Plan a test
A/B test sample size calculator
Visitors each version needs, and how many days that takes at your traffic.
A/B testing
Ecommerce A/B testing: test ideas by page, from real store tests
Ecommerce A/B testing tells you which change to your store actually brings more orders. This guide lists 35 test ideas by page type, each with the reason it can work, the metrics worth measuring, and 14 split tests Convertica ran on real stores, with the metric, length and confidence for every result.
What is ecommerce A/B testing?
Ecommerce A/B testing is a controlled experiment on an online store. Shoppers are split at random between the current version of a page, the control, and a changed version, the variant. Both run at the same time, so the change is the only systematic difference, and the two groups are compared on a store goal such as orders or revenue per visitor.
The method is the same as any A/B test, set out in full in the A/B testing guide. What makes a store different is what you measure and what gets in the way:
- The goal is money, not clicks. A change that brings more clicks to the cart can still lose orders. Store tests should be judged on order completion rate or revenue per visitor.
- Desktop and mobile often disagree. In eight of the 14 Convertica store tests below, one device moved and the other did not, so a combined number can hide a loss or a win.
- Promotions and seasons move the baseline. A sale, a holiday or a big email changes who visits. A test that overlaps one measures those shoppers, not your normal ones.
- The purchase is a funnel. A change on the homepage has to survive four or five more steps before it becomes an order, so changes near the order tend to show up in sales sooner.
Which metrics should an ecommerce A/B test measure?
An ecommerce A/B test should be decided on a metric that pays: order completion rate, which is orders divided by visitors, or revenue per visitor, which is revenue divided by visitors and so captures both how many people buy and how much they spend. Add to cart and checkout starts are useful secondary metrics that show where a change works.
Choose the primary metric before launch and write it down. Most Convertica store tests on this page report order completion rate, revenue per visitor or both. The ecommerce metrics guide shows how each one is calculated.
| Metric | How it is calculated | What it tells you | Use it as |
|---|---|---|---|
| Order completion rate | Orders / visitors in the test | Whether more shoppers buy | Primary |
| Revenue per visitor | Revenue / visitors in the test | Whether the store earns more per shopper, counting order value too | Primary, especially for price, bundle and shipping tests |
| Average order value | Revenue / orders | Whether buyers spend more per order | Secondary: it can rise while orders fall |
| Add to cart rate | Visitors who add to cart / visitors | Whether the product page persuades | Secondary, or primary only on low traffic with care |
| Checkout start rate | Visitors who start checkout / visitors | Whether the cart moves people on | Secondary |
| Subscription rate | Subscription orders / visitors | Whether a subscribe offer lands | Primary for subscription stores |
Ecommerce A/B testing ideas by page
The best ecommerce A/B testing ideas remove a doubt or a step at the moment a shopper decides. Below are 35 ideas grouped by page type: homepage, navigation and collection pages, product page, cart, checkout, and landing pages and offers. Each comes with the reason it can work and, where Convertica has tested that kind of change on a real store, a link to the case study.
Treat them as ideas to check against your own evidence, not changes to copy. A test that won on one store can lose on yours, because your shoppers, products and traffic differ. Analytics, session recordings, heatmaps and a free CRO audit tell you which of these problems your store actually has.
| Page type | What shoppers decide there | Ideas below | Convertica tests |
|---|---|---|---|
| Homepage | Whether this store is worth their time and where to go next | 5 | 2 |
| Navigation and collection pages | Which product to look at | 5 | 1 |
| Product page | Whether to buy this product, in which option, at this price | 10 | 8 |
| Cart page | Whether to carry on to checkout | 5 | 2 |
| Checkout | Whether to finish the order | 5 | 0 |
| Landing pages and offers | Whether the ad's promise holds and which offer to take | 5 | 1 |
Homepage A/B test ideas
Homepage A/B tests work best when they answer the first questions a new shopper has: is this store trustworthy, does it sell what I want, and where do I click next. The homepage is often a store's busiest page, which makes tests finish faster, but it sits furthest from the order, so judge it on orders and revenue, not clicks.
