This Time, It's Personal

Improving the caller experience, as it relates to contact centers, has traditionally focused on improving speech recognition, writing clear prompts, and creating efficient call flows. Vendors attest that automation should ideally help customers get what they want as quickly as possible, but is that always happening? Even the phrase "call flow" suggests a series of steps a caller must take before he can access the information he needs.

Additionally, as the benefits of automation become more apparent to enterprises, functionality becomes increasingly more intricate. On the one hand, this certainly relieves a lot of stress on an enterprise’s phone lines and empowers the caller to help himself. But if the caller doesn’t know how to help himself, or doesn’t know how to use the system, he’s likely to get frustrated and the interaction is a failure. This is bad news for everybody, particularly the enterprise, which knows that it takes less effort to keep existing customers happy and loyal than to find new ones.

Yet, the CFI Group’s Contact Center Satisfaction Index showed that one in five customers who call a company contact center don’t resolve their issues, and customers who hang up without resolution are twice as likely to defect.

Furthermore, the creation of such institutions as GetHuman.com—which is well-known for its company-by-company cheat sheet demonstrating how callers can bypass an enterprise’s IVR in favor of a human agent—attests to consumer dissatisfaction toward many automated contact centers. The problem with IVRs, according to Walt Tetschner, a GetHuman project leader and advocate, isn’t the fact of their existence but of their usability. "I don’t position GetHuman as being anti-IVR," he wrote in an email. "I like IVR for many applications. If it can do the job, it can be faster than waiting in a long queue. On the other hand, when callers say that they want to talk to a human, then let them get to a human. Making it difficult to get to a human is a top complaint that callers have."

JetBlue Airways is one enterprise that meets GetHuman’s standards; the airline is  recognized by the organization as a member of its Great Customer Service Club. But JetBlue’s IVR, which uses speech recognition technology from Aumtech, has limited functionality. Callers can automatically get flight information, such as arrival or departure times, but more complicated requests, like finding or booking flights, or three misrecognitions immediately transfer the caller to a human agent. (JetBlue implements an at-home agent program, with all agents based in Utah’s Greater Salt Lake area.)   

"It’s kind of a balancing act for us," says Nanette Ford, JetBlue’s director of reservations operations. "We want to make sure we didn’t inundate customers with a lot of functions that we would push. We wanted [callers] to be able to speak to a human being, [but] if they wanted to self-serve, they could."

JetBlue’s system revolves around a corporate philosophy that sees call center automation as an alternative option. This isn’t feasible or even preferable for every company. "After 25 years of designing self-service technologies, we know for certain that customer satisfaction (CSAT) can be increased by using speech in a well-designed application," says Dave Pelland, director of Intervoice’s Design Collective. "So if all things are equal, a new application will typically perform better, which can result in better CSAT if implemented using speech."

Of course, Pelland adds, all things are rarely equal, and vendors have to factor in application costs, the caller’s environment, and the application domain to determine the best way of interacting with a customer.

According to Jim Larson, an independent consultant and VoiceXML trainer, enterprises often forego the trouble of personalizing their systems. "They think one speech application fits all," he says.

Caller ID
Two of the biggest aspects for personalization, according to vendors and analysts, are implementing a system that is able to identify a caller and being able to determine caller skill level. For many vendors, this means tapping into a caller’s history and being able to access the necessary data across the enterprise.

In years past, enterprises tended to isolate their contact centers from the rest of the enterprise, effectively creating a barrier to keep callers away. Nowadays, vendors choose not to build stand-alone systems, but rather to deploy them across a services-oriented architecture where all the pieces can interact with one another.

"We have a lot of integration into canned services," says Roberto Pieraccini, SpeechCycle’s chief technology officer. Many of SpeechCycle’s clients are cable companies. "Even before the caller stops speaking, we know exactly who the caller is, what his history is, and what sort of subscription he has."

The call center effectively connects a variety of environments, from IT departments within the enterprise to the cable plant delivering Internet video. If there’s an outage, for instance, calls demanding to know why service has been lost will spike. A fully integrated contact center will be able to trace the problem to its source.

Additionally, a system should recognize repeat callers by keying into the automatic number identification or the dialed number identification service. While there are limitations to this, particularly if the caller is using a new phone that the system doesn’t recognize, it’s still beneficial for an enterprise to maintain a level of personalization for repeat callers. "We know if the person called in the past 24 hours," Pieraccini says. "And in that case, the interaction takes a completely different route. We recognize that you called before and called for the same problem. We recognize that we didn’t completely solve the problem."

Pieraccini likens it to going to a new doctor. "Imagine going to a doctor with the doctor not knowing anything about you. That’s very difficult. That’s why when you go to a doctor, you have to fill out all this paperwork. So we have to know everything about the customer, what has happened, and what happened before."

Being able to make changes in real time is crucial in personalizing the caller experience. Sometimes that’s as fundamental as regulating the speed of spoken prompts. Word-per-minute rates are static in most applications, calibrated to a happy medium determined by the vendor. The problem is that first-time users and power users often get treated the same way in an IVR. "It turns people off," says Daniel O’Sullivan, CEO of Interactive Digital. "You call the IRS, Social Security, your bank, whatever, and it’s that mundane monotone Yes, here’s your balance."

Interactive Digital’s Adaptive Audio listens to calls and, based on how quickly and accurately callers respond to the prompts, speeds up or slows down the speaking rate. 

