CN105704335B - Predictive outbound algorithm, interchanger dialing method and device based on dynamic statistics process - Google Patents

Abstract

The invention relates to a predictive outbound call algorithm based on a dynamic statistical process, a switch dialing method and a device. The device of the invention includes a data input module, a predictive outbound call algorithm module, a switch, a dynamic statistical module, a data storage module and a data output The display module and the data input module input initialization parameter data to the pre-test outbound algorithm module, the pre-test outbound algorithm module is connected to the switch for information interaction, the dynamic statistics module is connected to the data storage module, receives data from the data storage module, and performs dynamic statistics The module counts statistical data of different frequencies and transmits them to the pre-test outbound algorithm module, and at the same time displays the data through the data output interface. The invention makes statistics and predictions based on the historical data of dialed calls, dynamically determines the number of calls to be dialed at the next moment, improves the utilization rate of seats while not exceeding the target call loss rate as much as possible, achieves an optimal balance and completely replaces manual dialing.

Description

Predictive outbound call algorithm based on dynamic statistical process, exchange dialing method and device

technical field

The invention relates to the technical field of telephone outgoing calls, in particular to a predictive outgoing call algorithm based on a dynamic statistical process, an exchange dialing method and a device. Statistics, data mining, queuing theory, Markov chain, birth and death theory, and algorithm design and analysis technologies are used, especially the technologies of real-time resource allocation scheduling and feedback control optimization.

Background technique

Call centers can be divided into two types: inbound and outbound. The inbound call center provides passive services, while the outbound call center provides a way to actively serve customers, and provides an efficient way for enterprises to carry out marketing, questionnaires and other businesses. With the development of computer and communication technology, the call center system based on traditional technology has been difficult to meet the changing needs of the market, so automatic outbound calls are born. The automatic outbound call system actively initiates calls, filters out unreachable customers and successfully of customers are assigned to idle customer service personnel, thereby improving the utilization rate of customer service personnel.

There are three main types of existing outbound call algorithms: empirical outbound call prediction algorithms, integral model algorithms and the application of queue theory in prediction algorithms.

The empirical outbound call prediction algorithm dynamically adjusts the feedback rate value based on the recent outbound call results, and adjusts outbound call adjustment parameters (>1 for overspeed outbound calls, <1 for decelerated outbound calls), so that the adaptability of the algorithm becomes stronger. The disadvantage is that the distribution of the average idle time is unstable, and the call loss cannot be well controlled.

The integral model algorithm first examines the distribution of the connection sequence of calls: the connection rate distribution V(t), the disconnection rate distribution F(t), then there is Then examine the timing sequence of the call duration: the call duration distribution h(t), satisfying Then the probability that the call duration is longer than time t, H(O)=1, H(+∞)=0. Assuming that the system calls out the sequence C(t), the number of incoming calls at time t and the number of current calls at time t represent the number of calls that are currently in conversation, including calls that have already talked to the operator and calls that have been picked up by the user. The machine is not assigned to a call that the operator is waiting on. If it is required that the instant connection rate of the outgoing user after off-hook is g, that is, the call rate of the free agent after the outgoing user is off-hook, then there is N(t)*g<A, and at the same time, in order to fully utilize the agent resources, there is N (t)≥A, where A is the number of currently registered seats. That is, it can be obtained from the above: A≤N(t)<A/g. From this constraint, the exhalation time series C(t) of the system can be calculated. According to this outgoing call prediction algorithm, it is possible to perform self-feedback learning and dynamically adjust subsequent outgoing calls according to the results of historical calls, that is, past call time series and current system resource conditions. Although the above algorithm is indeed feasible and has played a great role in practical applications, it is found through analysis that there are still areas worthy of improvement. For example, is it the best choice to divide it into n batches on average in the period T? The average processing time is L/2 Is it a real situation or an ideal situation, is it static or fluctuating, and how much is it fluctuating? Or will the abnormality of the current situation affect the prediction of the next moment, and how stable is it?

