The most useful A/B testing examples are real tests with a control, a variant and a number.
Below are 25 from our archive across ecommerce, SaaS, lead gen and mobile apps.
The biggest wins came from changes to the offer or the flow: Acuity Scheduling swapped a free plan for a 14-day trial (+268.14% paid sign-ups), Britannica cut one registration page (+130% revenue) and HomEquity Bank split a form into seven steps (+154% conversion rate).
Eight of the 25 lost, and we explain why.
How we picked: every example is a test from the LiftArchive database, chosen for a clear change, a reported result and a named source.
We did not run these tests; the numbers are as published by whoever ran them.
Browse the rest in the full test database or more collections on the examples hub.
What makes a good A/B test example
A good example shows what changed, which metric moved, by how much, who ran it and anything that weakens the result.
We list the caveats for each one. To run your own, start with our guide on how to run an A/B test.
All 25 examples at a glance
| # | Example | What they did | Result | Type |
|---|---|---|---|---|
| 1 | Acuity Scheduling | Swapped free-forever plan for a 14-day Pro trial | +268% paid sign-ups | SaaS |
| 2 | Encyclopaedia Britannica | Removed one page from registration | +117% free trials, +130% revenue | SaaS |
| 3 | HomEquity Bank | Split a one-page form into seven steps | +154% conversion rate | Lead gen |
| 4 | Tennis gear brand | Mega menu with category descriptions and a testimonial | +$151,192/month (estimate) | Ecommerce |
| 5 | Knife store | Added a free recipe e-book to knife orders | +48.71% revenue per visitor | Ecommerce |
| 6 | Litehouse | Removed an email popup from the top product page | +9.80% conversion rate | Ecommerce |
| 7 | Dr. Squatch | Quantity selector defaulting to 2 | Up to +54% revenue per user | Ecommerce |
| 8 | Stone Creek Coffee | "Save X% per oz" labels on larger sizes | +41% conversion rate, AOV -10% | Ecommerce |
| 9 | NuFACE | Free-shipping threshold above the shop button | +90% orders | Ecommerce |
| 10 | BioLite | Explained contact fields in checkout | +18% transactions | Ecommerce |
| 11 | Tinkoff Bank | Let applicants send passport details later | +35.8% applications | Lead gen |
| 12 | Qualicorp | Removed repeated form fields | +19.31% completed requests | Lead gen |
| 13 | Teamleader | "No credit card required" on sign-up | +12.5% sign-ups | SaaS |
| 14 | Powtoon | Concrete storage numbers on pricing | +27.9% revenue | SaaS |
| 15 | Server Density | Packaged pricing tiers | +114% revenue, sign-ups -24.96% | SaaS |
| 16 | Zumba | Cut the trial from 14 to 7 days | +8% install-to-paid | App |
| 17 | Mojo | Annual plan shown as a monthly price | +45% Brazil, +26% Mexico, +8% USA | App |
| 18 | Vegepod | Grouped product info into accordions | -8% revenue per visitor (lost) | Ecommerce |
| 19 | Gym King | Star ratings on collection cards | -10% revenue per visitor (lost) | Ecommerce |
| 20 | Greyson Clothiers | Removed visual filters | -10.5% revenue (lost) | Ecommerce |
| 21 | Bingo Card Creator | Madlibs-style sign-up form | -22% trial sign-ups (lost) | SaaS |
| 22 | Paid content site | Shorter sign-up form | Free +16%, paid 0.97% to 0.72% (lost) | SaaS |
| 23 | Seminary donation page | Value copy before the donation ask | -28% donations (lost) | Lead gen |
| 24 | BoldVoice | Removed the paywall close button | More trials, net loss after refunds | App |
| 25 | Dipsea | Web checkout only, no in-app purchase | 27.0% to 18.1% conversion (lost) | App |
The 25 A/B testing examples
Examples 1 to 17 won. Examples 18 to 25 lost, or won on one metric and lost on the one that mattered.
1. Acuity Scheduling: a 14-day Pro trial instead of a free-forever plan
What they did: The main sign-up button used to open a free-forever account with limited features, and people had to upgrade inside the app.
The variant sent them to a 14-day trial of the full $10/month Professional plan, with card details entered up front.
Why it works: A time-limited trial of the full product shows people what they would lose, and taking the card up front removes a second decision later.
Result: +268.14% paid Professional sign-ups, run by Acuity Scheduling on VWO with a 99.99% chance to beat the original. Free sign-ups were not reported.
Copy this: If your free plan is generous, test a full-feature trial of your main paid plan against it, and measure paid accounts rather than sign-ups.
2. Encyclopaedia Britannica: one page cut from registration
MarketingExperimentsWhat they did: Britannica's subscription path had one more page than it needed.
