Method for controlling abandonment rate in outbound campaigns

[0001] Applicant claims benefit pursuant to 35 U.S.C. §119 and hereby incorporates by reference the following U.S. Provisional Patent Applications: “System and Methods for Optimal Outbound Dialing in Virtual Contact Center,” Ser. No. 61/426,992, filed Dec. 23, 2010; “System and Methods for Outbound Dialing in Multi-Campaign and Blended Environment of Virtual Contact Center,” Ser. No. 61/427,151, filed Dec. 24, 2010; “Method for Controlling Abandonment Rate in Outbound Campaigns,” Ser. No. 61/430,660, filed Jan. 7, 2011; and “Method for Estimation Impact of Calls in Dialer for Predictive Dialing,” Ser. No. 61/430,669, filed Jan. 7, 2011.

[0002] The technology described in this patent document relates generally to contact centers, and more particularly to dialing methods used by the contact centers for determining a number of outbound calls to generate.

[0003] One function typically performed by a contact center is conducting outbound campaigns, where agents of the contact center proactively contact prospective customers to make a sale or to achieve some other objective. To optimize agent utilization, a contemporary contact center often utilizes an automated dialing system to handle the dialing aspect of an outbound call. The automated dialing system, in general, is responsible for automatically making outbound calls to potential customers and routing only the successfully connected calls to agents. Thus, with the dialing process being automated by the system, agents can focus their time on interfacing with customers.

[0004] In accordance with the teachings described herein, methods and systems are provided for controlling and correcting abandonment rate in an automated contact center that uses a predictive dialing method for determining a number of calls to generate for dialing. The automated contact center comprises a queue where dialed calls that successfully connected await agent handling and an agent pool where calls are handled by agents. An abandonment rate is a proportion of calls in the queue that are abandoned.

[0005] A method for controlling abandonment rate may include: determining a mean agent occupancy, wherein the mean agent occupancy is used by the predictive dialing method to determine a number of calls to generate, wherein the abandonment rate is close to zero when the predictive dialing method uses the mean agent occupancy, and wherein agent occupancy is a proportion of agents in the agent pool that are occupied by calls; collecting empirical data on agent occupancy while the predictive dialing method uses the mean agent occupancy to determine a number of calls to generate; determining a target abandonment rate; determining a target agent occupancy, wherein a proportion of the agent occupancy empirical data having an agent occupancy equal to or higher than the target agent occupancy is within a predetermined range of the target abandonment rate; using the target agent occupancy to adjust the mean agent occupancy used by the predictive dialing method; using the adjusted predictive dialing method to determine a number of calls to generate for dialing; and generating calls based upon the number of calls to generate.

[0006] A method for correcting abandonment rate may include: collecting empirical data on agent occupancy and on abandonment rate while using the predictive dialing method, wherein agent occupancy is a proportion of agents in the agent pool that are occupied by calls; determining a normal mean agent occupancy, wherein the normal mean agent occupancy is used by the predictive dialing method to determine a number of calls to generate; determining a normal abandonment rate using the abandonment rate empirical data; monitoring a current abandonment rate while using the predictive dialing method; calculating a correcting abandonment rate when the current abandonment rate reaches a threshold; determining a correcting agent occupancy using the correcting abandonment rate, wherein a proportion of the agent occupancy empirical data having an agent occupancy equal to or higher than the correcting agent occupancy is within a predetermined range of the correcting abandonment rate; modifying the predictive dialing method to use the correcting agent occupancy instead of the normal mean agent occupancy; monitoring a correction-period current abandonment rate while using the modified predictive dialing method; resetting the modified predictive dialing method to use the normal mean agent occupancy instead of the correcting agent occupancy when the correction-period current abandonment rate is within a predetermined range of the normal abandonment rate; using the reset predictive dialing method to determine a number of calls to generate for dialing; and generating calls based upon the number of calls to generate.

[0017] An automated dialing system of a contact center may have multiple responsibilities. In addition to dialing calls and forwarding successful calls to agents, the automated dialing system may also have the responsibility of determining an optimal number of calls to generate. The number of calls is important because generating too few calls may result in underutilization of the agents, and generating too many calls may result in a backlog of calls that could cause customer dissatisfaction. In order to generate enough calls to adequately occupy the agents while not unreasonably adding to the backlog, it is desirable for the automated dialing system to take into consideration relevant system conditions, such as agent occupancy, when determining a number of calls to generate. More particularly, since a generated call typically would not reach an agent until a future time, it is desirable when determining the number of calls to generate to predict and consider relevant contact center conditions at the future time.

[0018] FIG. 1 is a block diagram of an example of an automated dialing system implemented by a virtual contact center. At a high level, the system includes an application server 101 that facilitates communication connections between customers 104 and agents 106 (agents 106 may include supervisors 107, administrators 108, and other human or non-human resources that interface with the customers 104). The system may have more than one instance of the application server 101, with each instance servicing, for example, a different tenant. However, for simplicity the ensuing description will focus on one instance of the application server 101.

[0019] To establish communication connections with customers 104, the application server 101 connects to a telephone server 102. The telephone server 102, which is responsible for third party call control, communicates with customers 104 through a network, such as a Public Switched Telephone Network (PSTN) 103. Although the particular example of FIG. 1 depicts using telephonic means to communicate with customers 104, other communication means may also be employed, such as Internet chat systems or cellular phone messaging systems.

[0020] The application server 101 is also communicatively connected with agents 106. FIG. 1. depicts the communication being established through the Internet 105, but any other communication means may be used as well (e.g., through a private or local network).