Application of queuing theory in predictive algorithms: queuing theory was first introduced by A.K. Erlang when he solved the problem of automatic telephone design. The prediction algorithm is summarized as solving two related problems: 1. In the case of maximizing the agent busy factor (agent utilization rate), the call loss rate is not allowed to exceed an acceptable maximum value ARmax. 2. In the case of reducing the call loss rate as much as possible, the agent busy factor should not be lower than an acceptable minimum level Umin. The algorithm is to meet the above two conditions to generate the current number N of the calls. The algorithm can collect or receive some parameters related to the system: the total number of agents m, the arrival rate of incoming calls p, the average service time T, and the hit rate ρ. The algorithm assumes that the outgoing flow obeys the Poisson distribution of parameter λ; the service time in this system obeys the general distribution of parameter μ. In this forecasting algorithm, the distribution of outgoing calls is defined as Poisson distribution, that is, the time between two adjacent dialed out calls obeys the exponential distribution. This is very different from the phenomenon in reality. In addition, the service time obeys the general distribution. It is found in the research that the service time will be concentrated in two stages, one is 3-10s, and the other is between 140-160s. It is not a standard general distribution, which will cause the model The calculation results have a large deviation. In addition, the algorithm does not take the current state of the agent into consideration, which will cause calls to be made according to the number of original data calculations when all the agents are suddenly busy, until the average time T is recalculated ( takes a long time), so there will be a long period of turmoil. In the case of a sudden decrease in service time for a period of time, it will also cause fluctuations in the sudden decrease in seat utilization.

It can be seen from the above analysis that most of the existing predictive outbound algorithms are coupled with the dynamic statistics part, or even have no dynamic statistics part. The most important determinant of the predictive outbound algorithm is the dynamic statistics part. Some of the existing outbound algorithms are calculated using theoretical formulas, requiring a mathematical model to be established in advance, and the subsequent parameter values are all calculated through theoretical formulas. ; Although this method does not use or seldom access the database, the establishment of the model is relatively fixed. For some businesses, this model no longer holds true, or the adaptability of this model is poor. Some of the existing outbound algorithms are also implemented through dynamic statistics, but this method requires a large amount of access to historical data for machine learning, training data or neural networks, etc. This method frequently accesses the database and requires a large number of Data, waste a lot of time, cause performance degradation, and it is difficult to make reasonable predictions without a large amount of data at the beginning.

Contents of the invention

In order to overcome the above-mentioned defects in the prior art, the purpose of the present invention is to provide a predictive outbound algorithm based on a dynamic statistical process, an exchange dialing method and a device.

In order to achieve the above object, according to a first aspect of the present invention, the present invention provides a predictive outbound algorithm based on a dynamic statistical process, which includes the following steps:

S1, start, set the target agent utilization rate AO, the maximum call loss rate AR _max , the number of exchanges T, the maximum waiting time TQ _max , the prediction interval △t, the maximum length of the waiting queue qL, the user's patience time 1/η, the user's unwillingness The probability p ₀ of joining the waiting queue, and the initial value of the number of outgoing calls λ, let the cumulative value of l=0.

S2. At time t, obtain the total number of current customer service N, the total number of idle customer service f, the number of calls j, the connection rate ρ, the average ringing time 1/β, the average call time 1/μ, and the waiting queue length q through the switch The value of , said t is a positive number;

S3, judge whether q<qL is satisfied, if so, go to step S4, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt;

S4, calculate the required minimum number of seats M, the calculation formula is:

S5, judging whether N≤M is satisfied, if so, go to step S17, otherwise go to step S6;

S6, the calculation system predicts the maximum outbound call value l, accumulatively calculates al=al+l, wherein,

S7, dynamically counting the call loss rate AR at the last moment and judging whether AR<AR _max is satisfied, if so, go to step S8, otherwise go to step S16;

S8, judging whether the current total number of idle customer service f>al is satisfied, if satisfied, then go to step S9, otherwise go to step S13;

S9, prediction result s=(f-j)/ρ-q+al, if al>1 then al is set to 0;

S10, calculate user waiting time Tq, Tq=q/(N*μ);

S11, judge whether Tq<TQ _max is satisfied, if it is satisfied, go to step S12, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt;

S12, update the prediction result s=s*ARrate, ARrate is the exhalation attenuation ratio of the system, ARrate=1-AR, AR is the call loss rate,

AR=p ₀ B(θ,N)+(1-p ₀ )λβρη, turn to step S18;

S13, the prediction result s=f-j-q+al, if al>1, set al to 0;

S14, calculate call loss rate AR, AR=p ₀ B(θ, N)+(1-p ₀ )λβρη

B(θ,0)=1,

S15, determine whether AR<AR _max is satisfied, if so, go to step S10, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt;

S16, prediction result s=f-j-q, go to step S18;

S17, prediction result s=f-j, turn to step S18;

S18, calculate the final number of dialed calls λ, λ=(int)min(max(s, 0), T);

S19, making λ calls outside, the algorithm sleeps until t+Δt time, go to step S2 for the next round of prediction, and update the current time to t=t+Δt; repeat the above steps until the application program terminates.

