For Alicia Lydecker, a 21-year-old graduating senior at Syracuse University, spring break was supposed to be the time of her life. She would finally be able to take the European trip she always dreamed about. Before she could ride the gondolas in Venice or admire the Eiffel Tower in Paris, though, she says she wanted to contact her banks and credit card company because she didn’t want any issues with making abnormal purchases. "I was calling because I was going out of the country, and I needed to inform them so that [my credit-card company] wouldn’t freeze my account," she recalls, on the assumption that sudden purchases made outside the U.S. would probably arouse suspicions.
Making sure overseas travel arrangements and school responsibilities are squared away is difficult enough, especially for Lydecker, who also works three jobs in addition to her full-time student status. So when she called up her credit-card company, she says she was less than thrilled to reach its interactive voice response (IVR) system. She immediately decided to use the touch-tone options instead of going the speech self-service route, because of prior experiences. "I just [used the buttons] because I’ve spoken [to the IVR] in the past and it only recognizes what I say half the time -- and I really didn’t want to be bothered," she laments, noting she always makes it a point to call from inside her dorm room where there is no outside noise interfering. So instead of being able to say "security" or "fraud department," Lydecker waded through the maze of push-button menu options -- and found nothing to her liking. "I ended up just waiting until the end and choosing the ‘speak to a service representative’ option," she recalls.
Lydecker was eventually able to get the information she needed, and she didn’t have any issues when purchasing food, drinks, and souvenirs as she trekked across Europe in March. But the fact that she had to spend her precious time avoiding her credit-card company’s speech-enabled service system is a common complaint many make of IVR systems. "If the speech option worked correctly, I think that would be much easier [than using touch tones]," Lydecker posits. "But I’m normally in a hurry, and I don’t want to deal with error prompts when the system can’t understand my voice."
Lydecker isn’t alone in her negative experience with IVR systems. Virtually everyone who has ever called one of these systems has at least one horror story to tell about how difficult it was to navigate through the menu or speak the magic response to proceed. Web sites such as GetHuman (www.gethuman.com) have been created just to allow consumers to sound off on the experiences they’ve had as well as ways to "get around" an IVR system. (See the Customer Centricity column, "GetHuman? Get Real," March 2007, for more.) According to GetHuman’s Web site, its goal is to "convince enterprises that providing high-quality customer service and having satisfied customers costs much less than providing low-quality customer service and having unsatisfied customers." The Web site has even gotten the attention of IVR designers like Dave Pelland, director of a design collaborative for Dallas-based automated information solutions provider Intervoice. "Companies started to get bad press based on these IVRs, and that has clearly changed their attitude," Pelland explains. "When [companies] come to us now, they’re much more open and willing to listen to us experts tell them how to...solve this problem."
Love IVR or hate it, automation and speech self-service are not going away for the foreseeable future. A recent Datamonitor study, "Understanding the Changing Role of IVR in Evolving Infrastructures," concludes that global investment in IVR licenses will increase from $475 million in 2006 to $845 million by 2012. So if companies are throwing millions of dollars into their IVR systems, voice user interface (VUI) design should be a major point of emphasis in order to ensure the investment is money well spent.
Many speech designers and engineers say that VUI design is a mixture of art and science, so while there are some must-avoid mistakes, simply crossing them off a list doesn’t guarantee success. Phillip Hunter, vice president of interaction design for New York-based contact center automation technology provider SpeechCycle, calls this the "recipe mentality." "If I can just go pull the right recipe from the cookbook, then I don’t have to know anything about food or temperatures to be able to produce it," he explains. "That might work if you’re cooking spaghetti or baking a cake, but it doesn’t work in voice design."
And merely sidestepping pitfalls may not be enough, says Susan Hura, principal at Atlanta-based VUI consultancy SpeechUsability. "You can avoid the mistakes and still design [the VUI] improperly," she says.
