ITEXPO begins in:   New Coverage :  Asterisk  |  Fax Software  |  SIP Phones  |  Small Cells

IP Contact Center Technology:
Eliminating The Risks (Part IX)

By Eli Borodow of Telephony@Work and
Kevin Hayden, TELUS Communications Inc.

• A special editorial series sponsored by Telephony@Work•


Over the last eight months we've focused on what inbound contact centers need to know before selecting an IP-based contact center solution. This month we'll be focusing on outbound technologies, specifically predictive dialing, in the context of IP-based solutions.

The Blended Contact Center
Leveraging idle agents on inbound projects to work on targeted dialing campaigns increases productivity and maximizes efficiency. A blended contact center enables agents to be switched seamlessly between inbound and outbound call campaigns and Web communications; maximizing the efficient use of all available agent resources.

In a blended environment, predictive outbound calls are triggered automatically when inbound call volumes fall below pre-set thresholds; and outbound campaigns are slowed or halted when inbound call volumes rise above pre-set thresholds.

A fully blended predictive dialing solution must be integrated-by-design for efficient call blending with the company's inbound technologies. Deploying a blended contact center can be easy or difficult depending on whether the overall solution is integrated-by-design or patched together after-the-fact. Some inbound technology vendors have aggregated their technology suites via corporate acquisitions and their solutions will often be no more integrated than if you were to buy the inbound and outbound technology solutions from different vendors. Other inbound technology vendors will OEM and re-brand third-party predictive dialers ' which can lead to tremendous operational difficulties both from a reporting perspective and when outbound and inbound functions of the organization are performed with the same pool of agents.

Predictive Dialing Considerations
In order to minimize agent idle time in between calls on an outbound campaign, predictive dialers 'predict' when the next agent on a campaign will become available based on average call lengths for agents in the active dialing campaign, and then proactively initiate calls to maximize agent efficiency.

To better accomplish this objective, predictive dialers also predict how many outbound calls must be made in order to reach a live person; factoring in up-to-date statistical data from the dialer on the percentage of successful live connections versus the percentage of no answers, busy signals and answering machines/voice mail ' all of which must typically be filtered out by the dialer. These statistical patterns dynamically drive the dialer's 'pacing algorithm,' which in turn determines the volume of outbound calls that should be made to address anticipated agent availability.

The problem with pacing algorithms, of course, is that their predictions simply represent educated guesses based on up-to-date historical data, and those guesses will be more or less accurate based on the size and validity of the statistical sample that drives the pacing algorithm.

In predictive dialing there is a delicate balance and a core tension between minimizing dropped call rates and maximizing agent talk time. If the pacing algorithm is too aggressive, live prospects will receive calls without an agent available on the other end. At that point, the call would likely be 'dropped' (hung up on by the dialer) or abandoned (hung up on by the prospect) because the dialer had guessed wrong and no agent was actually available to be connected to the prospect. On the other hand, if the pacing algorithm is too conservative, agents will spend more time idle and waiting for calls as the dialer seeks out live prospects. The key to maximizing efficiency and best resolving this tension is to scale the dialer to maximum levels in order to provide the pacing algorithm with the most statistically valid data based on the largest possible statistical sample.

The Scalability Challenge
The challenge for many larger organizations is that traditional dialer solutions often don't scale beyond 300 voice channels. Since multiple calls are generally made to find live prospects, the number of seats that can be supported by these systems in the real world is generally much less than that. As a result, most large organizations must run multiple dialers, often in the same locations, to meet their high-volume needs.

The problem is that traditional dialers do not allow multiple dialing nodes to work together and share data for maximum efficiency. While many technology vendors can aggregate administration and reports across dialer nodes, they typically cannot enable multiple dialer nodes to feed common data to a shared pacing algorithm; i.e., one that can be shared across diverse dialer nodes. As a result, each of the dialer nodes must make its own 'predictions' based on the subset of the statistical data specific to the individual dialer node.

In other words, traditional dialers will not allow multiple nodes to be unified in order to run on a common statistical sample. As a result, all dialer nodes are unable to make predictions on the aggregate sampling of all the nodes, and each dialer is an island that is limited to making predictions on the basis of only its own limited data sample.

The Benefits Of A Network-Based Software Architecture
Given the prediction efficiencies of leveraging the largest possible statistical sample, the goal of larger organizations running predictive dialers at scale is to have a single decision-making matrix control all dialing nodes, with each dialing node delivering its results to that matrix in real-time so that dialing decisions can be made taking all data from all dialing nodes into account. That is the benefit of a network-based software architecture in the context of predictive dialing. Companies that run dialing campaigns at scale on multiple stand-alone dialers can realize tremendous efficiency and productivity gains by migrating to network-based dialer solutions.

