Systems and methods for predicting churn in a multi-tenant system
Store a subscription dataset for each tenant of a multi-tenant system, each of the subscription datasets having a common data format native to the system, the subscription datasets including billing data and not behavioral data. Determine primary features from a particular subscription dataset. Derive secondary features from the primary features. Generate a churn prediction model based on the primary features and the secondary features. Obtain a second subscription dataset, the second subscription dataset comprising billing data that is more recent than the particular subscription dataset. Identify, using the particular churn prediction model and the second subscription dataset, one or more subscribers as a churn-risk. Report the one or more subscribers identified as a churn-risk.
1 . A multi-tenant system comprising:
one or more processors;
a plurality of tenant interfaces configured to support load balancing when multiple tenants of the multi-tenant system access the multi-tenant system concurrently, each tenant of the multiple tenants being associated with a subset of the plurality of tenant interfaces for load balancing; and
memory storing instructions that, when executed by the one or more processors, cause the multi-tenant system to perform:
storing a respective subscription dataset for the each tenant of the multiple tenants of the multi-tenant system, each respective subscription dataset having one or more common data formats native to the multi-tenant system, each respective subscription dataset including first billing data of a first subset of subscribers of multiple subscribers of the tenant and not including product or service usage data of the first subset of the subscribers of the multiple subscribers of the tenant;
obtaining, by a churn prediction engine, the respective subscription dataset for a particular tenant;
determining, by the churn prediction engine, one or more primary features from the particular respective subscription dataset of the particular tenant of the multi-tenant system;
deriving, by the churn prediction engine, one or more secondary features from the one or more primary features;
generating, by the churn prediction engine, a particular churn prediction model for the particular tenant based on the one or more primary features and the one or more secondary features, the churn prediction engine using a same machine learning algorithm capable of operating on the one or more common data formats native to the multi-tenant system so that the same machine learning algorithm can generate different churn prediction models for different tenants of the multi-tenant system;
obtaining, by the churn prediction engine, a second subscription dataset of the particular tenant of the multi-tenant system, the second subscription dataset comprising second billing data of a second subset of the subscribers of the particular tenant of the multi-tenant system that is more recent than the respective subscription dataset of the particular tenant of the multi-tenant system and not comprising product or service usage data of the second subset of the subscribers of the particular tenant;
identifying, by the churn prediction engine using the particular churn prediction model and the second subscription dataset of the particular tenant of the multi-tenant system, one or more particular subscribers of the second subset of the subscribers of the particular tenant of the multi-tenant system as a churn-risk; and
reporting the one or more particular subscribers of the particular tenant of the multi-tenant system identified as a churn-risk using at least one of the subset of the plurality of tenant interfaces associated with the particular tenant.
2 . The multi-tenant system of claim 1 , wherein the instructions further cause the multi-tenant system to perform:
determining, by the churn prediction engine, one or more different primary features from a different respective subscription dataset of a different tenant of the multi-tenant system;
deriving, by the churn prediction engine, one or more different secondary features from the one or more different primary features;
generating, by the churn prediction engine, a different churn prediction model based on the one or more different primary features and the one or more different secondary features;
obtaining, by the churn prediction engine, a third subscription dataset of the different tenant of the multi-tenant system, the third subscription dataset comprising billing data that is more recent than the different respective subscription dataset of the different tenant of the multi-tenant system;
identifying, by the churn prediction engine using the different churn prediction model and the third subscription dataset of the different tenant of the multi-tenant system, one or more different subscribers of the different tenant of the multi-tenant system as a churn-risk; and
reporting the one or more different subscribers of the different tenant of the multi-tenant system identified as a churn-risk.
3 . The multi-tenant system of claim 2 , wherein the one or more common data formats native to the multi-tenant system are independent of third-party data formats, and are not based on an integration of third-party data.
4 . The multi-tenant system of claim 2 , wherein the churn prediction engine is configured to use the same machine learning algorithm for the particular churn prediction model and the different churn prediction model.
5 . The multi-tenant system of claim 4 , wherein the same machine learning algorithm comprises a regression analysis.
6 . The multi-tenant system of claim 1 , wherein the one or more primary features include date of last payment and subscription status.
7 . The multi-tenant system of claim 6 , wherein the one or more secondary features include length of time since the date of last payment.
8 . The multi-tenant system of claim 6 , wherein the one or more secondary features include any of measures and statistics of plan information keywords.
9 . The multi-tenant system of claim 6 , wherein the one or more secondary features include any of measures and statistics of subscription billing amounts.
10 . The multi-tenant system of claim 6 , wherein the one or more secondary features include any of measures and statistics of subscription history.
11 . The multi-tenant system of claim 6 , wherein the one or more secondary features include any of features derived from use by subscribers of products from different tenants on the multi-tenant system.
12 . The multi-tenant system of claim 1 , wherein the respective subscription dataset of the particular tenant of the multi-tenant system includes a limited number of examples of churn.
