A reading nobody checked
It is not a record. It is a sample, assembled in the browser of whoever happened to visit, and almost anything that happens in a browser can lose a conversion or count it twice. A checkout change, a script error, a privacy setting, a refresh: each one quietly edits the total before you see it.
The work you booked lives somewhere else, in the system where your team writes it down the day it is sold. Those two numbers should be in the same neighborhood, and most owners have never held them side by side. The reconciliation takes an afternoon, and nobody does it because the screen has never asked to be doubted.
What one audit found
Search Engine Journal published an audit by Christian Goodrich of eight UK ecommerce brands, six of which allowed detailed matching between GA4 and store order records. Four of them, called A through D in the article, had significant tracking problems. GA4 recorded 81% of Retailer A's store web orders in Q4 2024, 41% in Q4 2025 and just 32% over the Christmas period.
Between Christmas Day 2025 and January 4, 2026, GA4 told that retailer its revenue was down 68% a day on the year before, while its own order records showed a fall of 13%. Retailer B's capture rate went from 94% of store web orders to 36% in a year. From October 28 to December 4, 2025, Retailer C's store recorded 2,215 orders and GA4 recorded no purchases at all.
The banner that expired
One of the causes was the consent banner itself. The consent plan covered a set number of page views a month, and once they ran out the banner stopped showing until the next billing date. Nothing broke and nothing alerted anyone, because data kept arriving, and the missing days read like a slump in sales.
The error runs the other way too. Retailer D showed a bot-driven spike of between 23,187 and 25,622 sessions on September 27 and 28, 2026, traffic that was never going to buy anything. Undercounting looks like a dead channel, overcounting looks like a winner, and in both cases the screen looks entirely normal while it is wrong.
What changes this week
The author's recommended check has three steps and belongs to the owner and whoever decides marketing, done once before any budget moves. Export weekly web orders for Q4 2024 and Q4 2025, leaving out marketplace, subscription and manual orders. Pull GA4 purchases for the same weeks as distinct transaction IDs on purchase events. Then divide GA4 purchases by store orders, week by week.
In that weekly ratio, the author says to look for step changes, near-zero weeks, and readings above 100%. For a service company the two columns have the same shape: booked jobs in your own system against the conversions the platform claims. What changes is the Monday conversation. A bad week on the dashboard stops being a reason to move money until somebody compares it to the work, and the channel you were about to cut gets judged on jobs you can name.