Patent Application
20080086690
In the present invention, a call is handled using a combination of an automated system and a human agent resulting in effective utilization of the human agent's time.
In an embodiment, a call is answered by an IVR system and the “health of the call”, as explained subsequently in this document, is maintained by the automated system. The “health of the call” may involve, for example, threshold scores associated with the speech recognition system, action classification scores and such parameters associated with other such sub-systems. Once the health of the call is detected to be bad, which may be determined by low threshold scores of for example, the sub-systems, unusual routes traversed by the call flow between the caller and the automated system and other such unusual conditions, the automated system seeks the help of a human agent. The history and the context of the call are presented to the human agent and the human agent provides the desired inputs to the automated system to enable the automated system to recover the call from the “bad call health” status. The automated system may then continue to handle the call without the help of the human agent.
In the proposed invention, a human agent may never interact with a caller directly, therefore the caller's interface remains unchanged for the duration of the call. Since the human agent is involved only for the duration required to rectify the bad call health by employing the corrective actions, the human agent is capable of handling multiple incoming calls simultaneously.
In the foregoing description, specific embodiments of the present invention have been described by way of examples with reference to the accompanying figures and drawings. One of ordinary skill in the art will appreciate that various modifications and changes may be made to the embodiments without departing from the scope of the present invention as set forth in the claims below. Accordingly, the specification and figures are to be regarded in an illustrative rather than a restrictive sense, and all such modifications are intended to be included within the scope of the present invention.
In one embodiment, the present invention provides a hybrid call handling method of interaction with a plurality of callers. In another embodiment, the present invention provides a hybrid call handling system. In yet another embodiment, the present invention provides a hybrid call handling system for detection and rectification of a bad call health status.
Referring to the drawings in general and to
The call history comprises a call log, an ASR score for each prompt, an Action Classification Score (ACS) for each prompt, and number of events for each prompt. The number of events includes help and no-match events. In the context of the present invention, the prompt is generally referred to, without any limitation, as a voice response provided by the IVR. The ASR score, a quantitative score, is computed based upon the number of words correctly recognized in the callers' speech. The action classification score, a quantitative score, is computed based on the ability of the automated system to understand the callers' query. A high ACS means that the automated system may handle the callers query while if the ACS is low, the call control is passed to the human agent. The no-match event is used by the automated system to comprehend a caller's response. In an embodiment, a no-match event may be maintained using a counter and the counter may be configured to increment by one each time the automated system is unable to understand a caller's response. However, it will be apparent to a person skilled in the art that other logics used to maintain the no-match event fall within the spirit and scope of the present invention, for example voiceXML may be used to for creating voice user interfaces that use automatic speech recognition (ASR) and text-to-speech synthesis (TTS). Similarly, a help event may be determined using a counter. The counter maintains a track of the number of times a caller asks for the automated system's help. For example, each time a caller requests for help, the counter is incremented by one. Thus, the value of the counter may help in determining the caller's understanding and ease with the automated system.
Each of the call parameters is assigned a threshold value prior to handling of the calls. The threshold value may either be a number or a context-dependent attribute. The call health status is determined based on whether the call parameters reside in the pre-assigned threshold values. If at least one of the call parameter does not fall within the pre-assigned threshold value, the call health status is referred to as bad call health. The bad call health status is detected at step 220. An unacceptable, or bad call health status, as it is generally referred to in the art, is typically quantified, without limitation, from a series of events, attributes, or behavioral data that is generally available to a person skilled in the art. These include, without limitation, the repetition of prompts by the system, a caller with a non-native accent, a change in the emotional state of the caller, wherein the caller may indulge in shouting or demonstration of anger and the like, a lower confidence level in the understanding of the caller query, and strange or unusual call routes in the call flow.
Next, the human agent is notified the bad call health status of the monitored call at step 225. In one embodiment of the current invention, the human agent intervenes using the machine interface to handle the monitored call as opposed to interacting with the caller directly, thereby providing a single automated interface to the caller. In another embodiment, notifying comprises transferring the call control of the monitored call to the human agent. In step 230, the human agent rectifies the bad call health by interacting with the caller using the machine interface. In one embodiment, the method of the present invention optionally provides for transferring the call control back to the automated system upon rectification of the bad call health status.