13 . A method being implemented by a computing system including one or more physical processors, storage media storing machine-readable instructions, and a plurality of tenant interfaces configured to support load balancing when multiple tenants of a multi-tenant system access the multi-tenant system concurrently, each tenant of the multiple tenants being associated with a subset of the plurality of tenant interfaces for load balancing, the method comprising:
storing a respective subscription dataset for the each tenant of the multiple tenants of the multi-tenant system, each respective subscription dataset having one or more common data formats native to the multi-tenant system, each respective subscription dataset including first billing data of a first subset of subscribers of multiple subscribers of the tenant and not including product or service usage data of the first subset of the subscribers of the multiple subscribers of the tenant;
obtaining, by a churn prediction engine, the respective subscription dataset for a particular tenant;
determining, by the churn prediction engine, one or more primary features from the particular respective subscription dataset of the particular tenant of the multi-tenant system;
deriving, by the churn prediction engine, one or more secondary features from the one or more primary features;
generating, by the churn prediction engine, a particular churn prediction model for the particular tenant based on the one or more primary features and the one or more secondary features, the churn prediction engine using a same machine learning algorithm capable of operating on the one or more common data formats native to the multi-tenant system so that the same machine learning algorithm can generate different churn prediction models for different tenants of the multi-tenant system;
obtaining, by the churn prediction engine, a second subscription dataset of the particular tenant of the multi-tenant system, the second subscription dataset comprising second billing data of a second subset of the subscribers of the particular tenant of the multi-tenant system that is more recent than the respective subscription dataset of the particular tenant of the multi-tenant system and not comprising product or service usage data of the second subset of the subscribers of the particular tenant;
identifying, by the churn prediction engine using the particular churn prediction model and the second subscription dataset of the particular tenant of the multi-tenant system, one or more particular subscribers of the second subset of the subscribers of the particular tenant of the multi-tenant system as a churn-risk; and
reporting the one or more particular subscribers of the particular tenant of the multi-tenant system identified as a churn-risk using at least one of the subset of the plurality of tenant interfaces associated with the particular tenant.
14 . The method of claim 13 , further comprising:
determining, by the churn prediction engine, one or more different primary features from a different respective subscription dataset of a different tenant of the multi-tenant system;
deriving, by the churn prediction engine, one or more different secondary features from the one or more different primary features;
generating, by the churn prediction engine, a different churn prediction model based on the one or more different primary features and the one or more different secondary features;
obtaining, by the churn prediction engine, a third subscription dataset of the different tenant of the multi-tenant system, the third subscription dataset comprising billing data that is more recent than the different respective subscription dataset of the different tenant of the multi-tenant system;
identifying, by the churn prediction engine using the different churn prediction model and the third subscription dataset of the different tenant of the multi-tenant system, one or more different subscribers of the different tenant of the multi-tenant system as a churn-risk; and
reporting the one or more different subscribers of the different tenant of the multi-tenant system identified as a churn-risk.
15 . The method of claim 14 , wherein the one or more common data formats native to the multi-tenant system are independent of third-party data formats, and are not based on an integration of third-party data.
16 . The method of claim 14 , wherein the churn prediction engine is configured to use the same machine learning algorithm for the particular churn prediction model and the different churn prediction model.
17 . The method of claim 16 , wherein the same machine learning algorithm comprises a regression analysis.
18 . The method of claim 13 , wherein the one or more primary features include date of last payment and subscription status.
19 . The method of claim 18 , wherein the one or more secondary features include length of time since the date of last payment.
20 . A non-transitory computer readable medium comprising computer readable instructions configured to, when executed, cause one or more physical processors to perform:
providing a plurality of tenant interfaces configured to support load balancing when multiple tenants of a multi-tenant system access the multi-tenant system concurrently, each tenant of the multiple tenants being associated with a subset of the plurality of tenant interfaces for load balancing;
storing a respective subscription dataset for the each tenant of the multiple tenants of the multi-tenant system, each respective subscription dataset having one or more common data formats native to the multi-tenant system, each respective subscription dataset including first billing data of a first subset of subscribers of multiple subscribers of the tenant and not including product or service usage data of the first subset of the subscribers of the multiple subscribers of the tenant;
obtaining, by a churn prediction engine, the respective subscription dataset for a particular tenant;
determining, by the churn prediction engine, one or more primary features from the particular respective subscription dataset of the particular tenant of the multi-tenant system;
deriving, by the churn prediction engine, one or more secondary features from the one or more primary features;
generating, by the churn prediction engine, a particular churn prediction model for the particular tenant based on the one or more primary features and the one or more secondary features, the churn prediction engine using a same machine learning algorithm capable of operating on the one or more common data formats native to the multi-tenant system so that the same machine learning algorithm can generate different churn prediction models for different tenants of the multi-tenant system;
obtaining, by the churn prediction engine, a second subscription dataset of the particular tenant of the multi-tenant system, the second subscription dataset comprising second billing data of a second subset of the subscribers of the particular tenant of the multi-tenant system that is more recent than the respective subscription dataset of the particular tenant of the multi-tenant system and not comprising product or service usage data of the second subset of the subscribers of the particular tenant;
identifying, by the churn prediction engine using the particular churn prediction model and the second subscription dataset of the particular tenant of the multi-tenant system, one or more particular subscribers of the second subset of the subscribers of the particular tenant of the multi-tenant system as a churn-risk; and
reporting the one or more particular subscribers of the particular tenant of the multi-tenant system identified as a churn-risk using at least one of the subset of the plurality of tenant interfaces associated with the particular tenant.