The test removed one whole page from the registration flow and changed nothing else.
Why it works: Every extra page is another place to quit.
Result: +117% free trial conversion, +61% paid subscriptions and +130% revenue, reported by MarketingExperiments.
Copy this: Count the pages between your first call to action and an account.
Test the flow with the least useful page taken out.
3. HomEquity Bank: a one-page form split into seven steps
What they did: HomEquity Bank, which sells reverse mortgages to Canadians aged 55 and over, replaced a one-page inquiry form with a seven-step form.
Why it works: A long form on one screen looks like work.
Asking one or two questions at a time makes each step feel small, and people who start tend to finish.
Result: +154% conversion rate, reported by HomEquity Bank.
A later test cut it from seven steps to four and won again.
Copy this: If your lead form has more than five fields on one page, test a multi-step version that opens with the easiest question.
4. Tennis gear brand: a mega menu with descriptions and a testimonial
What they did: Oliver Kenyon tested three mega menus for a large tennis and racket gear store.
The winner added one line under each category and a customer testimonial, with no images. A version with thumbnails lost to it.
Why it works: A line of plain description tells shoppers where to go faster than a picture, and the testimonial adds proof inside the menu.
Result: +$151,192 a month (an estimate, or $0.77 per session), run by Oliver Kenyon. No sample size was shared.
Copy this: Write one line under each menu category saying what's in it, and test that before you add thumbnails.
5. Knife store: a free BBQ recipe e-book with every order
What they did: A knife store added a "Free E-book" label and cover to the product gallery, and emailed buyers a BBQ recipe e-book after ordering.
Why it works: A gift that matches how people use the product adds value without cutting the price, and costs almost nothing to deliver.
Result: +48.71% revenue per visitor, run by Oliver Kenyon.
It ran around a seasonal sale and was the only number shared.
Copy this: Before you test a bigger discount, test a low-cost digital bonus tied to how people use the product.
6. Litehouse: removing the email popup from the best-selling product page
What they did: Litehouse, a solar lighting brand, removed the email popup that opened on every visit to its top product page.
Why it works: Someone on a best-seller page is close to buying, and a popup gets in the way.
Pre-test data showed 6,295 popup dismissals per 100,000 views.
Result: +9.80% conversion rate and +7.46% revenue per user, run by Oliver Kenyon's agency Conversionwise.
Confidence was 93.1%, below the usual 95% bar.
Copy this: Turn your popup off on your highest-intent pages and test it, counting both orders and lost email sign-ups.
7. Dr. Squatch: a quantity selector that defaults to two soaps
SplitBaseWhat they did: Dr. Squatch's soap pages had no quantity field.
SplitBase added one set to 2 by default.
Why it works: Most people keep the default, and for a product people use up, two matches what many would buy anyway.
Result: Up to +54% revenue per user for soap buyers, run by SplitBase.
"Up to" covers several versions of the test, not one clean run.
Copy this: If your product is consumable, test a default quantity of two against one, and watch return rates as well as revenue.
8. Stone Creek Coffee: "Save X% per oz" on larger bags
What they did: Blend Commerce added labels to the 2 lb and 5 lb sizes showing the saving per ounce.
Why it works: Shoppers don't do unit-price math.
A percentage saving makes the bigger size feel like the smart buy.
Result: +41% conversion rate, +26% revenue per visitor and +117% subscription orders per visitor, run by Blend Commerce. Average order value fell 10%.
Copy this: If you sell more than one size, show the per-unit saving on the larger ones.
9. NuFACE: "Free shipping over $75!" above the shop button
What they did: The page had a "Shop NuFace" button and no shipping offer.
The variant put "Free shipping over $75!" right above the button.
Why it works: Stating the threshold before people shop removes the shipping-cost worry and gives them a target to spend toward.
Result: +90% orders and +7.32% average order value at 96% confidence, run by Red Door Interactive on VWO.
This tested a new offer, not only a new message.
Copy this: Put your free-shipping threshold next to the main buy button, not only in the site banner.
10. BioLite: explaining why checkout asks for email and phone
SurefootWhat they did: Surefoot moved BioLite's email and phone fields into a labeled Contact Information block below the shipping address, with a line explaining how each would be used.
Why it works: People hesitate to give contact details without knowing why. A short reason removes the worry.
Result: +18% transactions, 14% less drop-off from step 1 to step 2 and +20% email capture, run by Surefoot.
Copy this: Add one line of helper text under every checkout field that might feel personal.
11. Tinkoff Bank: "Don't remember passport details?"
What they did: Tinkoff added a checkbox to its card application letting applicants send passport details later by phone or email.
Why it works: A field people can't answer on the spot ends the session.
Result: +35.8% application form submissions, run by Tinkoff Bank on VWO with more than 265,000 visitors over 23 days.