[0021] Connected in this manner, the application server 101 in operation can use the telephone server 102 to generate outbound calls over the PSTN 103 and establish telephonic connections with the customers 104. Once a connection is established, the application server 101 routes the connected call to an available agent 106 through the Internet 105. At a high level, the operation of generating outbound calls, establishing connection with customers 104, and routing successful connections to agents 106 is one of the basic functions of the automated dialing system.

[0022] At a more detailed level, when generating outbound calls, the application server 101 employs a campaign manager 112. The campaign manager 112 periodically generates, controls, and monitors all outbound calls during their lifecycles. The generated outbound calls may, for example, be represented as software objects and stored in a call container 117, which is accessible to the campaign manager 112 and an automated call distributor 110 (described below).

[0023] Before calls are generated, however, the campaign manager 112 determines an optimal number of calls to generate. As described above, the number of calls to generate is important for achieving and maintaining desirable system conditions (e.g., having an agent occupancy of 90%). To determine the number of calls to generate, the campaign manager 112 invokes a call director 111, which uses a dialing call calculator 116 to calculate the number of calls to generate. The dialing call calculator 116 is an implementation of one or more methods for calculating the number of calls to generate. Examples of the methods will be described in more detail below with respect to FIGS. 3-10. Once the dialing call calculator 116 determines the number of calls to generate and communicates that number to the call director 111, the call director 111 returns the number to the campaign manager 112.

[0024] Based upon the received number of calls to generate, the campaign manager 112 obtains a corresponding number of contact records that will be used for contacting customers 104. To obtain the contact records, the campaign manager 112 communicates the number of calls to generate to a campaign list manager 113. The campaign list manager 113 then accesses a contact center database 114, which stores contact records, and determines which contact records to retrieve and to return to the campaign manager 112. Once a set of contact records has been selected, the campaign list manager 113 communicates the contact records to the campaign manager 112.

[0025] When the campaign manager 112 receives the contact records, it attempts to establish communication with the customers 104 that are associated with the contact records. If, for example, the contact records include telephone numbers, the campaign manager 112 may use an embedded dialer 112 and the telephone server's 102 outbound capabilities to dial the telephone numbers. Once a call is dialed, the telephone server 102 then waits for a response.

[0026] When a call is not answered by a customer 104 within a predetermined time, or when a machine answers the call, the call may be terminated. On the other hand, when a call is answered by a customer 104, the telephone server 102 receives a corresponding event and informs an automated call distributor 110 to distribute the call to an available agent 106 with the appropriate skills for handling the call. In order to determine agent availability, the automated call distributor 110 communicates with a user control 115, which keeps track of agent availability via the Internet 105 connection with the agents 106. If the user control 115 indicates that an appropriate agent 106 is available to handle the call, the call is distributed to that agent 106. However, if no appropriate agent 106 is available, the automated call distributor 110 queues the call and waits for an agent 106 to become available. During the wait, either the customer 104 or the system may abandon the call. However, if the call is not abandoned and an agent 106 becomes available, then the automated call distributor 110 distributes the call to the agent 106. Once a call is distributed to an agent 106, a communication connection is established between the customer 104 and the agent 106.

[0027] Whether a generated call helps keep agents 106 busy or adds to the backlog of queued calls depends on the level of agent occupancy at the time when the call reaches the automated call distributor 110. If agents 106 are available at that time, then the call will occupy at least one agent 106; on the other hand, if no agents 106 are available, then the call will add to the backlog. Therefore, when determining a number of calls to generate, it is desirable to consider the likely agent occupancy at the future time period when a generated call would likely reach the automated call distributor 110. If agent occupancy is likely to be low, then more calls should be generated to keep agents 106 utilized; if agent occupancy is likely to be high, fewer calls should be generated to lessen the likelihood of call abandonment. This is the basic principal behind a predictive dialing method that a dialing call calculator 116 may implement to determine a number of calls to generate.

[0028] In order to estimate a future agent occupancy, the dialing calculator 116 analyzes the status of each call that is currently in the system and determines its likelihood of occupying an agent at the future time period. The different statuses of a call will be described below with respect to FIG. 2, and how the statuses factor into the estimation of future agent occupancy will be described more fully below with respect to FIG. 3.

[0029] The description thus far describes the application server 101 handling a single campaign. The application server 101, however, may also be implemented to handle multiple outbound campaigns that have different sets of requirements (e.g., the outbound campaigns could have different target abandonment rates, target agent occupancies, required agent skill sets, target customers, etc.). In such a multiple outbound campaign environment, the application server 101 may employ multiple campaign managers 112 and multiple campaign list managers to manage the multiple outbound campaigns and the associated contact records.

[0030] In addition to outbound calls, the application server 101 may also process inbound calls. When a customer 104 calls the contact center, the call travels through the PSTN 103 and reaches the telephone server 102. The telephone server 102 transfers the call to an Interactive Voice Response system (IVR) 109, where a determination is made as to whether the customer 104 is to speak with an agent 106. If the customer 104 is to speak with an agent, the IVR 109 routes the inbound call to the automated call distributor 110, where the call would be distributed to an available agent with the appropriate skills for handling the call.

[0031] It should be understood that the system blocks or components described herein may be programs that are implemented by software instructions executing on one or more processing devices. In addition, the software blocks may each be implemented in a single, integrated software application, or in separate software applications.

