TRACKING SPECIALIST — BARIS ASA · LONDON
Synctag
Start the check → TRACKING SPECIALIST — BARIS ASA · LONDON

← The field manual

MANUAL · CH.08

How to fix metric definitions so reports agree: a metric register in five steps

Filed

Baris Asa, Synctag, tracking specialist, London · Last checked:

When two reports both say “Revenue” and show different totals, the label is shared but the definition isn’t. The fix is a metric register: one written line per number, saying which question it answers, its source, exact field and owner, with every report built from that line. Five steps: list the numbers, give each one question, write the line, rebuild the reports, change only through the register.

This is what most teams mean by a single source of truth for KPIs: one written definition per number, not one dashboard. A tile (one chart or number box on a dashboard) is built from its line, so two people asking the same question in two places get the same number, or a difference that is written down.

Is this your problem? A five-minute check. Open the two reports that disagree. Click the number in each (in Looker Studio, edit mode, then the chart’s Setup panel) and write down the exact field, the date range and the time zone. Different fields under one label: a definition problem, and this chapter. Same field, same dates, different totals: the data underneath disagrees, which is F-302 or a GA4 reconciliation question, and no register fixes it.

In my experience it’s nearly always “Revenue”. It’s rarely only two versions.

StepWho does itLeaves behind
1. List the numbersThe business, with whoever runs trackingNumber list, where each one appears
2. One question eachThe owner of each numberQuestion and source per number
3. Write the lineWhoever runs trackingRegister, one line per number
4. Rebuild the reportsWhoever runs trackingTiles built from the register, locked
5. Change controlOwner, then trackingDated changes, annotated reports

“Whoever runs tracking” is the agency, freelancer or in-house analyst who looks after GA4 and the tags. The owner of a number is the person who acts on it. They decide what it means; tracking checks the definition can be built, and builds it.

This chapter is the method. Why the same label gives different numbers (four GA4 revenue fields, conversions summed across events, a visit’s channel against an action’s channel, time zones) is the R-501 page. Why a correctly defined number still loses rows to (other), thresholds or sampling is R-502 . Who builds the report itself: Looker Studio reporting.

Step 1: which numbers go in the register?

Every number someone acts on or argues about, and no others. Collect them from where decisions are made, not from the dashboard: the weekly marketing deck, the board pack, the finance sheet, the agency’s monthly report, the figure someone posts every Monday.

For each one, write where it appears and who reads it. The same word turns up more than once. “Revenue” in the board pack and “Revenue” on the marketing dashboard are two entries until step 2 says otherwise.

Leave out numbers nobody looks at. For an online store the register usually settles at ten to fifteen lines. A register of 80 lines is a data dictionary, and nobody keeps it up. A tile that isn’t on the list after step 1 is a candidate for deletion in step 4.

Step 2: which question does each number answer?

One number answers one question, and the question picks the source. When the argument was two questions sharing one label, this step ends it.

QuestionUsual source
How much did we sell? (finance total)The store or the finance system
Which channel brought the sale?GA4, one revenue field chosen on purpose
What is Google Ads bidding on?Google Ads conversions, as the account counts them
Is the site getting more visitors?GA4 users or sessions, one metric named

When two people need two answers, the number becomes two lines with two names: “Revenue (finance, store)” and “Revenue (marketing, GA4)”. Neither is wrong. They answer different questions, and a gap between them is expected.

The same goes for GA4 and Google Ads. “Purchases (GA4)” and “Conversions (Google Ads)” are two lines, each with its own stated gap. They count differently and credit different dates, so they rarely match. Why, and which gaps are normal: GA4 and Google Ads numbers don’t match. The register doesn’t make them agree. It stops the gap being argued about.

Settle disagreements by the question, not by which number is higher. If the owner can’t say what decision the number feeds, it doesn’t need a line yet.

Step 3: what goes on one register line?

Enough that someone who wasn’t in the room can rebuild the number and get the same result. Exhibit A: three lines from one register (illustrative, not a client).