At step 330, the call control is transferred to the human agent. At step 335, at least one call parameter is provided to the human agent to enable the human agent to rectify the bad call health status. The call parameters comprise at least one of the call flow, duration of the call, the caller's inputs, prompts played back to the caller, a call priority status, probability of a non-native caller accent, and the stability of a caller's emotional state. The human agent rectifies the bad call health status at step 340. The human agent to whom the call is transferred for rectification may be the same human agent who received the call initially. This human agent controls the flow of the call and guides the caller more actively based on the analysis of the call parameters. Alternatively, the call may be escalated for rectification to a different human agent. The different human agent may have skill sets or training better suited to deal with calls with bad call health status. The two human agents may be co-located or may interact with each other through a network. At step 345, a check is made to determine whether the bad call health is rectified. The step 340 is repeated iteratively until the bad call health status is rectified. However, if the bad call health is rectified, the call control is transferred back to the step 310.
Depending upon the applications and requirements, several embodiments of the present invention may be implemented.
In one of the embodiments of the present invention, whenever the bad call health is detected by the automated system, the automated system may raise an alert to get assistance from the human agent. The human agent may enter the required inputs, for example, misrecognized text or a different route for the call flow to follow. The automated system subsequently proceeds to the next process step with the caller.
In another embodiment, whenever the bad call health is detected by the automated system, the human agent takes over but the caller still interacts with the same interface. The human agent may enter the text that he might want to speak into the text-to-speech system which may get synthesized to the caller. In many situations, the human agent may speak predetermined phrases; therefore shortcuts may be used instead of typing the full text into the automated system. In other scenarios, the human agent may speak into the automated system and his speech may get recognized and further synthesized using the Text-to-Speech engine.
In yet another embodiment of the present invention, the call is always attended by the human agent although the human agent does not speak directly to the caller. The human agent always uses a text-to-speech system in the caller's accent to speak to the caller.
When CTI server 120 receives incoming call 105 routed via network 110 and switch 115, it directs incoming call 105 to ACD 125. ACD 125 routes incoming call 105 to one of automated systems 510 to 530 depending upon the availability of the automated system. The automated system receives incoming call 105. In accordance with an embodiment of the present invention, the automated system includes an IVR that interacts with the caller. Each received call has an associated call health status. The automated system monitors the call health status and if the call health status is bad, the automated system transfers the call control to human agent 135.
In accordance with another embodiment of the present invention, the automated system comprises a machine interface via which human agent 135 interacts with the caller. Human agent 135 monitors the machine interface and identifies bad call health status of the monitored call. Human agent 135 intervenes to provide human support using the call history. In either embodiment, once human agent 135 rectifies the bad call health status by taking the required corrective actions, human agent 135 transfers the call control to the automated system. Further, the human agent converts the call heath status as healthy.
Call navigating module 610 comprises a call context gatherer unit 625. The navigation of calls includes navigating the interaction between the IVR and the caller and gathering the context of the call. Call context gatherer unit 625 captures all prompts (in text) and the corresponding Automatic Speech Recognition (ASR) transcribed text.
Call monitoring module 615 monitors the call health status of each call being navigated for the entire call duration by monitoring at least one call parameters. The call parameters comprise at least one of the call flow pattern, the events generated by the call-flow (for example, no-match event, help event, no-input event), the duration of the call, the caller's inputs, speech recognition scores on the caller's inputs, the prompts played back to the caller, a call priority status, the probability of a non-native caller accent, and a change in the caller's emotional state as perceived by the IVR. Each of the call parameters is assigned a threshold value prior to handling of the calls. The threshold value may either be a number or a context-dependent attribute. The call health status is determined based on whether the call parameters reside within the pre-assigned threshold values. If at least one of the call parameter crosses the pre-assigned threshold value, the call health status is referred to as one of bad call health. Call monitoring module 615 comprises a call log analyzer unit 630.
