Get Adobe Flash player

Anatomy of a predictive dialler

Making, connecting and managing calls

A significant number of telephony services use outbound and blended predictive dialler systems. For many corporations incorporating proactive contact and notification strategies in their operations, through companies employing interactive voice messaging (IVM) (also referred to as ‘outbound IVR’) and voice broadcasting, to businesses whose prime purpose is debt collection, outbound technologies are a key component.

Outbound and blended campaigns enable enterprises to proactively reach their customers and prospects. For many, this plays a valuable role in ensuring a superior service outcome for customers and ensures efficient and effective use of agent resources. For companies whose primary business is to fill sales pipelines, streamline collections or aid telemarketing and fundraising efforts, the telephony application and software utilised is directly associated with revenue generation.

Whether your application is a predictive dialler, IVM, voice broadcasting or contact centre solution, a key requirement will be the ability to screen out all calls other than live calls – collectively, a technology that makes that happen is known as call progress analysis (CPA).

Making the call

The first stage in automated outbound dialling involves proactively contacting customers, account holders and prospects.

An outbound dialler is typically a server that has been preconfigured to make calls automatically, under control of a software application, to a list of customers or prospects, as per campaign scripts. A fully integrated dialler is used to increase productivity by anticipating (or predicting) the number of calls that need to be generated to ensure that a live call can always be connected to an account representative (call centre agent).

When calling consumers at home it is not untypical for only one in five calls to be answered by a live call recipient. Diallers solve this problem using mathematical modelling or virtual simulation to determine the appropriate number of calls to make at any one time to maximise agent productivity. This relieves agents of the dialling function and provides them with a steady flow of live contacts only.

Dialler technology increases an account representative’s ability to speak to the right party and increases the chances of getting a payment, promise-to-pay or other appropriate result.

Call connection

The next stage relates to connecting contacts with agents. In an ideal world, only live speakers would be connected to agents. In reality, the dialler will hit no-answers, busy signals, wrong numbers, answer machines and disconnects. To manage such outcomes, predictive diallers often deploy CPA, which essentially employs a number of algorithms that listen on the line and attempt to determine what is happening and who or what answers the call. Such software is designed to distinguish between busy signals and other network conditions in addition to answering machines and live speakers.

Some modules will detect the frequency, noise gate and energy of the audio signal and attempt to distinguish whether the call is answered by a human or if e.g., a voicemail greeting is being presented on the line. Additional media processing functions are used to detect information from the telephone network – DTMF and call progress tones, such as ringing, busy/engaged, voicemail (in theory), special information tones (SIT or ‘triple tones’) and fax tones. Other software, associated with the protocol in use, provides complete cause code functionality.

A further condition is apparent with the popularity, in some countries, of caller ring back tones (CRBT). Calling a number whose owner has subscribed to a CRBT service typically results in the caller (or dialler system) hearing a music tone instead of ringing.


A note of caution

The ultimate test is how quickly and accurately this analysis can differentiate between the various tones during call set-up, and a fax, modem, human “Hello” or an answering machine greeting post-connect, before taking the final step. 

As such, providers should be cautious when deploying such technology. Indeed, for certain campaigns it may be prudent for certain elements, such as live speaker detection, to be ‘turned off’ to protect the quality of the call, as they cannot be 100% accurate. In other words, the trade off between preventing the called party hearing silence or the call being abandoned, because it was thought to have been answered by a machine (the so called ‘false positive’), may be greater than the efficiency needs of the call centre.


Call management

Finally, the predictive dialler needs to act on the results of CPA. For example, if a network condition based on a triple tone has been determined, it will promptly abort the action and immediately attempt another call. If a call is answered by a human, it needs to be transferred to an available agent or, in the case of IVM, connected to some form of automated message playback. Note that strict rules about the nature of any automated message playback are enforced by the appropriate regulatory authorities.

Voicemail strategies – If CPA determines that the call was answered by voicemail, a pre-recorded or text-to-speech (TTS) synthesised message can be played, although this is not required by legislation and, typically, is not done. Such messages can be to inform the customer of the nature of the call or to provide the contact with a number to call, which might be considered good practice. After all, not leaving a voice message when a call is answered by an answering machine can be a cause of anxiety to the called party just as much as a silent call.

On the other hand, if the call is likely to be abandoned – there is no agent available to take a live answered call – a brief information message giving details about the call should be played. Note that such messages and other steps to reduce the impact of abandoned calls in this situation are often a regulatory requirement. Note also, that if there is no agent available, a call is classed as abandoned, even if an information message is played.

Contact verification – Intermediate steps, such as contact verification can also be inserted. This enhances the call centre application by only providing verified, right-party contacts to an account representative. This is achieved by using interactive voice response (IVR) and speech technologies – TTS and automatic speech recognition (ASR). When the system is connected to a live speaker, the IVR uses a combination of pre-recorded messages and TTS to announce the contact’s name and ask for the account holder as a means of verification.

Apart from credit/debit card activation, emergency broadcasts, and medical and other appointment reminders, using IVM in this way can also be used to cost-effectively get answers to non-revenue generating questions or to set up payment arrangements with some account holders.


Regulatory comment

In some countries the use of outbound dialling for IVM has recently come under the scrutiny of regulatory bodies, where it was considered as possible misuse. In the United Kingdom, at least, Ofcom has concluded that the use of IVM does not intrinsically constitute misuse. Indeed, as used for non-marketing purposes, it has received little adverse customer reaction and can be of widespread public benefit.

That notwithstanding, it is interesting to compare the differing approaches in diverse market regions. In the United States, for example, outbound voice broadcast has been used to good effect in electioneering campaigns associated with party presidential candidature. Whereas, in the United Kingdom some years ago, a political party fell foul of The Privacy and Electronic Communication Regulations by conducting a similar campaign that was branded a nuisance and led to it being threatened with prosecution.


Important factors

Across all of the identified analysis steps, three very important factors, which involve both pre- and post-connect processing, impact on the effectiveness or success of outbound dialling campaigns: accuracy in detecting tones; live speaker detection accuracy; and call connection time. These will be the subject of a future blog.