Each test on its own
Which headline wins, which image wins, which price wins — every arm scored as its own A/B test, aggregated across everything else running alongside it.
The winner isn't one change. It's the right combination.
Run many A/B tests at once and let Mtrix find the combination that wins together — then read it back per test, so you know which change earned its keep.
One experiment · every test inside it · every visitor in all of them
Most tools test one thing at a time. Mtrix runs them together.
A classic A/B test changes one thing and splits traffic evenly until it calls a winner. To learn your headline, then your hero image, then your price, you queue three tests back to back — weeks each. Mtrix runs them as one multi-armed experiment: every visitor is in all three at once, and the bandit steers traffic toward the versions pulling ahead, so you stop paying for the ones that are losing.
How the bandit fits A standard A/B test holds traffic allocation fixed to measure each version cleanly. The bandit does the opposite on purpose — it shifts traffic toward the versions that are winning, so the lift starts landing while the experiment is still running. Opt-in, off by default.
Read it as the winning combination — or as each test on its own.
Every arm is a self-contained A/B test, so Mtrix scores each one independently. Flip one toggle and the same experiment re-reads as the actual combinations across your arms — because the pair that wins together isn't always the two changes that win apart. No other testing tool lets you switch between the two.
The combination that wins together
The real combinations across your arms — so you can see the exact pairing that performs best, including the wins you only get when two changes line up.
Flip the Sorted ⇄ All Variants toggle in the report — same data, two lenses.
Why the combined read makes the call
Read per arm, Social proof wins your headline test at 96% and Lifestyle wins your image test at 92%. But the real winner is the pairing — Social proof · Lifestyle, 97% confident at $2.08 revenue per visitor, ahead of what either change earns on its own.
Two changes that each win their own test don't always win most together — and a middling change can become the winner only in the right pair. That's the decision you can't make testing one thing at a time.
More tests at once. Same honest read.
Running many arms together doesn't muddy the math. Mtrix corrects for testing several things at once, keeps a slice of traffic exploring every arm so a lucky combination can't run away with it, and tells you in plain language whether you have a winner: the chance each variant is best, and the risk if you pick it. Check it as often as you like — the read stays honest even when you look early and often. No black box, no stats degree required.
Peeking-proof Sequential testing keeps the result valid every time you open the dashboard, whether you're reading one arm or the whole combination.