> ## Documentation Index > Fetch the complete documentation index at: https://docs.replo.app/llms.txt > Use this file to discover all available pages before exploring further. # A/B Testing with Experiments > Run an A/B test with Experiments: split visitors between versions of a page, pick a goal, and keep the version that converts best. Run an A/B test on any page of your site and keep the version that converts best. A/B testing with Experiments splits the visitors to one page between the original and up to 4 new versions, counts how many of each group reach your goal, and tells you when one version has won. Visitors keep the same URL, so your ads and links don't change. ## Before you begin * **Publish your site.** Visitors are only split on your published site. Version pages must be published before the experiment can launch, and Replo offers to publish them for you at launch. * **Check that results can be counted.** The Replo pixel records which version each visitor saw and what they did next. If the **A/B tests** tab in [Insights](https://docs.replo.app/apps/insights.md) shows **A/B tests need the Replo pixel**, finish the pixel step in the Insights **Setup** tab first. ## Set up an experiment **1. Open Experiments** In the [Site Builder](https://docs.replo.app/apps/website-builder.md), open the page you want to test. Click **Settings** in the top bar and, under **This page**, choose **Run an experiment**. The setup opens in its own tab with that page already picked. To start from the list of all experiments instead, choose **Experiments** under **This site** in the same menu, then click **Create experiment**. If the page is already in a draft or live experiment, the menu item reads **View experiment** and opens that experiment. ![The Experiments list with All, Live, Draft and Ended tabs, one live and one draft experiment, and the Create experiment button](https://docs.replo.app/images/experiments-list.png) **2. Pick the page to test** Under **What do you want to test?**, search for the page and select it. It becomes the **Original**. A page can be in only one draft or live experiment at a time. Pages that already are show **In an experiment**; click one to open its experiment. For a [dynamic page](https://docs.replo.app/features/dynamic-pages.md) template, choose whether to test every page it renders or pick up to 1,000 of them. The pages you don't pick keep the original. ![New experiment: What do you want to test? with a page search, the site's pages, and In an experiment next to pages already being tested](https://docs.replo.app/images/experiments-pick-page.png) **3. Make version B** On the **Version B** card, choose how to make it: * **Copy the original**: Replo duplicates the page. Under **What should version B change?**, describe the change (optional). After you create the experiment, Replo makes that change in chat, on the copy only. * **Use another page**: pick a page you already built. To test more than one change, click **Add version C**. An experiment holds the original plus up to 4 versions (B to E). ![The Original and Version B cards side by side after Copy the original, with the What should version B change? box and Add version C](https://docs.replo.app/images/experiments-version-b.png) **4. Name the experiment** **Experiment name** fills in as "\[Page] experiment". Change it to something you'll recognize in the list later. **5. Set the traffic split** The **Traffic split** card gives every version an even share by default. Type a percentage for each version (1 to 99) or click **Split evenly** to reset. Once a goal is set, the card estimates how long the test will take, for example **About 14 days to a result at your traffic**. The version with the smallest share sets the pace, so an uneven split takes longer. > [!WARNING] > > The shares must add up to 100%. If they don't, the card shows **These add up to 90%. Make them 100%.** and the experiment can't be created. **6. Choose the goal** Under **Goal**, pick what counts as a conversion: * **Visits a page**: then pick the page next to **Counts when a visitor reaches**, for example a thank-you page. * **Clicks a button** * **Adds to cart** * **Starts checkout** * **Buys** Every version is judged on this one goal. ![The Traffic split card at 50% and 50%, the Goal options from Visits a page to Buys, and the optional Hypothesis box](https://docs.replo.app/images/experiments-split-goal.png) **7. Write a hypothesis (optional)** In **Hypothesis**, note what you expect and why, for example "If we move the bundle higher, more visitors will add it to their cart." It shows on the experiment's **Goal** card later. **8. Create and launch** The main button reads **Launch** when every version is already a published page and your site is ready to split traffic. Otherwise it reads **Create experiment**, which saves a draft and opens it. If the button is greyed out, the reason shows next to it, for example **Pick the page that counts as a conversion**. In the draft, click **Launch**. If a version isn't published yet, the button reads **Publish B and launch** (or **Publish versions and launch**): Replo publishes your site, then launches when the publish finishes. > [!WARNING] > > If you asked Replo to change a copied version, check the change in chat before you launch. Until Replo finishes, that version is an exact copy of the original, and the test compares two identical pages. Open it from the **Versions** tab with **Edit page**. ![A draft experiment's Versions: Original and Version B at 50% each, Version B marked Not published yet, and the Publish B and launch button](https://docs.replo.app/images/experiments-draft.png) **What success looks like:** the experiment shows a **Live** status, and a message confirms **Visitors see it in a minute or two.