Call log analyzer unit 630 analyses call logs, ASR score, Action Classification Score (ACS), and number of events. The number of events includes help and no-match events. The speech recognition score, a quantitative score, is computed based upon the number of words correctly recognized in the callers' speech. The action classification score, a quantitative score, is computed based on the ability of the automated system to understand to the callers' query. A high ACS means that the automated system may handle the callers query while if the ACS is low, the call control is passed to the human agent. The no-match event is used by call log analyzer unit 630 to comprehend a caller's response. In an embodiment, a no-match event may be maintained using a counter and the counter may be configured to increment by one each time the automated system is unable to understand a caller's response. However, it will be apparent to a person skilled in the art that other logics used to maintain the no-match event fall within the spirit and scope of the present invention, for example voiceXML may be used to for creating voice user interfaces that use automatic speech recognition (ASR) and text-to-speech synthesis (TTS). Similarly, the help event may be determined using a counter. The counter maintains a track of the number of times a caller asks for the automated system's help. For example, each time a caller requests for help, the counter is incremented by one. Thus, the value of the counter may help in determining the caller's understanding and ease with automated system 510.
Call monitoring module 615 further detects a bad call health status of the call on the basis of one or more pre-assigned threshold values of one or more call parameters and passes the call control to a bad call health alert unit 635 of call control module 620. For example, a bad call health is detected if the ASR score is below 45, action classification score is below 50 and number of no-match event is greater than 2.
Call control module 620 further comprises a call control transfer unit 640 and a call context presentation unit 645. Bad call health alert unit 635 alerts the human agent upon detection of the bad call health status. In one embodiment of the present invention, bad call health alert unit 635 comprises a visual alert unit to provide visual cues to the human agent about the bad call health status. In another embodiment of the present invention, bad call health alert unit 635 comprises an audible alert unit to provide audio cues to the agent about the bad call health status. Call control transfer unit 640 transfers the call control to the human agent after alerting the human agent upon detection of the bad call health status. Call context presentation unit 645 displays the call context, gathered in call context gatherer unit 625, of the call with the bad call health status. The call context is displayed in a form comprehended by a human agent. The human agent rectifies the bad call health status by employing corrective actions.
It will be apparent to a person skilled in the art that various modules disclosed in conjunction with the disclosed embodiments of the present invention are logically unique entities. The functions performed by one or more of these modules may be merged in to a single program code for implementation on a network node. Further, one or more of these modules can be clubbed together to perform the desired functionality, for example, the call navigating module and the call monitoring module may be clubbed together. Various implementations imbibing the teachings of the present invention will be apparent to one skilled in the art. All these implementation are deemed to lie within the spirit and scope of the present invention.
The human agent to whom the call is transferred for rectification may be the same human agent who received the call initially. This human agent controls the flow of the call and guides the caller more actively based on the analysis of the call parameters. Alternatively, the call may be escalated for rectification to a different human agent. This human agent may have skill sets/training better suited to deal with calls with bad call health status. The two human agents may be co-located or may interact with each other through a network.
In another embodiment, the present invention provides a hybrid call handling system comprising a call receiving module, a call monitoring module coupled to the call receiving module to monitor a call health status, and a call control module coupled to the call monitoring module.
In yet another embodiment of the present invention, the present invention provides a hybrid call handling system comprising a plurality of calls originating from a plurality of callers, the plurality of callers coupled over a first network; one or more automated systems comprising one or more modules for at least one of (1) navigating a plurality of calls received from the plurality of callers (2) monitoring a call health status for each of the plurality of calls being navigated for an entire call duration (3) controlling the call, the one or more automated systems coupled over a second network to one or more human agent terminals comprising at least one of a call health alert unit and one or more human agents for rectification of bad call health.
An exemplary implementation of the present invention is described using
While typical embodiments have been set forth for the purpose of illustration, the foregoing description should not be deemed to be a limitation on the scope of the present invention. Accordingly, various modifications, adaptations and alternatives may occur to one skilled in the art without departing from the spirit and scope of the present invention.