Aggregating Sites
Given the above noted limitations, 'traditional' predictive dialers also obviously can't allow multiple centers to be unified with a shared pacing algorithm based on a shared statistical sample. Here the goal is to aggregate the data from all sites in real-time to provide a unified statistical sample that can drive dialer decisions at all sites if they are calling into a common area.

With solutions that are designed to extend outbound dialing capabilities to remote centers, even agents working from home are empowered with the same capabilities as local agents in the contact center in which the infrastructure is deployed.

Companies should consider multitenant solutions where some centers will require stand-alone campaigns, autonomous data security and local control over their own campaigns. Economies of scale driven by shared licenses, hardware and phone lines is another benefit of this multisite deployment approach. Prudent technology buyers should also avoid vendors that support multidialer integration but add tremendous cost to the linkage between those different dialers. Such linkages are a no-cost capability in next-generation solutions that are designed for purpose.

Working Most Efficiently Within Legally Mandated Limits
Since predictive dialing has increasingly been the subject of both federal and state legislation in the U.S., contact centers need to ensure that the solutions they deploy comply with those legal mandates. Beyond compliance with do-not-call list regulations, companies are challenged to stay within mandated dropped call rates. The challenge arises from the fact that predictive dialers must stop 'predicting' and revert to a one-to-one available agent-to-call ratio once the legally prescribed limit has been exceeded. The need for accuracy in predictions has therefore become increasingly important, since efficiency decreases dramatically once the prescribed dropped call rate has been exceeded.

Call Progress Detection
Call progress detection is another key to effective predictive dialing. The quality of the call progress detection can vary greatly from one vendor to another. Call progress detection algorithms identify answering machines/voice mail, busy signals, no answers, pagers, out-of-service tones, etc. and distinguish them from live voices on the other end of the call. Since effective filtering is a key determinant of agent productivity, ensuring that your call progress detection algorithm is of high quality must always be of paramount importance. Some solutions, particularly some newer IP-based solutions, offer immature call progress detection algorithms and, in some cases, cannot filter out all types of call progress events (such that you might filter out busy signals but not answering machines and/or voice mail systems, etc.). This not only can dramatically impair efficiency, it also limits your options on how to deal with call progress detection events.

Another key to maximizing ROI is to make sure your dialer allows you to define different default actions for different types of call progress detection events, for each individual campaign. For example, if the dialer detects an answering machine or voice mail, for some campaigns you wouldn't want to send the call to a live agent but would choose instead to leave a pre-recorded message on that person's answering machine or voice mail. In other circumstances you might want a live person to leave a customized message. In another campaign, you might choose to abandon the call and have the dialer try again after a specified time interval. Never sacrifice the ability to define the dialer's reaction to call progress detection events on a campaign-by-campaign basis.

Many dialers don't offer unified administration. As a result, separate modules are often required for provisioning, managing, monitoring and analyzing campaigns; with many more interfaces and points of integration required for an inbound/outbound blended environment. For most blended contact centers, a diversity of functional trade-offs are also required because calls are handed off between systems instead of being handled by the same system ' which in many cases also has a negative impact on reporting. A mature predictive dialer shouldn't require a database administrator for list management, delays to process call tables or direct manipulation within the database.

Predictive dialing offers organizations compelling efficiencies in managing outbound call campaigns. Next-generation solutions can increase that efficiency and deliver compelling ROI by enabling diverse dialing nodes to share common sampling data and a common pacing algorithm; even when those dialing nodes are resident in different locations. Of course, the accuracy of call progress detection and the manageability of the technology solutions are also important considerations.

As with other aspects of a comprehensive IP contact center strategy, prudent technology buyers need to go beyond kicking the tires and take a good look under the hood of any proposed solution. Understanding the differences between vendor software architectures provides the most reliable approach for ensuring that your near- and long-term needs will be met most effectively.

If you've missed any of our first eight columns on IP contact center technology, simply e-mail us and we'll be happy to send copies to you.

Eli Borodow is the CEO of Telephony@Work, the leading provider of adaptive, multitenant IP contact center technology for contact centers and service providers. He can be reached via e-mail at [email protected].

Kevin Hayden is the Director of Integrated Contact Center Solutions at TELUS Communications Inc., a tier-1 telecommunications carrier in Canada and the Canadian leader in hosted contact center services. He can be reached via e-mail at [email protected].

[Return To The Table Of Contents ]

| More