AgentBIAgent with Business Intelligence for DTC sites

Back to the knowledge base

BIZ-4

What if several experiments run at once and their results conflict?

Published:2026-11-21Views:2

When two experiments change different stages of the same funnel at once, they contaminate each other: each shows a lift, but the combined picture does not add up. Check whether they touch the same users and path; where they overlap, either merge them into one experiment or stagger the time windows.

Record the affected page, metric and audience for each experiment in a registry and schedule to avoid overlap; if a segment turns out to be covered by two experiments, state in the record that attribution there is unreliable and do not use it to support a conclusion. Interactions between experiments can only be validated by a dedicated design, not read directly from mixed data.

This article covers scheduling and attribution discipline for parallel experiments only; it judges no result and decides no version for anyone. Write a conflict as a conflict, note flat numbers as flat, and do not use it to push more experiments or a change of direction.

2026-11-22