| Test idea | Why it can work | Tested by Convertica |
|---|---|---|
| A trust strip under the hero: shipping, reviews and payment security | New visitors look for reasons to trust a store they do not know before they browse | Homepage trust signals: +57.9% revenue per visitor on desktop in a 12-day test, at 93% confidence (below the usual 95%) |
| A visible search bar instead of a search icon | Shoppers who know what they want go straight to it instead of browsing menus | Search bar: +71.9% order completion rate on desktop at 98% confidence; mobile did not change; test length not published |
| A hero headline that names what you sell and why it is better | A slogan makes visitors work out what the store is; a plain statement lets them decide in seconds | Not tested in a published case study |
| Shortcuts to the top categories above the fold | Fewer clicks between landing and a product page, especially on mobile | Not tested in a published case study |
| The announcement bar: free shipping threshold against a discount code | The bar is seen on every page, so the offer it carries can change basket size or conversion across the store | Not tested in a published case study |
The homepage trust test is also a lesson in reading results: mobile reached 97% confidence, with order completion rate up 68.9% and revenue per visitor up 40.9%, while the desktop result sat at 93%. A result below the confidence level you set is a signal worth acting on carefully, not proof. More homepage patterns are in ecommerce homepage best practices.
Navigation and collection page A/B test ideas
Navigation and collection page tests help shoppers get from a broad category to the right product with less effort. The ideas below shorten the path, show the information shoppers compare on, and let a decided shopper act sooner. Measure them on orders, because a better collection page should send more people to products they then buy.
| Test idea | Why it can work | Tested by Convertica |
|---|---|---|
| Search on collection pages and the blog, not only the homepage | Shoppers who land deep in the store from search or an article can still find a product fast | Part of the search bar test above |
| A mobile menu that shows top categories at once | Nested menus hide the range; shoppers who cannot see a category assume you do not sell it | Not tested in a published case study |
| Filters shown on mobile instead of behind a button | Size, color and price filters cut a long list to what a shopper can buy | Not tested in a published case study |
| Product cards with rating, price and shipping information | Shoppers compare on these; showing them on the card saves a click to each product page | Not tested in a published case study |
| Default sort: best sellers against newest | The first products seen get most of the clicks, so the order of the grid changes what sells | Not tested in a published case study |
More on menus, search and filters is in ecommerce navigation best practices.
Product page A/B test ideas
Product page A/B tests are where most of Convertica's store tests have run, because the product page is where a shopper decides to buy. The strongest ideas answer the questions that stop a purchase at that moment: is it any good, which option do I need, what will it really cost, and when will it arrive.
| Test idea | Why it can work | Tested by Convertica |
|---|---|---|
| Reviews moved from the bottom of the page to just below the product | Proof seen while deciding does more than proof found after scrolling past the button | Reviews section: +42.38% orders (orders per visitor) for a bedding store on mobile, and +20.34% completed orders for a jewelry store on desktop, in 2-week tests at 98% confidence |
| Clear variant selectors: labeled options and visual swatches | Shoppers who cannot tell which option they are choosing, or see "No Selection", stop rather than guess | Variant selectors: +144.8% conversion rate in a three-week test at 98% confidence |
| A sticky Add to Cart bar on a cleaner page | The button stays in reach while shoppers read details and reviews | Product page redesign: +56% order completion rate on desktop at 98% confidence; mobile was neutral; test length not published |
| A free shipping badge beside the price | Shipping cost is a common reason to leave late in checkout; saying it is included at the price removes the doubt early | Free shipping badge: +60.3% revenue per visitor and +27% order completion rate on desktop; mobile showed no significant change; test length and confidence level not published |
| A free shipping threshold message: "you qualify" or how much more is needed | It answers the cost question and gives shoppers a reason to add an item | Free shipping threshold: +175.4% revenue per visitor and +88.3% order completion rate on desktop at 95% confidence; mobile showed no significant change; test length not published |
| Shipping, warranty and payment answers in pop-ups on the page | Shoppers get answers without leaving the page and losing their place | Factory to Home: +12.7% revenue per visitor and +18.1% order completion rate, desktop and mobile combined (96% confidence on desktop); test length not published |
| Packages shown as a swipeable comparison on mobile instead of stacked | Stacked options on a phone cannot be compared side by side, so shoppers scroll back and forth or give up | Mobile pricing table: +36% revenue per visitor and +43.9% order completion rate on mobile at 98% confidence; test length not published |
| Trust signals and an auto-ship offer on the product page | Repeat-purchase products can sell more subscriptions when the offer and the reasons to trust it are seen at the decision | JustThrive: +101.82% revenue, cumulative across two sequential split tests, 7 December to 24 March |
| An estimated delivery date under the price | A date is easier to act on than a shipping policy, and matters most for gifts and urgent buys | Not tested in a published case study |
| Gallery thumbnails shown instead of dots | Shoppers can see that more photos exist and jump to the one they need, such as size or detail | Not tested in a published case study |
What the product page tests have in common
Most of these winners did not add anything new. They moved information shoppers needed, such as reviews, shipping cost or the right option, to the point where the decision is made. That is a useful filter for your own ideas: if a shopper has to scroll, click away or guess to get an answer, the fix is a test candidate. The product page optimization guide covers the full page.
