Create Your First Experiment

Choose the right experiment type for what you want to change, then follow the creation walkthrough for that type. Start here if this is your first experiment in Elevate A/B Testing.

---

This page walks you through your first experiment end to end: choosing what to test, setting it up, previewing it, and knowing what to expect once it is running.

If you have not installed Elevate A/B Testing yet, start with Install Elevate A/B Testing and then Verify Your Installation.

Before you start

• Elevate A/B Testing is installed on your store.

• The theme extension is enabled on your live theme, if you plan to run a Price, Product, Product Image, Content, or Custom Code experiment.

• You know which page or product you want to test on.

Choose your first experiment

Any experiment type works as a first test. Some are simply easier to set up and easier to confirm.

If you want to	Start with	Why it is a good first test
Try a different page layout	Page Experiment	No theme extension needed. The change is easy to see.
Compare two landing pages you already have	Split URL Experiment	No theme extension needed. Nothing new to build.
Change wording, an image, or a style	Content Experiment	No code, and the visual editor shows you the result as you work.
Test a price	Price Experiment	The highest impact test type, and no product duplication is required.

Avoid making your first experiment a Theme Experiment. It changes your entire store at once, which makes any problem harder to isolate.

Pick something worth testing

A good first experiment has three properties.

It gets enough traffic. Test a page real visitors reach. A product page with steady traffic will produce a readable result far sooner than a page nobody visits.

The change is meaningful. A different headline, a different price, a different hero image. Small cosmetic changes rarely move anything, and you learn nothing from a flat result.

You have a hypothesis. Know what you expect to happen and why, before you launch. Elevate gives you a field for this during setup, and filling it in is what makes the result actionable later.

Set up the experiment

1. In Elevate, open Experiments.

2. Choose the experiment type you want from the list of cards.

3. Select Create Experiment.

From here, setup differs by experiment type. A Price Experiment asks you for price points. A Split URL Experiment asks you for URLs. Follow the walkthrough for the type you chose:

Experiment type	Setup guide
Price	Create a Price Experiment
Page	Create a Page Experiment
Split URL	Create a Split URL Experiment
Product	Create a Product Experiment
Theme	Create a Theme Experiment
Shipping	Create a Shipping Experiment
Product Image	Create a Product Image Experiment
Checkout	Create a Checkout Experiment
Content	Create a Content Experiment
Custom Code	Create a Custom Code Experiment

What is the same for every experiment type

Three steps appear in every experiment, whichever type you choose. Where they fall in the flow, and how many steps there are in total, depends on the type.

General info and goal

Experiment name. Required. Name it so you can identify it later without opening it. "PDP price test, hero product, March" beats "Test 3."

Description and hypothesis. Both optional, and both worth filling in. The hypothesis field is where you record what you expect to happen and why, before you launch. Writing it down is what stops you rationalizing a flat result later.

Assets. Optional. Attach mockups, briefs, or reference material so the context lives with the experiment. Images and documents up to 10 MB, videos up to 100 MB.

What do you want to measure. Choose one goal metric. This is the metric Elevate uses to decide the winner. Every other metric is still calculated and shown in your report, so you are not giving anything up by choosing one.

Six metrics are offered directly, with Revenue per visitor recommended. Select Browse all metrics for the full list, or to add a custom metric.

Category	Metric	What it measures
Conversion	Conversion rate	Clicks on the tested element, divided by views
Conversion	Add to cart rate	Visitors who added a product to their cart
Conversion	Cart to checkout rate	Carts that went on to start checkout
Conversion	Checkout start rate	Visitors who began checkout
Conversion	Checkout success rate	Visitors who completed their purchase
Revenue	Revenue per visitor	Net revenue divided by unique visitors, after discounts
Revenue	Average order value	Net revenue divided by number of orders
Revenue	Profit per visitor	Gross profit after COGS, shipping, and fees, divided by unique visitors
Subscriptions	Subscription percentage of orders	Orders that included a subscription

For a first experiment, Revenue per visitor is usually the right choice. It accounts for both how many people buy and how much they spend, so it will not mislead you the way conversion rate alone can when a cheaper variation converts better but earns less.

