Predictive communications system

Abstract

A computer implemented method and computer system for generating outbound calls in a call center in which a plurality of agents are to communicate respectively with outside parties through the outbound calls. The method includes determining a number of agents available for outbound calls, determining phone numbers respectively corresponding to the outside parties, creating an ordered list of calls corresponding to the phone numbers and periodically generating new call attempts by automatically dialing the phone numbers at a dynamic rate, wherein a number of new call attempts is based on a predictive algorithm.

Description

BACKGROUND

The disclosed technology relates to generating outbound communications from a call center. Call centers are computer systems that connect parties, such as potential customers, to persons or groups of persons, referred to as agents, that are able to assist the party. Customers, or other callers interact with a contact center using voice, email, text, and web interfaces to communicate with agent(s) through a computer network and one or more of text or multimedia channels. The agent(s) may be remote from the contact center and may handle communications with customers on behalf of an enterprise associated with the contact center.

The agent(s) may utilize computing devices, such as workstations, desktop computers, laptops, telephones, a mobile smartphone and/or a tablet. Similarly, customers may communicate using a plurality of devices, including a telephone, a mobile smartphone, a tablet, a laptop, a desktop computer, or other. For example, telephone communication may traverse networks such as a public switched telephone networks (PSTN), Voice over Internet Protocol (VoIP) telephony (via the Internet), a Wide Area Network (WAN) or a Large Area Network. The network types are provided by way of example and are not intended to limit types of networks used for communications.

In some implementations, agents may be assigned to one or more “queues”, and the agents assigned to a queue may handle communications that are placed in the queue by the contact center. For example, there may be queues associated with a language (e.g., English or Chinese), topic (e.g., technical support or billing), or a particular country of origin of the communication. When a communication is received by the contact center, the communication may be placed in a relevant queue, and the communication can be routed to one or more of the agents assigned to the relevant queue to handle the communication.

Further, outbound call campaigns, where the call center proactively contacts outside parties and connects the parties with agents, is a well-known form of telemarketing and/or information gathering. For example, the agents may originate calls to the customers or potential customers and ask survey questions or describe products and services that are for sale. In more sophisticated instances, a list of outbound calls is created and calls on the list are made automatically initiated by the call center computer(s) and, when answered, are routed/connected to an agent in a manner that can be similar to routing of incoming communications. It is desirable to balance the quantity and rate of outgoing calls with the number of available agents in a dynamic manner to minimize both agent idle time and the need to abandon calls when an agent is not available to handle the call. The dynamics of a call center are very difficult to forecast because of the many dynamic, and seemingly random factors that are involved. For example, it is difficult, if not impossible to predict the time that any specific call will take (e.g., the call recipient may have many questions). Therefore, the availability of agents is very dynamic. Further, the same group of agents could be handling incoming calls to the call center contemporaneously with the outgoing calls and/or the agents may be assigned to multiple distinct outgoing call campaigns.

The concept of “predictive modeling”, also known as “predictive analytics”, applies mathematical processes to predict future events or outcomes by analyzing existing patterns that are likely to be indicative of future results. Once historical data indicating previous activity has been collected, it can often be used to train statistical models known as “machine learning models” (ML). Predictive modeling has been used in connection with meteorology and weather forecasting, as well as in many business applications. For example, predictive modeling is often applied in online advertising and marketing to determine what kinds of products users might be interested in, and what they are likely to click on, based on previous activity on the Internet. Predictive modeling has also been used to detect email “spam” and to detect fraudulent activity, such as financial scams.

There are many known predictive analytics methodologies, including logistic regression, time series analysis, decision trees, and neural networks. A neural network is a machine learning model that processes large volumes of labeled data to detect correlations between variables in the data. The algorithm can then be used to make inferences about unlabeled data files that are similar in type to the training data set. Common algorithms for predictive modeling include: Random Forest; Gradient Boosted; K-Means; and Prophet

Various predictive modeling tools, which allow customized predictive models to be created, are well known. These tools include:

• • Sisense
  • Oracle Crystal Ball
  • IBM SPSS Predictive Analytics Enterprise
  • SAS advanced Analytics Predictive modeling considerations

In applying a predictive model, it is critical to define and acquire the correct data and define the workflow of the application in which the model is uses. Further, in outgoing call campaigns, as noted above, there are many unrelated variables, that render the application of learning models difficult and ineffective. For example, it is difficult to apply learning models to the various variables in a pragmatic manner, there is no standard source of quality training datasets For these reasons, predictive modeling has not been effectively applied to outgoing call campaigns.

