LuriaDocs
Dashboard

Lift and confidence, in plain English

What the lift number and the confidence level on your dashboard mean, why they move early on, and why Luria doesn't call winners fast.

plain textupdated 2026-08-19

Two numbers sit next to every test in Luria: lift and confidence. You don't need a statistics background to use them. You need two sentences and a few habits.

Lift

Lift is how much better the variant converts than your original, over the same period, on the same traffic.

If your original turned 2.0% of visitors into buyers and the variant turned 2.3% into buyers, the variant's lift is 15%. Not 0.3 percentage points, 15%: it converted 15% more of the people who saw it.

Three things make that number trustworthy:

  • Same period. Original and variant are shown at the same time, so a sale, a holiday, or a viral post hits both equally.
  • Same traffic. Visitors are split between them at random. Meta and Google and email visitors land on both.
  • Same goal. Lift is measured against the goal you set, purchases or leads, not clicks or scroll depth.

Lift can be negative. A variant with -8% lift is worse than your original, and Luria retires it.

Confidence

Confidence is how sure Luria is that the lift is real and not luck.

The dashboard shows it as a simple level rather than a decimal:

LevelWhat it meansWhat to do
EarlyThere's a number, but it could easily flip.Read it as "we're looking at this." Don't quote it.
LikelyProbably real. The direction is unlikely to reverse, the size may still shift.Reasonable to mention; keep the caveat.
ConfirmedLuria would bet on it. It has become the winner and gets most of the traffic.Quote it. Reuse the copy elsewhere.

Confidence is about the evidence, not the size of the effect. A small lift can be Confirmed; a huge lift can be Early.

Why early numbers swing

Flip a coin 10 times and you might get 7 heads. Flip it 1,000 times and you'll get very close to 500. Conversion rates behave the same way. In a variant's first few days, a couple of lucky purchases can make it look 40% better, and a quiet afternoon can make it look 20% worse. Neither is the truth; the truth shows up as the sample grows.

So if you see a variant at +35% on day two and +9% on day nine, nothing went wrong. The day-two number was noise; the day-nine number is closer to real.

Why Luria doesn't call winners early

It's tempting to ship the variant that's ahead on day two. Tools that do this create a specific kind of damage: they ship a "winner" that was never better, you redesign your email around it, and three weeks later your conversion rate is flat and you don't know why.

Luria waits until confidence reaches Confirmed before it calls a winner and moves the bulk of traffic to it. While it waits, it isn't idle: the variant that's ahead already gets a larger share of traffic, so you capture the upside in the meantime without locking in a mistake.

Why low traffic means longer

Confidence comes from conversions, not from calendar days. A store doing 3,000 sessions a month at a 2% conversion rate produces about 60 purchases a month to learn from. A store doing 600 sessions produces about 12. The second store will take roughly five times as long to reach the same confidence on the same-sized lift.

That's why the training timeline is written for ~3,000+ sessions a month and stretches proportionally below that. If your traffic is low, the honest expectation is fewer winners per month, each one still real. Ways to speed it up are in Improve training.

Why 2% Confirmed beats 20% Early

A 2% lift at Confirmed is money you can count on every month from now on. A 20% lift at Early is a number that might be 20%, might be 2%, and might be -5%. One of these can go in your forecast. The other cannot.

This matters when you compare Luria to your own past A/B tests or to other tools. A dashboard full of big early lifts feels exciting and is mostly noise. A dashboard with a few modest confirmed lifts that compound over months is what actually changes your revenue. Luria is built for the second kind.

FAQ

Next steps

On this page