Artificial intelligence (“AI”) integration with live chat

1. An interactive voice response (“IVR”) system comprising:

a telephony server that, in operation:

receives voice inputs submitted by a caller using a communication network;

provides an interface for the caller to interact with an artificial intelligence (“AI”) engine over the communication network; and

transmits to the caller, over the communication network, responses to the voice inputs generated by the AI engine; and

an application server hosting the AI engine that, in operation:

receives, from the telephony server, a first set of voice inputs generated by the caller;

applies a first machine learning model to the first set of voice inputs and generates a machine interpretation of the first set of voice inputs comprising identification of the caller and a current concern of the caller;

based on the machine interpretation of the first set of voice inputs, classifies the caller as requiring intervention by a human agent;

based on the classification, links a terminal accessible by the human agent to the application server; and

after linking the terminal to the application server:

receives, from the telephony server, a second set of voice inputs generated by the caller;

applies a second machine learning model to the second set of voice inputs and generates a machine interpretation of the second set of voice inputs, based on:

a target historical conversation determined to be associated with the current concern of the caller; and

a resolution applied to the target historical conversation;

intercepts responses to the second set of voice inputs provided by the human agent via the terminal with the machine interpretation of the second set of voice inputs by displaying to the human agent on the terminal the machine interpretation of the second set of voice inputs; and

wherein:

the intercepting reduces a duration of time the terminal is linked to the application server relative to a duration of time the terminal is linked to the application server without the intercepting;

the first machine learning model is utilized by the application server when the AI engine interacts exclusively with the caller;

the second machine learning model is utilized by the application server when the AI engine interacts with the human caller in parallel with the human agent; and

the second machine learning model is trained using machine interpretations generated by the AI engine and accepted, edited or rejected by the human agent.

2. The IVR system of claim 1 further comprising at least one database:

storing real-time transaction information associated with the caller; and

running an application program interface that provides the AI engine access to the real transaction information;

wherein the machine interpretations of the first and second sets of voice inputs are based on the real-time transaction information.

3. The IVR system of claim 1 :

wherein the AI engine detects rejection or acceptance by the human agent of the machine interpretation of the second set of voice inputs; and

AI engine is recursively trained using the rejection or acceptance and thereby reduces the number of times the AI engine links the terminal to the application server relative to the number of times the AI engine links the terminal to the application server without the recursive training.

4. The IVR system of claim 3 wherein the recursive training reduces the number of times the AI engine classifies the first set of voice inputs as requiring intervention by the human agent relative to the number of times the AI engine classifies the first set of voice inputs as requiring intervention by the human agent without the recursive training.

5. The IVR system of claim 3 , wherein the AI engine detects whether the human agent agrees with the classification of the first set of voice inputs as requiring intervention by the human agent;

wherein the first machine learning model is trained based on whether the human agent agrees with the classification.

6. The IVR system of claim 1 , wherein the AI engine intercepts the responses to the second set of voice inputs provided by the human agent by providing to the terminal predictive text, generated by the AI engine, that completes keyboard inputs entered by the human agent in response to the second set of voice inputs.

7. The IVR system of claim 1 wherein the AI engine generates the machine interpretation of the second set of voice inputs based on the machine interpretation of the first set of voice inputs.

8. The IVR system of claim 1 wherein the AI engine generates the machine interpretation of the second set of voice inputs based on the responses to the second set of voice inputs provided by the human agent via the terminal and the machine interpretation of the first set of voice inputs.

9. An interactive voice response (“IVR”) system comprising:

a telephony server that, in operation:

receives voice inputs submitted by a caller using a communication network;

provides an interface for the caller to interact with an artificial intelligence (“AI”) engine over the communication network and interact with the IVR system; and

transmits to the caller, over the communication network, responses to the voice inputs generated by the AI engine; and

an application server hosting the AI engine that, in operation:

receives, from the telephony server, a first set of voice inputs generated by the caller;

generates a machine interpretation of the first set of voice inputs;

based on the machine interpretation, classifies the first set of voice inputs as requiring intervention by a human agent;

based on the classification, links a terminal accessible by the human agent to the application server; and

after linking the terminal to the application server:

receives, from the telephony server, a second set of voice inputs generated by the caller;

generates a first set of machine responses to the second set of voice inputs using a first machine learning algorithm that is trained for interaction with the caller and the application server;

generates a second set of machine responses to the second set of voice inputs using a second machine learning algorithm that is trained for interaction with the human caller and the human agent; and

in parallel, provides the first set of machine responses to the caller and provides the second set of machine responses to the human agent;

wherein, providing the first and second sets of machine responses in parallel reduces a duration of the interaction between the caller and the IVR system relative an interaction duration without providing the responses in parallel.

10. The IVR system of claim 9 , wherein the AI engine applies a third machine learning model when the AI engine provides the set of machine responses for two or more callers linked to the human agent.

11. A method for leveraging artificial intelligence to integrate human and machine responses within an interactive voice response (“IVR”) system, the method comprising:

initiating an interaction with a human caller and an artificial intelligence (“AI”) engine;

providing voice inputs from the human caller to the AI engine;

receiving, from the AI engine, a predictive recommendation to transfer the human caller to a human agent;

initiating a hand-off procedure and transferring the human caller to the human agent; and

after transferring the human caller to the human agent:

continuing to provide the caller's voice inputs to the AI engine and providing responses to the caller's voice inputs generated by a first machine learning model that is trained for direct interaction between the AI engine and the caller;

providing, to the human agent, predictive responses to the caller's voice inputs generated by the AI engine using a second machine learning model that is trained for interaction with the human caller, the AI engine and the human agent; and

recursively training the AI engine by providing to the AI engine:

responses of the human agent to the caller's voice inputs; and

responses of the caller to the predictive responses generated by the AI engine and provided to the human agent;

wherein:

the recursive training reduces a frequency of when the AI engine generates the predictive recommendation to transfer the human caller to the human agent; and

providing the machine generated responses to the caller and human agent in parallel:

shortens a duration of the interaction by responding to the caller's inputs in parallel with the human agent; and

increases a number of concurrent interactions managed by the human agent by responding to the caller's inputs providing responses to the caller's voice inputs generated by a first machine learning model after the interaction is transferred to the human agent.

12. The method of claim 11 further comprising, after the human caller has been transferred to the human agent, the AI engine providing, in parallel, a first number of predictive responses directly to the human agent and a second number of predictive responses directly to the human caller;

wherein, the first number is greater than the second number.

13. The method of claim 11 further comprising recursively training the AI engine by the AI engine ingesting:

predictive responses generated by the AI engine and accepted by the human agent; and

predictive responses generated by the AI engine and rejected by the human agent.

14. The method of claim 11 further comprising:

providing, to the AI engine, keyboard inputs generated by the human agent; and

providing to the human agent predictive text, generated by the AI engine, that completes a message of the human agent embodied in the keyboard inputs.

15. The method of claim 11 , the transferring of the human caller to the human agent comprising:

providing the human agent with at least two or more historical interactions conducted by the AI engine with the human caller;

a predicative indicator of at least one target interaction most relevant to the interaction; and

based on the predictive indicator, loading into a computer system used by the human agent, a transaction history associated with the target interaction.

16. The method of claim 11 , further comprising training the AI engine by providing:

to the second machine learning model, the predictive responses generated by the AI engine and accepted or rejected by the human caller when the AI engine interacts exclusively with the human caller; and

to the first machine learning model, the predictive responses generated by the AI engine and accepted or rejected by the human agent when the AI engine interacts with the human caller and the human agent in parallel.