Interaction effects matter because modern products often run many experiments on the same users at once.
Running tests side by side is far faster than queuing them, so teams want to know when it is safe.
An interaction means the combined effect of 2 changes differs from the sum of their separate effects.
For example, a larger image and a longer description might each raise sales but together push the buy button off screen.
Independent assignment means each test is randomised separately, so every combination of versions appears.
With independent assignment, each test's average effect is still measured fairly, even if interactions exist.
Interactions can be checked by comparing results within each combination of versions.
Detecting an interaction needs far more data than detecting a single effect, so small ones go unnoticed.
In practice, strong interactions between unrelated tests appear to be uncommon.
Tests that touch the same part of the page, or the same decision, are the most likely to interact.
Those tests can be kept mutually exclusive, so no user sees both.
A factorial test combines changes on purpose, testing every combination in one design.
For 2 changes with 2 versions each, that gives 4 groups and shows both the separate and the combined effects.
Teams often group tests into layers, where tests in the same layer never share users.
The usual mistake is fearing all overlaps and running tests one at a time, which slows learning far more than interactions cost.