# Conversion growth and personalization proof
# Conversion growth
The ultimate goal of using personalization is to achieve conversion growth. Conversion growth is defined as the difference between the conversions of the control and personalized groups.
# Credible intervals of conversion growth
However, if the volume of accumulated data is not very large, we cannot distinguish real conversion growth from a random (lucky for us) distribution of conversions.
Therefore, we need to analyze not only the conversion growth value but also the credible interval of conversion growth:
For example, an 80-percent credible interval of conversion growth is a range (from and to some value) within which the original "real" conversion growth value lies with an 80% probability.
For a better understanding of the concept of the credible interval, see the corresponding article.
# Personalization proof
Personalization is considered proven when the chosen credible interval of conversion growth becomes fully positive (i.e., entirely greater than zero).
For example, a credible interval of conversion growth of [+2%; +26%] is fully positive, whereas [-1%; +23%] is not.
By default, we use a 90-percent credible interval of conversion growth.
However, depending on the traffic volume of your project and your goals, this interval can be made larger or smaller.
You can change the credible interval of conversion growth in the personalization settings:
Once proven, personalization is considered proven and will begin to be billed.
# Over what period is conversion growth assessed?
We proceed through the versions of the personalization model, starting from the last one. Then the last + penultimate, and so on.
Moving from the "end," the system will try to find the earliest combination of model versions that will yield a fully positive credible interval of conversion growth.
The main idea is to get the most current statistically significant personalization conversion growth.
For example, at the moment, personalization has 5 versions of the personalization model.
- The 5th version of the model gives a credible interval of conversion growth of [-15%; +34%].
- The 4th + 5th versions give a credible interval of conversion growth of [-3%; +24%].
- The 3rd + 4th + 5th versions give a credible interval of conversion growth of [+1; +17%].
This means that conversion growth will be calculated as the conversion growth for data starting from the 3rd version of the personalization model (and up to the 5th).
If we cannot get a fully positive credible interval of conversion growth, then this personalization is not proven. However, on the personalization statistics page by default, the versions of the models will be selected, the combination of which is closest to achieving a fully positive credible interval of conversion growth.
# Frequently asked questions
# What happens if the credible interval becomes positive and then negative again?
This is a perfectly acceptable scenario, especially when not much data has been accumulated.
Suppose we have proven personalization, as the chosen credible interval has become fully positive ([+2%; +26%]).
However, then, we received several conversions for the control group of the personalization, and the credible interval became [-0.5%; +24%].
In this case, personalization becomes "unproven" again and stops being billed.