Machine learning-driven servicing interface
Systems and methods for customizing business applications based upon user intent scores is described. A machine learning model trained to specifically predict when a user is likely to engage in a specific activity while interacting with the business application may be trained using data regarding prior interactions between a business application and a plurality of users. The machine learning model may thereafter provide a probability score for a particular user, the probability score indicating the likelihood that the user will engage in the specific activity for which the model has been trained to predict. The probability may be combined with a business value factor to produce a user intent score indicating the relative value of the user engaging in the specific activity. A business application comprising an app executing on a client device, a webpage, an automated menu at a call-in service center, or human operator interacting with the user at a call in the service center may be customized for the particular user based upon the user intent scores.
1. A computer-implemented method, comprising:
receiving, by a server and from a business application, a request for user intent scores for one or more user intents of a user, the request including information regarding user interactions with a reprogrammable customized automated audio menu of the business application, and wherein each user intent score corresponds to one of the one or more user intents;
determining, by the server, business value factors associated with each of the one or more user intents, wherein each of the business value factors corresponds to one of the one or more user intents, and each business value factor indicates a relative benefit of the user engaging in a specific activity versus other activities;
determining, by the server, organic relevance factors associated with each of the one or more user intents;
determining, by the server using a user identifier for the user, trained machine learning models;
utilizing, by the server, the trained machine learning models to determine probabilities of specific activities occurring, wherein the trained machine learning models are trained with data associated with the user, and each probability of the probabilities to indicate that a corresponding specific activity will occur relating to the one or more user intents based at least in part on the information regarding user interactions with the business application;
calculating, by the server, user intent scores for the one or more user intents, each of the user intent scores comprising a combination of the business value factor and organic relevance factors for each of the one or more user intents and a probability of the probabilities the corresponding specific activity will occur, each probability of a corresponding specific activity combined with a corresponding business value factor and corresponding relevance factors;
causing, by the server, communication of the user intent scores to the business application; and
modifying, by the server based on the user intent scores, the business application to provide an update for the reprogrammable customized automated audio menu.
2. The computer-implemented method of claim 1 , wherein the one or more user intents include a cross-sell propensity of the user acquiring a specific product, and the method comprising training the trained machine learning models using account history-related variables of an account of the user and digital clickstream data regarding the interaction of the user with the business application as input.
3. The computer-implemented method of claim 1 , wherein the one or more user intents include a feature recommendation indicating a feature of the business application that the customer is most likely to use next and a user intent prediction of a task that the customer is likely to initiate, and the method comprising training the trained machine learning models using application feature usage and recent transaction behavior.
4. The computer-implemented method of claim 1 , wherein generation of the user intent scores occurs in a batch mode and wherein the user intent scores are calculated based upon user data retrieved from a user data database for all or a subset of users and wherein user intent scores are updated for all or the subset of users and stored in a user intent scores database.
5. The computer-implemented method of claim 1 , wherein retrieving the user intent scores comprises retrieving the user intent scores from the user intent scores database.
6. The computer-implemented method of claim 1 , wherein generation of the user intent scores occurs in real-time and further wherein the user intent scores are based upon static user data retrieved from a user data database and are further based on user actions while interacting with the business application during a current session between the user and the business application.
7. A non-transitory, computer-readable storage medium containing instructions that, when executed by one or more processors of one or more servers, cause the one or more processors to:
receive, from a business application, a request for user intent scores for one or more user intents, the request including information regarding user interactions with a reprogrammable customized automated audio menu of the business application, and wherein each user intent score corresponds to one of the one or more user intents;
determine business value factors associated with each of the one or more user intents, wherein each of the business value factors corresponds to one of the one or more user intents, and each business value factor indicates a relative benefit of the user engaging in a specific activity versus other activities;
determine organic relevance factors associated with each of the one or more user intents;
determine trained machine learning models associated with the user;
utilize the trained machine learning models to determine probabilities of specific activities occurring, wherein the trained machine learning models are trained with data for the user, and each probability of the probabilities to indicate that a corresponding specific activity will occur relating to the one or more user intents based at least in part on the information regarding user interactions with the business application;
calculate user intent scores for the one or more user intents, each of the user intent scores comprising a combination of a business value factor and organic relevance factors for each of the one or more user intents and a probability of the probabilities the corresponding specific activity will occur, each probability of a corresponding specific activity combined with a corresponding business value factor and corresponding relevance factors;
cause communication of the user intent scores to the business application; and
reprogram, based on the user intent scores, the business application to provide an update for the reprogrammable customized automated audio menu.
