How it works
Koda sits between the visitor and your client's website — at the network edge, in over 300 locations worldwide. Here's the whole journey, and why it can't flicker.
From click to rendered page, with a test running:
Someone clicks a link to your client's site. Because the domain points at Koda (one DNS record), the request arrives at the edge location nearest that visitor — usually within a few milliseconds.
The edge looks up the site's test configuration. No active test? The request passes straight through, byte for byte. Koda in passthrough mode adds no meaningful delay and changes nothing.
First-time visitors are assigned by a fair 50/50 coin flip, then remembered with a cookie for 30 days. Returning visitors keep their side — nobody ever sees both versions of the page.
The original website responds exactly as it always does. Your client's server never knows a test is running and needs zero changes.
For B-visitors, Koda edits the HTML as it flows through — swapping a headline, recoloring a button — without holding the page back. A-visitors receive the untouched original. Total added compute: under one millisecond.
Whatever arrives is what the visitor sees. There is no "old version" in the browser to flash and no script racing to change things after the fact — the test already happened upstream.
That last step is the whole product: by the time the browser is involved, the decision is history.
Traditional testing tools (Optimizely, VWO, and similar) work inside the browser. The original page loads first, then their script wakes up and swaps the content. On a fast connection that swap takes a blink; on a slow phone it can take a second or more. Visitors see the old headline flash into the new one — researchers call it FOOC, the "flash of original content."
The industry's fix is the anti-flicker snippet: hide the entire page (a white screen) until the testing script finishes. That trades the flash for a delay — and page speed is itself a conversion factor, which means the testing tool is quietly damaging the very metric it's supposed to improve. It also drags down Core Web Vitals, which feed into Google rankings.
Koda doesn't have this dilemma. The rewrite happens at the network edge, before the browser receives a single byte. There is no original content in the browser to flash and nothing to hide. Zero flicker isn't a feature we optimized — it's a consequence of where the work happens.
Everything here is always on. There are no settings to forget.
Serving different content to Google than to humans is called cloaking, and sites get penalized for it. Koda detects crawlers — Googlebot, Bingbot, link previews, speed-test tools — and always serves them the unmodified page. Test content is invisible to search engines, which makes Koda safer than JavaScript tools, where crawlers can execute the test script.
If a test configuration is broken, or anything unexpected happens inside Koda, the visitor simply receives the original page. The failure mode of every error is "the website works normally." A test can never take a client's site down — which is the first question every agency's client will ask.
Browsers save pages to avoid re-downloading them. During a test that's dangerous: a saved copy of Variant B could be shown to an A-visitor. Koda rewrites the caching instructions on tested pages so browsers always check back — while images, styles, and scripts keep caching at full speed, so the site stays fast.
Each visitor keeps their assigned version for 30 days via a cookie tied to the specific test. Start a new test and assignments reset cleanly — old cookies can't leak into new experiments. Conversions are counted once per visitor per test, so refreshing a thank-you page can't inflate results.
Honesty section. Three current limits worth knowing before you commit:
Because Koda edits HTML while it streams, it can only target elements it can recognize on sight: IDs like #hero-title, classes like .buy-button, and plain tags. Selectors that depend on looking backwards or counting from the end of the page don't work. In practice: ask the client's developer to put an ID on anything you want to test.
Each hostname runs one test at a time. For agencies this is usually a feature — sequential tests give cleaner reads than overlapping ones — but if you need multi-test support, tell us; it's on the roadmap.
Tests are defined as a short configuration (see the agency guide) rather than a point-and-click editor. Early-access agencies get hands-on help writing their first tests.
Early-access agencies get a live demo and hands-on onboarding for their first client test.
Get early access