A/B testing and GDPR: what actually matters

First-party instead of third-party, your own server instead of someone else’s data centre.

What this article covers
  1. What an A/B test actually stores
  2. First-party instead of third-party
  3. Consent: a switch, not good intentions
  4. Five questions for any testing tool
  5. What remains

This article is not legal advice. It describes which technical decisions make the data protection difference in A/B testing, the legal assessment is made by your data protection officer or your lawyer.

What an A/B test actually stores

Technically an A/B test needs exactly two things: an identifier so the same visitor sees the same variation on their next page view, and counters for exposures and conversions. No names, no e-mail addresses, no profile.

The difference between privacy-friendly and problematic rarely lies in that identifier, but in where it ends up and who else gets to see it.

First-party instead of third-party

If the testing script loads from a foreign domain, every request is contact with a third party, including the IP address, before anything is even measured. If the same script runs on a subdomain of your own site, that contact stays with you.

Practical side effect: first-party delivery is restricted less often by browsers and ad blockers, so your measurement has fewer gaps.

If you need consent, it has to work technically. That means: before consent nothing happens, no identifier, no event, no request. A tool that already measures “anonymously” beforehand makes your documentation hard.

The clean approach is the other way round: the script lies dormant until the consent manager releases it. In abdriven that happens through the data-consent="required" attribute and a call to SplitTest.consent() once consent is given.

Five questions for any testing tool

  • Where is the data, and who besides us has access to it?
  • Is genuinely nothing stored or sent before consent?
  • Can the script be delivered through our own domain?
  • What is in the reports, aggregated numbers or individual visitors?
  • How long is raw data kept, and can we decide that ourselves?

What remains

Privacy-friendly A/B testing is not a question of intent but of architecture: your own server, your own domain, consent as a switch, aggregated analysis. Set it up that way and you will not have to explain anything at the next audit that you cannot prove.

Make your next change an informed one

Less guesswork.
More now we know.

One question is a good place to start.