In order to achieve the above object, according to a second aspect of the present invention, the present invention provides a method for dialing an exchange based on a predictive outbound algorithm of a dynamic statistical process, which is characterized in that it comprises the following steps:

Step 1, predicting the number of calls λ to be dialed at the next moment according to the predictive outbound algorithm according to claim 1;

Step 2, the exchange dials λ calls according to the prediction result;

Step 3, after the average ringing time 1/β, the β is a positive number, there will be ρ proportion of calls connected, and the remaining 1-ρ proportion of calls that are not connected are filtered out, and the ρ is greater than 0 a positive number less than 1;

Step 4, the number of connected people enters the control system, and the control system will allocate according to whether there are currently free seats;

Step 5, if there is an idle seat, assign an agent to serve it, and the call is successful if the service is completed; if there is no idle seat, there will be customers with a probability of p0 who are willing to join the waiting queue for waiting, and the remaining customers with a probability of 1-p0 will not Willing to join the waiting queue and give up to cause call loss;

Step 6, some of the customers in queue also give up and form a call loss due to waiting beyond their patient time 1/η, and only the rest will pass through the control system according to the idle service that has been served at present without exceeding their patient time 1/η. The agent assigns the agent service to it, and finally the service is completed.

In order to achieve the above object, according to a third aspect of the present invention, the present invention provides a device for running the predictive outbound algorithm based on the dynamic statistical process of the present invention, which includes a data input module, a predictive outbound algorithm module, a switch , a dynamic statistical module, a data storage module and a data output display module, the data input module inputs initialization parameter data to the predictive outbound algorithm module, and the predictive outbound algorithm module is connected to a switch for information interaction, and the dynamic The statistics module is connected with the data storage module, receives data from the data storage module, and the dynamic statistics module counts statistical data of different frequencies and transmits them to the predictive outbound algorithm module, and simultaneously displays the data through the data output interface;

The predictive outbound algorithm module includes a configuration parameter unit, an initialization parameter unit, a theoretical calculation data unit, a cycle statistics parameter unit, a predictive outbound algorithm unit, and a predicted result unit; the configuration parameter unit is used to receive switch configuration parameter information , the initialization parameter unit is used to receive various target parameter data input by the data input module, the theoretical calculation data unit is used to receive theoretical calculation but periodically changed data, and the cycle statistics parameter unit is used to receive the dynamic statistics module Various data collected with different statistical frequencies, the predictive outbound algorithm unit is used to implement the predictive outbound algorithm according to claim 1, and the predicted result unit is used to deliver the predicted result to the switch;

The dynamic statistics module includes an acquisition historical data unit, a statistical strategy unit, a statistical strategy unit, a unit for transferring statistical data to the predictive outbound algorithm module and a unit for transferring statistical data to the output interface, and the historical data acquisition unit stores It includes obtaining the data that has been written into the database and the data that is being served by the switch but has not yet been written into the database. The selection statistical strategy unit is used to select which statistical strategy to use. The statistical strategy unit includes a variety of specific statistical strategy.

The present invention has the following beneficial technical effects:

1. The present invention makes statistics and predictions based on the historical data of dialed calls, dynamically determines the number of calls that should be dialed at the next moment, improves the utilization rate of seats while not exceeding the target call loss rate as much as possible, achieves an optimal balance and completely replaces manual dial.

2. The outbound call algorithm of the present invention establishes a queuing theoretical model when the seats are not enough, while the ringing time, service time and statistical strategy use the statistical model of manual dialing data, and the model established is closer to the actual situation. In the same call loss The seat utilization rate will be greatly improved under the high rate. The outbound call algorithm of the present invention is based on queuing theory, but the initial parameters (such as service time, connection rate, etc.) are very different for different businesses, so these parameters can be well adapted to various businesses through dynamic statistics. When the number of seats is 10, the agent utilization rate is 65-70%, and the call loss rate is 3-4%; when the number of seats is 20, the seat utilization rate is 70-73%, and the call loss rate is about 2%. When the number of seats is 30, the agent utilization rate is 72-75%, and the call loss rate is 1-2%; when the number of seats is 40, the agent utilization rate is 80-85%, and the call loss rate is about 1%. When the number is 50, the agent utilization rate is 82-85%, and the call loss rate is less than 1%. With the increase of the number of seats, it becomes more and more stable, and the call loss rate is very small or even non-existent at the beginning, and the fluctuation is also very large. Small; when the number of seats is 60 or more, there is almost no call loss and the seat utilization rate is maintained at 80%. And it has a good effect on different business data.

