Traffic Allocation
How much traffic enters a test, how it is split between variants, when to use an uneven split, and why visitor assignment is sticky.
Traffic allocation decides two things: how much of your traffic enters a test at all, and how the visitors who do enter are split between the control and its variants. By default every visitor who matches your targeting enters, and the split is even — 50% / 50% for two experiences, roughly a third each for three.
How Weights Are Set
On the wizard’s Variants step, the Traffic allocation rows show each experience’s share of the test’s traffic. Drag a share or type a new value: the difference is pushed across the other experiences in proportion to what they held, so the total always stays at 100%. Split evenly resets every share to an equal split. Changes are saved as soon as you make them.
A test needs at least one variant besides the control. If you set every share but one to 0%, the Variants step warns that the split has only one experience.
How Much Traffic Enters the Test
The General step has a traffic slider: “How much of your traffic enters the test.” It runs from 1% to 100% and starts at 100%. Visitors left out by the slider see your original page and are not counted in the results. Whether a visitor is in or out is decided consistently, so the same visitor gets the same answer on every page view.
The share entering the test can’t be changed while the test is running. Pause it first.
When to Use a 50/50 Split
This is the right default for almost every test. Equal splits give you the fastest path to statistical significance because both variants accumulate samples at the same rate — you reach the threshold for the underdog variant as quickly as possible.
Use 50/50 when:
- You’re equally curious about both variants.
- You don’t have a strong reason to expect either side to be much worse.
- You want results sooner rather than later.
When to Use an Uneven Split
There are a few legitimate cases where you’d weight the split unevenly:
Risk-mitigated rollout
You’ve made a substantial change and you’re not sure it won’t break something. Set the new variant to 10–20% of traffic. If something goes wrong, the blast radius is small. If it does well, ramp it up by editing the weights.
The downside: you collect data slowly on the new variant, so you’ll wait longer for significance. That’s the trade.
High-cost variant
Your variant requires a paid third-party widget that bills per impression, or sends visitors to an experimental flow you’d rather not over-expose. Cap it at a smaller share.
Multi-arm with a star variant
You have a variant you’re confident will beat control and one or two long-shot variants. Weight the confident one heavier so the experiment can graduate it sooner, and let the long shots accumulate data more slowly:
What Doesn’t Help
- Changing weights mid-experiment to “let the winner accumulate faster” is a form of peeking and contaminates your data. Don’t.
- Setting different weights per device. Split Test Pro doesn’t support per-device weights. Use device targeting instead — run separate experiments per device segment if that matters.
- A 1% / 99% canary split for the first 24 hours, then a 50/50 split. Two different experiments stitched together. Pick one model and run it cleanly.
Visitor Assignment Is Sticky
Whatever split you set, an individual visitor’s assignment is deterministic and sticky. By default a visitor is assigned by their visitor ID: once they’re in Variant B, they stay in Variant B on every page and every return visit.
This is a feature: returning visitors see a consistent experience, and your conversion rates aren’t polluted by the same person seeing both versions.
When a test is activated by a custom event, the General step also offers Yes — for this session (the assignment lasts for the visit) and No — re-evaluate each page view. See Event Activation.
Weights can be changed only while the test is a Draft or Paused. While it is running, the Variants step’s split is read-only — click Pause, adjust the shares, then resume. A change only affects visitors who haven’t been assigned yet.
One Control, Several Variants
Every test is A/B/n: one control plus one or more variants, each a distinct version of the page. When you build with Splitty, it creates up to three variants against the control.
Next Steps
- Decide what each variant should change: Variant Types Overview.
- Run multiple experiments without traffic conflicts: Running Multiple Experiments.
- Learn what happens after a visitor is assigned: How CSS Variants Work.
