What 12 client sites actually measure

A measurement audit across a web agency's BigQuery portfolio · 7 projects · 2026-09-08

Agencies accumulate analytics projects one client at a time, and nobody ever looks at them side by side. This audit reads the schema and row counts of every active BigQuery project in a client portfolio and asks one question of each: can this site answer a business question at all?

It takes about half an hour to run, costs nothing to query, and — as the findings below show — tends to surface work worth doing that nobody knew was there.

1 of 7

Of 7 sites in the portfolio, only Store B records every step between a visitor landing and paying.

Agencies assume their clients’ analytics work. This audit checks, site by site, whether the data needed to answer a business question is actually being collected.

The inventory

Project	What it measures	Rows recorded	Purchases	Still recording?
Store B	Full payment journey (GA4) + Segment	15,686	11	live
Site 1	Some events, no purchases recorded	415,432	not tracked	live
Site 2	Page views only	42,360	not tracked	live
Site 3	GA4 with custom events + Segment	17,509	none	live
Site 4	GA4 defaults only + Segment	5,974	none	live
Site 5	GA4 with custom events + Segment	837	none	3 days behind
Site 6	Page views only	817	not tracked	live

The one site that measures the whole way to payment

Store B is the only site in the portfolio where the steps between landing and paying are all recorded. 891 visitors, 8 buyers, 11 purchases between 2026-05-03 and 2026-09-06. It tracks 12 distinct event types; the other Google Analytics sites record only the six that Google switches on by itself.

Step	People	Times
Viewed a product	not measured
Added to cart	not measured
Started checkout	32	68
Entered shipping details	15	18
Entered payment details	8	11
Purchased	8	11

Rows marked not measured are steps the site does not report. They are almost certainly happening — people cannot check out without seeing a product first — but they are invisible to us.

What stands out

1. The best-measured site records purchases but never the amount

Store B’s 11 purchases carry no value: total revenue reads 0. The site knows someone bought and not what they spent.

What this means: no revenue, no average order value, no return on any campaign — not because the data is thin, but because one field is missing from the tracking. A small fix turns the best-instrumented site in the portfolio into a reportable one.

2. The drop happens at shipping, and only one site can see it

32 people started checkout on Store B, 15 got as far as entering shipping details — 53% lost at a single step.

What this means: this is the kind of finding a client acts on the same week. No other site in the portfolio can produce it, because no other site measures those steps.

3. The busiest site measures nothing about the business

Site 3 had 1,470 visitors and records 12 event types, all of them Google’s defaults. Nobody configured anything.

What this means: the largest audience in the portfolio is the one least is known about. Traffic without instrumentation is not an asset; it is a missed measurement.

4. 4 sites stream raw Google Analytics data into BigQuery

These have Google's native GA4 export switched on — every event, every day, automatically, with data landing through 2026-09-07. Two of the three record only Google's defaults; Store B is the one where somebody set up the business events by hand.

What this means: no data collection needs to be built for those sites. The raw material is already there and current, and it is already being paid for.

What an audit like this leaves you with

Which sites are supposed to sell online? Some gaps are deliberate. Separating the deliberate ones from the broken ones turns a list into a work plan.

One project contained churn predictions built on an empty table. Models outlive the data they were built on, and nothing warns you when that happens.

Where to start. On volume alone, Store B is the only site in this portfolio where behavioural analysis would currently mean anything.

How this was produced. One query per project against BigQuery’s table metadata — no data scanned, no cost — feeding a script that renders this page. Re-running it on a new portfolio takes minutes.

Row counts are read rather than sampled, which matters: four of the twelve projects have tables that exist and are empty, and an audit that counts tables instead of rows reports those as healthy.

Not currently receiving data

Shown for reference — these sites recorded data before and stopped. Their history is kept in full for future analysis, nothing is deleted.

Project	What it measures	Rows recorded	Purchases	Still recording?
Store A	Full purchase funnel	152,337	247	stopped 162 days ago
Store C	Full purchase funnel	99,517	7	stopped 83 days ago