3. The present invention develops a method that divides the predictive outbound algorithm into two important modules: the predictive outbound algorithm module and the dynamic statistics module; thus it can be optimized from two directions, thereby avoiding the problem that only the optimization algorithm will appear. Bottleneck problem. And when optimizing the algorithm, you don't need to care about the specific statistics of the parameters, and you don't need to consider the implementation of the algorithm and whether it is useful to the algorithm when you calculate the parameters.

4. The dynamic statistical module is added to improve the adaptability of the algorithm, so that it can adapt to a variety of different businesses and the situation where the data fluctuates too much; the idea of combining the statistical model with the theoretical model is adopted to avoid the extreme situation of the theoretical model The defect that deviates from the lower limit makes the theoretical model closer to the statistical model.

5. Adopt the idea of combining feedback and forecasting, so that those who can predict accurately will become more accurate, and those who can predict less accurately can become more accurate through feedback, thus combining the advantages of forecasting and feedback.

Additional aspects and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and comprehensible from the description of the embodiments in conjunction with the following drawings, wherein:

Fig. 1 is a flow chart of predictive outbound algorithm in a preferred embodiment of the present invention;

Fig. 2 is a schematic flow diagram of a switch dialing method in a preferred embodiment of the present invention;

Fig. 3 is a schematic diagram of the composition and structure of a device for inventing a predictive outbound call algorithm based on dynamic statistics in a preferred embodiment.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar reference numerals designate the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be mechanical connection or electrical connection, or two The internal communication of each element may be directly connected or indirectly connected through an intermediary. Those skilled in the art can understand the specific meanings of the above terms according to specific situations.

The parameter meaning among the present invention is explained as follows:

Ringing duration: the time required to make a call, from the start of the call to the successful connection of a call or the identification that it cannot be connected.

Service time: The service time of the call after the call is successfully connected.

Finishing time: the time for making business records for customers (it can also be regarded as a part of service time).

Waiting queue length: If you are willing to wait online when you are connected to the phone and cannot get service, this queue length is the waiting queue length.

User patience time: the time from when the user joins the waiting queue to when he is unwilling to wait and actively gives up and hangs up the phone.

User waiting time: the time the user actually waits in the queuing process.

Maximum waiting time: the upper limit of the maximum time that all online customers are willing to wait.

Customer service utilization rate: the ratio of the sum of service hours of all customer service to the total online time of all customer service.

Call loss rate: The ratio of the number of calls that are connected but not served to the total number of connected calls.

According to the statistical analysis of the actual manual outbound dialing data, the present invention shows that the service time approximately obeys the negative exponential distribution, and the ringing time approximately obeys the generalized Poisson distribution.

The present invention provides a kind of predictive outbound algorithm based on dynamic statistical process, as shown in Figure 1, it comprises the following steps:

S1, start, set parameter values, including switch configuration parameters and initialization parameters, for example, the target agent utilization rate AO, the maximum call loss rate AR _max , the number of switches T, the maximum waiting time TQ _max , the prediction interval △t, The maximum length of the waiting queue qL, the user's patience time 1/η, the probability p ₀ that the user does not want to join the waiting queue, and the initial value of the number of outgoing calls λ.

In this embodiment, the workload is set, the maximum call loss rate AR _max = 0.02, the number of switches T = 10000, the target agent utilization rate AO = 0.8, the maximum waiting time TQ _max = 15s, the prediction interval time Δt = 5s, wait The maximum length of the queue (queue capacity) qL=2, the user's patience time 1/η=10s, the probability that the user does not want to join the waiting queue p ₀ =0.01 and other target parameters. The value behind the equal sign is the default value of the parameter, so that the accumulated value of l is al=0.

S2. At time t, obtain the total number of current customer service N, the total number of idle customer service f, the number of calls j, the connection rate ρ, the average ringing time 1/β, the average call time 1/μ, and the waiting queue length q through the switch The value of , said t is a positive number;

S3, judge whether q<qL is satisfied, if so, go to step S4, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt;

S4, calculate the required minimum number of seats M, the calculation formula is:

S5, judging whether N≤M is satisfied, if so, go to step S17, otherwise go to step S6;

S6, the calculation system predicts the maximum outbound call value l, accumulatively calculates al=al+l, wherein,

S7, the statistical strategy module uses the pre-stored statistical method to dynamically count the call loss rate AR at the last moment and judge whether AR<AR _max is satisfied, if it is satisfied, go to step S8, otherwise go to step S16;