[0032] FIG. 2 illustrates the lifecycle and statuses of outbound calls in the application server 101. When a call is dialed by the dialer 112, a dialing event 206 causes a call software object 205 representing the dialed call to be generated and stored in a call container 201. While the dialed call is waiting for a response, it is considered as having a status of being dialed 202 and may be referred to as “being in the dialer.” A number of results may cause termination 207 of the call at this time: the called number may be busy, the call may not be answered within a predefined time interval, the call may be answered by an answering machine or a fax machine, or a dialing error may have occurred. However, if the call is answered by a live person, the dialing is considered a success and the call's software object 205 would persist in the call container 201. The metric typically used for successful calls is called a hit ratio, which is defined as a fraction of successful calls to all generated calls.

[0033] A call that successfully connects to a customer then waits to be connected to an available agent. While waiting, the call is considered as having a status of being queued 203 and may be referred to as “being in the queue.” If at the time the call enters the queue 203 there is an appropriate agent available to the call, the call is immediately connected to the agent (i.e., the call spends zero time in the queue). However, if no appropriate agent is immediately available, the call is queued until an appropriate agent becomes available to it. While the call is being queued, the customer, not wanting to wait any longer, may disconnect the call. The system itself may also drop the connection to satisfy regulations relating to call waiting time. In either case the call is considered to be abandoned 208. The metric for measuring abandoned calls is called abandonment rate, which is defined as a percentage of abandoned calls to all successful calls that entered the queue.

[0034] When a call is connected to an agent, a conversation between the agent and the customer may ensue and last for a period of time, which is usually a stochastic value. During the conversation, the agent is considered as being occupied and therefore unavailable to handle other calls. Moreover, even when the conversation terminates, the agent may continue to be occupied for a period of time to process wrap-up work. After the agent completes any wrap-up work, the agent may then be considered unoccupied and thus available to handle new calls. From the time an agent becomes occupied by a call to the time the agent becomes unoccupied, the call is considered as having a status of being handled 204 and may be referred to as “being in the agent pool.” The duration of a call being in the agent pool may be referred to as agent handling time. It is when the agent becomes unoccupied after handling the call that the call is considered to be terminated or completed 209, at which time the call exists the system.

[0035] With the introduction of the concept of abandonment rate, objectives of the automatic dialing system may be described as optimizing agent occupancy and minimizing abandonment rate. These are contradictory goals, however, because when agents work close to the maximum occupancy (i.e., near 100%), waiting time for new calls may increase, which in turn may cause more calls to become abandoned. On the other hand, if a low volume of calls are generated to reduce abandonment rate, the number of calls may not keep the agents sufficiently occupied. Therefore, in practice a more practical objective is to generate an optimal number of outbound calls that would maintain agent occupancy within some predefined boundaries (e.g., between 80% and 95%) while keeping the abandonment rate less than some predefined value (e.g., less than 3%). The dialing methods described below are designed to help the automatic dialing system achieve this practical objective.

[0036] FIG. 3 is a flow diagram of an example predictive dialing method, which determines an optimal number of outbound calls to generate based upon an estimation of future agent occupancy. The predictive dialing method can be implemented by, for example, the dialing call calculator 116. As apparent from the above description with respect to FIG. 2, every call in the system has the potential of occupying an agent at a future time: calls in the dialer and the queue may enter the agent pool, and calls in the agent pool may remain in the agent pool. Thus, to estimate future agent occupancy, the predictive dialing method estimates the likely impact that each call currently in the dialer, queue, and agent pool will have on agent occupancy at the future time period. These impacts, respectively, will be termed dialer impact, denoted by N_D, queue impact, denoted by N_Q, and agent pool impact, denoted by N_A. Collectively, they will be termed call impact, which can be denoted and represented by the equation: N=N_D+N_Q+N_A.

[0037] The future time period of interest is the estimated moment when a currently generated call enters the agent pool. This is the time period of interest because agent occupancy at that time period will determine whether agents are available to handle the currently generated call. Therefore, if t₀ denotes the current time when the predictive dialing method is invoked and D denotes the average time for a generated call to enter the agent pool, the future time of interest can be denoted by t=t₀+D. Using this definition, the predictive dialing method's predictive objective can be characterized as predicting agent occupancy at time t=t₀+D, based upon all calls that are in the system at time t₀.

[0038] At 301 of FIG. 3's flow diagram, the predictive dialing method determines system data that will be used for estimating call impacts. This includes the average time D for a call to reach the agent pool, hit ratio h, and agent handling times. The data may be predetermined or dynamically generated based upon empirical data collected from the system or other systems.

[0039] At 302, 303, and 304, the predictive dialing method calculates the likelihood of each call in the system (i.e., each call in the dialer, the queue, and the agent pool) at time t₀ occupying an agent at time t. Although the example method depicted by FIG. 3 calculates the impacts in the order of dialer impact 302, queue impact 303, and agent pool impact 304, the impacts may be calculated in any order, including being calculated concurrently or in parallel. Furthermore, when calculating the dialer impact 302, queue impact 303, and agent pool impact 304, the impact contribution of each call in the system may also be calculated in any order, including being calculated concurrently or in parallel with each other.