SUMMARY

Applicants have identified specific parameters of influence, and workflows, in outbound call campaigns for which application of a learning model can be effective to improve the efficiency of call origination and routing in a call campaign computing system. A first disclosed implementation is a computer implemented method for generating outbound calls in a call center in which a plurality of agents are to communicate respectively with outside parties through the outbound calls, the method includes determining a number of agents available for outbound calls, determining phone numbers respectively corresponding to the outside parties, creating an ordered list of calls corresponding to the phone numbers, periodically generating new call attempts by automatically dialing the phone numbers at a dynamic rate, wherein a number of new call attempts is based on a predictive algorithm that includes;

• • (a) selecting a set of N number of calls that are next on the ordered list
  • (b) determinizing corresponding probabilities of pick up for each call in the set of calls;
  • (c) summing the corresponding probabilities of pick up to determine a pick up rate for the set of calls; and
  • (d) launching a number of calls in the set of calls that correspond to the number of available agents; and
  • repeating (a), (b), (c) and (d) until a desired number of calls have been launched.

A second disclosed implementation is a computer system including processors which execute code to accomplish a method of generating outbound calls in a call center in which a plurality of agents are to communicate respectively with outside parties through the outbound calls, the method includes determining a number of agents available for outbound calls, determining phone numbers respectively corresponding to the outside parties, creating an ordered list of calls corresponding to the phone numbers, periodically generating new call attempts by automatically dialing the phone numbers at a dynamic rate, wherein a number of new call attempts is based on a predictive algorithm that includes;

• • (a) selecting a set of N number of calls that are next on the ordered list
  • (b) determinizing corresponding probabilities of pick up for each call in the set of calls;
  • (c) summing the corresponding probabilities of pick up to determine a pick up rate for the set of calls; and
  • (d) launching a number of calls in the set of calls that correspond to the number of available agents; and
  • repeating (a), (b), (c) and (d) until a desired number of calls have been launched.

BRIEF DESCRIPTION OF THE DRAWING

The foregoing summary, as well as the following detailed description of the invention, will be better understood when read in conjunction with the appended drawings. For the purpose of illustrating the invention, there are shown in the drawings various illustrative embodiments. It should be understood, however, that the invention is not limited to the precise arrangements and instrumentalities shown.

FIG. 1 is a schematic illustration of a computer system for pacing outbound calls in a campaign.

FIG. 2 is an architecture and flow diagram of the application of a predictive model for predicting call picked up rate.

FIG. 3 is a graph showing the results of an analysis of the impact of the utilization of predicted values to select records in an outbound call campaign.

FIG. 4 is an architecture and flow diagram of utilization of predicted values to select records in an outbound call campaign.

FIG. 5 is an architecture and flow diagram of utilization of predicted values to allocate agents in an outbound call campaign.

FIG. 6 is an architecture and flow diagram of utilization of predicted values to cancel pending calls in an outbound call campaign.

DETAILED DESCRIPTION

Certain terminology is used in the following description for convenience only and is not limiting. For example, unless specifically set forth herein, the terms “a,” “an” and “the” are not limited to one element but instead should be read as meaning “at least one.” The terminology includes the words noted above, derivatives thereof and words of similar import.

As noted above, in an outbound calling campaign, it is desirable to have an agent ready and waiting when a call is answered by a party. However, it is also desirable to maximize the number of outbound calls per agent, i.e., to minimize agent idle time. To balance these conflicting objectives, it is known to automatically pace outbound calls, in cycles of calls originated, based on the number of currently available agents and historical “picked up rate” that indicates the number of calls that will likely be picked up (i.e. answered by a live voice), as a percentage of calls made for example. However, as noted above, there are a myriad of unrelated dynamic factors in ascertaining the best pacing for an outbound call campaign. As a result, known outbound call campaigns often have a high number of idle agents, and/or answered calls that must be abandoned without connection to an agent.

