How do we fix a customer churn rate and time period for churn flag definition?

In my example there are various customers who buys multiple products from a firm. The customer churn prediction should be done to identify churn and take action . What sort of analysis do we do to decide on the definition of a churn flag ? How should the customer rate and time period decided ? For eg : should it be month on month , qtr on qtr , year on year . Should it be >=10% , >=20% reduction in purchase ? What sort of analysis do we do to decide that ?