PRACTICES TO DESIGN BY
Experts suggest starting from Square One: Figure out the goals of the planned IVR, as well as the needs of the callers. This piece is so important to the VUI design puzzle that Pelland and Judi Halperin, a speech engineer at Basking Ridge, N.J.-based communications solution provider Avaya, both say their respective companies do a thorough presales process to ensure that all the cards are on the table before dealing with VUI design. That way, Halperin says, "from the get-go we’ve got the big stakeholders involved in both sides of the process. We optimize communication with the customer so we can start out properly and set expectations."
Pelland says that Intervoice also has a presales process, but says executives have to do more than just list their goals -- they must order them by importance, a step that could change the entire speech ballgame. "When we first get out there and do requirements-gathering, we actually have the client rank their business needs," Pelland explains. "For example, customer satisfaction, cost savings, agent time, and call completion -- can you rank these? You can start to see the thought process ahead of time, and it has worked really well."
Companies should then make sure they’re employing a legitimate designer with proven experience in designing successful VUIs. "VUI should be built by someone who not only has experience in building VUI, but some understanding of the business and what companies are trying to accomplish, so they can build to that user population," explains Elizabeth Herrell, vice president at Forrester Research. In searching for a designer, having a knowledge of linguistics in order to create grammatically correct (and genuine-sounding) prompts is key, according to Blade Kotelly, chief designer at Cambridge, Mass.-based information-access software company Endeca. But Kotelly, who also wrote the book The Art and Business of Speech Recognition, cautions against hiring someone solely because of linguistics skill. "Hire those that try to write scripts, people who are good at making presentations," he says. "If your designer can’t give you a professional presentation that makes you seem well-polished or matching the specific crowd or audience, he won’t write good prompts that will match the audience."
Having the proper designer in place can help with the next essential piece to a successful VUI design: mimicking the flow of natural conversation. This isn’t to say that designers should be incorporating slang and improper colloquialisms into speech prompts, but they should consider the little things that are sometimes overlooked during a typical conversation. Hura calls these essential tidbits discourse markers.
(See "The ABCs of IVR," below, for more terms of art.)
"[Discourse markers] are the little connector words that relate one piece of the conversation to the next piece," she explains. "You’re actually using discourse markers all the time by saying all right and OK. This is a very powerful cue, because what you’re telling me by saying OK and sure is I heard you, go ahead."
This is hardly an exhaustive compilation, but designers say that having a defined list of best practices regardless of company goals can harm the VUI design. "It’s a tough question, because it’s really hard to have an industry standard here -- it changes based on the business goals, customer goals, and domain," Pelland laments. "If we went to a banking customer and they asked us for our best practices, and then they go and tell us their number-one concern is customer satisfaction, not cost savings, it shuffles the whole deck. It’s really tough to show up and say, ‘Listen, here’s a handbook that says, "Here are your best practices, and the best way to do VUI." ’ "
DON’T RUSH THE FINAL PRODUCT
A tremendous mistake when designing a VUI for a company’s IVR system is to allow the company to dictate scheduling based on budgetary issues, thereby rushing the designers to throw together the VUI in order to make an arbitrary deadline. "One of the biggest things I feel affects the whole quality of an end solution is when the customer has tunnel vision in terms of its end date," Halperin says. "Companies fail to look at the big picture and apply appropriate user-centric design strategies.
[Companies] just basically want something documented and done, and don’t fully take the time -- or have the time -- to sit down and really understand what it is you’re trying to accomplish and why it might take more than a day-and-a-half [of work] to handle it."
SpeechCycle’s Hunter believes this is a common mistake caused by the culture of approving business projects. "By the time budgets are approved, and resources are allocated and so on, it’s eaten into a significant amount of time -- and someone has established that we will be in production by some date that doesn’t really have any allowance for what it actually takes to do the project." Halperin suggests reemphasizing and setting clear expectations from Day One to avoid this. "A lot of those expectations are taken care of up front, because only having two days [to build a speech system] is not possible if you want a quality product," she says.
EXPLAIN BENEFITS (AND RISKS)
It’s one thing to follow a company’s guidelines for an IVR solution, but another to just entirely acquiesce and build everything it says it wants. Halperin explains that, all too often, designers will accept all requests for different aspects of VUI design without pointing out the benefits and potential risks of implementing those requests. "For example, customers may say to kick out anybody who has this [particular] code on their account, and after you look at their data that’s 99 percent of their customers," she says.