As noted above, calls of outgoing call campaigns are accomplished in cycles in which a relatively small (with respect to the entire number of calls in the campaign) number of calls are made at one time based on the number of available agents to handle the calls and other parameters (referred to as “features” herein) of the call campaign. For example, it may be desirable that less than 50% of outgoing calls be abandoned by the call center (because there is not available agent when the call is answered, for example). The following formula can be applied to determine the number of calls that should be ringing currently:

• • number_new_attempts=max(number_attempts−number_pending_attempts, 0)
  • number_attempts=floor(min(ceiling(number_agents_available*1.0/pick_up_rate),
  • number_agents_available*MAX_DIALING_RATIO,
  • max(number_agents_available, floor((number_agents_available+allowed_aba)
  • Where:
  • pick_up_rate=total_answered_attempts/total_finished_attempts
  • allowed_abandonment=picked_up_attempts*ABANDONMENT_RATE−total_abandon
  • picked_up_attempts=all attempts where the record answered the call

The following table describes examples of various variables/data structures that can be used in an algorithm to determine a number of new calls for a cycle.

Reference                Description
number_pending_attempts  Number of attempts that are ringing and
                         waiting either for the customer to pick up,
                         orfor the agent to connect.
number_agents_available  Number of agents available for
                         outboundcalls.
pick_up_rate             Current customer pick up rate for a given
                         window.
total_answered_attempts  Attempts that were answered by a customer.
total_finished_attempts  Total number of attempts finished in a given
                         window.
MAX_DIALING_RATIO        The maximum dialing ratio per agent. For
                         predictive, it must be greater than 1 so
                         thatwe can attempt more calls than agents
                         available.
allowed_abandonment      The number of additional calls an automated
                         Dialer can originate while respecting the
                         desired abandonment rates/limits
total_finished_attempts  Total number of attempts finished in a given
                         window.
ABANDONMENT_RATE         Constant that defines the abandonment rate
                         for a given time window.
total_abandoned_attempts The number of times a call center
                         terminated a call before it was completed.
                         This may happen where there wasn't an
                         agent available when the contact answered
                         the call.

Disclosed implementations leverage Machine Learning models in new ways, and for new applications, to automatically learn from data and improve decision making for outbound call campaigns. For example, applicant has discovered that, by applying Machine Learning models to predict picked up rate, the efficiency of an outbound call campaign can be greatly improved with minimal computing resources. An outbound dialer system possesses large, classified datasets that can be used to train supervised predictive models. For instance, each call of a previous campaign has a corresponding system disposition and possibly a call disposition. Therefore, this data can be used to produce various predictive models to determine, for example, a probability of an individual call being picked-up, a probability of a busy call or a probability of a call being abandoned. Accordingly, these models can provide valuable information for various decisions that are critical to outbound call campaigns such as: calculation of the number of calls; selection and sorting of records; allocation of agents for pending calls; and cancellation of pending calls.

FIG. 1 illustrates an outgoing call campaign system 100 in accordance with disclosed implementations. As will be seen below, system 100 can be applied to originate outbound calls in a campaign in cycles and at a rate that improves efficiency of the campaign. System 100 may include one or more servers 102. Server(s) 102 may be configured to communicate with one or more remote computing platforms 104 according to a client/server architecture and/or other architectures. Users, such as those desiring to configure the features of a call campaign, may access system 100 via client computing platform(s) 104. System 100 can be integrated into a conventional call center architecture.

Server(s) 102 may be configured by machine-readable instructions 106. Machine-readable instructions 106 may include one or more instruction modules. The instruction modules may include computer program modules. The instruction modules may include one or more of agent determination module 108, phone number determination module 110, ordered list access module 112, call attempt module 114, predictive module 116 and/or other instruction modules. For example, modules can be provided for selection and sorting of records; allocation of agents for pending calls; and/or cancellation of pending calls, as described in detail below.

Agent determination module 108 can determine, or otherwise ascertain, a number of agents currently available for outbound calls. Phone number determination module 110 can determine, or otherwise ascertain, phone numbers respectively corresponding to the outside parties to be contacted in an outbound call campaign. Ordered list access module 112 can access an ordered list of calls corresponding to the phone numbers. Call attempt module 114 can periodically generate new call attempts, in cycles, by automatically dialing the phone numbers at a dynamic rate. Predictive module 116 can determine the dynamic rate based on a predictive algorithm that includes;

• • (a) selecting a set of N number of calls that are next on the ordered list
  • (b) determinizing corresponding probabilities of pick up for each call in the set of calls;
  • (c) summing the corresponding probabilities of pick up to determine a pick up rate for the set of calls;
  • (d) launching a number of calls in the set of calls that correspond to the number of available agents; and
  • repeating (a), (b), (c) and (d) until a desired number of calls have been launched.

