Contact Center Automation: A Well-Oiled, Intelligent Machine for Customer Service

By Iain McKay

Automation is a term that evokes images of heavy industry and machinery. So how does it apply to the contact center?  Well, it could perhaps be thought that the contact center operates like a clockwork system of cogs, shafts, and pulleys where every piece has a role to play. This makes the ideal situation the proverbial well-oiled machine where each component part, or cog, is well tuned for its role relative to its peers, taking input from one part of the system and passing it on to another after some transformation.

This clockwork machine analogy really only holds true for contact centers dealing with a high volume of single-issue calls, through one channel and with few exceptions. Realistically, today’s contact centers are not like that, and customer service is not like that. An enterprise’s inbound customer contacts come in a hard-to-model chaotic manner with peaks and troughs only loosely estimable according to temporal patterns, such as time-of-day, day-of-week, marketing campaigns, price rises, and system faults. Therefore, what may be a well-oiled machine one minute could be reduced to a badly performing, inefficient, bottleneck-laden system the next.

Automation Challenges: What makes contact centers so hard to automate? Certainly, the individual channels and processes of a contact center can be automated, and many already have been, but in a piecemeal manner. The trouble is that business processes for contact centers have been replicated in a variety of ways with subtle differences, with extra management overheads across different channels, and with no plan to glue the disparate channels together. Traditionally, there is no common process across channels, and each channel is largely unaware of contacts across the others.

Take self-service telephony, for example. This is often built as a silo application, designed with telephony in mind, managed by telephony specialists, and adhering to a rigid business process. Making changes to such a system is a complex and expensive undertaking, thanks to the proprietary scripting languages on interactive voice response (IVR) platforms. A caller to an IVR line might perform an identification process, carry out a transaction, or declare that they have some other need and then be routed to an agent in the contact center. This agent will likely have to identify the customer again and reestablish what they want to do. This is a well-documented source of customer frustration.

Bottlenecks: Alongside the traditional telephony channel, customers can contact companies via a Web interface, email, instant messaging, text-chat, and SMS (Short Message Service). Companies receiving inbound contacts from this array of channels have traditionally struggled to achieve a holistic, real-time view of the state of the contact center. They usually segment the agent cohort into teams to handle the separate channels. This results in bottlenecks across individual channels (a peak of inbound telephony calls, for example), while others are lying underutilized (such as agents awaiting text-chat). On the other hand, contact routing tends to be simple and single-channel focused.

Another problem with today’s contact center infrastructure is that agents cannot easily manage multiple simultaneous contacts. Taking one telephone call at a time makes for good customer service. However, there are other channels, especially those that require limited real-time interaction, such as email and text-chat, which may be combined.

In general, the contact center agent cohort is poorly modeled in terms of agent skill set, technical prowess, channel ability, experience, efficiency, or other measures of their prowess other than average handling time. This means that simple resource-allocation algorithms such as “round robin” do not try to pick the best agent for the task in hand, only the next available one, resulting in poorer customer service.

With all these complications, it’s understandable that automating the contact center is such a challenge. So how can contact centers overcome these complex and expensive problems?

Processes Are Key: At the core of the solution lies the contact center’s business processes. A majority of the processes in a contact center are, in essence, the same – the identification process, for example, is identical whether conducted over the phone, via email, or on the Web. If the same common processes can be executed across multiple channels, this reduces costs, channel silos, and management overheads. Having a common process across all channels also makes for a more useful experience from the customer’s point of view, breeding expectation and familiarity. A solution that has holistic knowledge of the various channels is able to provide a single view of the customer contact history–for example, allowing an agent to see that the customer used the Web interface, sent an email, got no response, and then started a text-chat session.

Another advantage of having an integrated process is that when phone calls handled by the IVR system are routed to agents, the agents can see the data entered or process attempted by the aborted IVR session.

Self-service telephony does work, and it can reuse the same business processes that drive other self-service channels, such as the Web. The key is to model the business processes and store them centrally so they can automatically be repurposed to whichever channel requires them. For example, any IVR platform with a VoiceXML interpreter can make use of the same underlying business process, such as “take credit card payment,” that appears on a website or the agent’s desktop.

Watch Your Grammars: Beyond consolidating business processes, there are other technologies that can help to automate the contact center. Natural language IVR, with speech recognition and synthesis, has matured and allows for far more natural user interaction, providing that there is adequate tooling for voice-channel and natural language specifics, most notably defining grammars that the speech-recognition engine should be listening for. There are systems that can automatically construct the grammars based on the customer’s position within a process, looking ahead and around for items in future or sibling processes. Tooling should also allow for the definition of the dialogue between the automated system and the caller: A customer is initially presented with more open, mixed-initiative prompts, drilling down to closed and directed system-initiative prompts.

The Virtual Agent: Thanks to these advances in natural language processing, there is now the ability to use nonhuman virtual agent technology in such a way that the agent cohort can consist of a blended pool of both human and virtual agents who are able to answer queries, perform processes, and route contacts between one another. Virtual agents are able to drive the IVR channel, engage in text-chat with customers on instant messenger channels, and parse incoming SMS messages and inbound emails, acting upon them automatically.

Virtual agents are far more numerous than their human counterparts, with far lower associated costs, but they are not perfect; they are only as intelligent as the dialogue and business-process modeling has allowed them to be. It is suggested that most inbound self-service contacts should be fronted by virtual-agent technology and then routed on to a human agent when it comes to the more complicated processes, such as complaints or exception handling. Indeed, technology is now at the point where customers can be engaged in text-chat with an agent and be unaware whether they are typing to a human or virtual agent at that particular point.

Automation Still Needs People: Making better use of contact center staff is a key enabler of a more automated operation. Agents can receive multiple pieces of customer interactions to work on. While bombarding agents with too many pieces of simultaneous work is counterproductive, it makes for more efficient use of agents when they are on a slow real-time channel such as text-chat. On such a channel, agents can handle two or three chats concurrently before the customer notices a slow-down in response times and the agents become frustrated with the cognitive load being placed upon them. A modern, process-aware contact router handles multiple channels and is able to prioritize these according to real-time needs. For example, inbound telephony is given a higher priority than email correspondence, and high-value customers can be given priority over others.

In a related benefit, if a process-aware routing engine delivers contacts to agents, the overall system can provide real-time management metrics. This can concern both the business processes, such as the number of up-sells or credit card payments in the last three minutes, and the contact statistics, like average handling time, number of inbound contacts across email, SMS, IVR, telephony, text-chat, and so on. This allows contact center management to be able to see the live state of health of the contact center machinery. Management can then choose to modify routing parameters according to the current business requirements.

A Well-Oiled Machine: All of these solutions for delivering an automated, multichannel contact center are available today. Contact centers need no longer suffer from inefficient silo-based operations, and customers need no longer give the same information multiple times within the duration of a phone call. Customers can even engage with virtual agents without necessarily realizing so, and human agents can be left to do what human agents do best while their virtual counterparts handle the majority of easily automatable processes. The contact center machine of today should feature many well-oiled cogs able to reconfigure themselves into the optimal position for maximum customer service and business benefit.

Dr Iain G McKay, BEng (Hons) CEng MIET, designs and builds intelligent contact center solutions at Graham Technology, specializing in contact center software and services.

[From Connection Magazine April 2007]

%d bloggers like this: