Before you click
The QA checklist is complete, including tagging confirmed on every surface and one real order
Products and price changes are what you intended
The traffic split matches how much exposure you are comfortable with
No product in this test is also in another active price test
Someone else knows the test is going live — particularly customer support
Set it live
Open the test and click Set live.
This is not instant-and-silent. Cooee gathers every published price test for the store, makes sure the Cart Transform Function exists, and only then writes the display data. If the transform cannot be created, the sync clears everything rather than half-publishing — so you will never end up displaying a price checkout cannot charge.
What happens immediately
The Cart Transform Function is installed and activated if not already present
The test data is written to Shopify — one metafield for the function, one for the theme
New visitors start being assigned to groups
Reporting begins collecting
Visitors already browsing are not retroactively enrolled. Enrolment happens on page load, so the test builds up over the first minutes rather than switching on for everyone at once.
First hour
In a fresh incognito window each time:
Both prices are appearing on the storefront
A cart still carries the test price through to checkout
No new errors in the browser console
Support has had no pricing complaints
Then place one more real order on the live test. QA was done pre-launch; this confirms the live configuration charges correctly. It is the single highest-value check you can make.
First 48 hours
This is when problems surface — odd device and browser combinations, a caching layer you had forgotten, an app that only runs on certain pages, a template nobody thought to tag. Watch for:
Support mentioning price confusion, or a price changing between pages
Traffic split roughly matching what you configured, once you have a few hundred visitors
Any product in the test showing zero views, which usually means its template is untagged
Do not read the results yet
Early numbers move violently and mean nothing. A variant can be 40% "ahead" on day one on the strength of two orders.
Give it two to four weeks
Let it cross at least two full weekly cycles — buying behaviour differs by day
Wait until there are enough orders per arm that a single order does not swing the totals
The strongest temptation in the first week is to stop a test that looks like it is losing. Resist it unless something is actually broken.
Where to look
Overview — the configured price changes, and each product's totals per arm
Daily report — how the test as a whole performed each day
Product report — per day, per product, per arm, with a product filter
Orders — orders attributed to the test
If something is wrong
Pause it if you need to investigate and intend to continue; Stop it if the test is compromised. A test charging the wrong price costs more than the time invested in it. See Ending a price test, then Troubleshooting price tests.