Using Answerphone Detect systems in a voip auto dialler environment

Answerphone detect systems reduce wasted agent time by detecting and dispositioning voicemail and answerphones without involving the agents. This saves around 20-30 seconds of agent time which when extrapolated over a month equates to a large cost saving. The additional benefit is that the time saved can be used to increase sales.

Some traditional ISDN answerphone detect (AD) systems can detect voicemail and answerphones using the signalling channel or "D channel" to quickly use the call set up signalling to detect voicemail or answerphones. Others use a set of rules mainly implemented on specialized hardware running on digital signal processors (DSPs). The optimization of the performance of these rules often resembles “black art” to determine, as best as one can, the outcome of an outbound call attempt. There are some drawbacks to these systems such as:

  • Poor day 1 accuracy - Unreliable in varying network conditions (eg wireless, low volumes, international calls).
  • Slow response times causing problems with OFCOM.
  • Time consuming in terms of tuning and optimisation.
  • Reliance on ISDN for call termination which is 15 times more expensive than voip.

The two most important factors in the AD process are as follows:

  1. 1. Accuracy
    If the AD system determines that a call is answered by a human while it is in fact a voice-mail greeting, the agent receiving the call will spend an amount of time handling the call before they become available to answer another call. While this may be under a minute or so, it becomes a considerable loss of productivity when accumulated over time and in large volume call centers, especially if you consider that less than 30% of outbound call attempts actually reach a live person. This loss of productivity is compounded even further by a loss of sales opportunity when dealing with the sales of products and services.
  2. 2. Time to Transfer to Next Available Agent
    When it takes too much time connecting to the agents, there is an increased risk of not being able to deliver the service at all, whether the AD system is accurate or not. Customers may hang-up on the call or the contact center operator may be faced with stiff penalties for non-compliance of nuisance calls regulations (in most countries, this typically needs to be under 3 seconds from the time there is a live answer). Note that these elements are not mutually exclusive. For example, a contact center operation may “tune” the AD system for faster detection but may have a negative impact on overall productivity because of an increased error rate. Conversely, if the AD system is tuned for better accuracy, it may take too much time before an agent can take the call, which can result in the customer dropping the line.

As an comparison the voip AD system designed by the voip market leader in open standards AD uses a different approach to call progress analysis. Instead of a complex set of rules, it uses statistical models based on Neural Networks to represent the potential outcome of an outbound call attempt. The result is a virtual machine that “learns” the behaviors that represent the different conditions representing call progress events. This method provides results with far superior accuracy and flexibility compared to traditional approaches:

Improved Accuracy and Response Time

Benchmarks have shown that the statistical approach reduces error rates by 50%, compared to leading traditional products while reducing the call transfer timing by 40%.

More Resilience Against Various Network Conditions

This approach provides superior robustness against volume variations, background noises and other network conditions. This is highly important because applications are being now deployed in more heterogeneous conditions than ever before. Also, since contact centers are starting to use automated dialing for a much different set of applications (callback, automated notification, proactive campaigns), as opposed to a single application, the AD system must allow for per–call optimization of the calling parameters.

Streamlined Tuning Process

The statistical approach helps streamline the tuning process. Data can be gathered from a deployment and used to re-train the statistical models. Tuning data from all sites helps to improve the baseline performance over time.

Simple Optimization and Tuning

The process of tuning AD parameters for a given deployment adds significant value, as it ensures that the AD system will provide the best possible operating points, given the specific deployment conditions. The statistical approach takes the guesswork out of the equation. The tuning process consists of gathering data from the field and using this data to further train the statistical model of the AD system. This ensures a smooth and predictable outcome. Furthermore, as more and more tuning is performed at different sites, the fi eld data collected will also serve to improve the performance of the base models.


Some traditional based AD systems work very well using ISDN signalling to accurately disposition answerphone machines and voice. However the trend today is a fast move towards SIP-based voip auto diallers. These diallers use SIP trunks to terminate calls through the public switched telephone network (PSTN) and thus bypass the ISDN network to achieve massive cost savings. A SIP-based AD system is therefore a pre-requisite. Choosing the right system is a critical decision especially with the recent changes to OFCOM regulations regarding AD usage but SIP-based AD systems definitely operate as effectively, if not more effectively then traditional systems.

For more information on AD systems please contact Hostcomm on 0808 168 4400.

Trusted by

Pharm N Case StudyLogo New V Case StudyFirst Data Logo 2018 N Case StudyHelp Lin N Case StudyShelter Logo N Case StudyHome Logic Icynene Case StudyOrbit Logo Purple 2 Case StudyInspire Logo 2 Case Study


PCI DSS Certified, TPS Telephone Preference Service, ICOCSA Supplier Member, Cyber Essentials