Copy this: Find the field people are least likely to know by heart and offer a "send it later" option.
12. Qualicorp: stop asking for the same details twice
What they did: Qualicorp's quote form asked again for name and date of birth on step 2. The variant dropped the repeats.
Why it works: Being asked twice makes a form feel broken and longer than it is.
Result: +19.31% visits to the request-received page and +24.68% Request Contact clicks over 14 days, run by VWO.
Copy this: Walk through your own form and delete every field that asks for something you already have.
13. Teamleader: "No credit card required" on the trial sign-up
What they did: Teamleader's free trial page didn't say whether a card was needed.
The variant added a "No credit card required" message on the Netherlands and Belgium sites.
Why it works: When unsure, people assume they'll be charged. Four words remove that doubt.
Result: +12.5% sign-ups: 48.7% to 54.8% in the Netherlands and 33% to 38.6% in Belgium, run by Teamleader on VWO.
Copy this: If your trial doesn't need a card, say so right next to the sign-up button.
14. Powtoon: "10 GB" instead of "unlimited" storage
What they did: Powtoon's pricing page showed 2 GB of storage on Pro and unlimited on Business.
VWO tested three versions, and the winner showed Business at 10 GB against Pro at 2 GB.
Why it works: A concrete number is easier to compare than "unlimited", and the gap between 2 and 10 makes the upgrade clear.
Result: +27.9% revenue, run by VWO.
Copy this: Replace vague plan limits with concrete numbers and test whether the bigger plan sells better.
15. Server Density: packaged pricing instead of per-server
What they did: Server Density replaced per-server pricing with packaged tiers starting at US$99/month.
Why it works: Packages save buyers from working out a price and set a higher floor, at the cost of fewer trials.
Result: +114% revenue ($394 to $833), run by Server Density on VWO, but free sign-ups fell 24.96% (6.92% to 5.19%).
With only 20 vs 15 paid conversions, the win is fragile.
Copy this: If buyers have to configure their own price, test two or three named packages, and track revenue, not trial starts.
16. Zumba: a 7-day free trial instead of 14 days
What they did: Zumba's app offered a 14-day free trial. The variant cut it to 7 days.
Why it works: A shorter trial brings the payment decision forward while people are still engaged.
Result: +8% install-to-paid conversion, reported by Zumba.
Copy this: Test a shorter trial and measure install-to-paid, not trial starts.
A wellness app test in our archive saw the same pattern: the 14-day trial got slightly more trial starts but fewer paid conversions.
17. Mojo: the annual plan shown as a monthly price
What they did: Mojo's paywall showed the annual plan as one yearly price.
The variant showed it as a per-month price, billed annually.
Why it works: A monthly figure feels smaller and is easier to compare. More people started trials and picked annual.
Result: +45% in Brazil, +26% in Mexico and +8% in the USA in new revenue per paywall impression, reported by Mojo.
Copy this: Show your annual plan's price per month and test it market by market, since the effect varied a lot by country.
18. Vegepod: product page accordions
What they did: Vegepod's product page showed all product information in full. Blend Commerce grouped it into accordions.
Why it lost: Mobile shoppers scroll rather than open sections, so hidden content went unread.
Result: Lost overall: -8% revenue per visitor and -2% conversion rate.
Desktop won (+26% revenue per visitor, +15% conversion rate) while mobile lost (-24% conversion rate, -28% revenue per visitor).
Copy this: Always split results by device before you call a test.
One number can hide a winner and a loser.
19. Gym King: star ratings on collection cards
What they did: Gym King added star ratings to the product cards on its collection pages. The test ran on Shoplift.
Why it lost: Only 60-70% of products had reviews, so many cards showed empty ratings.
An empty rating looks worse than no rating.
Result: Lost: -10% revenue per visitor. Gym King stopped it after 6 days, about £10,000 behind.
Copy this: Check your review coverage before showing ratings on listing pages, and hide the stars on products with none.
20. Greyson Clothiers: removing visual collection filters
What they did: Greyson removed the visual filters it had added to collection pages, to check they were earning their place.
Why it lost: The filters helped people narrow the range quickly.
Result: Lost: -5.3% engagement and -10.5% revenue for the version without filters, run by Greyson Clothiers on VWO.
The loss confirms the filters were worth keeping.
Copy this: Run removal tests on features you added without a test. A loss tells you to keep them.
21. Bingo Card Creator: a "madlibs" sign-up form
What they did: Patrick McKenzie tested a madlibs-style form, with the fields inside a sentence, against a standard form on Bingo Card Creator.
Why it lost: A sentence with blanks is harder to scan than a plain form.
Result: Lost: 27.55% vs 21.73% trial sign-ups, about -22% for the madlibs form, with 736 participants over 34 hours.
