The Five-Domain Dashboard

Before any of the building, I hit a wall that had nothing to do with the dashboard. The data connection I was handed to pull GA4 through Adverity didn't actually have access to these properties. The account it authenticated with listed dozens of other properties but not the ones I needed, so every attempt to connect the data source came back empty.

Rather than assume the properties were missing or misconfigured, I checked the authenticating account's property list directly and confirmed the gap: the properties existed, the connection just wasn't authorized for them. The fix was to create a fresh authorization through an account that did have access, which unblocked the whole build. It's a small thing, but it's the kind of thing that eats a day if you assume the data is broken instead of checking who's allowed to see it.

With access sorted, I built the report from an empty canvas: layout and grid, a shared theme, then the data source connections for each domain across GA4, BigQuery, and Search Console. From there the dashboard pages themselves, structured so each domain reads consistently and the whole thing holds together as one report rather than five separate ones bolted side by side.

That part was straightforward work. The part worth writing about is what happened when I started checking the numbers against the prior report.

The sessions figures came in noticeably higher than the previous month's report, across every domain. The traffic hadn't actually jumped, so something in how I was counting was inflating the totals. This is the trap with GA4 data pulled from BigQuery: the raw sessions field over-counts relative to what most people mean by a session, because of how GA4 segments events. If you take the field at face value, your numbers look plausible and are quietly wrong.

The prior report had been built to count sessions a particular way, so to agree with it I had to match that methodology rather than trust the raw field. I rebuilt the session metric as a calculated field with an event-name filter, so it counted sessions the same way the baseline did instead of the way the raw export presented them.

Rewriting the metric wasn't enough on its own. I had to show the new numbers actually landed on the old ones, so I validated all five domains against the prior report rather than spot-checking one and assuming the rest followed. New-user counts came in within a few percent across the board, and on the largest domain the match was down to a handful of users out of nearly seventeen thousand.

That kind of margin is the difference between a dashboard someone glances at and a dashboard someone trusts enough to make decisions from. Getting the largest domain to agree within three users out of roughly sixteen thousand eight hundred meant the methodology was right, not just close.

One more problem surfaced during validation. A URL-exclusion filter I'd set on the Search Console charts, meant to strip out a set of paths that shouldn't count, wasn't changing the numbers at all. The chart looked filtered and behaved as though it wasn't.

A filter that silently does nothing is worse than no filter, because it looks handled. I traced it to the filter configuration itself rather than the data: the kind of mismatch where the include-versus-exclude setting or the targeted field is slightly off, so the rule matches nothing and passes everything through untouched. Once corrected, I built the Search Console version as a deliberate twin of the GA4 exclusion so the two data sources were filtering the same way, and I mapped out which domain pages actually needed the exclusion and which didn't, rather than blanket-applying it and hoping.

The building was maybe a third of this job. The rest was refusing to trust a number until it agreed with a number someone already believed. GA4 data out of BigQuery will hand you plausible, confident, wrong totals if you take the raw fields at face value, and a filter will happily sit on a chart doing nothing while looking like it works. The work that made this dashboard worth shipping was the part nobody sees in the final report: checking every domain against the baseline until the methodology was provably right rather than merely reasonable.