Halperin suggests gently pushing back on certain requested specifications. While it’s important to stress that it’s possible to fill a given request, not mentioning the consequences of potential actions can be deadly to the VUI design. Plus, executives will respect the fact that a designer wants to explain the ins and outs. "It’s a give-and-take, but rolling over, playing dead, and becoming a secretary by just writing it down really does not do justice to the final solution," she states.
THE PROOF IS IN THE PROMPTS
The beauty of writing text for actual reading, be it on Web sites or in print, is that, to a certain extent, you can be as lengthy or as brief as the situation demands.
However, when designing scripts for auditory comprehension, prompts that may read just fine to the eye may be disastrous for callers looking for a quick solution -- and that begins with the introduction prompts. "People put too much time into setting introduction prompts -- almost always too long," Kotelly says. "This mode doesn’t allow for the same kind of expression, and [this mode] shouldn’t be used for it because it’s not useful to people."
Hura says short prompts are actually the beauty of the automated phone system. "When people call up automated systems, they want their problem solved as quickly as possible," she says. "That’s what it comes down to: They’re not calling for a lovely customer experience or to have a better relationship with your organization." Having long introductory prompts can be disastrous for a VUI, because unlike text meant to be read, not heard, you can’t skip ahead.
"People don’t mind automation, but they really mind poor automation," Hunter states. "With a person, they can interject and say, ‘Excuse me, I don’t need to hear all that -- can we get on with what I need?’ You can’t really do that with a machine, and as a result people feel stuck, trapped, and slowed down." Besides introductory prompts, Halperin also suggests condensing security and legal disclaimer prompts as much as legally possible. "[Legal disclaimers] can also seriously degrade an application, even though they may be required at the beginning of the call," she says. "They may have to be there from a legal standpoint, but as far as callers are concerned, it’s daydream time. They stop listening to the prompts and start multitasking, and then they miss the next prompt because they think it’s part of the disclaimer."
The phrasing of the actual questions is also important. "People don’t ask questions in the right order," Kotelly laments, using the example of the too-similar Where are you going to? and Where are you leaving from? The words "from" and "to" should be placed at the beginning of such prompts, he says -- having the differentiating bit at the beginning allows callers to answer quickly. (See our two-part feature series, "Speak Up!" [December 2007] and "Listen Up!" [January 2008], for more on this.)
HIDING ACCESS TO LIVE AGENTS
Sometimes automation can’t give all the help that’s necessary, and live agents are needed to rectify the situation. A major flaw is trying to hide these live agents by making it difficult to stop the automation and immediately go to an operator. "Eventually companies will realize they’re going to get a 40 to 60 percent completion rate in the IVR depending on what it’s set up to do," Herrell points out, adding that many customers still have to (or want to) go to a human, but companies make it next to impossible to find live agents -- which infuriates callers. "All you’re doing [by hiding access to live agents] is making people antagonistic toward using the automated system," she says.
Experts differ on how exactly to make it easier for callers in an IVR to get to live agents, but the sentiment is the same: Live agents must be readily available. "Zero should always, always go to an agent," Kotelly says. Hura notes that it should be easy to access a live agent, but that companies should ask a few more questions so the IVR system can send the caller to the proper live agent the first time. "Rather than just offering an agent right up front, what you can do is recognize when callers say Agent [or] Operator, or press 0 -- but then ask them just the information that you would need in order to do the correct routing," she says.
TEST, TEST, TEST
The only way to truly find out if your VUI design is on the right track is to actually take it for a spin with real users. "You do your best and try to follow principles, but the fact is that none of us creating these systems have the exact same mindset as those end users," Hura says. Pelland gives an example of a VUI he was helping to design for a United Kingdom-based voicemail system: After building a prototype and having all the U.K.-based executives test the system, testers provided it for real users, and found no one was able to access their voicemail messages. "Thank God we did a usability test," he recalls. "In the U.K., people don’t use contractions when they’re speaking, and the command chosen in the U.S. to listen to the messages was What’s it say? When we did usability testing with people off the street...every single one of them failed -- it was one of those humbling moments as a designer when you realized you messed up." However, the lesson was not learned in vain: "That’s my favorite example of why usability testing is important," he says. "I usually present that story in the ‘Why you must usability test before you ship’ spiel."
