Of 1,724 real A/B test results in LiftArchive, 1,447 (84%) won, 112 (6%) lost and 165 (10%) came out flat.
That is far higher than the win rate companies measure across every test they run, which is usually 10% to 33%.
The gap is publication bias: practitioners share their wins more often than their losses, so read the numbers below as “what public tests look like”, not your odds on the next test.
Method and sample
The data covers 1,724 tests from 391 practitioners, agencies, testing tools and companies, each credited and linked to its original source.
Ideas without a result are left out.
Flat means the test was shared as no clear difference or not significant.
The numbers update every time the database changes; this version is from November 2026.
Win, loss and flat split by vertical
| Vertical | Tests | Won | Lost | Flat |
|---|---|---|---|---|
| Ecommerce | 1219 | 82% | 6% | 12% |
| Lead-gen | 220 | 94% | 3% | 3% |
| SaaS | 205 | 84% | 11% | 5% |
| Mobile apps | 80 | 88% | 12% | 0% |
Sources for the company win rates are on our A/B testing statistics page (Kohavi, Deng and Vermeer 2022; Optimizely 2023).
Typical lift size
When a shared test won, the median reported lift was +12.5% revenue per visitor (n = 357), +12.6% conversion rate (n = 696) and +6.2% average order value (n = 151).
Medians are used because a few huge lifts would drag an average upward.
Win rate by page type
Page types with at least 15 shared tests:
| Page type | Tests | Won |
|---|---|---|
| Product page (Ecommerce) | 492 | 80% |
| Collection page (Ecommerce) | 143 | 73% |
| Site-wide (Ecommerce) | 141 | 88% |
| Homepage (Ecommerce) | 133 | 81% |
| Cart (Ecommerce) | 96 | 85% |
| Landing page (Lead-gen) | 94 | 96% |
| Paywall (Mobile apps) | 71 | 86% |
| Landing page (Ecommerce) | 67 | 90% |
| Navigation & menu (Ecommerce) | 60 | 77% |
| Checkout (Ecommerce) | 58 | 91% |
| Landing page (SaaS) | 51 | 94% |
| Lead form (Lead-gen) | 43 | 93% |
| Homepage (SaaS) | 30 | 93% |
| Pricing page (SaaS) | 26 | 81% |
Win rate by element
Ecommerce elements with at least 15 shared tests, highest win rate first:
Limits: what these numbers can't tell you
Shared tests are a biased sample: wins get posted, losses mostly don't.
Many tests report one headline metric, so a conversion-rate win may hide a revenue loss.
Sample sizes and run times vary, and few posts report both.
Use this page to pick what to test, then let your own test decide.
Cite this data
Free to quote with a link back to this page.
LiftArchive, “A/B test win rate: what share of 1,724 real tests won, lost or were flat”, November 2026. https://liftarchive.com/research/ab-test-win-rate/