Skip to main content

Experimentation

EDS has built-in experimentation -- A/B and multivariate testing, plus audience targeting -- without any third-party tool. Variants are resolved at the CDN edge, so visitors never see a flicker of the control before the variant takes over.

How a test is set up​

Define experiments in a spreadsheet (the "experiments" sheet in your content source):

ExperimentPageVariantSplit
hero-test/home/experiments/hero-v250
pricing-cta/pricing/experiments/pricing-cta-b50
  • Page -- the URL being tested
  • Variant -- an alternative page that replaces the original for the test group
  • Split -- percentage of traffic that sees the variant

Variant pages are authored exactly like the original page -- same blocks, same authoring flow -- just with different content.

Audiences​

Target content to specific audiences using the "audiences" sheet:

AudienceCondition
mobiledevice = mobile
returningreturning-visitor = true
germanygeo.country = DE

Combined with experiments:

ExperimentPageVariantSplitAudience
hero-test-de/home/experiments/hero-de100germany

Audiences resolve at the edge, so geographic and device targeting happens before the HTML is even sent to the browser. No client-side JavaScript is required to swap content.

Multivariate​

Add multiple variant rows for the same experiment and EDS will distribute traffic across them according to the splits:

ExperimentPageVariantSplit
hero-mvt/home/experiments/hero-a33
hero-mvt/home/experiments/hero-b33
hero-mvt/home/experiments/hero-c34

Tracking results​

EDS does not write results back into the spreadsheet. You wire your analytics tool of choice (Adobe Analytics, GA4, Plausible) and read the experiment / variant assignment from a meta tag or a data-experiment attribute that the EDS runtime adds to the <body>.

A typical setup in delayed.js:

scripts/delayed.js
const exp = document.body.dataset.experiment;
const variant = document.body.dataset.experimentVariant;
if (exp && window.analytics) {
    window.analytics.track('experiment-exposure', { exp, variant });
}

Stickiness​

EDS sets a cookie when a visitor is assigned to a variant, so they see the same variant on subsequent visits until the experiment ends or the cookie is cleared.

Concluding an experiment​

Two options:

  1. Bake the winner in -- replace the original page's content with the winning variant, then remove the experiment row.
  2. Promote the variant -- delete the original page, redirect to the variant URL.

Option 1 keeps URLs stable and is the default.

Limitations​

  • Variant pages must live in the Content Bus -- you can't run an experiment against an external URL.
  • Splits are per-experiment, not stacked. If you run two experiments on the same page, visitors are assigned independently to each.
  • Server-side personalisation tied to authenticated identity is out of scope -- use the hybrid pattern for those pages.

See also​