** ## How visitors are split Each visitor is assigned a version using your traffic split. They keep that version on every visit, and the address bar shows the page's normal URL while they see their version. Copied versions live under `/v/` and are hidden from search engines. To see a version the way visitors do, open the experiment's **Versions** tab, open the menu on a version's card, and choose **Preview on your live site**. Preview visits are left out of results. ## Read the results Open the experiment from **Settings** > **Experiments**. A live or ended experiment opens on its **Results** tab, which shows: * **Leader**, **Goal**, **Lift** and **Chance to win** at the top. * A table with each version's **Visitors**, **Conversions**, **Conversion rate** and **Chance to win**. * A banner that says what to do next, such as **Gathering data** or **Version B won, so you can end this experiment**. Click **Open in Insights** to open the same report in Insights, or **Write up results** to have Replo add a written summary to it. Every experiment is also listed in the **A/B tests** tab in Insights. ![A live experiment's Results tab: the banner says the 25% off headline is ahead with an 81% chance to win, the Leader, Goal, Lift and Chance to win cards, each version's visitors and conversions, and the conversion rate over time](https://docs.replo.app/images/experiments-results.png) ### How a winner is called Numbers show from the first visitor, but Replo doesn't call a result until every version has at least 100 visitors and 10 conversions and the experiment has run for 7 days. Until then the banner says **Gathering data** and estimates how long is left. After that: * A version **wins** when it has at least a 95% chance of beating the original. * The **original wins** when every other version has a 5% chance or less. * If neither happens, the banner tells you to keep it running. If detecting a 10% lift would take more than 60 days at your traffic, it says **No clear difference, so you can end this experiment**: try a bolder change. ## End an experiment **1. Click End experiment** On a live experiment, click **End experiment**. **2. Pick the version to keep** Select the version every visitor sees from now on. If you keep a version other than the original, visitors to the page's URL get that version. ![The End experiment dialog: pick Original or 25% off headline to keep, an Outcome menu set to No clear winner, and a What did you learn? box](https://docs.replo.app/images/experiments-end.png) **3. Record the outcome** **Outcome** fills in for you: "Version B won" when you keep a version, **No clear winner** when you keep the original. Change it to **The original won** or **Stopped early** if that fits better, and add a note under **What did you learn?**. Then click **End experiment**. The experiment moves to the **Ended** tab of the list. To test the same versions again, open its menu and choose **Run again**. > [!NOTE] > > You can delete a draft or an ended experiment, but not a live one. Deleting an ended experiment where you kept a version other than the original sends visitors back to the original. ## Troubleshooting > [accordion] Launch is greyed out > > Hover the button to see why. **Version B's page is missing** means the version's page was deleted: create it again or delete the draft. **Version B will be deleted at your next publish** means the page is set to be removed: restore it first. > [accordion] Saved, but couldn't connect the experiment to your site > > The experiment saved, but visitors aren't split until it's connected. Click **Try again** on the message. > [accordion] Couldn't publish, so the experiment didn't launch > > The publish that launch started failed. Fix the publish (see [Publishing](https://docs.replo.app/publishing.md)), then click launch again. If you see **Your site is already publishing**, wait for that publish to finish and launch again. > [accordion] A live experiment says a version's page is missing > > Visitors aren't being split while a version has no live page. Create the page again or end the experiment. > [accordion] The traffic split says No traffic on this page yet > > Replo has no traffic for this page to estimate from, so the card shows no time estimate. It doesn't block the experiment; you can still create and launch it. ## Next steps > [card] [Insights](https://docs.replo.app/apps/insights.md) > > See every A/B test with your other reports. > [card] [Publishing](https://docs.replo.app/publishing.md) > > Publish your site so versions can go live. > [card] [Act on Your Analytics](https://docs.replo.app/analytics.md) > > Decide what to test next from your traffic and sales. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.