A/B tests
Declaring the variant shown, the exposure event and the variant written on the payment.
Declaring the variant
Create the test in Distribution → Events: its id (slug, 1 to 40 a-z 0-9 _ characters) and its 2 to 4 variants (1 to 20 characters) never change afterwards. Draw the variant yourself, keep it for the visitor, then declare it before showing it:
window.firstmrr?.variant("pricing_page", variant)The exposure (ab:exposure, or exposure on the server) says a visitor saw a variant; then every event of that visitor carries their assignments. A visitor seen with two variants of the same test on the same day is left out of the results.
On the payment
Write fm_variant = "pricing_page:b" beside the other provenance keys; several tests are joined by a comma (4 at most). A malformed value is ignored as a whole, never half read.
Results (visitors, signups, payers, revenue per variant) only show a winner from 100 visitors per variant, once its probability of being best exceeds 95% corrected for the number of variants (97.5% for two).