[0040] At 302, the dialer impact is determined based upon the probability of each call in the dialer at time t₀ occupying an agent at time t=t₀+D. Let C_D denote all calls in the dialer at time t₀, let c_jεC_D denote a call in the dialer, and let σ_j≦t₀ denote the moment at which call c_j entered the dialer. The duration that call c_j has spent in the dialer by time t₀ can be expressed as t₀−σ_j. Since the average duration for a call to reach an agent is D and call c_j has already spent t₀−σ_j time units in the dialer, call c_j is expected to reach the agent pool after D−(t₀−σ_j) time units, or at time t₀+D−(t₀−σ_j). Thus, in order for c_j to remain in the agent pool at the future time t=t₀+D, c_j would need to stay in the agent pool for an additional duration of at least t−(t₀+D−(t₀−σ_j))=(t₀+D)−(t₀+D−(t₀−σ_j))=t₀−σ_j. The probability of call c_j being in the agent pool at time t can be estimated by the probability of a call in the agent pool taking more than t₀−σ_j time units to complete, and can be expressed as: P_j^D=1−F_A(t₀−σ_j), where P_j^D denotes the probability of call c_j remaining in the agent pool (i.e., not completing at time t) and F_A(x) is the probability distribution function of agent handling time. However, in cases where call c_j has been in the dialer for more than D time units (i.e., t₀−σ_j>D), the call is considered as entering the agent pool at time t₀. As such, the probability of call c_j remaining in the agent pool at time t can be estimated by the probability of a call in the agent pool taking more than t−t₀=D time units to complete, and can be expressed as: P_j^D=1−F_A(D). Therefore, the probability of being in the agent pool at time t can be represented by the following equation:

[0041] $\begin{array}{cc}{P}_j^D=\{\begin{array}{cc}1-F_A\left(t_0-{\sigma }_j\right),& \mathrm{if}{t}_0-{\sigma }_j\le D\\ 1-F_A\left(D\right),& \mathrm{otherwise}\end{array}& \left(1\right)\end{array}$
Calculating P_j^D for each call in the dialer at time t₀ and summing the results yields the dialer impact. However, since only some of the calls in the dialer will successfully connect to customers, the dialer impact is adjusted by the hit ratio. Thus, the dialer impact can be computed using the following equation:

[0042] $\begin{array}{cc}{N}_D=h·\sum _{c_j\in C_D\left(t_0\right)}{P}_j^D& \left(2\right)\end{array}$

[0043] At 303, the method calculates the queue impact N_Q. Calls in the queue are considered to be entering the agent pool at time t₀. Thus, if a call is to remain in the agent pool by time t, the call would have to stay in the agent pool for at least t−t₀=D time units. As such, the probability of a call in the queue at time t₀ being in the agent pool at time t can be estimated by the probability of a call in the agent pool taking more than D time units to complete, which is expressed as: 1−F_A(D). Applying the 1−F_A(D) probability calculation to each call in the queue at time t₀ and aggregating the results yields the queue impact N_Q. Thus, the queue impact can be computed using the following equation:
N_Q=|Qc|·(1−F_A(D))  (3)
where |Qc| represents the number of calls in the queue at time t₀.

[0044] At 304, the method calculates the agent pool impact N_A. The calculation of agent pool impact N_A includes calculating the probability of a call in the agent pool at time t₀ remaining in the agent pool at time t, while taking into consideration that the call has already been in the agent pool for some duration. To express the calculation in mathematical notation, let C_A denote all calls in the agent pool at time t₀, let c_iεC_A denote a call in the agent pool, and let τ_i≦t₀ denote the moment at which call c_i entered the agent pool. The duration that call c_i has already spent in the agent pool by time t₀ can be expressed as t₀−τ_i. Thus, if call c_i is to remain in the agent pool at time t, which is D time units later from the current time t₀, then the total duration that call c_j would have spent in the agent pool at time t can be expressed as t₀−τ_i+D.

[0045] With the above definitions, the probability of a call in the agent pool at time t₀ remaining in the agent pool at time t can be estimated by the probability of a call not completing within t₀−τ_i+D duration, given that the call has not completed in t₀−τ_i duration. This probability can be expressed in standard conditional probability notation, P(A|B)=P(A∩B)/P(B), as: P(call not completing within t₀−τ_i+D duration|call has not completed within t₀−τ_i duration). Given the definition of event A (i.e., call not completing within t₀−τ_i+D duration) and event B (i.e., call has not completed within t₀−τ_i duration), the union of A∩B can be simplified to be event A. This is because calls that are not completed within t₀−τ_i+D duration (i.e., event A) necessarily are not completed within the shorter t₀−τ_i duration (i.e., event B). Thus, the conditional probability of a call being in the agent pool at time t can be simplified as: P(A|B)=P(A)/P(B).

[0046] Probability calculations again uses the probability distribution function of agent handling time, F_A(x). Since F_A(t₀−τ_i+D) is the probability of a call completing within t₀−τ_i+D duration, the probability of a call not completing within t₀−τ_i+D duration (i.e., P(A)) can be expressed as 1−F_A(t₀−τ_i+D). Similarly, P(B) can be expressed as 1−F_A(t₀−τ_i). Thus, the probability of a call c_i not completing at time t can be calculated using the following formula:

[0047] $\begin{array}{cc}{P}_i^A=P\left(A|B\right)=\frac{P\left(A\right)}{P\left(B\right)}=\frac{1-F_A\left(t_0-{\tau }_i+D\right)}{1-F_A\left(t_0-{\tau }_i\right)}& \left(4\right)\end{array}$
Aggregating the P_i^A value of each call in the agent pool at time t₀ yields the following equation for the agent pool impact:

[0048] $\begin{array}{cc}{N}_A=\sum _{c_i\in C_A\left(t_0\right)}{P}_i^A& \left(5\right)\end{array}$

[0049] Having determined the dialer impact N_D, queue impact N_Q, and agent pool impact N_A, the predictive dialing method at 305 determines the total call impact N. As described above, call impact N represents an estimated total number of calls that would be in the agent pool at time t, and is defined as:
N=N_D+N_Q+N_A  (6)