S8, judging whether the current total number of idle customer service f>al is satisfied, if satisfied, then go to step S9, otherwise go to step S13;

S9, prediction result s=(f-j)/ρ-q+al, if al>1 then al is set to 0;

S10, calculate the user waiting time Tq, Tq=q/(N*μ), wherein q is the number of people in the current queue, if q is 0, it means that there is no waiting client at present;

S11, judge whether Tq<TQ _max is satisfied, if it is satisfied, go to step S12, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt;

S12, update the prediction result s=s*ARrate, ARrate is the exhalation attenuation ratio of the system, ARrate=1-AR, AR is the call loss rate, AR=p ₀ B(θ,N)+(1-p ₀ )λβρη, Go to step S18;

S13, the prediction result s=f-j-q+al, if al>1, set al to 0;

S14, calculate call loss rate AR, AR=p ₀ B(θ,N)+(1-p ₀ )λβρη, telephone call loss can be divided into those who do not want to enter the waiting queue and those who are willing to enter the waiting queue Part, the first half only occurs when i=N, assuming θ=λβρ(1-p ₀ )/μ is the input load of successfully connected calls, according to the Erlang-B formula, it can be calculated when N customer service personnel are not idle The probability π _N is the call loss rate. This probability can be obtained by iteration through the Erlang formula:

B(θ,0)=1,

S15, determine whether AR<AR _max is satisfied, if so, go to step S10, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt;

S16, prediction result s=f-j-q, go to step S18;

S17, prediction result s=f-j, turn to step S18;

S18, calculate the final number of dialed calls λ, λ=(int)min(max(s, 0), T);

S19, making λ calls outside, the algorithm sleeps until t+Δt time, go to step S2 for the next round of prediction, and update the current time to t=t+Δt; repeat the above steps until the application program terminates.

In this embodiment, in step S7, the dynamic statistics of the call loss rate AR at the previous moment is to divide the number of calls lost at the stage of call loss from the moment t1 before the current moment to the current moment stage by the number of all calls connected at this stage, The t1 is a positive number. In a more preferred embodiment of the present invention, t1 is a time between 15 minutes and 45 minutes, preferably 20 minutes.

If the selected interval is too small, there may be too little data in the time period to reflect the real situation. If the selected interval is too large, it cannot reflect the change of parameters in real time, and when the data fluctuation is obvious, it is not enough to reflect the real fluctuation situation.

Compared with the traditional predictive outbound algorithm, the present invention has the following advantages: the theoretical model of queuing when the seats are not enough is established through the theory of birth and death and queuing theory, and the statistical model of manual dialing data is used for ringing time, service time and statistical strategies , the established model is closer to the actual situation, and the seat utilization rate will be greatly improved under the same call loss rate; Dynamic statistics module. Therefore, it can be optimized from two directions, and there are many optimization schemes in both directions, thereby avoiding the bottleneck problem that only occurs when the algorithm is optimized; and when optimizing the algorithm, it is not necessary to care about the specific statistics of the parameters, and it is not necessary to consider the implementation of the algorithm when calculating the parameters and whether it is useful to the algorithm; the dynamic statistical module is added to improve the adaptability of the algorithm, so that it can adapt to a variety of different businesses and the situation where the data fluctuates too much; the idea of combining the statistical model with the theoretical model avoids the theoretical In extreme cases, the model deviates from the larger defect, so that the theoretical model can be better approached to the statistical model; the idea of combining feedback and prediction is adopted to make those that can predict accurately become more accurate, and those that are less accurate can be passed through feedback become more accurate, thus combining the advantages of prediction and feedback.

The automatic outbound call system is a very complicated call system, especially when there are certain requirements for customer service utilization and call loss rate. Therefore, a predictive outbound call algorithm should not only consider the above-mentioned influencing parameters but also consider the actual operating state parameters of the system. Therefore, the present invention proposes an outbound call algorithm combined with theoretical model simulation prediction and system real-time operating status, so that the customer service utilization rate and call loss rate of the outbound call system meet certain requirements.

