Machine-learning models to facilitate user retention for software applications
An ordered combination of machine-learning models may be used to identify users who are likely to abandon use of an application, predict the reasons why those users are likely to abandon, and identify intervening actions that the application can perform to reduce the probability that the users will abandon the application. For example, a first machine-learning model determines a retention-prediction value indicating a probability that the user will complete a target action in the application before a session terminates. If the retention-prediction value satisfies a threshold condition, a second machine-learning model determines a reason why the session is likely to terminate before the user completes the target action. A third machine-learning model determines an intervention action for the application to perform to increase the probability that the user will complete the target action before the session terminates.
1. A system comprising:
one or more processors and memory storing one or more instructions that, when executed on the one or more processors, cause the system to:
send one or more web pages for display to a user via a network during an interaction session between the user and an application, wherein the one or more web pages include elements for collecting response data from the user;
receive, via the web pages, response data from the user,
collect, via the application, additional data that characterizes user behavior during the interaction session;
generate a composite data set from the response data and the additional data;
determine, via a first machine-learning model based on the composite data set, a retention-prediction value indicating a probability that the user will complete a target action in the application before the interaction session terminates;
determine that the retention-prediction value satisfies a threshold condition;
determine, via a second machine-learning model based on the composite data set and the retention-prediction value determined by the first machine-learning model, a predicted reason the interaction session is likely to terminate before the user completes the target action;
determine, via a third machine-learning model based on the composite data set and the predicted reason determined by the second machine-learning model, an intervention action for increasing the probability that the user will complete the target action before the interaction session terminates; and
perform, via the application, the intervention action.
2. The system of claim 1 , wherein the instructions, when executed on the one or more processors, further cause the system to:
determine, via a fourth machine-learning model based on the composite data set, a next action the user is anticipated to perform in the application; and
alter at least one aspect of the one or more web pages to facilitate user performance of the next action.
3. The system of claim 2 , wherein the instructions, when executed on the one or more processors, further cause the system to:
fetch, via the application, data related to the next action determined via the fourth machine-learning model.
4. The system of claim 1 , wherein determining the intervention action via the third machine-learning model includes: inputting a set of input features that includes the retention-prediction value into the third machine-learning model.
5. The system of claim 1 , wherein the instructions, when executed on the one or more processors, further cause the system to:
receive, via the web pages, updated response data from the user;
collect, via the application, updated additional data that characterizes user behavior during the interaction session;
generate an updated composite data set that includes both the updated response data and the updated additional data; and
determine, via the first machine-learning model based on the updated composite data set, an updated retention-prediction value indicating an updated probability that the user will complete the target action in the application before the interaction session terminates.
6. The system of claim 5 , wherein the instructions, when executed on the one or more processors, further cause the system to:
subtract the retention-prediction value from the updated retention-prediction value to determine a difference;
divide the difference by a time interval to determine a rate of change; and
determine an updated time interval based on the rate of change.
7. The system of claim 1 , wherein performing the intervention action includes:
open, via the application, a messaging interface; and
establish a network connection with a live support agent to allow the user to communicate with the live support agent through the messaging interface.
8. A non-transitory computer-readable storage medium containing instructions that, when executed by one or more processors, perform an operation comprising:
sending one or more web pages for display to a user via a network during an interaction session between the user and an application, wherein the one or more web pages include elements for collecting response data from the user;
receiving, via the web pages, the response data from the user;
collecting, via the application, additional data that characterizes user behavior during the interaction session;
generating a composite data set from the response data and the additional data;
determining, via a first machine-learning model based on the composite data set, a retention-prediction value indicating a probability that the user will complete a target action in the application before the interaction session terminates;
determining that the retention-prediction value satisfies a threshold condition;
determining, via a second machine-learning model based on the composite data set and the retention-prediction value determined by the first machine-learning model, a predicted reason the interaction session is likely to terminate before the user completes the target action;
determining, via a third machine-learning model based on the composite data set and the predicted reason determined by the second machine-learning model, an intervention action for increasing the probability that the user will complete the target action before the interaction session terminates; and
performing, via the application, the intervention action.
9. The non-transitory computer-readable storage medium of claim 8 , wherein the operation further comprises:
determining, via a fourth machine-learning model based on the composite data set, a next action the user is anticipated to perform in the application; and
altering at least one aspect of the one or more web pages to facilitate user performance of the next action.
10. The non-transitory computer-readable storage medium of claim 9 , wherein the operation further comprises:
fetching, data related to the next action determined via the fourth machine-learning model.
11. The non-transitory computer-readable storage medium of claim 8 , wherein determining the intervention action via the third machine-learning model includes:
inputting a set of input features that includes the retention-prediction value into the third machine-learning model.
12. The non-transitory computer-readable storage medium of claim 8 , wherein the operation further comprises:
receiving, via the web pages, updated response data from the user;
collecting, via the application, updated additional data that characterizes user behavior during the interaction session;
generating an updated composite data set that includes both the updated response data and the updated additional data; and
determining, via the first machine-learning model based on the updated composite data set, an updated retention-prediction value indicating an updated probability that the user will complete the target action in the application before the interaction session terminates.
13. The non-transitory computer-readable storage medium of claim 12 , wherein the operation further comprises:
subtracting the retention-prediction value from the updated retention-prediction value to determine a difference;
dividing the difference by a time interval to determine a rate of change; and
determining an updated time interval based on the rate of change.
14. The non-transitory computer-readable storage medium of claim 8 , wherein performing the intervention action includes:
opening, via the application, a messaging interface; and
establishing a network connection with a live support agent to allow the user to communicate with the live support agent through the messaging interface.
15. A method comprising:
sending one or more web pages for display to a user via a network during an interaction session between the user and an application, wherein the one or more web pages include elements for collecting response data from the user;
receiving, via the web pages, the response data from the user;
collecting, via the application, additional data that characterizes user behavior during the interaction session;
generating a composite data set from the response data and the additional data;
determining, via a first machine-learning model based on the composite data set, a retention-prediction value indicating a probability that the user will complete a target action in the application before the interaction session terminates;
determining that the retention-prediction value satisfies a threshold condition;
determining, via a second machine-learning model based on the composite data set and the retention-prediction value determined by the first machine-learning model, a predicted reason the interaction session is likely to terminate before the user completes the target action;
determining, via a third machine-learning model based on the composite data set and the predicted reason determined by the second machine-learning model, an intervention action for increasing the probability that the user will complete the target action before the interaction session terminates; and
performing, via the application, the intervention action.
16. The method of claim 15 , further comprising:
determining, via a fourth machine-learning model based on the composite data set, a next action the user is anticipated to perform in the application; and
altering at least one aspect of the one or more web pages to facilitate user performance of the next action.
17. The method of claim 16 , further comprising:
fetching, data related to the next action determined via the fourth machine-learning model.
18. The method of claim 15 , wherein determining the intervention action via the third machine-learning model includes:
inputting a set of input features that includes the retention-prediction value into the third machine-learning model.
19. The method of claim 15 , further comprising:
receiving, via the web pages, updated response data from the user;
collecting, via the application, updated additional data that characterizes user behavior during the interaction session;
generating an updated composite data set that includes both the updated response data and the updated additional data; and
determining, via the first machine-learning model based on the updated composite data set, an updated retention-prediction value indicating an updated probability that the user will complete the target action in the application before the interaction session terminates.
20. The method of claim 19 , further comprising:
subtracting the retention-prediction value from the updated retention-prediction value to determine a difference;
dividing the difference by a time interval to determine a rate of change; and
determining an updated time interval based on the rate of change.