[0050] At 306, the call impact N is compared to the total number of agents in the agent pool, denoted by m, to determine whether at time t there would likely be any unoccupied agents. If N is equal to or greater than m, that means that at time t the agents are likely to be at full capacity handling calls and would not have capacity to handle additional calls. Thus, the predictive dialing method at 308 returns n_predictive=0 to indicate that no new calls should be generated. On the other hand, if N is less than m, then it indicates that at time t it is likely that some agents would be available to handle calls. In particular, the estimated number of unoccupied agents at time t is m−N. However, more than m−N calls may be generated, since a portion of the generated calls would fail to connect to customers. Thus, taking hit ratio h into consideration, the predictive dialing method at 307 may return an estimated number of calls to generated based upon the equation:

[0051] $\begin{array}{cc}{n}_{\mathrm{predictive}}=\lfloor \frac{m-N}{h}\rfloor & \left(7\right)\end{array}$
Combining the two scenarios indicated by 307 and 308, the optimal number of calls to generate can be expressed as:

[0052] $\begin{array}{cc}{n}_{\mathrm{predictive}}=\{\begin{array}{c}\lfloor \frac{m-N}{h}\rfloor ,\mathrm{if}m>N\\ 0,\mathrm{otherwise}\end{array}& \left(8\right)\end{array}$
A predictive dialing method using equation 8 would likely maintain an average agent occupancy of 100%. Alternatively, to maintain an average agent occupancy of u_mean, equation 8 can be rewritten as:

[0053] $\begin{array}{cc}{n}_{\mathrm{predictive}}=\{\begin{array}{c}\lfloor \frac{u_{\mathrm{mean}}·m-N}{h}\rfloor ,\mathrm{if}{u}_{\mathrm{mean}}·m>N\\ 0,\mathrm{otherwise}\end{array}& \left(9\right)\end{array}$

[0054] The above equation for determining the optimal number of calls to generate may be modified to account for fluctuations in abandonment rate. For example, a clear indication that a system's abandonment rate should be reduced is when the current system abandonment rate R has reached or exceeded a predetermined allowable maximum abandonment rate, denoted by MAR. To actively reduce the abandonment rate, rather than using the predictive dialing method described above, a progressive dialing method may be used. The progressive dialing method is appropriate because it assumes the worst-case scenario, which is that each call in the system at time t₀ will be in the agent pool at time t. In other words, the method does not take into consideration that calls in the system at time t₀ may terminate or exit the system before the future time t. Thus, under the progressive dialing method the call impact is represented by the equation N₀=|Dc|+|Qc|+|Ac|, where N₀ denotes the call impact, and |Dc|, |Qc|, and |Ac| denote the number of calls in the dialer, the queue, and the agent pool, respectively, at time t₀. Alternatively, the equation for N₀ could also include the a ratio h to account for dialer calls that fail to reach customers, thus making the equation: N₀=h·|Dc|+|Qc|+|Ac|.

[0055] Once the call impact N₀ is determined, it is compared to m, the total number of agents in the agent pool, to determine the number of calls to generate. Thus, the equation for determining the number of calls to generate under the progressive dialing method can be expressed as:

[0056] $\begin{array}{cc}{n}_{\mathrm{progressive}}=\{\begin{array}{c}\lfloor \frac{m-N_0}{h}\rfloor ,\mathrm{if}m>N_0\\ 0,\mathrm{otherwise}\end{array}& \left(10\right)\end{array}$

[0057] As indicated above, the progressive dialing method may be used when R≧MAR; but when R<MAR, the predictive dialing method described above is used. If a more aggressive approach to reducing abandonment rate is desirable, the number of calls calculated by the predictive dialing method (i.e., n_predictive) may be reduced based upon the abandonment rate R. An example of a corrected predictive dialing equation may be expressed as follows:

[0058] $\begin{array}{cc}{n}_{\mathrm{corrected}}=\{\begin{array}{c}{n}_{\mathrm{predictive}}-\frac{n_{\mathrm{predictive}}-n_{\mathrm{progressive}}}{\mathrm{MAR}}·R,\mathrm{if}R<\mathrm{MAR}\\ n_{\mathrm{progressive}},\mathrm{otherwise}\end{array}& \left(11\right)\end{array}$
According equation 11, when the current abandonment rate R is zero, the predictive dialing method (i.e., n_predictive) will determine the number of calls to generate. If, however, the current abandonment rate R is more than zero but less than MAR, n_predictive is reduced based upon the current abandonment rate R. Finally, if the current abandonment rate R is equal to or greater than MAR, then the number of calls to generate will be determined by the progressive dialing method (i.e., n_progressive).

[0059] Whichever dialing method is used, the outcome is returned to the campaign manager 112, and a corresponding number of calls is generated.

[0060] When calculating the dialer impact, the above described example of a predictive dialing method assumes that a call stays in the dialer for a deterministic time. However, because events that causes calls to exist the dialer (e.g., customers answering calls) are random, dialing time in practice is a stochastic value. Therefore, to account for the stochastic nature of dialing time, an alternative method for calculating dialer impact may be used.

[0061] FIG. 4 is a flow diagram of an example method for calculating dialer impact while accounting for the stochastic nature of dialing time. For practicality and ease of implementation, the method uses discrete empirical data. At 401, discrete empirical data is collected on in-dialer duration of successful calls, in-dialer duration of all calls, and agent handling time. The data is used to construct probability distribution and density functions, which may be represented and stored as arrays of length N. For example, P_S=[ps₁, ps₂, . . . , ps_N] represents a probability density function of in-dialer time for successful calls; F_D=[fd₁, fd₂, . . . , fd_N] represents a probability distribution function of in-dialer time for all calls; and F_A=[fa₁, fa₂, . . . , fa_N] represents a probability distribution function of agent handling time. This procedure may be repeated periodically to ensure relevance of the data.

