Reporting nobody trusts

Symptom and cause

Symptom

Dashboards exist, but every meeting argues about whose number is right. The report gets opened once, then ignored.

Cause

No agreed definition per metric: "revenue", "conversions" and "sessions" mean different things in different tiles, pulled from sources that don't agree.

Where it's caught

Two dashboards showing different totals for the same thing; metrics with no documented definition or owner.

Why the same metric shows different numbers

Most "which number is right?" arguments are not tracking faults. They are six choices made differently in different tiles, usually by different people, and never written down. Each choice is legitimate on its own. The fault is that nobody can say which one a tile made.

Which revenue. GA4 has four fields with revenue in the name. Total revenue adds purchases, in-app purchases, subscriptions and ad revenue. Purchase revenue is purchases minus refunds. Gross purchase revenue is purchases before refunds. Item revenue counts items only, minus refunds, so the tax and shipping on an order are left out (whether VAT is left out depends on what the item prices carry). A tile called Revenue can be any of the four, or the store's own figure, which is a fifth definition. The exhibit below is three tiles called Revenue reading three different fields.

Which count. Key events in GA4 is the number of times users triggered any key event. Without a filter on event name, a tile labelled Conversions adds purchases, sign-ups and anything else marked as a key event into one total. Ecommerce purchases and Transactions are separate metrics. Google Ads conversions are a separate system again, with their own rules.

Which users. GA4 reports show active users, people who engaged in the date range. Total users is a different metric, counting every user who triggered any event. A tile labelled Users built on Total users runs higher than the GA4 report next to it.

Which channel. Session-scoped dimensions, such as Session source / medium in the Traffic acquisition report, credit the session's source on Google's paid and organic last-click basis, which ignores direct when an earlier source exists. Event-scoped dimensions, such as Source / medium without the prefix, follow the property's attribution model, data-driven by default. Revenue by channel can differ between two tiles that both show the right total.

Which day. GA4 uses the property's reporting time zone as the day boundary, so a store or ad account on another time zone splits a day differently. Changing the time zone only affects data going forward. Recent days also move: GA4 processing can take 24 to 48 hours, some data arrives up to 7 days late, and key event credit can shift for up to 12 days. Looker Studio caches GA4 data for 12 hours by default, so a dashboard can be half a day behind the GA4 screen next to it.

Which rows. Sampling and thresholding remove or estimate rows without saying so on the tile. That is its own code, R-502.

None of these gaps is a fault by itself. Two tiles with different definitions and different labels are fine. Two tiles with the same label and different numbers, or a number nobody can define, is R-501.

Tile	Field it reads	30 days	Verdict
Weekly marketing · Revenue	GA4 Purchase revenue	£128,400	one definition
Board pack · Revenue	GA4 Item revenue (items only)	£104,900	another
Finance sheet · Revenue	Store net sales export	£131,700	a third
Metric register	none on file	n/a	no one owns Revenue

three tiles called Revenue, no definition on file

How to catch it

1. Start with the argument itself. Take the two numbers people disagree about and, for each, write down the source, the exact field, the filter and the date range. In Looker Studio, open the chart in edit mode and read the metric in the Setup panel; the tile title does not count. Same field, filter and dates with different numbers points at a data or sampling problem (R-502). Different fields under the same label is R-501.
2. List every tile labelled Revenue across the dashboards people actually open. GA4 offers Total revenue, Purchase revenue, Gross purchase revenue and Item revenue. Note which one each tile reads, or whether it reads the store or the finance export instead. More than one field under one label is the fault.
3. Check every tile labelled Conversions or Key events. If it uses the Key events metric with no event-name filter, it adds every key event together: purchases, sign-ups, anything marked. Compare it with Ecommerce purchases for the same range.
4. Check the channel tiles. Session source / medium is session-scoped and paid and organic last click. Source / medium with no prefix follows the attribution model, data-driven by default. Two channel tables built on different ones disagree by channel and agree on the total.
5. Check the dates. GA4's property details in Admin show the reporting time zone; compare it with the store and the ad accounts. Then compare totals over ranges that end at least two days ago (a rule of thumb from GA4's 24-48 hour processing, not a Google figure), and channel splits over ranges that end at least 12 days ago, because key event credit can shift for that long. A gap that only exists for recent days is timing.
6. Ask who owns each number. If nobody can give the definition of the revenue tile in one sentence, or two people give different ones, that is the finding. A metric with no written definition and no owner is R-501 whatever the data says.

What putting it right involves

1. Write the metric register before touching a dashboard. One line per number the business uses: name, one-sentence definition, source, exact field, filters, time zone, owner, and the normal gap to the store or ad platform. Keep it to the ten or so numbers that are argued about. It belongs in the same place as the tracking documentation (F-501).
2. Pick one source per question. Finance revenue usually comes from the store or the finance system. Channel comparison comes from GA4, with one revenue field and one scope chosen on purpose. The register says which question each number answers, so a gap between them is expected and stated, not argued.
3. Rebuild the tiles to the register. Rename every tile to say what it is, for example "Purchase revenue (GA4, net of refunds, session channel)". Filter key event tiles to named events. Remove tiles nobody can define. If a definition is wrong at the source, such as the purchase value itself, fix that first (F-302); a register cannot rescue a wrong value.
4. Fix the day. Set the GA4 reporting time zone to match the store if they differ, knowing the change applies from that day on. Show the "data to" date on the dashboard and exclude incomplete days from weekly comparisons. In Looker Studio, set the data source's freshness to suit how often the report is read.
5. Put the normal gaps on the dashboard. A line under the revenue tile such as "GA4 runs below the store: consent and blockers" stops the same argument every week. If the gap moves outside its usual range, that is a tracking fault to file, not a reporting one.
6. Give each number an owner and a review date. Any new tile or changed definition goes through the register first. Check after one full reporting cycle: the same question asked in two places should return the same number, or a stated, explained difference.