In some implementations, server(s) 102, remote computing platform(s) 104, and/or external resources 124 may be operatively linked via one or more electronic communication links. For example, such electronic communication links may be established, at least in part, via a network such as the Internet and/or other networks. It will be appreciated that this is not intended to be limiting, and that the scope of this disclosure includes implementations in which server(s) 102, remote computing platform(s) 104, and/or external resources 124 may be operatively linked via some other communication media.

External resources 124 may include sources of information outside of system 100, external entities participating with system 100, and/or other resources. In some implementations, some or all the functionality attributed herein to external resources 124 may be provided by resources included in system 100. For example, external resources can include a database of a call center system. As used herein, the term “module” may refer to any component or set of components that perform the functionality attributed to the module. This may include one or more physical processors during execution of processor readable instructions, the processor readable instructions, circuitry, hardware, storage media, or any other components.

FIG. 2 illustrates the application of a predictive model to improve the calculation of the number of calls in each cycle of an outbound call campaign. The use of a predictive model in this manner allows the number of potentially launched calls in a cycle to be calculated based on the predicted probability of each individual potential call, or a set of calls, being picked-up, not merely an overall percentage of calls that are expected to be picked up. The sum of the predicted probabilities of the potentially launched calls can be matched to the number of available agents to ascertain the number of calls to launch at any particular time. For example, the expected number of picked-up calls should be similar to the expected number of available agents to minimize agent idle time and minimize the number of abandoned calls. This allows the campaign algorithm to look at the calls in the campaign list and determine which specific calls to launch in a cycle based on the number of available agents. As a simple example, if the next four calls on a campaign list have respective probabilities of 10%, 20%, 50%, and 20%, and there is only one available agent, these four calls can be launched and one pickup will be expected (because the summed probability is 100% or 1 call).

The summing of call pickup probabilities can be periodically or continuously accomplished to select the specific calls, and the quantity thereof, to launch at any given time. In other words, the utilization of the predicted picked-up rate based on probabilities of pickup of each call allows improvement of both the average number of picked-up calls per cycle (related with agents' idle time) and the intended abandonment rate when compared with historical picked up rate and maximum dialing ratio. As shown in FIG. 2, a call campaign can be created at creation phase 210 in a conventional manner. Typically, the campaign will define the number of calls to be made the phone numbers to call, the acceptable abandonment rate, and other parameters that specify the campaign characteristics.

At prediction phase 220, a predictive model (any one of various known predictive algorithms can be utilized) is applied to each call, or a subset of calls, in the campaign to assign an individual probability of pick up for each of the calls analyzed by the predictive model. The individual probabilities of pick up can be stored in a database in association with the corresponding calls (e.g., phone numbers). At call launch phase 230, the individual probabilities generated by the predictive model can be applied to launch calls in the campaign in multiple cycles.

In particular, at call launch phase 230, a next number X of call records is read, at step 231, from the database storing the call campaign features. The number X of records can be adjusted in various ways and can be a predetermined fixed integer, such as 100. At step 232, the corresponding predicted pick up probabilities for each of the X calls is read from the database. At step 234, the probabilities are summed and a calculated suggested number of calls to be launched is generated. At 235, the maximum acceptable abandonment rate, the number of currently available agents and other parameters are retrieved, and an allowable number of calls is estimated based on these parameters. At 236 it is determined whether the allowable number of calls is greater than the suggested number of calls and, if so, the calculated suggested calls are launched. If not, the allowable number of calls can be launched step 230 can be repeated with a different value of X until the calculated suggested calls are less that the allowable number of calls.

Since the required latency for each cycle is very reduced, the application of the predictive model may have some tasks executed before each cycle. The predictive model can produce the features some time before in order to facilitate the procedure and to reduce required computing resources. For example, prediction phase 220 can be accomplished at any time if the data, parameters, and assumptions used by the predictive model are not out of date. To make the calculation of the number of calls even faster predicted probabilities can be generated each time the record list or other features of the campaign are updated in order to make predictions available for each cycle. Therefore, it is only necessary to extract the predicted values, and generate a cumulative sum thereof, once for each cycle. However, it may be desirable to have real-time access to the next records in order to produce a more accurate estimate of the number of launched calls.

