CN117676018A - Predictive outbound method and system for prepositive intent filtering AI dialogue robot - Google Patents

Abstract

The invention discloses a predictive outbound method and a predictive outbound system for a prepositive intent filtering AI dialogue robot, wherein the method initiates an outbound by starting an outbound application server; the client is connected, enters an AI dialogue robot interaction process, and filters out the intention client; the intention clients are transferred to the seats; the intention customer interacts with the seat to complete the outbound call. The number of the outbound calls initiated by the outbound application server uses the predictive outbound call calculation call completing rate, transfer rate and other parameters; calculating a predicted idle seat according to the seat state and the interaction time length; calculating a reserved seat according to the previously initiated outbound volume; calculating a predicted outbound volume; and (5) counting the idle condition and call loss rate of the seat, and calculating an up-regulation or down-regulation superposition coefficient. By calculating the filtering and predictive outbound method of the front AI dialogue robot, the access rate of the client is improved, the service scale is better enlarged, the seat utilization rate is effectively improved, and the call loss rate of the client is reduced.

Description

Predictive outbound method and system for prepositive intent filtering AI dialogue robot

Technical Field

The invention belongs to the technical field of artificial intelligent voice conversation robots and calling systems, and particularly relates to a predictive outbound method and a predictive outbound system of a prepositive intent filtering AI conversation robot.

Background

Telephone call centers can be classified into an incoming type and an outgoing type. The incoming call center provides passive services, while the outgoing call center provides a way to actively serve clients, and provides an efficient way for enterprises to develop services such as client services, marketing, questionnaires, and the like. With the development of computer and communication technologies, call center systems based on conventional technologies have been difficult to meet market changing demands, and automatic outbound occurs. The automatic outbound system actively initiates a call, filters clients which cannot be connected, and distributes the successfully connected clients to idle customer service, so that the seat utilization rate of customer service personnel is improved.

The number of calls made by the predictive outbound algorithm is typically greater than the number of idle agents at the time, because the call completion rate is typically much less than 1, thus greatly improving agent utilization over manual dialing and dialing without predictive functionality. However, if the number of calls is excessive, the call is made but no agent is forced to give up the service, which is called as call loss, and the proportion of call loss, namely the call loss rate, must be suppressed in practice, otherwise, great harassment is caused to the user, so that the algorithm needs to balance the agent utilization rate and the call loss rate.

With the rapid development of artificial intelligence and man-machine interaction technology, a voice conversation robot has become a main implementation form of the current intelligent chat system, and has shown high application value in various fields of national defense, military, business, education and the like. The voice conversation robot is applied to an automatic outbound system, filters unintentional clients, and can remarkably improve the seat utilization rate of customer service personnel.

Disclosure of Invention

Aiming at the defects of the prior art, the technical problem to be solved by the invention is to provide a predictive outbound method and a predictive outbound system for a pre-intent filtering AI dialogue robot, which can improve the calling efficiency, reduce the seat waiting time, fully predict various states of calling, call a calling system according to predicted data, fully distinguish the states of various calling systems under the prediction condition, fully match the calling resources, furthest reduce unreasonable personnel matching, save the resources, save the call cost and improve the user experience.

In order to solve the technical problems, the invention provides a predictive outbound method of a prepositive intent filtering AI dialogue robot, which comprises the following steps:

(1) Transferring the call line to an intention filtering AI conversation robot or seat by using the call line transferring function of the transferring and predicting external call application server;

(2) Carrying out timing calculation on the interval time parameters of the switching and predictive outbound application server through an outbound prediction algorithm to obtain the final predicted outbound volume;

(3) The final predicted outbound volume is proportionally distributed to a plurality of line dialing servers; the line dialing server can work in parallel and initiate a call on a line;

(4) If the call is directly transferred to the intention filtering AI conversation robot, the intention filtering AI conversation robot carries out conversation operation on the conversation and completes the recognition of the intention of the client;

(5) When the intention filtering AI conversation robot identifies the intention conversation, the conversation line is connected to the agent, if the agent is busy and the waiting time exceeds a threshold t, the switchover prediction external calling application server continues to switchover the AI conversation robot to take over, continues to communicate, waits for the agent to be idle, continues to switchover the agent, and the switchover can be circulated all the time.

Specifically, the timing calculation includes calculating a plurality of parameters, idle agents, reserved agents, predicted outbound volume and superposition coefficients in three working modes of the call.