Cart page A/B test ideas
Cart page A/B tests aim to move more shoppers from a full cart to a started checkout. The cart is where the total first appears and where doubts about returns, delivery and payment surface, so the best ideas show those answers next to the checkout button and remove steps for shoppers who have already decided.
| Test idea | Why it can work | Tested by Convertica |
|---|---|---|
| Shipping and refund policy badges under the checkout button | The worry about returns comes just before paying; a badge with a pop-up answers it without leaving the cart | Policy visibility: +21.8% order completion rate on desktop at 97% confidence; mobile was neutral; test length not published |
| Add to Cart that goes straight to the cart | When most orders hold one item, a stop on the product page or a drawer is an extra step | Homewares store: +47.7% conversion rate on mobile in a 19-day test at 98% significance; +26.9% conversions overall with the winner rolled out to mobile only |
| A free shipping progress bar in the cart | Shows how far the shopper is from the threshold, which can raise order value | Not tested in a published case study |
| Express checkout buttons in the cart | Saved payment details skip the forms; on some stores they also distract, which is why it is worth testing both ways | Not tested in a published case study |
| Cross-sell placement: in the cart against after the order | A suggestion in the cart can raise order value or pull attention from the checkout button | Not tested in a published case study |
The homewares test shows why a device split matters: the store rolled out the winner to mobile only. More cart ideas are in the shopping cart optimization guide, and the cart abandonment rate guide shows how to measure the drop.
Checkout A/B test ideas
Checkout A/B tests reduce the effort and surprises between a started checkout and a confirmed order. Shoppers here have already chosen, so the ideas below cut fields, show the full cost early and keep reassurance next to the pay button. Some platforms limit what you can change in checkout, so check what your platform and testing tool allow first.
| Test idea | Why it can work | Tested by Convertica |
|---|---|---|
| Guest checkout offered before account creation | A forced sign-up is extra work at the moment a shopper wants to pay | Not tested in a published case study |
| Fewer fields, with address lookup | Every field is a chance to stop, especially on a phone keyboard | Not tested in a published case study |
| Shipping cost and taxes shown before the payment step | A total that grows at the last step feels like a trick and sends shoppers away | Not tested in a published case study |
| Accepted payment methods shown at the start of checkout | Shoppers who do not see their preferred method may leave before they reach it | Not tested in a published case study |
| An order summary and returns line beside the pay button | Shoppers can confirm what they are buying and their way out without going back | Not tested in a published case study |
Convertica has no published checkout split test yet, so these ideas rest on conversion reasoning rather than our own results. The checkout optimization guide covers the full flow.
Landing page and offer A/B test ideas
Landing page and offer tests decide whether paid traffic turns into orders and which offer a shopper takes. Shoppers from an ad arrive with one promise in mind, so the page has to keep it and keep the next step in reach. Offer tests change what is sold, such as bundles or how a price is shown, so measure them on revenue per visitor.
| Test idea | Why it can work | Tested by Convertica |
|---|---|---|
| A call to action bar fixed to the bottom of the screen on mobile | On a long landing page the buy button scrolls away; a fixed bar keeps the next step one tap away | Sticky CTA: +252.9% rate of order completion on mobile in a 20-day test at 98% confidence; desktop did not change significantly, so it went live on mobile only |
| A headline that repeats the ad's promise | When the page matches the click, shoppers know they are in the right place | Not tested in a published case study |
| Bundles or multi-packs against single units | A bundle can raise order value; too many choices can also slow the decision | Not tested in a published case study |
| Price shown per unit or per dose | A smaller unit price can make a large pack easier to justify | Not tested in a published case study |
| An anchor price or the most popular package marked | A reference point and a suggested choice make comparing packages quicker | Not tested in a published case study |
Price and offer tests have their own risks, covered in the price testing guide. For pages built for ads, see landing page A/B testing.
