If you are running an A/B test to select the best template or subject line, you can access the detailed analysis under the Reports > A/B Test Results tab of the campaign studio. Depending upon the metric you are optimizing for (for example, clicks, impressions, or a custom goal), the report shows which variation had the best performance with respect to the control along with the confidence level of the results.
In the A/B test report, Blueshift will indicate whether the results were conclusive, i.e. statistically significant with a confidence level = 90%, and if there was a clear winner based on the results. The report will also show the confidence level in case your team has a different criteria for statistical significance.
If you are using automatic winner selection to automate your A/B tests, remember that the results are not guaranteed to be statistically significant.
To learn more about A/B testing, or to set up A/B testing for a trigger, see A/B Testing.
During A/B testing, if you want to compare the winning variation from a previous A/B test against new variations, we recommend that you first archive the winning variation and then clone it. Compare the cloned variation against the new variations in your new A/B test. This ensures that the new A/B test is isolated from your previous A/B test and that the results from your previous A/B test do not skew the results of your new A/B test.
To view the A/B test report, go to the Reports tab for the campaign.
- Select the Trigger for which you want to view the A/B test report.
- Select the Control variation against which other variations are to be compared. By default, the first variation is selected as the control and all the other variations are compared to it.
- Select the Metric that you want to optimize for. By default, the unique click rate is used as the metric to compare the performance of the different variations.
- Select the Date Range for the report. For example, you can view the A/B test results for the time period during which the winner was selected for automatic winner selection.
- The default start time is the campaign start time. If the campaign start time is earlier than 10.01.2019, then 10.01.2019 is selected as the start time for any A/B test analysis.
- The default end time for an active campaign is the current date (today's date). For a campaign that is paused or completed, the default end time is the campaign end date + 7 days.
You can also view the A/B Test results for archived triggers by selecting the Show Archived Triggers option.
In the following example, the A/B Test results for the trigger Send an email are displayed. The trigger has four variations of which Hello World is selected as the baseline or control variation. The Metric that is selected for the report is Clicks.
The details for all four variations are displayed. We are the world is the winning variation.
The following details are available in the A/B test report.
The total number of users to whom the particular variation was sent at the end of the selected time period.
For example, if the time window for the campaign is from June 1 to July 31, the count includes all unique users who were sent the particular variation from the start of the campaign until July 31.
In the example, the variation Hello World was sent to 5699 unique users, whereas the variation We are the World was sent to 377885 unique users.
The number of unique users to whom the particular variation was sent and who completed the selected metric during the time period.
For example, if the time window for the campaign is from June 1 to July 31, the count includes all unique users who to whom the particular variation was sent and who completed the selected metric from June 1 to July 31.
In the example, 155 unique users who received the variation Hello World clicked on it compared to 11865 unique users who received the variation We are the World.
The number of users that converted, i.e. completed the selected metric, as a percentage of the total number of users in the group.
Conversion = Unique Completed/Unique Users
For example, if the time window for the campaign is from June 1 to July 31, the Unique Completed count (x) includes all unique users to whom the particular variation was sent and who completed the selected metric from June 1 to July 31. Conversion for this time window is x/Unique Users.
In the example, for the variation Hello World, of the 5699 unique users who received the variation, only 155 unique users clicked on it. Hence Conversion = (155/5699) * 100 = 2.72%.
For the variation We are the World, of the 377885 unique users who received the variation, 11865 unique users clicked on it. Hence Conversion = (11865/377885) * 100 = 3.14%.
The Lift% for a variation is calculated by comparing the conversion for that variation with the conversion for the baseline or control variation.
Lift%variation = (Conversionvariation/ConversionControl variation) - 1
If the Lift % > 0, it is an improvement. If the Lift % < 0, it is a degradation.
In the example, the variation Hello World is set as the control variation. variation Hello Universe has a Lift % of -13.893%, whereas the variation We are the World has a Lift % of 15.445%. Hence the Lift % for the variation We are the World is an improvement over the control variation.
The statistical likelihood or probability (p’%) that the improvement (or degradation) observed from the A/B test is correct. So if you were to infinitely repeat this test, you would observe the same improvement p’% of the time.The 𝝌2 test (chi-squared test) is used to calculate the confidence level.
p’ = 1 - p(𝝌2)
|When the Confidence Level is higher than 90%, this indicates that the results are statistically significant. However, you can use a higher or lower confidence level based on your A/B test objectives.
|The range of values within which the conversion for the group will lie p’% (i.e. confidence level percentage) of the time.
|The highest performing variation is selected as the winner as long as the results are statistically significant.