Profit per visitor only works if you have added COGs. Without cost data, Elevate cannot calculate profit.

Traffic allocation

This step has two separate controls. They are easy to confuse and they do different things.

How much traffic should each variation receive. Splits the visitors who enter this experiment between your control and your variations. The default is even. Leave it there for your first experiment, because an even split gives you a readable result in the shortest time.

Isolate traffic for this experiment (Matually Exclusive Groups). Optional, and off by default. Turning it on reserves a share of your total store traffic for this experiment alone. Visitors in that reserved group do not enter any other experiment, so nothing can interfere with your results.

Use isolation when you are running several experiments at once and want certainty that they are not affecting each other. If this is your only running experiment, leave it off.

Isolated traffic is shared across your store. If other experiments already have traffic reserved, less is available to this one, and Elevate shows you the maximum you can allocate.

Audiences

By default, every visitor is eligible for the experiment. Narrow it only when you have a reason to.

The step offers a set of quick filters:

Filter	Options
Device type	All devices, Desktop, Tablet, Mobile
Visitor type	All visitors, New visitors only, Returning visitors only
UTM parameters	utm_source, utm_medium, utm_campaign, utm_content
Traffic source	All, or a specific source
Country	All, or specific countries

Custom audiences

For anything more specific, select Create Custom Audience. You build a rule from three parts: a dimension, a rule such as Equals, and a value.

Dimension	What it targets	Example value
Device Type	Specific device types	Mobile
Visitor Type	New or returning visitors	New
Country	Specific countries	Canada
Market	Specific Shopify markets	Europe
Traffic Source	Where the visit came from	Google
Page URL	Any part of the page URL	/products
Referring Domain	The domain the visitor arrived from	facebook.com
Entry Page	The first page URL the visitor landed on	/products/shoes
utm_source	Ad platform or website	facebook, tiktok, google
utm_medium	Type of ad platform	social, paidsocial, search, cpc
utm_campaign	Name of the campaign	black-friday-sale
utm_content	Name or type of the specific ad	black-friday-video
Customer Authentication	Logged in or logged out visitors	Logged in

Visitor Type and Customer Authentication are not the same thing. Visitor Type is about whether someone has been to your store before. Customer Authentication is about whether they are signed in to a customer account. A returning visitor may well be logged out, and a first time visitor can log in.

A note on narrowing

Every filter you add shrinks the pool of visitors entering the experiment, which means it takes longer to reach a trustworthy result. Targeting mobile visitors from paid social in Canada is a precise question, but on most stores it is also a question you will wait months to answer.

For a first experiment, leave targeting at its defaults and let the whole audience in.

Preview before you launch

Always preview before launching. Previewing is how you catch a broken layout, a price that did not swap, or a variation that looks identical to the control.

Check each variation and confirm:

• The control looks like your normal storefront.

• Each variation shows the change you configured, and only that change.

• Nothing else on the page has shifted or broken.

• The variation looks correct on mobile as well as desktop.

If a variation looks identical to the control, see Verify Your Installation before launching.

Launch and wait

Once you launch, the experiment moves to Running and starts collecting data.

Do not stop it early. This is the single most common mistake in A/B testing. Results swing wildly in the first days, and a variation that looks like a clear winner after 48 hours frequently is not one. Elevate uses a Bayesian model to tell you when a result is trustworthy. Wait for it.

Do not change the experiment while it is running. Editing variations mid experiment mixes two different tests into one dataset, and the result becomes unreadable.

Check in, but do not act yet. It is fine to watch the numbers. Just do not draw conclusions from them until Elevate indicates a winner has been identified.

What to do next

• Reports Overview explains what each number in your results means.

• Statistical Significance explains how Elevate decides a winner and when to trust it.

• Ending an Experiment covers how to stop an experiment and apply the winner.

• Audience Targeting covers limiting an experiment to specific visitors, once you are ready for that.