What is a Dialer?
Contact centers perform a wide spectrum of services such as reservations, sales, customer service, and technical support.
There are two types of call processing in contact centers: inbound and outbound. Inbound calls are originated outside the call center by customers, and then reach the call center where they are routed to operators, called agents. Outbound calls, conversely, are originated inside the call center, reach customers, and then are processed by agents. A typical application of outbound calls is a campaign in which customers are called with the aid of specialized call center equipment such as dialers, call progress detectors, answering machine detectors, etc.
There are several approaches to outbound dialing such as manual, preview, progressive, predictive and power dialing.
Let’s see the differences among them:
- progressive dialing where calls are generated only when agents become available and the number of calls is equal to the number of available agents
- predictive dialing, where the number of generated calls is based on prediction of how many agents will be available at the time when calls pass the dialer and on the estimated success rate of reaching the called party)
- power dialing, whichdiffers from a progressive or predictive dialer for it utilizes a manually-configured calls-to-agent ratio
- preview dialing, where the dialer selects a customer record from a call list and proposes this call to an agent, who can accept it and start the call or refuse it.
Nowadays, the outbound services have changed dramatically. As a rule the calls have become more targeted and friendly, end users have the possibility to enlist in Do-Not-Call lists to prevent themselves from receiving unsolicited calls and there are regulations regarding the maximum time a called party is allowed to be kept waiting for an agent. For example, outbound campaigns are used as: reminders of appointments and payments in a friendly manner, notification on the status of orders, useful up-sell offers, etc. Moreover, targeted outbound calls play the role of proactive notifications and therefore can reduce the number of inbound calls.
This paper addresses the problem of calculating outgoing telephony calls for predictive and progressive dialing regimes. Roughly speaking, the problem is how to define when to launch another outbound call so that agents will have enough work and customers are not left waiting long in a queue for an available agent.
This task belongs to the area of queuing system theory in which there are two main approaches: analytical and simulation.
Statement of the Problem
In this section, we consider the model of a Campaign Manager that will be referred to hereafter as the System (see Figure 1). This model is represented in terms of queuing systems. It is comprised of the following elements: a call generator, a dialer, the Call Progress Detecting, a queue, and a set of agents. Typically agents may also process inbound calls, which should be taken into consideration. The problem is to define a strategy for call generation by the call generator in such a way that the following conditions are met:
- All agents should have enough work, i.e., agent utilization should reach at least some pre-specified rate, this rate will be referred to as busy factor.
- The rate of customers who have not waited for processing and have left the queue should not be higher than some specified, called the abandonment rate.
It has been shown that these two requirements might not always be satisfied simultaneously.
Therefore we will treat these requirements as two separate problems:
- The abandonment rate should be maintained at an admissible level, while maximizing the agent’s busy factor.
- The agents’ busy factor is maintained at an admissible level while the abandonment ratio is as low as possible.
The generator has access to all system parameters which include the total number of agents m in the System (note that the number may vary), arrival rate of inbound calls λi, average handling time Ts, and a hit rate p that is a probability of a dialed telephone call being answered by a live person.
Some Dialing Methods
Predictive Dialing
We assume the outbound call flow to be a Poisson process with parameter λ. The service time (duration of a conversation) in our model has a general distribution with parameter μ=1/Ts. We also suppose that all customers are impatient and will hang up if they have to wait because all agents are busy. Note that for outbound calls this is quite close to the reality. This results in zero queue length. Thus we have a M/G/m/m queuing system in terms of queuing theory. Let ρ=λ/μ denote a traffic offer. Then the Erlang-B formula that is known as Erlang loss formula gives us a probability of finding all agents busy. The main idea of the predictive method is to obtain a dialing rate λm which is used by the outbound dialer to generate an outbound call flow and reach one of the optimization goals, i.e. either abandonment rate or busy factor.
Progressive Dialing
Usually the progressive dialing method works for pure outbound scenario with no inbound traffic. It includes the following two main steps:
- Each time when the system changes its state (an agent switches the status from ready to busy or vice versa, or a new outbound call is generated) the preliminary number N of new outbound calls to be dialed is calculated by the formula:
N = NR – ND (5)
where NR – number of agents ready to take a call, ND – number of already dialed outbound calls which are still in the dialer in the status of call progress detection. - If (5) gives a positive value, dial N outbound calls. This method is one of simplest because it does not need sophisticated mathematical theory for its implementation. In some situations it looks rather conservative and gives relatively low agent occupancy. Most frequently it is used during a start-up phase of an outbound campaign in order to collect all necessary data to prepare dialing in more advanced modes.
Choosing an Efficient Method
The predictive method is considered to be more efficient than the progressive one. However, for small agent groups (5-20 agents) this may not be the case. Indeed, as we will show below that our progressive method gives a constant agent busy factor while the abandonment rate is zero. With the predictive method, if we try to decrease abandonment rate, the agent occupancy will be also decreased turning to zero. At this point, the agent busy factor in predictive dialing will be less than the agent busy factor in progressive dialing. Thus for certain number of agents and certain values of abandonment rate, the progressive method is better than the predictive one. This phenomenon could be explained as following: The progressive method takes into consideration the current state of agents while the predictive method does not. Obviously this state becomes relevant for a small number of agents. The abandonment rate for the progressive method is always equal to zero.
We can conclude that the busy factor in the progressive method does not depend on the number of agents. Here we also can see the conditions under which the progressive method – which as appears to be a very primitive one – might be more efficient than the predictive one. The progressive method has an advantage over the predictive method if the agent handling time is much longer than the time an outbound call is typically kept in the dialer under relatively high hit rate.