If a caller is a frequent user, then the nature of the automated interaction should change. This is difficult to do within the rigid structure of preset call flows and dialogue designs. Voxify, for instance, with clients that include Continental Airlines—in his ASR Newsletter, Tetschner described the airline’s outbound passenger check-in as "a superb application"—Rite Aid Pharmacies, and NFL Shop, eschews static IVR applications written in VoiceXML. 

"If one guy is writing static VXML apps," says Voxify CEO John Gengarella, "he has to write ridiculous amounts of code for all the potential traversals of that call flow." By contrast, call flows generated dynamically based on data gleaned from a caller’s history and profile provide a more intimate experience. It also gives the enterprise the opportunity to up-sell or cross-sell based on past buying trends.

"You can’t do that writing a boatload of VXML code," Gengarella continues. "You can’t have dynamic call flows consistent with current data if you’re literally trying to write code lines for those traversals."

Swiss telecommunications provider Swisscom uses VoiceObject’s customer care portal—a high-volume voice portal handling nearly 12 million calls a year for Internet, landline, mobile, and television services. Last October, Swisscom project manager Christian Rosenberger stated in a press release his company’s need to personalize the caller experience by implementing "reusable dialogues, per-caller personalization, and sophisticated reporting capabilities." 

VoiceObjects implements an object-oriented architecture, which makes it easier to apply changes to the call flow swiftly and frequently. The downside is that it’s more difficult for the system to interact with other channels in the enterprise.

"Some contact centers are making a big effort to integrate what they call multiple channels, where one channel could be a telephone channel, another could be a chat channel, and another could be a Web-based channel," Larson says. Of course to do this, not only does all the data have to be available, it also has to be transferable from one area to the next.

In May, Intervoice launched its Contact Portal, which allows a customer to interact with an enterprise using the mode of his choosing—phone, email, or SMS—and to alternate between self-service or live agents. 

In the past, implementations that connected the entire enterprise have been cost-prohibitive, requiring investments in IVRs that connect to live agents, phone switches, and computer telephony integration. The advent of Voice over Internet Protocol has enabled costs to drop. Thus, data can pass over the same lines on the same network as phone packets, which drives easier implementations. "It’s no longer a project to send that data," says Kevin McPartlan, Intervoice’s vice president of product direction for enterprise products.

Specifics about the data needed vary by customer, but in general, vendors need to understand how the various channels in an enterprise interact with one another. Andrea Holko, Intervoice’s senior vice president of global consulting services, maintains a discovery methodology from self-service all the way through live assistance. One problem, she acknowledges, is that the call center can only directly affect some of those channels. She describes a meeting with the vice president of customer experience for a wireless company: "His goal and our goal is to sit together and get alignment across all the channels that we’re going to be impacting—the voice channel and call center technology and some subscriber services. It’s a consistency method where we execute the same level of detail to the other channels and implement [the solution] on the channel we actually control."

Making Sense
Working in the speech technology industry, it’s important to take into account that just because an application can be speechified doesn’t mean that it should. A case in point: the many horrible apps created simply by adding speech to a former touchtone system. It’s important to add speech in a way that makes sense, according to Holko. In certain financial institutions where information is largely numeric, touchtone makes sense. For security reasons, callers prefer to enter their information quietly with a keypad rather than vocally, Pelland says.

Alternatively, a locater application shouldn’t be touchtone. "You’ll want to kill yourself," Holko says. In other words, it makes sense to speechify a locater. "We use the human factor to determine what makes sense and what’s the easiest way to get from A to B."

Steve Springer, Nuance’s senior director of user interface design, emphasizes that customers care more about achieving their ends than the means that enable the them to get there. "When they start to get frustrated is when the how becomes much more front and center," he says. Therefore, it’s important to never lose track of what the caller ultimately wants to do.

One of the first systems that Springer worked on nearly 10 years ago was for an airline application originally designed to sell tickets. He discovered afterward that people tended to call inquiring about arrival and departure times—information that wasn’t available over the back end.

Now when Springer works with an enterprise customer to design an optimal caller experience, the first thing he and his team try to do is model as closely as possible the caller’s mentality. "That obviously varies from industry to industry and from company to company," he says. And within companies, that mentality can vary seasonally or even daily.

It’s difficult to measure something as abstract as the caller experience. Most vendors use surveys, which are reliable to an extent; however, they’re also expensive, they don’t always reach a particularly large sampling of the caller base, and it can still be difficult to tell why a caller is dissatisfied.

Is the caller irate, for instance, because he had a poor interaction with a clunky IVR, or because he hates being ripped off by his cable provider? In May, SpeechCycle unveiled the Caller Experience Index (CEI) and began working with researchers at Carnegie Mellon University to optimize the metric. The problem with most metrics, according to SpeechCycle CEO Zor Gorelov, is that vendors judge performance in a variety of areas within an application, which diminishes accuracy.

The CEI is designed to systematically produce an overall numerical score that enterprises can use to gauge their progress over time. But regardless of whether this particular metric is adapted by the speech industry as a whole, Pieraccini emphasizes the importance of having the ability to measure overall customer satisfaction. "If one wants to lose weight, the first thing they have to do is buy a scale," he says. "You can’t do anything if you can’t measure your progress."

Ultimately, just because automation rates increase doesn’t mean that customer satisfaction will. Their growth  must be mutual.