In another implementation, a predictive model can be applied to provide a more efficient selection of records and thus can avoid the unnecessary launch of many calls that will not be answered. Such a selection may improve the picked-up rate and agents' idle time by reducing the total number of calls with a minimum impact on the number of picked-up calls. Moreover, it allows a more efficient utilization of the resources by minimizing the degradation of caller IDs (e.g., less busy calls) and by making the cycles more productive. The probability of a call being picked-up is a logical criterion for the selection of records for a call campaign cycle. Therefore, those records with a very low probability could be excluded. FIG. 3 shows the results of an analysis of the impact of the utilization of predicted values to select records.

As shown in FIG. 3, the average daily number of calls has a sharp decline at the lowest minimum probability scores while the average daily number of picked-up calls has a minimum impact. This selection allows the system to discard a higher percentage of calls to be launched than the percentage of picked up calls that will be missed. For example, the utilization of 0.2 as a threshold allowed the system to increase the picked up rate from 45.48% to 60.2%.

Sorting records is also a task that could benefit from the utilization of the predictive model. The predicted probabilities of a call being picked-up could be a valuable basis to sort records. For example, if the objective is to schedule the best time to call, prediction can be made for different times and the best schedule can be selected according to these predictions. If the objective is to obtain a less volatile picked-up rate, records can be sorted in order to obtain a smoother distribution of the picked-up rate over time. If the goal is to first launch the calls with the highest probability of being picked-up and leave the ones with the lowest probability to the end, the records can be sorted in descending order according to the predicted probability. FIG. 4 shows the application of a predictive model to improve the selection and sorting of records for an outgoing call campaign.

As shown in FIG. 4, a call campaign can be created at creation phase 410 in a conventional manner. Creation phase 410 can be similar to, or the same as, creation phase 210 in FIG. 2. In predicting phase 420, a predictive model (any one of various known predictive algorithms can be utilized) is applied to each call, or a subset of calls, in the campaign to assign an individual probability of pick up for each of the calls analyzed by the predictive model. Predicting phase 420 can be similar to predicting phase 220 of FIG. 2. As noted above with respect to FIG. 2, the individual probabilities of pick up can be stored in a database in association with the corresponding calls (e.g., phone numbers).

The predictive model used for sorting can be similar to the model applied in the calculation of the number of launched calls as described above. The applied features can also be similar. However, this predictive model should be applied to the complete list of records when it is necessary to select and sort calls for a campaign. Also, this model may be applied for different times when it is necessary to schedule the best time to call. Call records are selected and sorted in sorting phase 430. Sorting phase 430 includes the primary steps of selecting records above a threshold/minimum score at 421 and sorting the records in accordance with objectives of the call campaign at 422. As noted above, the individualized call picked up rate is one metric that can benefit more from the application of learning models. The selection of records allows the system to discard a significant number of calls with minimum impact on picked-up calls. The agents' idle time can also benefit from the improvements made on the picked-up rate.

Agent reservations can be recalculated across cycles to guarantee a “blanket” effect for all dialed records across all campaigns. Campaigns with ongoing dials will have agents reserved for those possible calls. Conventionally, measures such as historical picked-up rate or maximum dialing ratio, which are not the most adequate measures are used for agent reservations. These measures are not ordinarily adjusted to the characteristics of current pending calls (e.g., attempt number, campaign, historical behavior of the customer) and to current ringing time. Thus, the utilization of a predictive model of the probability of a call being picked-up may allow a more effective allocation of agents by producing a more accurate estimate of the number of picked-up calls. Therefore, such an application can improve agents' idle time and abandonment rate by diminishing the difference between picked-up calls and allocated agents. The impact on these metrics are significant when there are more agents (increasing agents' idle time) or less agents (causing more abandonment then necessary).

FIG. 5 illustrates the application of a predictive model to allocate agents for pending calls. The creation of the campaign and the utilization of the predictive model is similar that applied in the calculation of the number of calls. Both share the same features except for the ringing time that is provided in real time. Therefore, the creation of features is shared by both models. Also, the creation of the predictions is done at the same time (whenever a record list or features are updated) and these predictions are applied at each cycle. The prediction is accomplished for each recordID and possible ringing time. These two attributes are the input parameters required to extract predictive probabilities later in each cycle. However, it is necessary to train a different predictive model for this implementation because this application also uses ringing time as a parameter/feature. The implementations described above do not use ringing time because they are applied before each call is launched. Ringing time is used as a feature due to its predictive value and also because we do have this information available in each cycle for each pending call. At 520, the number of allocated agents is calculated based on the predicted probabilities for each ringing time. The utilization of the predictive model to allocate agents can improve intended abandonment and agents' idle time.

