Proof, not promises
Every number on this page carries its measurement window, its sample size and its confidence level. A percentage without those three is not a result, it is a headline.
Where we hold a result back, we say so and why.
How to read these
What appeared to be leaking, and the evidence that pointed there.
Exactly what changed, not a category of work.
Tool, window, sample, confidence level, primary metric.
Short windows, mixed signals, and anything the data cannot settle.
Our full measurement methodology covers the confidence standard, how projections are calculated, what happens when a test loses and what we exclude.
Every one of these is assessed on mobile first, because that is where most of your traffic decides.
Offer and message clarity
Primal Queen
Largest session sample in our test history. Conversion rate was the primary metric; revenue per visitor rose 3%, which did not clear the decision threshold on its own.
Offer and message clarity
Primal Queen
Lifted conversion rate by 22% and revenue per visitor by 22%, across 299,077 sessions at 90% confidence.
Offer and message clarity
Canada Welding Supply
Lifted conversion rate by 14% and revenue per visitor by 12%, across 102,230 sessions at 90% confidence.
Summary only. The detail page waits until two internal measurements of this test agree.
Projected monthly contribution is extrapolated from the test window. It is not revenue banked, and we will not present it as though it were.
Results are specific to the cited experiments. Not every test wins, and projected contribution depends on the measurement window and store economics.
The gaps
Cart and checkout, mobile experience, speed and technical drag, and measurement integrity are all things the Diagnostic inspects and all things we have fixed. None of them currently has a result that clears our publication standard, so none of them appears above.
We also hold back results where two internal measurements disagree, and we have permanently excluded our single largest recorded result because a one-day window on a small sample is more likely to be an artefact than a finding.
Those absences are the reason to trust what is here.
Across 6 clients and 49 winning tests, our experiments have generated between $7K and $470K in projected additional monthly revenue per win, measured at 90% statistical confidence.
Every case here started as a store nobody could rank the problems in.