Ecommerce A/B test results from 14 Convertica case studies
These are split tests Convertica ran on online stores and published. Every result is copied from the case study with its metric, its test length where one was published, and its confidence level. dScryb is a membership store; the other 13 sell products online.
| Case study | Page | What changed | Stated result | Test length | Confidence |
|---|---|---|---|---|---|
| Homepage trust signals | Homepage | Shipping, reviews and security strip under the hero | +57.9% revenue per visitor, desktop; +68.9% order completion rate, mobile | 12 days | 93% desktop, 97% mobile |
| Search bar | Homepage, collections, blog | A visible search bar | +71.9% order completion rate, desktop; mobile did not change | Not published | 98% |
| Reviews section | Product page | Reviews moved up to just below the product, two stores | +42.38% orders, bedding store, mobile; +20.34% completed orders, jewelry store, desktop | 2 weeks | 98% |
| Variant selectors | Product page | Labeled panels and swatches in place of plain option rows | +144.8% conversion rate | Three weeks | 98% |
| Product page redesign | Product page | Cleaner page with a sticky Add to Cart and Buy Now bar | +56% order completion rate, desktop; mobile neutral | Not published | 98% |
| Free shipping threshold | Product page | Message showing whether the cart qualifies for free shipping | +175.4% revenue per visitor and +88.3% order completion rate, desktop; mobile no significant change | Not published | 95% |
| Free shipping badge | Product page | "Free Shipping Included" badge next to the price | +60.3% revenue per visitor and +27% order completion rate, desktop; mobile no significant change | Not published | Not published |
| Factory to Home | Product page | Shipping, warranty and payment FAQs in pop-ups | +12.7% revenue per visitor and +18.1% order completion rate, desktop and mobile combined | Not published | 96% on desktop |
| Mobile pricing table | Product page | Stacked packages turned into a slider on mobile | +36% revenue per visitor and +43.9% order completion rate, mobile | Not published | 98% |
| JustThrive | Product page | Trust signals and auto-ship offer, then a second variation against the winner | +101.82% revenue, cumulative across two sequential split tests | 7 December to 24 March, two tests | 98% significance stated for test 1 |
| Policy visibility | Cart | Policy badges with pop-ups under the checkout button | +21.8% order completion rate, desktop; mobile neutral | Not published | 97% |
| Homewares store | Product page to cart | Add to Cart sends shoppers straight to the cart | +26.9% conversions overall, with the winner rolled out to mobile users | 19 days | 98% significance |
| Sticky CTA | Landing page | A call to action bar fixed to the bottom of the screen | +252.9% rate of order completion, mobile; desktop no significant difference | 20 days | 98% |
| dScryb | Search results | Calls to action and access levels in membership search results | +49.4% paid membership sign-ups | 17 days | 98% |
Results like these are not a forecast for your store: each one came from a specific store, its shoppers and its problem. Every case study, including the lead generation and affiliate tests, is on the CRO case studies page.
What do these ecommerce A/B tests teach?
Four patterns run through Convertica's store tests: the winners answered a shopper's doubt at the point of decision, desktop and mobile often moved differently, a winner became the new control for the next test, and some results were reported honestly below the usual 95% confidence. Each one changes how you should plan and read your own tests.
- Answer the doubt where the decision happens. Reviews, shipping cost, policies and clear options all won by appearing next to the button rather than lower down or on another page.
- Read each device on its own. The search bar, product page redesign, policy, free shipping and sticky CTA tests each moved one device and not the other. The homewares and sticky CTA winners were rolled out to mobile only.
- Test against the winner. JustThrive's second test ran against the first winner, not the original page, and the two compounded to the cumulative revenue result.
- Report the confidence you got. The homepage trust test's desktop result reached 93%, which the case study states as below the usual 95%. Say so, rather than rounding it up into a win.
How to run an ecommerce A/B test in 7 steps
To run an ecommerce A/B test, find where shoppers leave, write a hypothesis that names the evidence, the change and the metric, work out the sample you need, build and check the variant on every device, run it in whole weeks, then read the result by device before you roll out the winner. The steps below set out each stage.
