampling and to API quotas. A dashboard that worked yesterday can return partial or empty data today because a quota was exhausted — often by the dashboard being open in several tabs.

There is a "data is sampled" indicator. It is small, and nobody looks at it.

Calculated fields that divide

conversions / sessions computed at row level and then averaged is not the same as total conversions divided by total sessions. The first weights small rows equally with large ones.

This is the single most common cause of a conversion rate on a dashboard disagreeing with the same conversion rate in GA4, and the dashboard is the one that is wrong.

What to do

Fewer charts. A dashboard with 30 charts is not being read.

Label the source of every number. "Conversions (Google Ads)" and "Key events (GA4)" side by side, never blended into "Conversions".

Reconcile once, at build time, against the source interface. If the dashboard says 412 and GA4 says 380, resolve it before anyone starts making decisions with it.

Date-stamp the refresh. A cached dashboard showing yesterday's data with no indication is how a meeting ends up discussing a number that has already changed.

Read next

Reading about it is slower than looking

Run the free audit on your own site. Public data, about ten seconds, no account.

Audit my site

Stop reporting what happened.

Start telling clients what to change — and prove it moved.

Get started
Beta · help shape it

What would make this worth paying for?

Everything on this site works today. It is early, so the roadmap is still genuinely movable — tell us what is missing and we will build it. Testers who give us something useful get a lifetime deal on whatever plan they end up on.

No account needed. Email is optional.