Line, as named on the tileQuestionSource and exact fieldOwner
Revenue (finance, store)How much did we sell?Store net sales exportFinance
Revenue (marketing, GA4)Which channel brought the sale?GA4 Purchase revenue, by Session default channel groupHead of ecommerce
Conversions (Google Ads)What is Ads bidding on?Google Ads Conversions, purchase action onlyPaid media lead

Each line also carries four more fields:

  • Filters: what is included and excluded, written out. “None” is an answer.
  • Date rules: the reporting time zone, the week (Monday to Sunday, say), and the rule that only weeks ended at least two days ago are compared. Google says GA4 processing can take 24 to 48 hours and reports may change in that time.
  • Normal gap: the gap measured, not hoped for. GA4 Purchase revenue usually runs below the store’s net sales (declined consent, blocked scripts), but can run above it when refunds never reach GA4 or purchases are counted twice. Write the range from the last baseline check (U-001). A gap that moves outside its range is a tracking fault to file, not a reporting one.
  • Last changed: the date, the reason, and who agreed.

Exact field is where most second definitions hide. Write the field as GA4 or the ad platform names it, never the tile title. Google’s definitions:

GA4 fieldWhat it adds up
Gross purchase revenuePurchases, in-app purchases and subscriptions, before refunds
Purchase revenueThe same, minus refunds
Total revenuePurchase revenue plus ad revenue (all net of refunds)
Item revenuePrice times quantity, minus refunds: excludes tax and shipping

For a website without app or ad revenue, the first three mostly differ by refunds. Item revenue differs by tax and shipping. A Conversions line also names its event: GA4’s Key events metric counts every event marked as a key event (an event GA4 treats as a conversion), so an unfiltered tile adds purchases, sign-ups and anything else marked.

Keep the register beside the tracking record, in a sheet or document the business owns. What the tags send and what the reports mean are two halves of one record (F-501 ).

Step 4: how are the reports rebuilt to the register?

The result to aim for: every dashboard reads one shared definition, and nobody can quietly change it in one report. In Looker Studio (renamed Data Studio in April 2026; the settings are the same), that means building the definition into a reusable data source, not into each chart. Google says changes to a reusable data source affect every report that uses it. A data source embedded in one report has to be edited inside that report, and each copy drifts.

  • Name the fields. Rename each field in the data source to its register name, so the tile title matches the line.
  • Write the description. Field descriptions can only be added in the data source. Paste the line’s one-sentence definition there, and turn on Show field descriptions on table charts so readers see it.
  • Lock it. Turn off Field editing in reports on the data source, so a report editor can’t change field properties in one report. That closes a common route to a second definition, not every route: filters, blends and date settings added in a report still change the number. Check those tile by tile against the line.
  • Show the date. Put a “data to” date on every page. Looker Studio refreshes GA4 data every 12 hours by default (1, 4 or 12 can be set) and Google Ads data every 12 hours, fixed. That cache sits on top of GA4’s own 24 to 48 hours of processing, which is why the line’s date rule compares closed weeks.
  • Label what can move. A note under any tile that can show (other), held-back rows or sampling, so a missing row has a stated reason (R-502 ).
  • Delete the rest. A tile with no register line goes. If someone misses it, it comes back through step 2.

Check who owns each data source and whose credentials it uses. With owner’s credentials, everyone sees the data through one person’s access, so the reports depend on that login staying. If it belongs to someone leaving, the fix is in the handover chapter.

Building the reports to the register is the job on the Looker Studio reporting page, if nobody in-house has the time.

Step 5: how does a definition stay fixed?

By changing it only through the register. A definition will change: a refund flow gets connected, a new key event is marked, the time zone is corrected. What breaks trust is changing the tile first and the record never.

  • Register first. The new definition, the date it starts, the reason, who agreed. The old line stays on record, marked as replaced, so last year’s numbers can still be read.
  • Annotate the date. GA4 annotations put a note on report line graphs, for a single date or a range. Creating one needs the Analyst role or above on the property; titles take 60 characters, descriptions 150. Put the same note on the dashboard page.
  • Expect a break, not a correction. Google says changing the reporting time zone only affects data going forward, and can leave a flat spot or a spike on the day. Marking an event as a key event affects reports from that point and doesn’t change historic data. Charts that cross the date show a step. The annotation says why.
  • New tiles go through step 2. A request for one more number gets a question and a line before it gets a chart.

