Glossary · Éprouver

Variant

A variant is one of the versions compared in an A/B test — the one in place, or the one set against it. Every click is assigned to one variant and one only, and that assignment is recorded with the click.

Tester deux versions, et ne désigner un gagnant qu’au-dessus du seuil. The forty terms of the trade

What it is for

It is the unit all results refer back to. Without a variant recorded on the click, you know a test took place but not who won: sales arrive later, and nothing then ties them to the version that produced them.

How it is defined

A name, a destination, and a weight — the share of traffic it should receive. The weight need not be equal: you can send 90% of traffic to the version in place and 10% to the new one, which lengthens the test but limits the risk.

The classic mistake

Adding a third and a fourth to go faster. The opposite happens: each variant divides the traffic, and you then need far more clicks in total to reach a threshold on any one of them.

A/B test

Two variants, one single difference

Version A

  • The page in place.
  • 50% of traffic.
  • The reference: we keep serving it.

Version B

  • One element changed.
  • 50% of traffic.
  • It has to beat A, not match it.

Everything else is identical. That constraint is what gives the result meaning.

The whole test, from setup to verdict

What Subtraq does with it

The variant is written onto the click at redirect time, not inferred afterwards. That is what lets a sale arriving three weeks later still be tied to the right version.


Read next

Un terme seul ne sert à rien. Voici ceux qui le rendent utilisable.

Statistical significance

Read the definition

The product's features

Open the page

« Variant », une fois posé sur de vraies ventes

A test that picks a winner on twelve clicks moves a budget onto noise. The product's thresholds are public: thirty clicks per variant at least, and 95% confidence before announcing anything.

See the features