Revenue per visitor
The number that actually pays for traffic. Mtrix runs the statistics on it per variant, so you never ship a version that converts more and earns less.
Prove which version makes more money.
Run A/B, multivariate, and split-URL tests on the store you already run — no rebuild, no data team. Every variant reads out in the numbers that pay the bills, and when the result is sure, Mtrix calls the winner in plain language.
Set it up, run it, call it, ship it — all in one place.
Pick the page, build the variant, choose who sees it, and launch — the same afternoon you thought of it. Mtrix runs the statistics and calls the winner in plain language. No warehouse to wire up, no analyst to wait on, no developer ticket to file.
Same object, the whole way — no tools to stitch together.
Test the things that actually move revenue.
Not just button colors. Test the offer, the product page, the cart, the checkout — four test types that cover every surface between a click and a sale.
A/B test
Pit one version of an element against another and let the better one win.
A store example
A sticky add-to-cart bar vs. none on your product page.
A higher click-through rate doesn't pay your ad bill. Revenue does.
Mtrix computes every metric per variant, then runs the statistics on the three that decide the business: revenue per visitor, conversion rate, and add-to-cart. So a "winner" here is the version that makes more money — not the one that gets more clicks.
Conversion rate
The classic, done right: significance you can trust, read in plain language.
Add-to-cart rate
The earliest honest signal of intent, tested per variant so you can read a result before checkout volume catches up.
One report, from first read to rollout.
Every experiment lands on one report: each variant read out in store metrics, the chance it's the best, a traffic split you can change while it runs, and the rollout call when it's decided. Here's how you work it, top to bottom.
01
Read the numbers like a pro
Every variant reports in the numbers you already run the store on — visitors, conversion, add-to-cart, checkout, average order, revenue. Scroll across the report and they're all there, per variant. No "engagement score" to decode: the columns are your P&L.
02
Know the chance it's the best
Flip the report to its significance view and every variant gets a verdict in plain English: the chance it beats control, the risk if you pick it anyway. 96% reads as ship it; 8% reads as stop hoping. No p-values to interpret.
03
Change the split mid-flight
Allocation isn't locked at launch. Type a new split, hit save, done — feed the variant that's pulling ahead, starve the one that's losing money, and the test keeps running. No restart, no lost data.
04
Decide the rollout
When the winner is real, the report says so — and shipping it is one click from the same screen. Roll the winner out to 100% of traffic, watch the revenue line move, and start the next test.
01
Read the numbers like a pro
Every variant reports in the numbers you already run the store on — visitors, conversion, add-to-cart, checkout, average order, revenue. Scroll across the report and they're all there, per variant. No "engagement score" to decode: the columns are your P&L.
-
01
Read the numbers like a pro
Every variant reports in the numbers you already run the store on — visitors, conversion, add-to-cart, checkout, average order, revenue. Scroll across the report and they're all there, per variant. No "engagement score" to decode: the columns are your P&L.
-
02
Know the chance it's the best
Flip the report to its significance view and every variant gets a verdict in plain English: the chance it beats control, the risk if you pick it anyway. 96% reads as ship it; 8% reads as stop hoping. No p-values to interpret.
-
03
Change the split mid-flight
Allocation isn't locked at launch. Type a new split, hit save, done — feed the variant that's pulling ahead, starve the one that's losing money, and the test keeps running. No restart, no lost data.
-
04
Decide the rollout
When the winner is real, the report says so — and shipping it is one click from the same screen. Roll the winner out to 100% of traffic, watch the revenue line move, and start the next test.
Stop losing sales to the losing version while you wait.
Split traffic evenly and you make half your shoppers keep hitting the worse version for the whole length of the test — a tax you pay in lost sales just to learn something. Turn on the multi-armed bandit and Mtrix shifts more traffic to the version that's winning as the data comes in, so you capture the lift while the test is still running.
Mtrix tilts traffic toward the leading version as the data arrives, so less of your traffic is wasted on the loser — and the lift starts landing before the test ends.
Set any variant's share of traffic yourself and change it live — dial a risky variant down to 10%, or push a clear winner to the front, without relaunching.
Opt-in. Off by default for a standard even split.
Traffic share over the test
Day 1 · 50 / 50
Bandit on — traffic follows the winner
Day 14 · 80 / 20
Set it by hand — live
Control 50% · Variant B 50%
Even split — 50% of your traffic still sees the version that's losing.
A winner overall can be a loser for paid traffic. Now you can see it.
Slice any chart in your experiment 27 ways — by UTM campaign, by referrer, by country and city, by device, and by the other tests a shopper was already in. So you don't just learn that a version won — you learn who it won with, and where it quietly lost. That's the read most brands need a data scientist and a warehouse to get. Here it's a dropdown.
Break down by
- + landing page, new vs. returning, browser… 27 in all
All traffic
Looks like a tie
Broken down by Other experiments seen
Also in: PDP Price Test
B loses
Not in any other test
B wins
The "winner" only won where it collided with another live test.
See why the winner won — not just that it won.
A test result on its own is a number with no story. On Mtrix the experiment lives next to everything else: replay the real sessions on the winning variant, see how fast each version loaded, catch the error that sank the loser — all from the same platform. So you don't just ship the winner, you understand why it won, and you know what to test next.
A winner, with the reason attached.