As shown in Figure 2, utilizing the predictive outbound algorithm based on the dynamic statistical process of the present invention, the exchange dialing method comprises the following steps:

Step 1, predict the number of calls λ to be dialed at the next moment according to the predictive outbound algorithm of the present invention;

Step 2, the exchange dials λ calls according to the prediction result;

Step 3, after the average ringing time 1/β, only the calls with a ratio of ρ are connected, and the remaining calls with a ratio of 1-ρ are not connected and are filtered out;

Step 4, the number of connected people enters the control system, and the control system will allocate according to whether there are currently free seats;

Step 5, if there is an idle seat, assign an agent to serve it, and the call is successful if the service is completed; if there is no idle seat, there will be customers with a probability of p0 who are willing to join the waiting queue for waiting, and the remaining customers with a probability of 1-p0 will not Willing to join the waiting queue and give up to cause call loss;

Step 6, some of the customers in queue also give up and form a call loss due to waiting beyond their patient time 1/η, and only the rest will pass through the control system according to the idle service that has been served at present without exceeding their patient time 1/η. The agent assigns the agent service to it, and finally the service is completed.

As shown in Figure 3, the present invention operates the device of the predictive outbound algorithm based on the dynamic statistical process, which includes a data input module, a predictive outbound algorithm module, a switch, a dynamic statistical module, a data storage module and a data output display module, The data input module inputs initialization parameter data to the predictive outbound call algorithm module, and the predictive outbound call algorithm module is connected to the switch for information interaction, and the dynamic statistics module is connected to the data storage module to receive data from the data storage module, and the dynamic statistics module counts different The statistical data of the frequency is transmitted to the predictive outbound algorithm module, and the data is displayed through the data output interface at the same time. The predictive outbound algorithm module is used to predict the call result at the next moment based on the historical data of the previous moment, the dynamic statistical results of the intermediate data, preset parameters (including target parameters) and theoretical data, and decide whether to call according to the predicted result The phone still accumulates parameters to the next moment. The dynamic statistics module is used to conduct statistics on the parameters required by the algorithm or intermediate results based on historical data, and to display some parameters for measuring algorithm indicators in real time and dynamically.

The predictive outbound algorithm module includes a configuration parameter unit, an initialization parameter unit, a theoretical calculation data unit, a period statistical parameter unit, a predictive outbound algorithm unit, and a forecast result unit. The configuration parameter unit is used to receive switch configuration parameter information (including the number of switches, the maximum waiting time, the current number of people in line, the number of currently available seats, etc.), the initialization parameter unit is used to receive various target parameter data input by the data input module, and perform theoretical calculations The data unit is used to receive theoretically calculated but periodically changed data (including theoretical AR and the like, all data calculated by formulas but not obtained by statistical methods are theoretically calculated data), and the periodic statistical parameter unit is used for Various data collected by the dynamic statistical module with different statistical frequencies are received, the predictive outbound algorithm unit is used to implement the predictive outbound algorithm, and the predicted result unit is used to transmit the predicted result to the switch.

The dynamic statistical module includes obtaining historical data units, selecting statistical strategy units, and statistical strategy units (statistical strategies are pre-stored in the statistical strategy unit, such as statistics according to the week, month or holiday, classified according to the morning and afternoon, and statistics according to each time period of the day According to the strategy of accumulative statistics, etc.), the unit that transmits statistical data to the predictive outbound algorithm module and the unit that transmits statistical data to the output interface, the historical data acquisition unit stores the data that has been written into the database and the switch For the data that is being served but has not yet been written into the database, the statistical strategy unit is used to select which statistical strategy to use, and the statistical strategy unit includes a variety of specific statistical strategies.

The device of the predictive outbound algorithm based on dynamic statistics of the present invention can be embedded in an existing switch, or can provide an interface with the switch for separate deployment. Those of ordinary skill in the art can understand that the process of realizing the predictive outbound algorithm based on dynamic statistics of the present invention can be completed through program design, and the specific program can be stored in a device based on a predictive outbound algorithm based on dynamic statistics In the readable storage medium, the program executes the corresponding steps in the above method when executed. The storage medium may be, for example, ROM/RAM, magnetic disk, optical disk and hard disk.

The present invention can regard the outgoing call system as a queuing model, regard the number of calls in the system as a function that changes with time, then this system is a time-continuous Markov chain, and the system sets the maximum number of calls (workload ) is W, assuming that the call center has N identical seats and allows a queuing queue length of c and T switches. After data mining and theoretical analysis of a large number of manual dialing data of the company, the number of outbound calls is modeled as a Poisson distribution with a parameter of λ, the ringing time of the phone is a Poisson distribution with an average value of 1/β seconds, and the average number of calls received is a Poisson distribution. The pass rate is ρ, and the probability that the connected customer does not enter the waiting queue (hang up directly) is p ₀ in the case of no service, and the patient time of the customer is 1/η second, and the service time of the connected customer obeys the mean Negative exponential distribution of 1/μsec.

