ENServing
/lp/homepage32,180visitors on this URL2.1 Test
Run a real test in minutes, without a developer
Server-side splits. Bayesian readouts. No flicker.
Decide before HTML ships
Variants are chosen server-side so visitors never see a flash of the control.
Ship when probability is clear
Bayesian and frequentist views on the same data - so the method does not decide the answer.
1.0
What you can test
Element-level changes and full-page variants use the same mechanism, so you can start small and scale the ambition.
1.1 Multivariate
Compare variants where conversion happens
Attach experiments to production pages - not a parallel sandbox that dies in a deck.
Measure real lift, not raw noise
Separate real visitors from junk hits so probability to win stays decision-grade.
2.0
How traffic splitting works
Assignment happens on the server, before the page is rendered. Sticky sessions, configurable weights, bot filtering built in.
Server-side assignment by default
Variant choice happens before HTML ships. No flicker, no client race.
Readouts you can ship on
Bayesian probability to win plus frequentist significance on the same traffic.
3.0
Will this cause flicker?
No. LiftLander decides the variant server-side and sends the final HTML in the first response. There is nothing to swap.
Swap after paint
- Original page paints first, then a script rewrites it
- Visible flash while the control is replaced
- Extra render-blocking request on every page
Decide before paint
- Assignment happens before the response is sent
- Final HTML arrives already correct
- No extra request, nothing to swap
Server-side assignment means the first paint is the final page.
4.0
Reading a result
Probability to win, expected lift, and risk of loss, in one view.
Risk and lift in the same view
Open a test detail to see probability to win, expected lift, and risk of loss without switching methods.
Promote without a rebuild
Winning variants become the page. Traffic splitting stops within seconds.
Related
Keep testing in context
FAQ
Testing questions
As a rule of thumb, 1,000 conversions per variant gives a reliable read on a 10% relative lift. Below that, the Bayesian readout will still show a direction and a credible interval rather than a false verdict.
At least two full business cycles, usually 14 days, so weekday and weekend behaviour are both represented. LiftLander will not mark a test conclusive before one full cycle even if the probability threshold is met early.
Yes, with the Bayesian readout and a longer window. You can also pool traffic across languages for a directional read, then confirm per market.
No measurable impact. Assignment happens server-side before the response is sent, so there is no extra client request and no render-blocking script.
Yes. A test can run per language or across all languages of a page, with results split per market either way.
No. Variants are served on the same URL with no separate indexable addresses, and translated pages carry correct hreflang annotations.
Yes. Tests on different pages are independent. Tests on the same page are automatically made mutually exclusive unless you explicitly allow overlap.
The winning variant becomes the page, the test is archived with its full result history, and traffic splitting stops within seconds.
Ship your first test this week
Free for 14 days, no card required. Larger teams can book a walkthrough instead.