Troubleshoot: Experiment Traffic Not Splitting Correctly

Last updated: June 2026

Overview

If traffic is not distributing between experiment variants as expected, the most common causes are using a direct funnel URL instead of a Smart Link, browser caching behavior, low session volume, or a variant still in draft status. This article walks through each scenario.

Check 1: Smart Link Not in Use

The ad campaign is pointing to a direct funnel URL.

Experiments only split traffic through Smart Links. A direct funnel URL always loads the same variant. Confirm your ad campaign destination is the Smart Link URL, not a direct link to either funnel.

Check 2: Browser Caching

Returning visitors are being served the same variant repeatedly.

This is expected behavior. Jurni uses cookie-based assignment to ensure each visitor sees a consistent experience across visits. This means the split is not 50/50 per page load, but 50/50 across unique visitors over time.

Check 3: Low Traffic Volume

The split looks uneven at low session counts.

At very low volumes, statistical variance naturally creates the appearance of uneven splitting. Splits even out significantly once each variant reaches 500 or more sessions. Avoid drawing conclusions from small sample sizes.

Check 4: Variant Is in Draft

One variant funnel is still in draft status.

A draft funnel cannot receive experiment traffic. Open both variant funnels and confirm both are published before the experiment goes live.

Frequently Asked Questions

Can I change the traffic split after an experiment is live?

Yes, but doing so resets the statistical significance calculation. Only adjust the split if you have a strong reason to, and note that prior data will no longer be part of the calculation.

Why is one variant getting 0 sessions?

The most likely cause is that the variant funnel is still in draft status. Confirm both variants are published.

How do I know if my experiment is getting enough traffic?

Aim for at least 200 sessions per variant before drawing any conclusions. For statistical significance, 500 or more sessions per variant is recommended.