Now the outbound system process can be regarded as a probabilistic state transition process under the time-continuous Markov chain, X(t) is the total number of calls in the system at time t, including the number of calls in service and in the waiting queue. Then the state space E={0,1,2,...,N+c} owned by the time-continuous Markov chain at this time, for many queuing processes, the arrival of a customer (causing state plus 1) can be Expressed as "live", the departure of a customer (causing the state minus 1) can be expressed as "dead", then when X(t)=k, when the state changes at the next moment, it can only become k±1, so the entire The process becomes an independent random process with "birth" probability λ _k and "death" probability μ _k - a birth-death process.

in:

Then the probability distribution {π ₀ ,π ₁ ,…,π _N+c } of the system reaching a steady state can be calculated according to the "birth and death" process, where ∑ _i∈E π _i = 1, according to the steady state in unit time The average number of times to enter the state and leave the state is equal, and the formula can be listed:

After obtaining the probability distribution of the system in each state, some indicators to measure the performance of the system can be calculated:

① The expected TL of the system waiting queue length can be obtained according to the probability distribution of the system in this state:

TL＝∑ _{i∈E, i＞N} (iN)π _i

② The customer's average waiting time E[W] expectation when he did not get the service and did not give up waiting:

③ Customer service seat utilization expectation U:

The core of the usual telephone predictive dialing system is the predictive outbound algorithm. This system develops a method that divides the predictive outbound algorithm into two important modules: the predictive outbound algorithm module and the dynamic statistics module. Therefore, it can be optimized from two directions, and there are many optimization schemes in both directions, thus avoiding the bottleneck problem that only occurs in the optimization algorithm. And when optimizing the algorithm, you don't need to care about the specific statistics of the parameters, and you don't need to consider the implementation of the algorithm and whether it is useful to the algorithm when you calculate the parameters.

The number of calls dialed through the predictive outbound algorithm is usually greater than the number of available seats at that time, because the call connection rate is usually much smaller than 1, so compared with manual dialing and dialing without predictive function, it can greatly improve the agent utilization rate , but if there are too many calls, the call will be connected but no agent will be forced to give up the service. This situation is called "call loss". In practice, the proportion of call loss must be suppressed, that is, the call loss rate, otherwise the user Causes great harassment, so the algorithm must balance between the agent utilization rate and the call loss rate. Nevertheless, as long as the algorithm makes a reasonable prediction, it will still achieve a high seat utilization rate and produce a good prediction effect. But this requires that the algorithm can establish a better model, and develop a set of theories (including queuing theory) for the changes of various parameters, and then adapt to various complex situations. It is not enough to build a good forecasting model alone, because forecasting must be based on historical data, what and how much historical data is needed, and how to analyze the data to get a reasonable value, etc. This is the role of the dynamic statistics module Well, the present invention only relates to predictive outbound algorithms.

In the description of this specification, descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific examples", or "some examples" mean that specific features described in connection with the embodiment or example , structure, material or characteristic is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although the embodiments of the present invention have been shown and described, those skilled in the art can understand that various changes, modifications, substitutions and modifications can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the invention is defined by the claims and their equivalents.

Claims (5)

