Working with cohorts

[sumaiyakhatun27]

User behavior in a service depends on the duration of interaction with the product. New and regular customers react differently to changes, such as the rollout of new features or the influence of marketing mechanics, so their return rate (retention) can vary greatly.

To accurately forecast orders, it is better to analyze data by cohorts. By cohorts macedonia mobile database we mean groups of users united by certain characteristics, for example, by the date of the first purchase.

First, we examine customers by cohort and look at the following sections:

month of first purchase;

platform (iOS, Android, Web);

acquisition channel (organic, paid channels, etc.).

For each of them, a separate forecast of orders for a given period is built, and then the data is combined to obtain the total volume of orders.

Order forecasting

We distinguish 3 groups of cohorts:

old (more than 8 months of observations) - predicted based on accumulated data;

relatively new (less than 8, but not less than 3 months) - we build a forecast based on the first part of the curve and supplement it with median values ​​from similar old cohorts;

new (less than 2 months) - we use average indicators for similar cohorts of the past.