The 90-day line.
Ninety days after a structural relaunch, someone opens the dashboard and the line is down. The room does the math on what the work cost. This is that dashboard, with the review date handed to you: drag it across the year and watch one unchanged curve get called a failure, then a wash, then a win. Nothing about the work moves. Only the day you looked.
Organic sessions · trailing 7 days vs baseline
−18.0%Day 90
“The redesign didn’t move the numbers.”
This is the meeting most structural work dies in. The line is real; the conclusion is early.
The curve never changes. Drag the review date and the same work reads as a failure on day 90 and a success on day 240. Whoever sets the measurement window writes the verdict, which is why that window is worth negotiating before launch rather than after.
Bench notes
- The curve
- Two systems, modelled separately and added. Search: thousands of URLs move at once, so crawling, redirect trust, and re-ranking take months, and the loss lands well before the gain. Habit: everyone who had learned the old structure lost it on launch day, and habit research puts re-formation at 66 days on average with a tail past eight months. Split the drivers and you can see the total is just their sum.
- The instrument
- The chart reports what an analytics tool reports: the trailing seven days against the pre-launch average, weekly seasonality and all. Everything past the review date is missing rather than faint, because on that day it genuinely does not exist yet. The rest of the year is a button, which is a luxury the meeting never has.
- What to notice
- The verdict sentence changes long before the work does. The trough is deepest around the first month, the line is still underwater at the quarterly review, and it crosses back over baseline months after the decision to keep going or kill it has already been made.
- What it is not
- A forecast. The shape is honest to how structural work behaves and the numbers are invented for this bench, not lifted from any engagement. Use it to argue about measurement windows, not to promise a number.