When collecting data from these usability tests, it’s important to gather both quantitative (behavioral) and qualitative (opinion) data. Statistics alone won’t tell you as a designer what’s flawed and what isn’t, Hura says. "You ask callers questions to give their opinions on whether or not they felt like they were being misunderstood frequently, or whether the system usually understood what they had to say," she says. "By balancing what users think with actual numbers, you can know whether or not you need to take action in any specific case."
THE FUTURE FOR IVR
Making sure VUI designs are done properly the first time is becoming even more important now that IVR systems are becoming critical to companies’ customer service repertoire, and spending on IVR ports is expected to rise steadily through 2012. Just because more companies are planning to incorporate IVR systems doesn’t necessarily mean people will just accept them as a necessary evil. Hunter says that the makers and designers of IVR systems are viewing the telephone as more than simply one avenue through which customers reach out when they require assistance -- and this changing philosophy could help improve not only VUI designs, but the entire IVR system framework. "The IVR automation industry is beginning to see [the phone] as two separate channels: a human channel and an automated channel over the phone," he says.
Citing a parallel with live Web chat being offered online, Hunter believes the same insight can -- and should -- be leveraged in IVR systems. "We need to start having that same focus on the phone -- this idea that you might reach automation first but that reaching a person should be an immediate and easy option for these front-door systems," he says. "So as soon as someone calls, there may be a clear option, a list of things we can do for you on the phone, or By the way we can also direct you to a person if you’ll tell us simply what you’re calling about. Then we’re putting power back into the hands of the callers. I actually think that will increase the adoption of automation because it doesn’t feel so imposed."
SIDEBAR: EXPANDING IVR MARKETS
The IVR influence is spreading across the globe. New research from Datamonitor shows global spending on IVR licenses will jump from $475 million in 2006 to $845 million by 2012. Regions with the largest compound annual growth rate (CAGR) through 2012:
- Asia-Pacific: 18 percent CAGR
- Caribbean and Latin America: 14 percent CAGR
SIDEBAR: The ABCs of IVR
Behavioral Data -- the quantitative statistics collected during usability testing that determine what callers are doing at different stages of the IVR system (e.g., "86 percent of callers have correct recognition at a particular prompt in the IVR," or "There’s a 14 percent error rate at this prompt").
Differentiators -- do sweat the small stuff -- such as where to place the words from and to in a prompt. That can make a huge difference in a caller understanding immediately what the prompt is alluding to, or confusion leading to a possible error.
Discourse Markers -- the small words and phrases such as OK, all right, and sure. These words help to orient callers and keep them proceeding through the IVR system.
Interactive Voice Response (IVR) -- a phone technology allowing a computer to detect voice and touch tones using a normal phone call.
Opinion Data -- questions asked to callers during usability testing to rationalize the behavioral data collected in order to determine whether or not changes truly need to be made to the IVR.
Prompts -- the questions or directions posed to callers when navigating an IVR system, such as From where are you traveling? or Press 5 for more options.
Speech Self-Service -- the caller using her voice to proceed through the IVR system. So when the system asks her to say security, she can say security aloud and get where she needs to go.
Touch-Tone Self-Service -- the caller presses a button on his telephone when prompted, as opposed to verbalizing the prompt. For example, he can press 1 for billing inquiries, press 2 for an account balance, and so on.
Usability Testing -- the phase in which actual callers test IVR systems before general release for a company’s use. A critical point in the design phase, usability testing
can catch unforeseen flaws that would otherwise wreak havoc on the IVR’s success.
Voice User Interface (VUI) -- the term describing the interaction with computers through a voice/speech platform in order to initiate an automated service or process. A VUI is often referred to as the heart of any speech application.
Editorial Assistant Christopher Musico can be reached at cmusico@destinationCRM.com.