-
Find where shoppers leave
Use your analytics funnel from product page to order, then session recordings and heatmaps on the step with the biggest drop. The test idea should come from what shoppers do on your store, not from a list of other stores' wins.
-
Write a hypothesis
Name the evidence, the change, the shoppers and the metric. For example: because recordings show mobile shoppers scrolling past the price to look for shipping costs, showing free shipping beside the price will raise mobile order completion rate.
-
Pick one primary metric
Order completion rate or revenue per visitor for most tests, with add to cart and checkout starts as secondary metrics. Write it down before launch so the winner cannot be chosen afterwards.
-
Work out the sample and the length
Enter your baseline and the smallest lift worth detecting into the A/B test sample size calculator, then divide by the daily shoppers who reach the page. How long to run an A/B test shows the full method.
-
Build the variant and check it everywhere
Check the variant on phones, tablets and desktop browsers, with every product option, a discount code and an empty cart. A variant that breaks for some shoppers produces a result about the bug, not the idea.
-
Run it in whole weeks and leave it alone
Split traffic evenly, avoid sales and holidays where you can, and do not change the page, prices or tracking during the test. Do not stop the first day the variant looks ahead.
-
Read the result by device, then roll out or retest
Check significance with the statistical significance calculator, look at desktop and mobile separately, and roll out only where the variant won. Then test the next idea against the new winner.
How much traffic does an ecommerce A/B test need?
An ecommerce A/B test needs enough shoppers on each version to tell a real lift from chance, and that number comes from your own conversion rate and the size of lift you want to detect. There is no fixed minimum for every store. Lower conversion rates and smaller lifts need far more visitors, which is why small changes on a quiet store can take months.
The table shows a worked example, not a target: a page where 2% of visitors order, one variant against the control, 95% confidence and 80% power. Replace the 2% with your own rate in the calculator.
| Smallest lift to detect | Visitors per version | 1,000 visitors a day | 5,000 visitors a day |
|---|---|---|---|
| 10% (2% to 2.2%) | 80,682 | 24 weeks: change the plan | 5 weeks |
| 20% (2% to 2.4%) | 21,109 | 7 weeks | 2 weeks |
| 30% (2% to 2.6%) | 9,798 | 3 weeks | 1 week |
Every figure comes from the same formula as the sample size calculator, rounded up to whole weeks with a one-week minimum. Anything over 8 weeks is marked "change the plan".
What to do when your store has too little traffic to test
- Test a bigger change. A rebuilt product page or a new shipping offer can move more than a button label, and a bigger difference needs a smaller sample.
- Test where the traffic is. Your best-selling product pages and the cart see more shoppers than a long-tail product.
- Test a template, not one page. A change to the product page template runs on every product at once, which pools the traffic.
- Fix the obvious without a test. Broken variant selectors, missing shipping costs and slow pages should be fixed, not tested.
Common ecommerce A/B testing mistakes
The most common ecommerce A/B testing mistakes are calling a winner on clicks instead of orders, stopping as soon as a variant looks ahead, testing through a sale, reading only the combined device number and launching a variant that breaks for some shoppers. Each one can turn a losing change into a false win.
| Mistake | Why it misleads | Do this instead |
|---|---|---|
| Judging on clicks or add to cart only | More carts can mean fewer orders if the change pushes undecided shoppers forward | Decide on order completion rate or revenue per visitor |
| Stopping when the variant looks ahead | Early results swing; stopping on a good day makes false winners more likely | Run to the planned sample in whole weeks |
| Testing through a sale or holiday | Sale shoppers behave differently, so the result may not hold afterwards | Schedule around promotions or extend the test |
| Reading only the combined number | A win on one device can hide a loss on the other | Read desktop and mobile separately |
| Not checking the variant on every device and option | A broken option or a page flash measures the bug, not the idea | Check every option, device and browser before launch |
| Changing the page during the test | The result mixes two different changes | Freeze the page, prices and tracking, or restart |
| Copying another store's winner | Different shoppers, products and problems | Use other tests for ideas, then test on your own evidence |
What do you need to run ecommerce A/B tests?
To run ecommerce A/B tests you need a testing tool that splits shoppers and records orders and revenue, analytics to find where shoppers leave, a way to build and check variants on every device, and enough traffic to reach a sample in weeks rather than months. The dScryb test, for example, ran in VWO with traffic split equally.