[0062] When the predictive dialing method is invoked, block 402 calculates the individual impact that each call in the dialer at time t₀ may have on the agent pool. The calculations may be made using the following or equivalent equation:

[0063] $\begin{array}{cc}p=\sum _{i=B}^{E+1}\frac{{\mathrm{ps}}_i}{1-{\mathrm{fd}}_i}·\left(1-{\mathrm{fa}}_{E-i}\right)& \left(12\right)\end{array}$
where B≧0 is an integer corresponding to t₀, and E corresponds to t.

[0064] Once the individual impacts of each call is calculated, the method, at 403, calculates the dialer impact N_D by summing the individual impacts. The dialer impact N_D is then used by the predictive dialing method (e.g., equations 6 and 9) to calculate the number of calls to generate.

[0065] As indicated above, a predictive dialing method that uses equation 9 is in effect maintaining the system's average agent occupancy to be at a predetermined value u_mean. Due to the direct relationship between agent occupancy and abandonment rate, maintaining the average agent occupancy at a certain level means that the abandonment rate is also roughly maintained at an associated level. Therefore, in a contact center where a maximum abandonment rate is specified, it may be desirable to determine an optimal value for u_mean such that the abandonment rate associated with the u_mean value does not exceed the maximum abandonment rate.

[0066] FIG. 5 is a flow diagram of an example of a method for determining an optimal average occupancy while keeping abandonment rate within a predetermined boundary. The first stage of the method involves collecting agent occupancy data while the system is in a “safe mode” where abandonment rate is zero or close to zero. At 501, the method sets the system in a “safe mode” regime by setting the mean occupancy u_mean of the predictive dialing method to: T_AHT/(T_AHT+T_D), where T_AHT is an average agent handling time and T_D is an average dialing time (both of which may be derived from empirical data). While the system is operating in “safe mode,” empirical data is collect at 502 on agent occupancy and the data is used to generate a discrete density function of agent occupancy. An example of the agent occupancy density function is illustrated in FIG. 6 as curve 601.

[0067] In the next stage, the objective is to determine an adjustment value, denoted by Δ, for the mean agent occupancy u_mean while keeping abandonment rate within a target maximum abandonment rate. At 503, the method determines the target maximum abandonment rate, denoted by A_max. The target abandonment rate A_max is then used at 504 to determine a target agent occupancy, denoted by u_target, such that a proportion of agent occupancy empirical data having agent occupancy equal to or greater than the u_target is equal to or within a predetermined range of A_max. As illustrated in FIG. 6, determining u_target(604) is equivalent to finding an agent occupancy 604 where the area 603 under the curve 601 that is to the right of the agent occupancy 604 is equal or negligibly close to the target abandonment rate A_max.

[0068] With reference again to the flow diagram of FIG. 5, once u_target is determined, the method at 505 calculates the adjustment value Δ, which is the difference between 100% occupancy and u_target (i.e., Δ=100%−u_target). In FIG. 6, Δ is represented by label 602.

[0069] At 505, the adjustment value Δ is used to adjust the original mean agent occupancy used by the predictive dialing method. In particular, the adjusted mean occupancy, u_mean′, may be calculated by using the following formula: u_mean′=u_mean+Δ. FIG. 7 illustrates the effect of the adjustment. Changing the predictive dialing method's mean occupancy from u_mean (705) to u_mean+Δ (715) effectively shifts the agent occupancy density function curve of the predictive dialing method from the left curve 701, which is associated with u_mean (705), to the right curve 711, which is associated with u_mean+Δ (715), by a distance of Δ (702). Consequently, the left area 703, which is associated with the left curve 701, is shifted outside of the 100% occupancy threshold to be in the right area 713, which is associated with the right curve 711. Since the right area 713 is above 100% occupancy, it represents a likely abandonment rate when the predictive dialing method uses u_mean+Δ as the mean agent occupancy. The likely abandonment rate is expected to be within the target abandonment rate A_max because the right area 713 is expected to be the same size as that of the left area 703, and that the left area 703 was determined based upon the condition that it be within or near the target abandonment rate A_max. The result of the foregoing is that the average agent occupancy maintained by the predictive dialing method is optimized (i.e., from u_mean to u_mean+Δ) while keeping the abandonment rate within the target maximum value A_max.

[0070] The above example method for determining optimal agent occupancy implicitly assumes that calls arriving to busy agents are abandoned. In other words, the method assumes that calls are abandoned if they arrive at an agent pool with 100% occupancy. However, depending on the system, calls may stay in a queue for some time (e.g., 2 seconds) before being removed and thus impacting the abandonment rate. Therefore, the adjustment value for u_mean may be higher than Δ because of the additional calls buffered by the queue. The additional calls that may be queued before being removed by the system can be represented by μT, where μ=1/T_AHT is a rate of call handling and T is the maximum time that a call can stay in the queue before being removed by the system. Therefore, an alternative expression for the adjusted optimal agent occupancy u_mean′ may be expressed by the following:
u_mean′=u_mean+Δ+μT  (13)

[0071] A predictive dialing method using u_mean′ is expected to maintain the system's average agent occupancy at u_mean′ while keeping the abandonment rate within A_max. However, in practice unexpected events (e.g., an agent or a group of agents disconnecting from the system) may cause the abandonment rate to jump to an abnormal abandonment rate A₀ that significantly exceeds A_max.