Specifically, the parameter calculation of the connection rate and the transfer rate specifically includes:

firstly, dividing a calculation working mode into three working modes of real-time calculation, interval calculation working modes and fixed values;

the real-time calculation working mode is used for counting historical outbound records, and the parameters of the historical outbound records are calculated according to invalid records filtered by setting a minimum threshold value min_effect_duration and a maximum threshold value max_effect_duration, wherein the parameters are the call completing rate, the transfer rate, the average calling duration, the average AI interaction duration, the average waiting duration, the average call duration and the average post-call processing duration;

if the obtained history outbound record number is smaller than a preset threshold parameter n, the working mode of the last effective calculation is adopted or the working mode of the last effective calculation is changed into a fixed value;

the fixed value working mode adopts preset default parameters;

the whole time interval of the interval calculation working mode adopts a real-time working mode and a fixed value working mode.

Further, the calculating the idle seat is to check the seat states one by one in the seat queue to which the current seat belongs, specifically includes:

(4.1) adding 1 to the idle seat if the seat is in the idle state;

(4.2) if the seat is in a busy state and the busy time is longer than m times of the average post-processing time, the seat is not considered to belong to the range of the to-be-effective seat, and the seat is skipped;

(4.3) when the seat is in a post-call processing state or a small rest state, adding the value of the average waiting time length to the average post-call processing time length, and subtracting the value of the current state time length to be P; when P is smaller than the value Q added by the average AI interaction time length added by the average call time length, adding 1 to the idle seat;

(4.4) when the seat is in a ringing state or a call state, adding the average waiting time length to the average post-call processing time length, and adding the average call time length to the average waiting time length, wherein the value obtained by subtracting the current state time length is M; and when M is smaller than the average calling duration plus the average AI interaction duration N, adding 1 to the idle agent.

Further, the calculating the reserved seats, that is, calculating the number of seats to be reserved from the current call list, wherein the calls can be classified into three types:

(5.1) after receiving the transfer instruction, namely completing the AI interactive call, reserving the number of seats according to the ratio of 1:1;

(5.2) the number of the reserved seats is carried out according to the transfer rate of the call which is connected;

and (5.3) carrying out the reserved seat number according to the call completing rate and the transferring rate on the unanswered calls.

Specifically, the calculating and predicting the outbound volume is as follows: predictive outbound volume= (idle agent-reserved agent)/call completing rate/transfer rate; i.e. the idle agent minus the reserved agent value divided by the call completing rate and then divided by the transfer rate.

Further, the calculating the superposition coefficient specifically includes:

(7.1) calculating the number of seats of which the idle time length of the seats exceeds a parameter max_idle_seconds, and when the number reaches an up-regulation reference proportion raise_date_scale of the total number of the idle seats, up-regulating the superposition coefficient proportionally;

(7.2) acquiring the number trans_failed_count of the current transfer seat, and proportionally reducing the superposition coefficient according to the reduction reference proportion reduce_date_scale of the number of the failed and the number of the current idle seats;

(7.3) the up and down ranges can specify the adjusted interval range;

and (7.4) after calculating the superposition coefficient, multiplying the current predicted outbound volume by the superposition coefficient to obtain the final predicted outbound volume.

Specifically, the dispatch of the outbound call is to dispatch the final predicted outbound call quantity to a plurality of line dialing servers in proportion;

(8.1) proportional calculating the number of lines of the line dialing servers divided by the number of bus lines of the line dialing servers, and the minimum number is 1;

(8.2) completing the allocation when the number of allocations is equal to the predicted outbound volume.

The invention also provides a predictive outbound system of the prepositive intent filtering AI dialogue robot, which comprises the following modules:

and (3) a switching module: by using the call line transfer function of the transfer and prediction outbound application server, transfer the call line to the intent filtering AI conversation robot or agent,

the calculation module: carrying out timing calculation on the interval time parameters of the switching and predictive outbound application server through an outbound prediction algorithm to obtain the final predicted outbound volume;

and a calling module: the final predicted outbound volume is proportionally distributed to a plurality of line dialing servers; the line dialing server can work in parallel and initiate a call on a line;

and an identification module: if the call is directly transferred to the intention filtering AI conversation robot, the intention filtering AI conversation robot carries out conversation operation on the conversation and completes the recognition of the intention of the client;

and the transfer seat module is used for: when the intention filtering AI conversation robot identifies the intention conversation, the conversation line is connected to the agent, if the agent is busy and the waiting time exceeds a threshold t, the switchover prediction external calling application server continues to switchover the AI conversation robot to take over, continues to communicate, waits for the agent to be idle, continues to switchover the agent, and the switchover can be circulated all the time.