When the expected number of picked-up calls is much higher than the expected number of available agents, it is very likely that some of these calls will be abandoned. It would be useful to cancel some of these pending calls to avoid abandonment. The predictive models can also be applied to support the decision to cancel pending calls. Thus, it may be determined how many pending calls should be cancelled based on the predicted number of picked up calls and predicted number of available agents for the same time period. For example, if it is predicted that only one agent will be available, only the set of calls that will have one expected picked up calls can be kept and all the others can be cancelled. To be able to do it, we need a predictive model regarding a pending call being picked up (the same as the one applied to the allocation of agents) and another to predict the number of available agents in a specific time window. The later model could be trained by applying data about agents availability and activity. A predictive model of the number the available agents could be trained by applying data about agents availability and activity. The output should be the expected number of available agents in a specific time window. The predictive model applied in the allocation of agents could be used to estimate the number of picked-upcalls in the same time windows. Based on both predictions, the algorithm would decide which pending call should be canceled. FIG. 6 illustrate the utilization of a predictive model to cancel pending calls. At 620, the number of pending calls that will be picked up is calculated based on the predicted probabilities for each ringing time and other variables. At each cycle, the expected number of picked up calls among pending calls will be calculated based on the extracted predicted probabilities of being picked up and the extracted predicted number of available agents. Then, based on both predictions (expected number of picked up calls and expected number of available agents) it is decided the number of calls to be canceled. Therefore, it should only be kept the set of calls that will have a similar number of picked up calls to the expected number of available agents for the same time window. This improves intended abandonment rate with minimum impact on agents' idle time values.

A predictive model is also very useful to select/exclude calls in a campaign. The picked up rate increases as the applied minimum predicted value to select calls also increases. This selection allows the system to discard a higher percentage of calls to be launched than the percentage of picked up calls that will be missed. For example, the utilization of 0.2 as a threshold increased the picked up rate from 45.48% to 60.2% in experiments conducted by the applicant.

The predictive model is also valuable to produce a more accurate estimate of the picked up rate of the next calls in order to be applied in the calculation of the number of calls to be launched. This estimate adjusts to the variance of real picked up rate and it is substantially more accurate than the existing estimate (historical picked up rate of each campaign). Applicants' experiments resulted in a large improvement in the mean absolute error (from 0.165 to 0.088) and in the root mean squared error (from 0.206 to 0.116).

The utilization of the predicted picked-up rate improves both the average number of picked-up calls per cycle (related with agents' idle time) and the abandonment rate when compared with historical picked up rate and maximum dialing ratio (1, 2 and 3). The only exception is “power dialing”, i.e. dialing a phone number right after a previous call is completed, for the abandonment rate but it has a high cost on the average number of picked up calls per cycle. For example, the median abandonment rate obtained by the historical picked-up rate, maximum dialing ratio 3 and maximum dialing ratio 2 were respectively 2.44, 4.31 and 2.26 times higher than the obtained by the predicted picked-up rate. Despite this large advantage on the abandonment rate, the median average number of picked-up calls per cycle obtained by the predicted picked-up rate was also higher than the median of the historical picked-up rate (+4.3%), maximum dialing ratio 2 (+12.1%) and power dialing (+77%) and very similar to maximum dialing ratio 3 (−1.2%).

It will be appreciated by those skilled in the art that changes could be made to the embodiments described above without departing from the broad inventive concept thereof. It is understood, therefore, that this invention is not limited to the particular embodiments disclosed, but it is intended to cover modifications within the spirit and scope of the present invention as defined by the appended claims.