[0072] FIG. 8 is a chart illustrating what may happen to the abandonment rate over time once it jumps to A₀ (810). The top curve 801 represents the abandonment rate returning to its normal level A_ok (811) over time if the predictive dialing method is not adjusted to correct the abnormal abandonment rate A₀ (810). Although eventually the abandonment rate would return to normal levels, the time required may not be acceptable.

[0073] To have the abandonment rate return to normal within a shorter time, such as by time t_k (813), the mean occupancy used by the predictive dialing method may be set to a level that is less than its normal level. The effect of lowering the mean occupancy is that the system would have a corresponding lower expected abandonment rate, denoted by A_k (812), where 0≦A_k≦A_ok. The benefit of having a lower expected abandonment rate is that the system's abandonment rate would decrease at a faster rate to reach A_k, as illustrated by the bottom curve 802. In the process of reaching A_k, the bottom curve 802 may cross the normal abandonment rate A_ok at time t_k. Once the abandonment rate returns to normal, the predictive dialing method's mean occupancy may set back to the normal level.

[0074] FIG. 9 is a flow diagram of an example method for correcting abandonment rate. At 901, the method collects empirical data on agent occupancy and abandonment rate during normal system operations. The collected empirical data is used to determine the normal mean occupancy u at 902 and the normal abandonment rate A_ok at 903. At 904, the method monitors the current abandonment rate. At 905, if the current abandonment does not exceed a threshold, the method continues to monitor the current abandonment rate at 904. However, if at 905 the method determines that the current abandonment rate A₀ exceeds the threshold, then the method calculates a correcting abandonment rate A_k at 906, using the following equation:

[0075] $\begin{array}{cc}{A}_k=\frac{A_{\mathrm{ok}}·m·\mu ·u·t_k+A_{\mathrm{ok}}{N}_0-A_0{N}_0}{m·\mu ·u·t_k}& \left(14\right)\end{array}$
where N₀ is a number representing all calls generated before the current moment, m is the total number of agents, μ=1/T_AHT is the system's average call handling rate, and t_k is a desired duration for the abandonment rate to return to normal.

[0076] At 907, the correcting abandonment rate A_k is used to determine an associated correcting mean occupancy u_k. This may be accomplished by using the method described above for calculating mean occupancy (i.e., the method described with respect to FIG. 5), using A_k as the target abandonment rate. At 908, u_k is set as the predictive dialing method's mean occupancy to begin the expedited process for correcting abandonment rate. During the correction process, the method at 909 continues to monitor the current abandonment rate. At 910, if the current abandonment rate remains outside a predetermined range of the normal abandonment rate A_ok, the method returns to 900 and continues to monitor the current abandonment rate. However, if the current abandonment rate is within the predetermined range of the normal abandonment rate A_ok, then the objective of correcting the abandonment rate is achieved. Having corrected the abandonment rate, at 911 the method switches the mean occupancy of the predictive dialing method back to the normal mean occupancy u, thereby returning the system to its normal regime.

[0077] The example predictive dialing method described so far is with respect to a single outbound campaign. Nevertheless, the method is scalable to handle blended (e.g., inbound and outbound campaigns) and multiple campaigns. These more complex operating environments, however, introduce additional challenges to the predictive dialing method. For example, in a blended and multi-campaign environment, agents may process inbound calls or other types of activities where arrival rate cannot be controlled, and agents may participate in more than one campaign. Furthermore, where several outbound campaigns compete for the same agents or resources, certain campaigns may have priority over others. For example, if two outbound campaigns are serviced by the same group of agents, a business requirement may specify that one outbound campaign is to have twice the agent workload or the number of generated calls as the other. A predictive dialing method in a blended and multi-campaign environment would need to take these additional considerations into account when determining the number of calls to generate.

[0078] FIG. 10 is a flow diagram of an example method for determining the number of calls to generate for an outbound campaign in a blended and multi-campaign environment. At 1001, empirical data is collected on system operations for each campaign, such as agent handling time, hit ratio, average dialing time, and average answering time. This empirical data will be used in the calculations described below.

[0079] At 1002, the method estimates a number of agents that should be reserved for non-outbound activities, such as inbound calls and auto-dial calls. A distinction between non-outbound activities and outbound activities is that the flow of non-outbound activities is outside of the control of the virtual contact center, and that non-outbound activities typically take higher precedence than outbound activities.

[0080] Non-outbound activities are characterized by activity/call arrival rate, denoted by 2, and average agent handling time, denoted by T_AHT. The product λ·T_AHT is referred to as traffic offer, which characterizes non-outbound workload. Using these definitions, the estimated number of agents needed, denoted as mⁱ_in, to process activities/calls of an non-outbound campaign i can be expressed by the formula:

[0081] $\begin{array}{cc}{m}_{\mathrm{in}}^i=\frac{{\lambda }^i·T_{\mathrm{AHT}}^i}{u^i}& \left(15\right)\end{array}$
where λⁱ, Tⁱ_AHT, and uⁱ are respectively the arrival rate, average agent handling time, and agent occupancy of the non-outbound campaign i. If the number of non-outbound campaigns handled by the system is denoted by I, the total number of estimated agents needed for processing all non-outbound activities can be expressed by:

[0082] $\begin{array}{cc}\sum _{i=1}^I{m}_{\mathrm{in}}^i=\sum _{i=1}^I\frac{{\lambda }^i·T_{\mathrm{AHT}}^i}{u^i}& \left(16\right)\end{array}$