1. A predictive outbound algorithm based on dynamic statistical process, is characterized in that, comprises the steps: S1, start, set the target agent utilization rate AO, the maximum call loss rate AR _max , the number of exchanges T, the maximum waiting time TQ _max , the prediction interval △t, the maximum length of the waiting queue qL, the user's patience time 1/η, the user's unwillingness The probability p ₀ of joining the waiting queue, and the initial value of the number of outgoing calls λ, let the cumulative value of l=0, and l is the maximum outgoing call value predicted by the system; S2. At time t, obtain the total number of current customer service N, the total number of idle customer service f, the number of calls j, the connection rate ρ, the average ringing time 1/β, the average call time 1/μ, and the waiting queue length q through the switch The value of , said t is a positive number; S3, judge whether q<qL is satisfied, if so, go to step S4, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt; S4, calculate the required minimum number of seats M, the calculation formula is: S5, judging whether N≤M is satisfied, if so, go to step S17, otherwise go to step S6; S6, the calculation system predicts the maximum outbound call value l, accumulatively calculates al=al+l, wherein, S7, dynamically counting the call loss rate AR at the last moment and judging whether AR<AR _max is satisfied, if so, go to step S8, otherwise go to step S16; S8, judging whether the current total number of idle customer service f>al is satisfied, if satisfied, then go to step S9, otherwise go to step S13; S9, prediction result s=(f-j)/ρ-q+al, if al>1 then al is set to 0; S10, calculate user waiting time Tq, Tq=q/(N*μ); S11, judge whether Tq<TQ _max is satisfied, if it is satisfied, go to step S12, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt; S12, update the prediction result s=s*ARrate, ARrate is the exhalation attenuation ratio of the system, ARrate=1-AR, AR is the call loss rate, AR=p ₀ B(θ,N)+(1-p ₀ )λβρη, Go to step S18; S13, the prediction result s=f-j-q+al, if al>1, set al to 0; S14, calculate call loss rate AR, AR=p ₀ B(θ, N)+(1-p ₀ )λβρη B(θ,N) is the probability π _N when N customer service personnel are not free; S15, determine whether AR<AR _max is satisfied, if so, go to step S10, otherwise go to step S2 for the next round of prediction at the time of t+Δt, and update the current time to t=t+Δt; S16, prediction result s=f-j-q, turn to step S18; S17, prediction result s=f-j, turn to step S18; S18, calculate the final number of dialed calls λ, λ=(int)min(max(s, 0), T); S19, making λ calls outside, the algorithm sleeps until t+Δt time, go to step S2 for the next round of prediction, and update the current time to t=t+Δt; repeat the above steps until the application program terminates. 2. The predictive outbound algorithm based on the dynamic statistical process according to claim 1, characterized in that, in the step S7, the dynamic statistics of the call loss rate AR at the previous moment is calculated from the time t1 before the current moment to the current moment The number of calls lost in a stage is divided by the number of all calls connected in this stage, and t1 is a positive number. 3. The predictive outbound algorithm based on dynamic statistical process according to claim 2, characterized in that, said t1 is the time between 15 minutes and 45 minutes. 4. according to the switchboard dialing method based on the predictive type outgoing call algorithm based on dynamic statistical process according to claim 1, it is characterized in that, comprising the steps: Step 1, predicting the number of calls λ to be dialed at the next moment according to the predictive outbound algorithm according to claim 1; Step 2, the exchange dials λ calls according to the prediction result; Step 3: After the average ringing time is 1/β, the β is a positive number, and the calls with a ratio of ρ are connected, and the remaining calls with a ratio of 1-ρ are not connected and are filtered out, and the ratio of ρ is greater than 0 a positive number less than 1; Step 4, the number of connected people enters the control system, and the control system will allocate according to whether there are currently free seats; Step 5, if there is an idle seat, assign an agent to serve it, and the call is successful if the service is completed; if there is no idle seat, there will be customers with a probability of p0 who are willing to join the waiting queue for waiting, and the remaining customers with a probability of 1-p0 will not Willing to join the waiting queue and give up to cause call loss; Step 6, some of the customers in queue also give up and form a call loss due to waiting beyond their patient time 1/η, and only the rest will pass through the control system according to the idle service that has been served at present without exceeding their patient time 1/η. The agent assigns the agent service to it, and finally the service is completed. 5. a kind of device of the predictive type outgoing call algorithm based on dynamic statistics process described in running claim 1, is characterized in that, comprises data input module, predictive type outgoing call algorithm module, switch, dynamic statistics module, data storage module and The data output display module, the data input module inputs initialization parameter data to the predictive outbound call algorithm module, the predictive outcall algorithm module is connected to the switch for information interaction, the dynamic statistics module is connected to the data storage module, and receives The data of the data storage module, the statistical data of different frequencies are counted by the dynamic statistical module and transmitted to the predictive outbound algorithm module, and the data is displayed through the data output interface at the same time; The predictive outbound algorithm module includes a configuration parameter unit, an initialization parameter unit, a theoretical calculation data unit, a cycle statistics parameter unit, a predictive outbound algorithm unit, and a predicted result unit; the configuration parameter unit is used to receive switch configuration parameter information , the initialization parameter unit is used to receive various target parameter data input by the data input module, the theoretical calculation data unit is used to receive theoretical calculation but periodically changed data, and the cycle statistics parameter unit is used to receive the dynamic statistics module Various data collected with different statistical frequencies, the predictive outbound algorithm unit is used to implement the predictive outbound algorithm according to claim 1, and the predicted result unit is used to deliver the predicted result to the switch; The dynamic statistics module includes an acquisition historical data unit, a statistical strategy unit, a statistical strategy unit, a unit for transferring statistical data to the predictive outbound algorithm module and a unit for transferring statistical data to the output interface, and the historical data acquisition unit stores It includes obtaining the data that has been written into the database and the data that is being served by the switch but has not yet been written into the database. The selection statistical strategy unit is used to select which statistical strategy to use. The statistical strategy unit includes a variety of specific statistical strategy.