The beneficial effects of the invention are as follows:

after the system and the algorithm are adopted, the transfer and forecast external call application server starts to initiate external call, and parameters such as the call completing rate, transfer rate and the like are calculated; judging how many idle agents participate in the service in the round; calculating the number of agents to be reserved according to the number of outbound calls initiated in a plurality of previous rounds; calculating a predicted outbound volume therefrom; counting the idle condition and call loss rate of the seat, and calculating an up-regulation or down-regulation superposition coefficient; and the outbound application server distributes the final predicted outbound quantity adjusted by the superposition coefficient to the line dialing server for outbound. The system and the algorithm can improve the calling efficiency, reduce the seat waiting time, fully predict various calling states under the condition of the same customer service, call the calling system according to the predicted data, fully distinguish the states of various calling systems under the predicted condition, fully match the calling resources, furthest reduce unreasonable personnel matching, save the resources, save the call cost and improve the user experience.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description, serve to explain the principles of the application.

Fig. 1 is a schematic diagram of the overall structure of a call system predictive outbound system according to the present invention;

FIG. 2 is a schematic diagram of an implementation flow of a call system predictive outbound system according to the present invention;

FIG. 3 is a timing diagram illustrating an implementation flow of a predictive outbound system for a calling system according to the present invention;

fig. 4 is a schematic diagram of a process for calculating the predicted outgoing call volume of a calling system according to the present invention.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present application.

The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used in this application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items.

Example 1

Please refer to fig. 1-3

Fig. 1 is a schematic diagram of the overall structure of a call system predictive outbound system according to the present invention;

fig. 2 and 3 are a schematic diagram and a timing chart of an execution flow of a call system predictive outbound system according to the present invention;

the invention provides a predictive outbound system of a prepositive intent filtering AI conversation robot, which comprises a switching and predictive outbound application server, an intent filtering AI conversation robot, a line dialing server and an agent.

The intent filtering AI conversation robot applies conversation technology to communicate with clients to realize client intent recognition.

The switching and predicting external calling application server realizes a prediction algorithm, calculates at regular time, distributes the call to the line dialing servers, and a plurality of the line dialing servers work in parallel to initiate the call on the line.

The transfer and prediction external call application server has a call line transfer function and transfers a call line to the intention filtering AI conversation robot or the seat. And the line dialing server connects the call line to the intention filtering AI conversation robot. And if the seat is busy, the waiting time exceeds a threshold t, and the switching and predicting external call application server continues to switch the AI conversation robot to take over, continues to communicate, waits for the seat to be idle, and continues to switch the seat. So that multiple rounds can be tried.

Example 2

Please refer to fig. 4

Fig. 4 is a schematic diagram of a process for calculating the predicted outgoing call volume of a calling system according to the present invention.

The invention also discloses a predictive outbound algorithm which calculates at regular time when the interval parameter interval of the transit and predictive outbound application server is 10 seconds. The call predictive outbound algorithm comprises the following steps:

s1: and calculating parameters such as the call completing rate, the transfer rate and the like. In this embodiment, the parameter calculation modes include the following:

the calculation working mode supports three working modes of real-time calculation, interval calculation and fixed value. And calculating historical outbound records of the working mode statistics in real time, filtering invalid records according to a set minimum threshold value of min_effect_duration=40 seconds and a set maximum threshold value of max_effect_duration=240 seconds, and calculating the connection rate, the transfer rate, the average call duration, the average AI interaction duration, the average waiting duration, the average call duration and the average post-call processing duration. If the obtained history outbound records are less than 500, the default parameters or the parameter values effectively calculated last time are specified. The fixed value working mode adopts preset parameters. The interval calculation working mode adopts a real-time working mode in part of time intervals, and adopts a fixed value working mode in part of time intervals;

s2: and calculating the idle seat. In this embodiment, the method for calculating the predicted idle seat is as follows:

checking the position states one by one in the current position queue:

(2.1) if the seat is in the idle state, adding 1 to the idle seat;

(2.2) if the seat is in the busy state and the busy time period is 2 times longer than the average post-processing time period, the seat is not considered to belong to the range of the to-be-effective seat, and the seat is skipped;

(2.3) if the seat is in the state of [ post-processing and small rest ], [ average post-processing duration ] + [ average waiting duration ] [ current state duration ] < [ average calling duration ] + [ average AI interaction duration ], adding 1 to the idle seat;

(2.4) if the seat is in the state of [ ringing and talking ], [ average post-processing duration ] + [ average waiting duration ] + [ average talking duration ] [ current state duration ] < [ average calling duration ] + [ average AI interaction duration ], [ idle seat is added with 1.