8. The non-transitory, computer-readable storage medium of claim 7 ,
wherein the one or more user intents include a cross-sell propensity of the user acquiring a specific product; and
the processor to train the trained machine learning models using account history-related variables of an account of the user and digital clickstream data regarding the interaction of the user with the business application as input.
9. The non-transitory, computer-readable storage medium of claim 7 ,
wherein the one or more user intents include a feature recommendation indicating a feature of the business application that the customer is most likely to use next and a user intent prediction of a task that the customer is likely to initiate; and
the processor to train the trained machine learning models using application feature usage and recent transaction behavior.
10. The non-transitory, computer-readable storage medium of claim 7 ,
wherein generation of the user intent scores occurs in a batch mode and wherein the user intent scores are calculated based upon user data retrieved from a user data database for all or a subset of users and wherein user intent scores are updated for all or the subset of users and stored in a user intent scores database.
11. The non-transitory, computer-readable storage medium of claim 7 ,
wherein retrieving the user intent scores comprises retrieving the user intent scores from the user intent scores database.
12. The non-transitory, computer-readable storage medium of claim 7 ,
wherein generation of the user intent scores occurs in real-time and further wherein the user intent scores are based upon static user data retrieved from a user data database and are further based on user actions while interacting with the business application during a current session between the user and the business application.
13. The non-transitory, computer-readable storage medium of claim 7 ,
wherein the machine learning models are periodically retrained based upon an evaluation of a quality of estimates generated by the machine learning models.
14. A system comprising:
one or more servers, the one or more servers comprising:
a memory to store instructions; and
processing circuitry, coupled with the memory, operable to execute the instructions that when executed enable the processing circuitry to:
receive, from a business application, a request for user intent scores for one or more user intents regarding a user interacting with reprogrammable customized automated audio menus of the business application, and wherein each user intent score corresponds to one of the one or more user intents;
retrieve business value factors for each of the one or more user intents, wherein each of the business value factors corresponds to one of the one or more user intents, and each business value factor indicates a relative benefit of the user engaging in a corresponding specific activity versus other activities;
determine trained machine learning models associated with the user;
determine probabilities with the trained machine learning models, wherein each of the trained machine learning models is trained with data associated with the user, and each probability of the probabilities to indicate that the user will engage in the corresponding specific activity related to the one or more user intents while interacting with the business application;
calculate the user intent scores for the one or more user intents, each of the user intent scores comprising a combination of the business value factor and a probability of the probabilities that the user will engage in the corresponding specific activity;
send, to the business application, the user intent scores quantifying user engagement for each corresponding specific activity; and
reprogram the business application to provide an update for the reprogrammable customized automated audio menus based on the user intent scores.
15. The system of claim 14 wherein:
the one or more user intents include a cross-sell propensity of the user acquiring a specific product; and
the machine learning models are trained using account history-related variables of an account of the user and digital clickstream data regarding the interaction of the user with the business application as input.
16. The system of claim 14 wherein:
the one or more user intents include a feature recommendation indicating a feature of the business application that the customer is most likely to use next and a user intent prediction of a task that the customer is likely to initiate; and
the machine learning models are further trained using application feature usage and recent transaction behavior.
17. The system of claim 14 wherein generation of the user intent scores occurs in a batch mode and wherein the user intent scores are calculated based upon user data retrieved from a user data database for all or a subset of users and wherein user intent scores are updated for all or the subset of users and stored in a user intent scores database.
18. The system of claim 17 wherein retrieving the user intent scores comprises retrieving the user intent scores from the user intent scores database.
19. The system of claim 14 wherein generation of the user intent scores occurs in real-time and further wherein the user intent scores are based upon static user data retrieved from a user data database and are further based on user actions while interacting with the business application during a current session between the user and the business application.
20. The system of claim 14 wherein the machine learning models are periodically retrained based upon an evaluation of a quality of estimates generated by the machine learning models.