Claims

1. A computer implemented method for generating outbound calls in a call center in which a plurality of agents are to communicate respectively with outside parties through the outbound calls, the method comprising: determining a number of agents available for outbound calls;determining phone numbers respectively corresponding to the outside parties;creating an ordered list of calls corresponding to the phone numbers;periodically generating new call attempts by automatically dialing the phone numbers at a dynamic rate, wherein a number of new call attempts is based on a predictive algorithm, wherein the predictive algorithm includes: (a) selecting a set of N number of calls that are next on the ordered list(b) determinizing corresponding probabilities of pick up for each call in the set of calls;(c) summing the corresponding probabilities of pick up to determine a pick up rate for the set of calls; and(d) launching a number of calls in the set of calls that correspond to the number of available agents; andrepeating (a), (b), (c) and (d) until a desired number of calls have been launched.

2. The method of claim 1 further comprising, prior to step (d) determining if the picked up rate is too high and if the picked up rate for the set of calls is too high for the number of available agents, returning to step (a) and reducing N.

3. The method of claim 2, wherein the picked up rate is determined to be too high if it results in an abandonment rate that is higher than a predetermined threshold.

4. The method of claim 1, wherein the predictive model is trained to provide a more efficient selection of records to thereby reduce the number of calls that will not be answered.

5. The method of claim 1, wherein the predictive model is trained to sort records to thereby increase efficiency of the call campaign.

6. The method of claim 1, wherein the predictive model is trained to allocate agents for pending calls.

7. The method of claim 1, wherein the predictive model is trained to cancel pending calls that are not likely to served by an agent.

8. The method of claim 1, wherein the predicative model is trained with a data set derived from call disposition records from at least one previous call campaign.

9. The method of claim 8, wherein a number of new call attempts is based on a predictive algorithm which predicts at least one of a probability of an individual call being picked-up, a probability of an individual call being busy, or a probability of an individual call being abandoned.

10. The method of claim 1, wherein creating an ordered list of calls corresponding to the phone numbers includes determining corresponding probabilities of pick up for each call in the set of calls and sorting the records based on the corresponding probabilities.

11. The method of claim 1, wherein creating an ordered list of calls corresponding to the phone numbers includes determining corresponding probabilities of pick up for each call in the set of calls and selecting the records based on the corresponding probabilities.

12. A computer system for generating outbound calls in a call center in which a plurality of agents are to communicate respectively with outside parties through the outbound calls, the method comprising: at least one computer processor; andat least one memory having instructions stored thereon which, when executed by the at least one computer processor, cause the at least one computer processor to carry out a method including: determining a number of agents available for outbound calls;determining phone numbers respectively corresponding to the outside parties;creating an ordered list of calls corresponding to the phone numbers;periodically generating new call attempts by automatically dialing the phone numbers at a dynamic rate, wherein a number of new call attempts is based on a predictive algorithm, wherein the predictive algorithm includes:(a) selecting a set of N number of calls that are next on the ordered list(b) determinizing corresponding probabilities of pick up for each call in the set of calls;(c) summing the corresponding probabilities of pick up to determine a pick up rate for the set of calls; and(d) launching a number of calls in the set of calls that correspond to the number of available agents; andrepeating (a), (b), (c) and (d) until a desired number of calls have been launched.

13. The system of claim 12 further comprising, prior to step (d) determining if the picked up rate is too high and if the picked up rate for the set of calls is too high for the number of available agents, returning to step (a) and reducing N.

14. The system of claim 13, wherein the picked up rate is determined to be too high if it results in an abandonment rate that is higher than a predetermined threshold.

15. The system of claim 12, wherein the predictive model is trained to provide a more efficient selection of records to thereby reduce the number of calls that will not be answered.

16. The system of claim 12, wherein the predictive model is trained to sort records to thereby increase efficiency of the call campaign.

17. The system of claim 12, wherein the predictive model is trained to allocate agents for pending calls.

18. The system of claim 12, wherein the predictive model is trained to cancel pending calls that are not likely to served by an agent.

19. The system of claim 12, wherein the predicative model is trained with a data set derived from call disposition records from at least one previous call campaign.

20. The system of claim 19, wherein a number of new call attempts is based on a predictive algorithm which predicts at least one of a probability of an individual call being picked-up, a probability of an individual call being busy, or a probability of an individual call being abandoned.

21. The method of claim 12, wherein creating an ordered list of calls corresponding to the phone numbers includes determining corresponding probabilities of pick up for each call in the set of calls and sorting the records based on the corresponding probabilities.

22. The method of claim 12, wherein creating an ordered list of calls corresponding to the phone numbers includes determining corresponding probabilities of pick up for each call in the set of calls and selecting the records based on the corresponding probabilities.