S3: and calculating a reserved seat. In this embodiment, the prediction reservation agent calculation method is as follows:

from the current call list, the number of seats to be reserved is calculated, and the calls can be classified into three types:

(3.1) after receiving the transfer instruction, namely completing the AI interactive call, reserving the number of seats according to the ratio of 1:1;

(3.2) the number of the reserved seats is carried out according to the transfer rate of the call which is connected;

and (3.3) carrying out the reserved seat number according to the call completing rate and the transferring rate on the unanswered calls.

S4: and calculating a predicted outbound volume. In this embodiment, the calculation formula of the predicted outbound volume is:

predicted outbound call= (idle agent-reserved agent)/call completing rate/transfer rate

S5: and calculating a superposition coefficient. In this embodiment, the superposition coefficient calculation method is as follows:

(5.1) calculating the number of seats with the idle time length of the seats exceeding a parameter max_idle_seconds=30 seconds, and when the number reaches an up-reference proportion, raise_datum_scale, of the total number of the idle seats, up-regulating the superposition coefficient proportionally;

(5.2) acquiring the number trans_failed_count of the current transfer seat, and proportionally reducing the superposition coefficient according to the reduction reference proportion reduce_date_scale of the number of the failed and the number of the current idle seats;

(5.3) the up and down can specify the adjusted interval range;

and (5.4) after calculating the superposition coefficient, multiplying the current predicted outbound volume by the superposition coefficient to obtain the final predicted outbound volume.

S6: outgoing calls are assigned. In this embodiment, the method for predicting the outbound volume is as follows:

the final predicted outbound volume is proportionally distributed to a plurality of line dialing servers.

(6.1) proportional calculating the number of lines of the line dialing servers divided by the number of bus lines of the line dialing servers, wherein the minimum number is 1;

(6.2) completing the allocation when the number of allocations is equal to the predicted outbound volume.

After the system and the algorithm are adopted, the transfer and forecast external call application server starts to initiate external call, and parameters such as the call completing rate, transfer rate and the like are calculated; judging how many idle agents participate in the service in the round; calculating the number of agents to be reserved according to the number of outbound calls initiated in a plurality of previous rounds; calculating a predicted outbound volume therefrom; counting the idle condition and call loss rate of the seat, and calculating an up-regulation or down-regulation superposition coefficient; and the outbound application server distributes the final predicted outbound quantity adjusted by the superposition coefficient to the line dialing server for outbound. The system and the algorithm can improve the calling efficiency, reduce the seat waiting time, fully predict various calling states under the condition of the same customer service, call the calling system according to the predicted data, fully distinguish the states of various calling systems under the predicted condition, fully match the calling resources, furthest reduce unreasonable personnel matching, save the resources, save the call cost and improve the user experience.

Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the application pertains.

It is to be understood that the present application is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof.

Claims (9)

1. A predictive outbound method for a pre-intent-to-filter AI-dialogue robot, the method comprising the steps of:

(1) Transferring the call line to an intention filtering AI conversation robot or seat by using the call line transferring function of the transferring and predicting external call application server;

(2) Carrying out timing calculation on the interval time parameters of the switching and predictive outbound application server through an outbound prediction algorithm to obtain the final predicted outbound volume;

(3) The final predicted outbound volume is proportionally distributed to a plurality of line dialing servers; the line dialing server can work in parallel and initiate a call on a line;

(4) If the call is directly transferred to the intention filtering AI conversation robot, the intention filtering AI conversation robot carries out conversation operation on the conversation and completes the recognition of the intention of the client;

(5) When the intention filtering AI conversation robot identifies the intention conversation, the conversation line is connected to the agent, if the agent is busy and the waiting time exceeds a threshold t, the switchover prediction external calling application server continues to switchover the AI conversation robot to take over, continues to communicate, waits for the agent to be idle, continues to switchover the agent, and the switchover can be circulated all the time.

2. The predictive outbound method for a pre-intent-to-filter AI-dialogue robot of claim 1 wherein the timing calculations include calculating parameters, idle agents, reserved agents, predicted outbound volume, and superposition coefficients for three modes of operation of a call.

3. The predictive outbound method of the pre-intent-filtering AI-dialogue robot according to claim 2, wherein the parameter calculation of the call completing rate and the transfer rate is specifically as follows:

firstly, dividing a calculation working mode into three working modes of real-time calculation, interval calculation working modes and fixed values;

the real-time calculation working mode is used for counting historical outbound records, and the parameters of the historical outbound records are calculated according to invalid records filtered by setting a minimum threshold value min_effect_duration and a maximum threshold value max_effect_duration, wherein the parameters are the call completing rate, the transfer rate, the average calling duration, the average AI interaction duration, the average waiting duration, the average call duration and the average post-call processing duration;

if the obtained history outbound record number is smaller than a preset threshold parameter n, the working mode of the last effective calculation is adopted or the working mode of the last effective calculation is changed into a fixed value;

the fixed value working mode adopts preset default parameters;

the whole time interval of the interval calculation working mode adopts a real-time working mode and a fixed value working mode.

