Predicting site visit based on intervention

1. A computer system, comprising:

a processor;

a memory device holding an instruction set executable on the processor to cause the computer system to perform operations comprising:

identifying a first user of users of a website visits the website at a rate less than a second specified threshold;

only in response to identifying that the first user visits the website at the rate less than the second specified threshold:

determining a first probability that the first user will visit the website within a specified time window if the first user is provided an intervention at a specified time;

determining a second probability that the first user will visit the website within the specified time window without being provided the intervention; and

determining a difference between the first and second probability; and

in response to determining the difference is greater than a first specified threshold, providing the intervention at the specified time.

2. The computer system of claim 1 , wherein the second specified threshold is one or two times per month.

3. The computer system of claim 1 , wherein determining the first probability includes using an accelerated failure time model constrained by a Weibull distribution.

4. The computer system of claim 3 , further comprising training the accelerated failure time model based on historical session counts of the user, a count of interventions received in a specified time period, and a last date the user visited the website.

5. The computer system of claim 4 , wherein the accelerated failure time model is further trained based on whether the user includes an app installed on a mobile device and push notifications are enabled for the app.

6. The computer system of claim 4 , wherein the first probability is determined offline and updated based on updated historical session counts of the user, updated count of interventions received in a specified time period, and updated last date the user visited the website.

7. The computer system of claim 4 , wherein training the accelerated failure time model further includes training based on censored and uncensored intervention events, wherein a censored intervention event is followed in time by another intervention event before a website visit and an uncensored intervention event is followed in time by a website visit before another intervention event.

8. A method comprising:

identifying a first user of users of a website visits the website at a rate less than a second specified threshold;

only in response to identifying that the first user visits the website at the rate less than the second specified threshold:

determining a first probability that the first user will visit the website within a specified time window if the first user is provided an intervention at a specified time;

determining a second probability that the first user will visit the website within the specified time window without being provided the intervention; and

determining a difference between the first and second probability; and

in response to determining the difference is greater than a first specified threshold, providing the intervention at the specified time.

9. The method of claim 8 , wherein the second specified threshold is one or two times per month.

10. The method of claim 8 , wherein determining the first probability includes using an accelerated failure time model constrained by a Weibull distribution.

11. The method of claim 10 , further comprising training the accelerated failure time model based on historical session counts of the user, a count of interventions received in a specified time period, and a last date the user visited the website.

12. The method of claim 11 , wherein the accelerated failure time model is further trained based on whether the user includes an app installed on a mobile device and push notifications are enabled for the app.

13. The method of claim 11 , wherein the first probability is determined offline and updated based on updated historical session counts of the user, updated count of interventions received in a specified time period, and updated last date the user visited the website.

14. The method of claim 11 , wherein training the accelerated failure time model further includes training based on censored and uncensored intervention events, wherein a censored intervention event is followed in time by another intervention event before a website visit and an uncensored intervention event is followed in time by a website visit before another intervention event.

15. A non-transitory machine-readable storage medium embodying instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

identifying a first user of users of a website visits the website at a rate less than a second specified threshold;

only in response to identifying that the first user visits the website at the rate less than the second specified threshold:

determining a first probability that the first user will visit the website within a specified time window if the first user is provided an intervention at a specified time;

determining a second probability that the first user will visit the website within the specified time window without being provided the intervention;

determining a difference between the first and second probability; and

in response to determining the difference is greater than a first specified threshold, providing the intervention at the specified time.

16. The non-transitory machine-readable storage medium of claim 15 , wherein the second specified threshold is one or two times per month.

17. The non-transitory machine-readable storage medium of claim 15 , wherein determining the first probability includes using an accelerated failure time model constrained by a Weibull distribution.