Compare options in A/B testing tools. Shopify stores have their own choices and limits, covered in Shopify A/B testing.
Getting help with ecommerce A/B testing
If you would rather have a team plan, build and read your tests, see A/B testing services. Convertica quotes A/B testing one by one as a custom proposal, so you can ask for an A/B testing proposal there. To find out first which problems on your store are worth testing, start with the free CRO audit: it checks a page for people and for AI agents and gives you every finding, with why each one matters.
Free calculators for planning and reading store tests
Plan the size and length of your next store test before launch, and check the result when it ends. Each calculator shows its formula and a worked example.
Check a result
Statistical significance calculator
Whether a finished test result is bigger than chance.
Measure
Conversion rate calculator
Your baseline order completion rate from visitors and orders.
A/B testing services
Want your A/B tests run for you?
Convertica runs A/B tests for clients, quoted one by one as a custom proposal. Monthly traffic comes first, because it decides whether a test on your site can reach a clear result, and how long it takes.
The split tests in Convertica's case studies were run by its team for past clients, and each result is published with its metric, timeframe and source.
Short answer
Ecommerce A/B testing shows two versions of a store page to different shoppers at the same time and compares orders or revenue per visitor. Start with the pages closest to the purchase: product page, cart and checkout. Test one change backed by evidence, run to a planned sample in whole weeks, and read desktop and mobile separately.
- Measure orders and revenue per visitor, not clicks, so a test cannot win on a number that does not pay.
- The product page, cart and checkout sit closest to the order, so changes there tend to show up in sales first.
- In Convertica's store tests, desktop and mobile often moved differently, so read each device before you roll out.
- A winning test is a new control: test the next idea against the winner, as JustThrive did.
- Low traffic makes small changes slow to test, so test bigger changes or fix clear problems without a test.
Questions about ecommerce A/B testing
What is ecommerce A/B testing?
Ecommerce A/B testing is a controlled experiment on an online store: shoppers are split at random between the current page and a changed version, and the two groups are compared on a store goal such as order completion rate or revenue per visitor. Because both versions run at the same time, the change is the only systematic difference.
What should I A/B test first on my online store?
Test first where shoppers are closest to buying and where your own data shows them leaving: usually the product page, the cart or the first checkout step. Pick a change backed by evidence, such as reviews shoppers cannot find or shipping costs they only see at checkout, rather than a color or a font.
What are some good A/B test ideas for an ecommerce website?
Good ecommerce A/B test ideas answer a shopper's doubt at the moment of decision: reviews moved up the product page, shipping costs shown beside the price, clearer variant selectors, policies near the checkout button, a visible search bar and a sticky Add to Cart button on mobile. Convertica has tested each of these on real stores.
Which metric should an ecommerce A/B test use?
Use a metric that pays: order completion rate (orders divided by visitors) or revenue per visitor, which combines conversion and order value. Track add to cart and checkout starts as secondary metrics to see where a change works, but do not call a winner on clicks alone.
How long should an ecommerce A/B test run?
Run an ecommerce A/B test until each version reaches the sample size you planned before launch, in whole weeks, and at least one full week so weekdays and weekends both count. Convertica's store tests with a published length ran from 12 days to more than five weeks; a test that would need months should be changed rather than started.
Can small online stores run A/B tests?
Yes, but a small store has to test bigger changes on its busiest pages, or a test can take months to reach a reliable answer. Work out the sample with a sample size calculator before you start. Where traffic is too low, fix clear problems directly and keep testing for the questions you cannot answer otherwise.
Does Shopify support A/B testing?
Shopify stores are usually A/B tested with a testing app or a testing tool's script added to the theme, which splits visitors between versions and reports orders. What you can change, especially in checkout, depends on your plan and the tool, so check both before you design a test.
Should I A/B test during a sale or holiday?
Avoid it where you can. Shoppers during a sale, a holiday or a big email push often behave differently from your normal traffic, so a result from those weeks may not hold afterwards. If a test overlaps one, extend it so normal weeks dominate, or note the overlap when you read the result.
What is the difference between A/B testing and split URL testing in ecommerce?
An A/B test changes elements on the same page address, while a split URL test sends shoppers to two different addresses, which suits a whole new page or flow. Both split traffic at random and compare the same goal. Empire Flippers' rebuilt submission process, for example, was split-URL tested before rollout.
