Create a smart link and an experiment

A smart link is an address on your domain, such as https://go.yourstore.com/spring, that you put in your ads. On its own it sends every visitor to one destination. An experiment on the link splits its visitors between two or more destinations and compares them.

You need a connected domain first. Links can be created while the domain is still pending; they go live once it is active.

Create a smart link

1. Open Links and select New link (or Add link on the domain’s panel).
2. Fill in the form:
   • Domain — the domain the link lives on. It can’t be changed later.
   • Path — the part after the domain, starting with /, for example /spring. Letters, digits and . _ ~ - only; case matters.
   • Name — only your team sees it.
   • Destination — where visitors go, for example https://yourstore.com/collections/spring. Leave it empty to send them to the fallback URL.
   • Fallback URL — used when there’s no destination, or the destination can’t be served. Your store’s home page is a good choice.
3. Select Create link.

Destination rules

• Every destination and fallback URL must be a full address starting with https://. Addresses are checked when you save; anything else is refused with the reason next to the field.
• A destination can’t contain a username or password.
• No destination can be on the link’s own domain: not the Destination, not the Fallback URL, and not an experiment variant’s destination. A link on go.yourstore.com that sent visitors to another address on go.yourstore.com would send them back through a link in a loop. Use your store’s own addresses (yourstore.com) instead.

What visitors get

Visitors are redirected to the destination with every query parameter of the link kept — the utm_ parameters and click ids your ad platform adds arrive on your page unchanged. GrowthFlow adds two parameters of its own, gf_cid (the click) and gf_var (the variant), which is how an order is matched to the click later.

Live within a minute

Saving publishes every link on the domain to the edge. The change is live within a minute. On Links, each link shows whether the edge serves it as saved:

• Live · v3 — served as saved (v3 is the domain’s published configuration version).
• Not live yet — saved, but not on the edge yet. A Publish now button on the domain’s panel publishes again.
• Off — the link is archived and no longer served.

Changing a link’s path breaks the old address for everyone who already has it; the form warns you before you save. Archive stops serving a link without deleting it; Unarchive brings it back.

Create an experiment

1. Open Experiments and select New experiment.
2. Fill in the form:
   • Link — the smart link whose visitors are split.
   • Name — only your team sees it.
   • Hypothesis — what you expect to happen, and why.
   • Variants — one row per page, from 2 to 5: a name, a weight and a destination URL. Use Add variant for more rows.
3. Mark one variant as the Control (Set as control on its row), usually the page you use today.
4. Set the weights: whole percentages that add up to 100. The form shows the total. A 50/50 test is two variants at 50; a cautious test of a new page might be 80/20.
5. Select Create draft.

A draft reaches no one: the link keeps sending visitors to its own destination. You can edit it or Delete draft freely.

Start an experiment

On the experiment page, select Start, check the split in the dialog, and select Start experiment. From then on every visitor to the link is assigned a variant according to the weights, and keeps that variant on every later visit (the assignment comes from the visitor cookie, so the same visitor on the same browser always gets the same page).

The change is live within a minute, like any link change. The Overview panel shows the Allocation version and whether the experiment is On the edge.

Why changing weights reassigns visitors

While an experiment runs, you can still change destinations and names: those changes go live without reassigning anyone.

Changing a weight, or adding or removing a variant, is different. Assignment depends on the weights, so new weights would quietly move some visitors from one page to another and mix two different tests in the same numbers. Instead, GrowthFlow asks first — Reassign every visitor? — and when you select Reassign and save:

• every visitor is assigned again under the new weights;
• the experiment’s allocation version goes up by one (1 → 2);
• results start a new segment. On the experiment page, the Allocation selector in Results shows each version separately, so numbers from different splits are never added together.

Select Keep editing to go back without saving. Decide the weights before you start, and change them only when you mean to start a new segment.

When a variant can’t be served

If a variant’s Health shows Unhealthy · fallback or Disabled · fallback, the visitors assigned to it are sent to the link’s fallback URL instead. They keep their assignment, so nothing shifts between the other variants, and those visits are left out of the results.

Pause, resume and end

• Pause sends every visitor to the control until you resume. Assignments are kept, and results pause.
• Resume splits visitors again with the assignments they had.
• End experiment sends the link’s visitors back to its own destination (or its fallback URL if it has none). An ended experiment can’t be restarted; the link can get a new experiment.

Troubleshooting

• “must start with https://” under a destination or the fallback URL. Use the full address, including https://. Plain http:// pages can’t be destinations.
• “Destination can’t be on go.yourstore.com itself: visitors would be sent back through this link in a loop.” (or the same for Fallback URL or a variant’s destination). The address points at the link’s own domain. Use the page’s address on your store instead, for example https://yourstore.com/collections/spring.
• The weights can’t be saved. They must be whole numbers that add up to 100, the control needs a weight above 0, and at least two variants need a weight above 0. In a draft, every variant needs a weight of at least 1 — delete a variant you don’t want instead of setting it to 0.
• “This experiment changed since you opened it.” Someone else saved it in the meantime. Close the form and reopen it to see the latest version.
• The link shows “Not live yet” for more than a minute. Select Publish now on the domain’s panel. If the domain isn’t Active, links wait for it: see Connect your domain.
• Visitors still see the old page after you started. Changes are live within a minute; a visitor who opened the page earlier may also have it cached in an open tab.
• There is no “New experiment” button. An experiment needs an active (not archived) link. Create one on Links first.