When it comes to IVR design, are you seeing the forest for the trees?
Customers who call a contact center are often required to navigate an IVR phone tree with myriad menus before being placed on hold and eventually reaching a live agent. At some progressive companies, a more pruned conversational system enables customers to request help by simply speaking into the phone. With one problem and two solutions, companies have to ask which is best for them: a phone tree–designed system or a conversation—driven design?
How May I Help You?
The conversation-driven design, also known as the "How may I help you?" interface, is starting to uproot phone tree–designed IVRs. To be sure, the tried and true phone tree designs have been around for decades and are a known and sometimes good, sometimes not so good entity. With these systems, customers call into a contact center and navigate a menu of options similar to branches of a tree before they ever reach a live agent.
"Half the time, when you let people skip a bunch of steps in the phone tree, they're not sure what to do and they press zero," says Bill Meisel, president of TMA Associates and executive director of the Applied Voice Input Output Society. "Companies realize that [this design] frustrates the customer and that it also doesn't solve the problem unless it's a simple issue. Someone once described a classic phone tree IVR as feeling as if you're pressing a button with your tongue."
David Pelland, director of UX innovations and design at Genesys, says that phone trees are known as discrete recognition menus, whereas conversational design uses natural language to let callers say what they want in their own way.
"With phone trees, you'd have to press one or press two in the days before speech recognition," Pelland says. "You'd call in and you'd get literally a tree, as if someone drew it on a board."
Getting Schooled in Design
According to Dena Skrbina, senior director of solutions marketing at Nuance Enterprise Division, there are two types of phone trees—touchtone and speech, also known as directed dialogue. The entire concept of a phone tree, she says, is a series of menu choices, but both of these ask the caller a series of questions to determine why she is calling.
With a touchtone tree, a caller may get a menu that says, "For billing, press one." Once the caller presses one for billing, she will be asked another set of questions.
The other type of tree involves directed dialogue technology, which has the same structure as a touchtone tree but allows the caller to speak his response. The menu might say something such as "Please tell us why you are calling," and the caller might say something like "Account status."
"You can repeat something that the prompt said, in which case you go down another level and...get a similar prompt," Skrbina says. "Eventually, you'll get to some self-service functionality or an agent who can handle the call."
Directed dialogue, Skrbina explains, uses speech recognition, but it's very directed. "If you think about the structure of these menus...they both use that same structure. However, one is more convenient than the other because it's hands-free—users can listen to a prompt and speak their answers."
The difference with conversational design is that it uses natural language. With natural language, callers can say the reason for their call in their own words. Instead of a very specific question that would take a caller down a defined path, the prompt might say something like "Hi, thanks for calling. How may I help you?" The caller then would describe in her own words what the issue is.
"Natural language understands that and gets you to the right place," Skrbina says. "With a tree paradigm, you probably had to have been asked three or four questions to get you to the right place. With natural language, you condense the trees and provide a more natural experience."
With a touchtone phone tree, there is also the issue of the zero-out rate, in which callers press zero right away when they get to a menu.
"That [zero-out rate] is about nineteen percent," Skrbina says. "When you move into a more conversational interface, it drops by four percentage points. A four percent increase in the number of people who can potentially automate makes business sense, and we see companies move toward this."
While conversational design can have more complex technical issues for deployment, it is far easier for customers to use, Skrbina points out.
Bill Scholz, president of AVIOS and speech consultancy NewSpeech, says that the conversational approach combines speech recognition, natural language, and dialogue management, which breaks down the steps involving the design process.
The system first asks callers to say what they want. That response is then processed by a speech recognizer. The output from the speech recognizer is subjected to natural language analysis, where it is examined for various concepts and rules, and, after that has been done, it goes to a dialogue manager, which tries to manage the interaction between the computer and the caller.
The dialogue manager identifies what tasks or requests that person is calling about, which gets passed to a task manager. Then it goes in the other direction. The task manager will either typically provide information, or have some kind of question, which goes back to the user.
The output from the task manager will then go to a response selection mechanism, determining what sort of response should be delivered to the caller. That goes to a natural language generator, which turns this into a string of words. Finally, a text-to-speech engine will render that to the caller.
If you don't want to settle for one design or the other, there is an option of combining phone trees with conversational design, says Deborah Dahl, principal at Conversational Technologies, chair of the Multimodal Coordination Group, and cochair of the Hypertext Coordination Group at the World Wide Web Consortium.
"You can start out by offering people a few common options, such as 'Press one for customer service.' Then [the system] could send you into a more conversational interaction," Dahl explains.
If It's Not Broken, Don't Fix It
Early adopters of conversational design include airlines, financial services, telecommunication companies, and, increasingly, utilities, all companies that have cash at hand to deploy such a substantial overhaul of their IVR design.
"Those companies that have been using IVR the longest are those same companies doing the most innovation," Skrbina says. "We're seeing it more in utilities, which at one time didn't have as much motivation to deliver great customer service," but now consumers have more choices. "This is an incentive to improve the customer experience."
Even with such conversational design advances, phone tree–based IVRs are still the norm for most SMBs. The reasoning? They work, Pelland says.
"For most companies, it actually solves their problems," he says. "Most IVRs are small enough that they don't have to ask you more than two or three questions and you can get through them. The bang for the buck [for conversational design] isn't there."
Indeed, another speed bump to conversational design's adoption is that it can be expensive to implement and tune on an