Churn prediction with machine learning

Abstract

Disclosed is a churn prediction system that predicts with a high level of accuracy which users will and which users will not stop opening the app over a 30-day time period. To this end a model is created using historical event data where the churn-related behavior of each user is known. New event data is then applied to the model to determine the likelihood of each user churning in the future. With these prediction scores a user is then qualified as falling into one of three classifications: low-risk, medium-risk, or high-risk of churn.

What is claimed is: 
     
         1 . A computer-implemented method, comprising:
 receiving a plurality of identifiers from a plurality of business customers, each identifier associated with a user device of a user of a business customer, each business customer having a set of identifiers;   receiving a selection of a churn definition for each business customer, the churn definition being specific to each business customer;   collecting historical event data describing events associated with the plurality of identifiers, the events related to events occurred in connection with usage of mobile applications of users of the business customers;   using the historical event data to train, for each business customer, a machine learning model based on the churn definition specific to each business customer, the machine learning model predicting user-churn likelihoods, a user-churn likelihood predicting a likelihood of a user churning according to the churn definition;   segmenting, for each business customer, the users based on the user-churn likelihoods; and   sending notifications to the users based on the segmenting.   
     
     
         2 . The method of  claim 1 , wherein the machine learning model is a decision tree. 
     
     
         3 . The method of  claim 1 , wherein the machine learning model uses a gradient boosting algorithm. 
     
     
         4 . The method of  claim 1 , further comprising:
 generating churn scores for the users of each business customer, wherein the user-churn likelihoods are based on the churn scores; and   presenting one or more churn scores in a graphical user interface.   
     
     
         5 . The method of  claim 1 , further comprising:
 presenting a suggestion to one of the business customers to take an action to reduce the user-churn likelihood of one or more users.   
     
     
         6 . The method of  claim 1 , further comprising:
 transmitting a notification to one of the users responsive to the user-churn likelihood of said one of the users exceeding a threshold.   
     
     
         7 . The method of  claim 1 , further comprising:
 presenting suggestions to one of the business customers to take different actions on different users based on the user-churn likelihoods of the users.   
     
     
         8 . A system comprising:
 one or more processors; and   memory coupled to the one or more processors, the memory configured to store computer code comprising instructions, the instructions, when executed by the one or more processors, causes the one or more processors to:
 receive a plurality of identifiers from a plurality of business customers, each identifier associated with a user device of a user of a business customer, each business customer having a set of identifiers; 
 receive a selection of a churn definition for each business customer, the churn definition being specific to each business customer; 
 collect historical event data describing events associated with the plurality of identifiers, the events related to events occurred in connection with usage of mobile applications of users of the business customers; 
 use the historical event data to train, for each business customer, a machine learning model based on the churn definition specific to each business customer, the machine learning model predicting user-churn likelihoods, a user-churn likelihood predicting a likelihood of a user churning according to the churn definition; 
 segment, for each business customer, the users based on the user-churn likelihoods; and 
 send notifications to the users based on the segmenting. 
   
     
     
         9 . The system of  claim 8 , wherein the machine learning model is a decision tree. 
     
     
         10 . The system of  claim 8 , wherein the machine learning model uses a gradient boosting algorithm. 
     
     
         11 . The system of  claim 8 , wherein the instructions, when executed, further causes the one or more processors to:
 generating churn scores for the users of each business customer, wherein the user-churn likelihoods are based on the churn scores; and   presenting one or more churn scores in a graphical user interface.   
     
     
         12 . The system of  claim 8 , wherein the instructions, when executed, further causes the one or more processors to:
 present a suggestion to one of the business customers to take an action to reduce the user-churn likelihood of one or more users.   
     
     
         13 . The system of  claim 8 , wherein the instructions, when executed, further causes the one or more processors to:
 transmit a notification to one of the users responsive to the user-churn likelihood of said one of the users exceeding a threshold.   
     
     
         14 . The system of  claim 8 , wherein the instructions, when executed, further causes the one or more processors to:
 present suggestions to one of the business customers to take different actions on different users based on the user-churn likelihoods of the users.   
     
     
         15 . A non-transitory computer readable medium configured to store computer code comprising instructions, the instructions, when executed by one or more processors, causes the one or more processors to:
 receive a plurality of identifiers from a plurality of business customers, each identifier associated with a user device of a user of a business customer, each business customer having a set of identifiers;   receive a selection of a churn definition for each business customer, the churn definition being specific to each business customer;   collect historical event data describing events associated with the plurality of identifiers, the events related to events occurred in connection with usage of mobile applications of users of the business customers;   use the historical event data to train, for each business customer, a machine learning model based on the churn definition specific to each business customer, the machine learning model predicting user-churn likelihoods, a user-churn likelihood predicting a likelihood of a user churning according to the churn definition;   segment, for each business customer, the users based on the user-churn likelihoods; and   send notifications to the users based on the segmenting.   
     
     
         16 . The non-transitory computer readable medium of  claim 15 , wherein the machine learning model is a decision tree. 
     
     
         17 . The non-transitory computer readable medium of  claim 15 , wherein the machine learning model uses a gradient boosting algorithm. 
     
     
         18 . The non-transitory computer readable medium of  claim 15 , wherein the instructions, when executed, further causes the one or more processors to:
 generate churn scores for the users of each business customer, wherein the user-churn likelihoods are based on the churn scores; and   present one or more churn scores in a graphical user interface.   
     
     
         19 . The non-transitory computer readable medium of  claim 15 , wherein the instructions, when executed, further causes the one or more processors to:
 present a suggestion to one of the business customers to take an action to reduce the user-churn likelihood of one or more users.   
     
     
         20 . The non-transitory computer readable medium of  claim 15 , wherein the instructions, when executed, further causes the one or more processors to:
 transmit a notification to one of the users responsive to the user-churn likelihood of said one of the users exceeding a threshold.