Copy this: Don't copy a design trend because another site uses it. Test it on your own traffic.
22. Paid content site: a shorter sign-up form that cost paid members
What they did: MarketingExperiments tested a shorter first sign-up page for a paid-content membership site: only name, email and password, new button text and a cleaner layout.
Why it lost: Easier sign-up brought in more free members but fewer payers.
Result: Free sign-ups rose about 16%, but paid memberships fell from 0.97% to 0.72% (33 to 26) and revenue dropped 6.1% in the test week. The counts are small.
Copy this: Pick your main metric before the test, and make it the one closest to revenue.
23. Seminary donation page: value copy before the ask
What they did: NextAfter tested a donation page that first explained the seminary's value and quoted well-known figures before asking for a gift. The control went straight to the ask.
Why it lost: People on a donation page have already decided. Extra persuasion slows them down.
Result: Lost: -28% donations, run by NextAfter and reported as statistically significant.
Copy this: Match the amount of copy to how motivated visitors are when they arrive. On high-intent pages, get to the form.
24. BoldVoice: a hard paywall with no close button
What they did: BoldVoice removed the close (X) button from its onboarding paywall, so users had to start a trial to continue.
Why it lost: Forcing a trial brings in people who never meant to pay. They ask for refunds later.
Result: Lost: more trial starts, but refunds rose by more than the gain, reported by BoldVoice. No figures were given.
Copy this: Judge paywall tests on revenue after refunds, measured once cohorts have had time to renew or cancel.
25. Dipsea: web checkout only instead of in-app purchase
RevenueCatWhat they did: RevenueCat ran a test with Dipsea where every subscribe button on the paywall sent users to a web checkout, with no Apple in-app purchase option.
Why it lost: Leaving the app to pay added friction that the lower fee didn't make up for.
Result: Lost: initial conversion fell from 27.0% to 18.1% and paid conversion from 6.3% to 5.3%.
Net proceeds per user went from about $2.09 to $1.96 at a 30% store fee.
Copy this: If you move app payments to the web, keep in-app purchase as an option and compare net proceeds, not fees saved.
Patterns across the winners
Removing a step beat adding a feature
Britannica cut a page, Qualicorp cut repeated fields, Litehouse cut a popup and Tinkoff let people skip a field for now.
Look for the step or field your visitors would happily lose.
The offer moved more than the layout
The largest lifts came from changing what people get: a full trial instead of a free plan at Acuity, a free-shipping threshold at NuFACE, packaged tiers at Server Density, a free e-book at the knife store.
Colors and layouts are the smallest lever. Our price testing guide covers how to test offers safely.
Answering a doubt at the right moment
BioLite explained why it wanted a phone number, Teamleader said no card was needed and the tennis store put a testimonial inside the menu.
Each one put an answer next to the moment the question comes up.
Framing the number, not changing it
Stone Creek Coffee showed the saving per ounce, Mojo showed the annual plan per month and Powtoon swapped "unlimited" for "10 GB".
The price stayed the same; the way it read changed.
Examples that lost (what not to copy)
The eight losses fall into three groups, and each one is a common mistake.
- The wrong metric. The paid content site and BoldVoice both "won" on sign-ups or trial starts and lost on money. Pick the metric closest to revenue before you start, and read our guide to A/B testing metrics.
- Averages hiding segments. Vegepod's accordions won on desktop and lost on mobile.
- Copying a trend without checking your own data. Gym King's star ratings failed because many products had no reviews, and the madlibs form was a design fad. Dipsea's web checkout saved fees and lost buyers.
How to test this on your own site
- Start from a problem, not an example. Find where people drop off, then pick the example that fits. Write it down as a hypothesis.
- Check you have the traffic. Use the sample size calculator before you start. If the answer is months, see our guide to testing with low traffic.
- Pick one main metric. For ecommerce that is usually revenue per visitor; for SaaS, paid accounts; for lead gen, qualified leads; for apps, revenue after refunds.
- Run full weeks. Our guide on how long to run an A/B test explains why.
- Read it properly. Check statistical significance, then split by device and new vs returning visitors.
For the software side, see our list of A/B testing tools.
If you would rather have someone plan and run tests for you, Oliver Kenyon ran several of the winning ecommerce tests above, including the mega menu and the popup removal.
You can see more of their work on their practitioner page.
More real results
A few more tests from the archive that follow the same patterns: fewer steps, clearer progress and answers placed where people need them.
Blend CommerceBlend CommerceWatch out for
- Headline numbers without a sample size. Several results above, including the mega menu's dollar figure, came without sample size or test length.
- Bundled changes. When several things change at once, you know the package worked but not which part.
- The winner's curse. Big lifts from small tests tend to shrink after rollout.