ENServing
/lp/homepage32,180visitors on this URL2.2 Test
Test combinations on live traffic, without a developer
Server-side assignment. Bayesian readouts. No flicker across elements.
Change several elements as one package
Headline, creative, and CTA share a single assignment so visitors see a coherent offer.
Same stats model as A/B
Bayesian and frequentist views on the same traffic - so the method does not decide the answer.
1.0
Test combinations, not just single swaps
Change headline, image, and CTA together when the offer is a package. Multivariate keeps combinations on the same page and the same goal.
2.1 A/B testing
Attach combinations to live pages
Experiments run on production URLs - not a parallel sandbox that dies in a deck.
Measure real lift across the matrix
Junk traffic is filtered before probability to win is calculated, so combination readouts stay decision-grade.
2.0
How combination assignment works
Visitors are assigned server-side to a consistent combination before HTML ships. Sticky sessions and bot filtering are built in.
Decide before the response paints
Combination choice happens before HTML ships. No flicker, no client race across elements.
Pages, variants, and goals stay linked
Every combination hangs off the same page object so results roll up cleanly when you promote a winner.
3.0
Read a multivariate result
Probability to win, expected lift, and risk of loss across combinations - the same Bayesian and frequentist views as A/B.
Combination readouts you can ship on
Open a test detail to see probability to win, lift, and risk of loss for the leading package.
Promote without a fork
Winning combinations become the page. Traffic splitting stops within seconds.
FAQ
Multivariate questions
Use multivariate when you need to change several elements together and care about the interaction - for example headline plus hero plus CTA as one offer package. Use A/B when a single change is enough.
More combinations need more visitors. As a rule of thumb, plan for roughly 1,000 conversions per combination for a reliable 10% relative lift. The Bayesian readout still shows direction earlier.
No. Assignment happens server-side before the response is sent, so visitors receive the full combination in the first HTML paint.
Yes. You can run per language or across locales, with results split by market either way. Stale translations are flagged on the test detail view.
Ship your first combination test this week
Free for 14 days, no card required. Larger teams can book a walkthrough instead.