4. The predictive outbound method for a pre-intent-filtering AI-dialogue robot according to claim 1, wherein the calculating the idle agents is to check the agent states one by one in the agent queue to which the current agent belongs, specifically:

(4.1) adding 1 to the idle seat if the seat is in the idle state;

(4.2) if the seat is in a busy state and the busy time is longer than m times of the average post-processing time, the seat is not considered to belong to the range of the to-be-effective seat, and the seat is skipped;

(4.3) when the seat is in a post-call processing state or a small rest state, adding the value of the average waiting time length to the average post-call processing time length, and subtracting the value of the current state time length to be P; when P is smaller than the value Q added by the average AI interaction time length added by the average call time length, adding 1 to the idle seat;

(4.4) when the seat is in a ringing state or a call state, adding the average waiting time length to the average post-call processing time length, and adding the average call time length to the average waiting time length, wherein the value obtained by subtracting the current state time length is M; and when M is smaller than the average calling duration plus the average AI interaction duration N, adding 1 to the idle agent.

5. The predictive outbound method for a pre-intent-to-filter AI-dialogue robot of claim 1 wherein the computing reserved agents, i.e., the number of agents to be reserved from a current list of calls, is classified into three categories:

(5.1) after receiving the transfer instruction, namely completing the AI interactive call, reserving the number of seats according to the ratio of 1:1;

(5.2) the number of the reserved seats is carried out according to the transfer rate of the call which is connected;

and (5.3) carrying out the reserved seat number according to the call completing rate and the transferring rate on the unanswered calls.

6. The predictive outbound method for a pre-intent-to-filter AI-dialogue robot of claim 1, wherein the calculating the predicted outbound volume is specifically as follows: predictive outbound volume= (idle agent-reserved agent)/call completing rate/transfer rate; i.e. the idle agent minus the reserved agent value divided by the call completing rate and then divided by the transfer rate.

7. The predictive outbound method for a pre-intent-to-filter AI-dialogue robot of claim 1, wherein the calculated superposition coefficients are specifically:

(7.1) calculating the number of seats of which the idle time length of the seats exceeds a parameter max_idle_seconds, and when the number reaches an up-regulation reference proportion raise_date_scale of the total number of the idle seats, up-regulating the superposition coefficient proportionally;

(7.2) acquiring the number trans_failed_count of the current transfer seat, and proportionally reducing the superposition coefficient according to the reduction reference proportion reduce_date_scale of the number of the failed and the number of the current idle seats;

(7.3) the up and down ranges can specify the adjusted interval range;

and (7.4) after calculating the superposition coefficient, multiplying the current predicted outbound volume by the superposition coefficient to obtain the final predicted outbound volume.

8. The predictive outward call method for a pre-intent-filtering AI conversation robot as claimed in claim 1 wherein the assigning outward call is by scaling the last predictive outward call volume to a plurality of the line dialing servers;

(8.1) proportional calculating the number of lines of the line dialing servers divided by the number of bus lines of the line dialing servers, and the minimum number is 1;

(8.2) completing the allocation when the number of allocations is equal to the predicted outbound volume.

9. A predictive outbound system for a pre-intent-to-filter AI-dialogue robot, the system comprising:

and (3) a switching module: by using the call line transfer function of the transfer and prediction outbound application server, transfer the call line to the intent filtering AI conversation robot or agent,

the calculation module: carrying out timing calculation on the interval time parameters of the switching and predictive outbound application server through an outbound prediction algorithm to obtain the final predicted outbound volume;

and a calling module: the final predicted outbound volume is proportionally distributed to a plurality of line dialing servers; the line dialing server can work in parallel and initiate a call on a line;

and an identification module: if the call is directly transferred to the intention filtering AI conversation robot, the intention filtering AI conversation robot carries out conversation operation on the conversation and completes the recognition of the intention of the client;

and the transfer seat module is used for: when the intention filtering AI conversation robot identifies the intention conversation, the conversation line is connected to the agent, if the agent is busy and the waiting time exceeds a threshold t, the switchover prediction external calling application server continues to switchover the AI conversation robot to take over, continues to communicate, waits for the agent to be idle, continues to switchover the agent, and the switchover can be circulated all the time.