Early access. Every feature on this site carries its status. Where we are today

How we know

Every answer shows what it rests on.

No invented numbers. Here is how RevealWhy is built to decide what to say, and when to say nothing. Each part carries its status.

How sure we are

Four steps, and an honest fifth.

  1. 01

    A hunch

    One signal, or a prediction alone.

  2. 02

    Likely

    Several signals agree, or people said it in interviews.

  3. 03

    Confirmed

    Checked against your other pages.

  4. 04

    Proven

    Only a finished A/B test.

And when the data is thin: “can't tell yet”, with when there will be enough.

  • One certainty ladder everywhere: a hunch, likely, confirmed, proven Building
  • When there isn't enough data yet, we say when there will be Building
  • Findings from interviews alone are capped at "likely" Building

What every finding states

The evidence travels with the answer.

  • Every finding states how many visits it rests on, and how many were partial Building
  • Evidence on every answer, and "can't tell yet" when we can't Building
  • The 3–5 moments worth watching, at the exact second Building

Statistics, in plain words

Look as often as you like. It stays honest.

The tests we're building use statistics that stay valid however often you check them. A fix is to be compared with pages you didn't change over the same days, so a sale week doesn't look like a win; early wins shrunk toward what usually happens; and money counted only from orders you confirmed.

  • A/B tests with honest statistics Building
  • Every change checked against your other pages: worked, didn't, or can't tell yet Building
  • No inflated wins Building
  • Money only from confirmed orders Building

Attention

How attention will be measured.

The method we're building splits time on screen across what competed on that screen, and only for screens people actually had in front of them (the method). How well this matches real eye tracking will be published here: [result pending]. Before that test runs, we'll publish the result that would make us drop the claim.

Inside the product, each attention view will show how far the predicted attention agrees with how long each element was on screen for your visitors, across the whole site: a correlation, with the number of views it rests on, or “not enough data”. That compares the prediction with on-screen time, not with eye tracking. The 0.55 target shown beside it is the bar set for agreement with real eye tracking, given for reference, so a number above it on on-screen time doesn't mean the eye-tracking bar is met.

  • Attention counted only where people really looked Building
  • Attention accuracy against real eye tracking, published with its pre-registered bar Building
  • Each attention view shows how far the prediction agrees with on-screen time on your site, and how many views that rests on Building

Ask a question, and see the evidence.