[0083] Once the estimated number of agents needed for non-outbound campaigns is determined, the method, at 1003, determines an estimated number of agents available to outbound campaigns. The general idea is that any agent in excess of the number of agents needed for non-outbound campaigns is considered as being available to outbound campaigns, unless the agent is only able to handle non-outbound activities/calls. More specifically, let M denote the total number of agents and let Mⁱⁿ denote the total number of agents that only handle inbound activities. The number of agents available to outbound campaigns, denoted as M^out, may be expressed as:

[0084] $\begin{array}{cc}{M}^{\mathrm{out}}=M-\max \left(\sum _{i=1}^I\frac{{\lambda }^i·T_{\mathrm{AHT}}^i}{u^i},M^{\mathrm{in}}\right)& \left(17\right)\end{array}$

[0085] At 1004, the estimated number of agents available to the outbound campaigns (i.e., M^out) is then partitioned among the outbound campaigns. The partitioning may be performed based upon a predetermined ratio, which may be defined in terms of agent workload or generated calls. If N denotes the number of outbound campaigns, let R^w={r₁^w, r₂^w, . . . , r_N^w} denote a workload ratio, let R^c={r₁^c, r₂^c, . . . , r_N^c} denote a generated call ratio, and let T={T₁, T₂, . . . , T_N} denote average agent handling times, where r_j^c, r_j^w, and T_j correspond to an outbound campaign j. Thus, the effective number of agents available to an outbound campaign j can be determined by the following formula if workload ratio is used:

[0086] $\begin{array}{cc}{m}_j^w=M^{\mathrm{out}}·\frac{r_j^w}{\sum _{i=1}^N{r}_i^w},0<j\le N& \left(18\right)\end{array}$
Similarly, the effective number of agents available to the outbound campaign j can be determined by the following formula if generated call ratio is used:

[0087] $\begin{array}{cc}{m}_j^c=M^{\mathrm{out}}·\frac{T_j/r_j^c}{\sum _{i=1}^N{T}_i/r_i^c},0<j\le N,r_j^c\ne 0& \left(19\right)\end{array}$

[0088] At 1005, different methods may be invoked for different outbound campaigns to determine an optimal number of calls to generate for each outbound campaign. One such method is a modified version of the above described predictive dialing method. The modified method differs in that the statistical and empirical data used in the calculations are gathered from and specific to the outbound campaign j. Furthermore, instead of using the total number of agents in the system (i.e., m) to assess agent availability, the effective number of agents available to the outbound campaign j is used (i.e., m^w_j or m^c_j). Also, when calculating call impact, the method may take into consideration only calls in the system that are associated with campaign j. For example, the calculation of campaign j's dialer impact N_D^j, queue impact N_Q^j, and agent pool impact N_A^j may only take into consideration calls in the dialer, queue, and agent pool, respectively, that require an agent skill set handled by the campaign j. Incorporating these differences into equation 10 of the predictive dialing method yields the following formula for determining the number of calls to generate for the campaign j:

[0089] $\begin{array}{cc}{n}_{\mathrm{predictive}}^j=\{\begin{array}{c}\lfloor \frac{m_j^w·u_{\max }-N^j}{h}\rfloor ,\mathrm{if}{m}_j^w·u_{\max }>N^j\\ 0,\mathrm{otherwise}\end{array}& \left(20\right)\end{array}$
where N^j=N_D^j+N_Q^j+N_A^j, u_max is the maximum agent occupancy for the campaign, and m_j^w may be replaced by m_j^c.

[0090] It should be understood that similar to the other processing flows described herein, one or more of the steps and the order in the flowchart may be altered, deleted, modified and/or augmented and still achieve the desired outcome.

[0091] This written description uses examples to disclose the invention, including the best mode, and also to enable a person skilled in the art to make and use the invention. The patentable scope of the invention may include other examples. Additionally, the methods and systems described herein may be implemented on many different types of processing devices by program code comprising program instructions that are executable by the device processing subsystem. The software program instructions may include source code, object code, machine code, or any other stored data that is operable to cause a processing system to perform the methods and operations described herein. Other implementations may also be used, however, such as firmware or even appropriately designed hardware configured to carry out the methods and systems described herein.

[0092] The systems' and methods' data (e.g., associations, mappings, data input, data output, intermediate data results, final data results, etc.) may be stored and implemented in one or more different types of computer-implemented data stores, such as different types of storage devices and programming constructs (e.g., RAM, ROM, Flash memory, flat files, databases, programming data structures, programming variables, IF-THEN (or similar type) statement constructs, etc.). It is noted that data structures describe formats for use in organizing and storing data in databases, programs, memory, or other computer-readable media for use by a computer program.

[0093] The computer components, software modules, functions, data stores and data structures described herein may be connected directly or indirectly to each other in order to allow the flow of data needed for their operations. It is also noted that a module or processor includes but is not limited to a unit of code that performs a software operation, and can be implemented for example as a subroutine unit of code, or as a software function unit of code, or as an object (as in an object-oriented paradigm), or as an applet, or in a computer script language, or as another type of computer code. The software components and/or functionality may be located on a single computer or distributed across multiple computers depending upon the situation at hand.

[0094] It should be understood that as used in the description herein and throughout the claims that follow, the meaning of “a,” “an,” and “the” includes plural reference unless the context clearly dictates otherwise. Also, as used in the description herein and throughout the claims that follow, the meaning of “in” includes “in” and “on” unless the context clearly dictates otherwise. Finally, as used in the description herein and throughout the claims that follow, the meanings of “and” and “or” include both the conjunctive and disjunctive and may be used interchangeably unless the context expressly dictates otherwise; the phrase “exclusive or” may be used to indicate situation where only the disjunctive meaning may apply.