Experiments

An experiment is a controlled test where two or more versions of a page compete for real traffic. Koryla uses the results to tell you which version performs better.

Creating an experiment

From your dashboard, click New Experiment and fill in the details:

Name

An internal name — only you see this. Use something descriptive like "Pricing page headline test" or "Checkout button color Q2".

Base URL

The URL path the Worker should intercept. Examples:

/ — homepage
/pricing — pricing page
/blog/my-post — a specific post

The Worker matches requests using pathname.startsWith(base_url_path), so /pricing would also match /pricing/ but not /price.

Variants

Every experiment needs at least two variants:

Control

The control is your original page. It acts as the baseline for comparison. Set a traffic weight for it — typically 50% in a 50/50 test.

The control variant's target URL is the same as the base URL (no rewrite happens).

Challenger variants

Each challenger variant points to a different URL on your site. Koryla rewrites the request to that URL at the edge.

For example:

Base URL: /pricing
Control target: /pricing (no rewrite)
Variant A target: /pricing-v2 (rewrites to this page)

The target pages for your variants must exist on your site. Koryla just routes traffic to them — you need to create the actual pages.

Traffic weights

Weights are relative numbers. Koryla normalizes them so they don't need to add up to 100:

Setup	Control	Variant A
50/50 split	50	50
80/20 split	80	20
Three-way	33	33 / 34

For a new experiment, start with a 50/50 split unless you have a reason to limit exposure to a new variant (e.g. a risky change).

Experiment status

Status	Description
`draft`	Created but not yet running. No traffic is split.
`active`	Running. The Worker is assigning visitors to variants.
`paused`	Temporarily stopped. Existing cookie assignments are respected but no new assignments happen.
`completed`	Finished. Traffic returns to the control (or winner, if you update the site).

Reading results

Note: Full analytics dashboards are on the roadmap. Currently, results flow to your connected analytics destination (GA4, PostHog, etc.).

In GA4 or PostHog, look for events named experiment_viewed (or your configured event name) with properties:

experiment_id — the Koryla experiment UUID
experiment_name — the name you set in the dashboard
variant_id — which variant was served
variant_name — variant name

Filter and compare conversion rates between variants using your analytics tool's exploration or funnel features.

As a rule of thumb, run experiments until you have statistical significance — typically at least 95% confidence that the difference isn't random. Most A/B testing calculators can tell you if you have enough data.

Practical guidelines:

Run for at least 2 weeks to account for day-of-week variation
Aim for at least 100 conversions per variant before drawing conclusions
Don't stop early just because one variant is winning — the early leader often loses

After the experiment

Once you have a winner:

Update your site to use the winning variant as the default (rename the page, update your CMS, etc.)
Set the experiment status to completed in Koryla
The Worker will stop intercepting traffic within 60 seconds

Don't delete experiments — they serve as a historical record of what you tested and what worked.