You are here:
Visualizing Experiment Performance
Review experiment performance, compare variations against your control variation, and see whether the experiment has a clear winner. The Data Visualization tab shows you which variations won and which need more data or a revised strategy.
What You See
The Data Visualization tab includes:
| Component | Description |
|---|---|
| Best Variation | The variation with the highest average on the selected metric. |
| Lift vs. Control | How much better or worse the best variation performs compared to the control. |
| Probability to Beat Control | The likelihood (0–100%) that the best variation outperforms the control. |
| Probability to Beat All | The likelihood that the best variation outperforms all other variations. |
| Metric picker | Compares variations across different metrics. Selecting a metric updates the summary cards and chart to match. |
| Performance by Variation chart | Shows each variation's estimated performance range against your control variation's baseline. Variations to the right perform better; variations to the left perform worse. |
| Daily Enrollment chart | A stacked area chart showing how many participants joined each cohort per day, based on the allocation weights you set when you created the experiment. Flat, even bands confirm the experiment is distributing traffic as expected. |
Interpret Results
The Performance by Variation chart shows a probability range for each variation, where the variation's true performance most likely falls based on the data collected so far:
| Component | Description |
|---|---|
| 95% Credible Interval | The wide bar. The system is 95% confident that the true performance falls within this range. |
| 50% Credible Interval | The narrow bar. This is the range where the true performance is most likely to be. |
| Center dot (best estimate) | The most probable performance value for this variation. |
| Baseline range | The gray background band marking your control variation's range. |
When a variation's range sits entirely to the right of the baseline range with no overlap, the system flags it as a winner:
- A green Winner Found badge means the system is confident a variation outperforms the control. Roll out this variation to all users.
- An orange Inconclusive badge means no variation delivered a significant impact. Run the experiment longer, or revise your strategy and test again.
Løste denne artikel dit problem?
Giv os besked, så vi kan forbedre os!