Review the register after one full reporting cycle, then whenever the person acting on a number changes. The test: two people asking the same question in two places get the same number, or a stated difference.

What does a register not fix?

A register defines numbers. It can’t make a wrong number right.

  • The value is wrong at collection. A purchase sent with the wrong value, currency or tax treatment flows into every tile. That’s F-302 , fixed at the source first.
  • The count is inflated. The wrong events marked as key events is F-104 . A correct label on an inflated count is still inflated.
  • GA4 and the store disagree on the total itself, outside the normal gap. That’s a reconciliation question: GA4 reconciliation.
  • Rows are grouped or withheld. Same definition, fewer rows: R-502 , and the note on GA4 thresholding.

Common questions

How do you agree metric definitions across a team?

Start from the decision, not the data. List the numbers people act on, give each one a single question and an owner, then write one line per number: source, exact field, filters, date rules, normal gap. Two teams asking two questions get two lines with two names. The reports are rebuilt to the lines, and any change goes through the register first.

What is a metric register?

A table with one line per number the business acts on: its name as shown on the dashboard, the question it answers, the source, the exact field, filters, date rules, the normal gap to the store or ad platform, and an owner. For an online store it usually settles at ten to fifteen lines. It sits with the tracking documentation, and every report tile is built from it.

Is a metric register the same as a single source of truth?

It is the useful version of one. A single source of truth for KPIs is rarely one dashboard or one database. It is one written definition per number, with the source named. Finance revenue can come from the store and channel revenue from GA4, as long as each has its own line and name.

Where should metric definitions live in Looker Studio?

In a reusable data source, as field names and field descriptions, with Field editing in reports turned off. Google says field descriptions can only be added in the data source, and changes to a reusable source reach every report using it. Filters and blends added in a report can still change a number, so check them against the register.

What happens to old reports when a definition changes?

They keep the old definition up to the change date. Record the new line with its start date and the reason, keep the old line on record as replaced, and annotate the date in GA4 and on the dashboard. Google says a reporting time zone change only affects data going forward, so expect a visible step rather than a corrected history.

Who should own a metric definition?

The person who acts on the number: finance for the sales total, the head of ecommerce or marketing for channel revenue, whoever runs paid media for what Google Ads bids on. Whoever runs tracking checks the definition can be built and builds it. Without a named owner, a definition changes each time someone edits a chart.

Also filed as

  • how to agree metric definitions
  • metric definitions document
  • KPI definitions for marketing reporting
  • single source of truth for marketing metrics
  • data dictionary for GA4 dashboards
  • two reports show different revenue

Faults explained here

Sources

  1. Analytics: Dimensions and metrics (Gross purchase revenue = purchases + in-app purchases + subscriptions; Purchase revenue = the same minus refunds; Total revenue = purchases + in-app purchases + subscriptions + total ad revenue minus refunds; Item revenue = price x quantity minus refunds, excluding tax and shipping; Key events: times users triggered a key event; checked 9 Nov 2026)
  2. Analytics: Mark events as key events (affects reports from time of creation, doesn't change historic data)
  3. Analytics: Data freshness (processing can take 24 to 48 hours; reports may change in that time)
  4. Analytics: Set up Analytics for a website and/or app (changing the time zone only affects data going forward; may leave a flat spot or a spike)
  5. Analytics: About annotations (Analyst or above to create; single date or range; shown on reports with line graphs; title 60 and description 150 characters; checked 9 Nov 2026)
  6. Google Ads: Understand your conversion tracking data (conversions reported by click date)
  7. Looker Studio: Edit a data source (changes to a reusable data source affect every report that uses it; embedded data sources are edited in their report)
  8. Looker Studio: Model your data (field descriptions only in the data source; Show field descriptions on table charts; turn off Field editing in reports)
  9. Looker Studio: Manage data freshness (Google Analytics connector every 12 hours by default, 1, 4 or 12; other Google ad and measurement products every 12 hours, fixed)
  10. Looker Studio: Data credentials (owner's credentials use one person's authorisation; Make me the owner)
WHO RUNS IT

Baris Asa, a London tracking specialist working as Synctag, writes the metric register and builds the reports on it for ecommerce brands and the agencies behind them.

Looker Studio reporting →