Experiments overview
Prove what works instead of guessing — what experiments are, when to run them, and how they fit the decide-test loop.
Experiments overview
Optimisation without testing is just opinion. Experiment → Experiments is where you prove a change actually works before rolling it out — and capture what you learn.
Why experiment
- Avoid false wins — a metric can move for reasons unrelated to your change. A controlled test isolates the effect.
- De-risk big swings — test a bold idea on part of the account before betting the whole budget.
- Build evidence — turn a tentative recommendation into a proven one.
When to run one
Run an experiment when a recommendation is high-impact but uncertain — a new bid strategy, a budget reallocation, a creative overhaul, or a landing-page change. For clear, low-risk fixes, just apply them.
The flow
- Create an experiment from a hypothesis.
- Track it on the calendar.
- Read the readout when it concludes.
- Save the takeaway to learnings.
For questions you cannot run as a clean split test, use the Causal Lab.