Customers have adapted to speaking with voice user interfaces (VUIs) to buy products, lodge complaints, or make travel plans. They understand that interacting with a VUI is not the same as chatting with a friend, but they do expect to be heard.

Though current VUI technology is a long stretch from normal everyday interaction, companies that design the dialogues for those systems must strike a balance between conforming to user expectations and system capability restraints. By harmonizing the client's infrastructure requirements with enduser expectations, those same companies are finding new ways to deliver more effective solutions.

One such firm is Interactive Northwest Inc. (INI), based in Portland, Ore. It works with clients and IT vendors to build an infrastructure focused on design while integrating voice response and speech-recognition technologies with existing business systems.

Among its clients is NTELOS, a digital wireless PCS provider in the Mid- Atlantic. NTELOS turned to INI to build an interactive voice response application allowing its wireless customers to receive minute usage information without talking to a live agent.

A few years ago, NTELOS acquired another wireless provider that enabled its wireless customers to dial #225 to access account information. Following the acquisition, NTELOS needed to continue this service, but did not have the systems in place to support it in an automated manner. As a result, all account information inquiries ended up at the NTELOS call center.

INI's solution combined Avaya's Definity servers and Conversant script builder applications running on the Conversant platform. It included an application detector, allowing the IVR to read incoming digits from the Definity vector into the application, and an account information IVR application with a host interface via a proxy NT server to an AS/400 database. The system gives callers three options:
1. receive current minute usage via a text message to their mobile phones;
2. hear their account information over the phone; or
3. make a payment over the phone.

The solution met company and customer goals by improving customer service, lowering caller wait times and avoiding a potential 30 percent increase in staff.

"We start out by trying to make sure the goals of the caller and the goals of the business are not incongruous," says Gary Van Gordon, vice president of interactive solutions at INI. "We start with what the customer is calling about as the starting point."

Van Gordon says it's easy to get overwhelmed with what the system is going to say and recommends taking a step back to focus on the big picture of how the customer wishes to be treated. As part of that process, INI has coached clients to ensure they leave a provision for human contact.

"Any business that thinks it is going to put in self-service to eliminate all need for human contact would be making a serious mistake," he explains.

Spencer Stern, senior manager at Virchow, Krause & Co., a business and IT consulting firm, agrees. "In these days of email, people often prefer not to talk to people," he says. "But you always need to provide the option."

Intervoice, a Dallas-based solution provider that integrates voice automation platforms, software, applications, and services, focuses on striking the right balance. It helps clients make the appropriate tradeoffs for getting the highest possible percentages of call completion and task completion while meeting timelines and budgets.

"Anyone can build an application that has a hundred percent call completion rate, but it won't necessarily have a hundred percent task completion rate for the user," says Dave Pelland, director of Intervoice's Design Collaborative. "Often a company will talk about customer satisfaction and a world-class user interface, but the only measurements they use for success on the project are call completion rates, time, and effort. On the other hand, their users don't usually care where their call completes, as long as they've successfully completed their task."

To meet these tasks, Intervoice executives say the key is to build a solution around open standards such as VoiceXML, CCXML, and other Web technologies. This makes it easier to maintain an implementation. It also allows for easy integration with other third-party systems and software, including rules-based engines and BEA Systems applications.

"The benefit of open standards is that you're not locked into a particular brand or type of hardware," Pelland says. "You have flexibility to choose."

Another key is supporting logging and metrics. Since maintaining a speech application can require specialized talent to analyze recognizer output, determine problem areas, and suggest new dialogues and grammar changes, a VUI integrator needs to be able to get a complete picture of the running system. Most, if not all, standards-based systems have this support today, but it still requires attention, Pelland says. 

Measuring Results
Implementinga VUI is not a situation where you put the system in, turn it on andwalk away. Rather, the rollout is an iterative process that involvesrecording and observing how people are using the system, Van Gordonsays.

It's critical not only to tracespots where a caller may be getting confused, but also to recognizeways the system can become more efficient.

INI suggests that VUIs getre-examined every 12 to 18 months. The systems have progressivelyimproved as the technology has become more cost-effective andorganizations have become more familiar with best practices.

"The tools have madeit easier, but they don't provide inherent knowledge," Van Gordon says."It's rare to hit a home run right out of the gate."

There continues to be a wedgebetween a company's intent with its VUI systems and what real peopledo. "Most companies have no idea what people are doing in theirsystems," says Walter Rolandi, founder of The Voice User Interface Co.

Behind the technology, there is a psychological element involved in creating successfulVUI systems. Rolandi says VUIs should not sound too human because itsets unrealistic expectations. Pretending that a machine has a peppypersonality will irritate customers. Instead, he suggests a structureddialogue approach.

Matt Wright, president of RightVoice, a company that specializes in making the information andservices of enterprises, telecommunications networks, and the Internetsound crisp and professional, agrees that VUIs can't be too over thetop. "A lot of user interfaces can be too happy or cheery, and thatquickly goes from cute to annoying," he says. "Between each message,changing pitch is an important way to introduce a new idea in a waythat doesn't fatigue the listener. Less is always more."

For nitty-gritty transactions suchas getting prescription refills, users don't need a warm and invitingvoice. They want to complete the call quickly and effectively. Time ismoney. In a huge system fielding millions of calls, every secondimpacts the number of ports needed. Changing tone or inflection can cutcosts.

While customers have grownaccustomed to getting basic service on the Internet, the telephonepresents different expectations. "On a telephone, the precedentculturally is to talk," Rolandi says. "If they are calling with aproblem, like an error in their [electric] bill or too much money beingtransferred from one account to another, and they are not happy aboutit, they often need to talk to a person. But people will gladly usespeech automation if it is quicker and easier than doing the same thingwith a person."

Setting the appropriate tone iscritical. The decision to use a male or female voice may depend on thebrand or service. If a company is selling cosmetics, it may want to usea woman's voice that is slow and deliberate and puts the customer atease, Stern says.

"The voice, grammar, andterminology you choose all have to be aligned around the existing brandor it will be confusing to customers," he says.

"If you want to increase usercomfort level, you need to implement a system that accounts for dialectissues," he continues. "While you need to be focused on bottom-linenumbers, you also have to focus on customer retention and loyalty."

Rolandi suggests that if any ofthe call center reps are not doing anything, a service call should getrouted to that person. "People today are better informed about thelimitations of technology," he explains. "But in a lot of instances,they need to talk to a person. In these circumstances, the thing to dois to quickly get them to one."

That's as true of VUI as it is forhuman interactions. Successful decision tree construction comes down tothe details. Decisions need to be limited to a few choices or thecustomer will get frustrated and hang up. Conditions need to be splitinto the appropriate categories, such as ordering, billing, repair, andservice. Creating the right algorithm is the difference between successand failure. "It's about creating a natural call flow that delivers asmoother experience for the caller," INI's Van Gordon says.