Zeroing in on Opting Out
As DTE Energy knows all too well, speech isn’t always the answer. In fact, it can be a business problem’s root cause.
The Detroit-based energy provider, like so many other companies, thought that a speech-only interactive voice response (IVR) system for its 9 million yearly callers would not only reduce call-handling rates, but also improve service. Its slick system, however, ran into two major, unforeseen socioeconomic problems along the way. First, because of a statewide economic depression affecting its customers’ ability to pay bills on time, the provider’s toll-free number received a higher-than-normal volume of calls tied to payment questions that a machine couldn’t answer. Second, Michigan began to see an influx of Middle Eastern immigrants, many of whom have accents that confused the speech recognizer. The result: Satisfaction rates plummeted while call-handling times increased.
Speech alone was not the solution. DTE Energy needed to give its customers more choices. For the company, it came in the form of keypad-operated dual-tone multifrequency (DTMF) capabilities layered on top of natural language and directed dialogue.
According to Patrick Duffy, DTE Energy’s customer service manager, 60 percent of the system’s users now use touchtone, and the average call-handling time has been reduced by 30 seconds.
"People will use the mode that best suits the situation they’re in," Duffy says. "Folks who are calling on cell phones and driving are more comfortable using speech; likewise [people] don’t want to say bank or credit-card numbers out loud in public settings."
Given DTE Energy’s situation, is a purely speech-driven IVR ever the answer? Despite improvements and innovations in natural language processing, many analysts think not. While many companies have strayed from DTMF-only interfaces, most industry consultants and analysts today champion the bimodal approach—placing speech and touchtone side by side. In fact, most callers don’t like being forced to use a company’s self-service system, a complaint lodged by 76 percent of more than 4,500 end users responding to a Genesys Telecommunications Laboratories survey last spring. That pressure ultimately led survey participants to a poor customer service experience, and many said that it would affect future dealings with a company.
This comes as a low blow at a time when companies face increasing pressures to cut costs, improve customer interactions, and stay on the cutting edge of speech in their contact centers. In addition, companies must carve a distinct place in the market through a unique IVR, while at the same time showing end users that working with a machine does not need to be a frustrating experience.
Drew Kraus, research vice president of enterprise communications applications at Gartner Research, says some companies with poor IVR systems have found ways to avoid facing these challenges altogether. "Companies have a level of customer angst, as [customers] aren’t sure if they’re entering the right information or option, and then they get frustrated," Kraus states. "After a while, the company trains customers how to opt out of an IVR, cutting [their] losses and teaching the customer the expedient way."
But it does not have to be this way. Putting in the hours and demonstrating a commitment to an enterprisewide solution just is not enough. Customer opt-outs will happen, but that should not mean designing a system that leaves end users rushing for the zero key. A strong deployment means keeping a constant ear to the ground and listening to customers, and designing a system that is always one step ahead of the end user. With the right combination of technologies, customer insight, testing, and monitoring, opt-outs will decrease.
But before this process can start, a company must learn why customers abandon its system. And companies should care. The more customers zero out of an IVR system, the more money companies spend on agents and lose on their significant investment in the application.
According to vendors and analysts alike, the primary reasons end users transfer to an operator stem from a handful of areas: poor system design, lack of faith in the system’s capabilities, poor error handling and recognition, and a real IVR nightmare—infinite loops.
Getting to the ‘Why’ Factor
During a session about phone service at the 2007 Nuance Conversations user conference, presenter Stephen Springer asked attendees to raise their hands if they had ever used the following phrases in their IVRs: Please listen carefully as our menu options have changed; This call may be monitored for quality assurance and training purposes; or Please visit our Web site for faster service or lower rates. A low rumble of chuckles filled the room, and nearly everyone in attendance begrudgingly raised their hands. Springer, Nuance’s senior director of user interface design, went on to explain that companies must not only go beyond traditional methods to give customers top-notch service, but that IVR dialogues should not be weighed down with time-consuming information, most of which a customer already knows prior to the call or doesn’t need to know to complete his transaction.
Operating a contact center in the Care 2.0 world means assuring callers that their needs are important by designing an IVR that is efficient and saves them time. Still, many companies face the challenge of uprooting older systems; usually, this means switching from DTMF to a speech-enabled IVR with directed dialogue or natural language, or a bimodal application similar to DTE Energy’s. Making that migration, however, is tricky and can actually decrease customer satisfaction rates if done poorly.
When examining IVR opt-out rates, a corporation could take pointers from a May 2007 report published by Moira Dorsey for Forrester Research. In the report, titled "Best and Worst of Phone Self-Service Design," Dorsey examined 16 IVR systems across four industries, and identified four major design flaws: value (missing essential content), navigation (inefficient task flow), presentation (poor production quality), and trust (no access to human assistance at key points).
To demonstrate a system’s value to the customer, the IVR must enhance the company’s existing outlets for customer care, whether it’s chat, a Web site, or video. In a multimodal landscape, an increasing number of customers come to the contact center after having thoroughly examined these other help outlets, and expect an answer to complex problems through the aid of an agent. But before the end user can get there, an IVR must be designed so information and options are presented clearly and logically. Aside from designing an IVR from a customer, not company, perspective, organizations may also start by automating simple transactions to speed up the process for the end user.
"[The system] must be clear, concise, and mutually exclusive," according to Mark Camack, vice president at Marketing Strategies International. "If those are the bellwethers of your menu construction, with a hierarchy of sequencing based upon volume and importance to the customer, you can lead the horse to water."
That’s how it works at the typical bank’s branch offices. When a customer enters a bank, odds are that he’s making the trip to deposit a check, not to inquire about mortgage rates. Thus, sites are designed with tellers in front to perform mundane activities, and a handful of specialists in separate cubicles to provide more detailed information. This structure should not end in the bank’s brick-and-mortar sites, but all too often it does.
In the same way, when a customer calls into a financial outlet’s contact center, he is most likely dialing in to check a bank balance, transfer funds, or find a branch’s hours of operation. Why use 20 seconds of the beginning dialogue discussing new loan rates or vacation packages?
"Customers, when they get into an interface, do on occasion get frustrated," says Tom Chamberlain, director of business process marketing at Aspect Software. "Companies, as they go through the design process, have had a more inside-out view of the design. They view it more from a company and not a customer perspective. They cannot properly anticipate what those customer needs are, and [this] forces consumers to navigate it by themselves."
Or, in DTE Energy’s case, the company’s old IVR was not just failing because of poor recognition rates, but also in system architecture. Customers simply could not accomplish tasks quickly with a system designed from the company’s point of view. "The structure of the menus wasn’t intuitive to our customers, making it difficult for some to locate the option they wanted," DTE’s Duffy explains. "The first step in our IVR required that all customers, except gas leak emergencies, identify themselves to the IVR before proceeding with their transaction, meaning that they were 30 to 60 seconds into the transaction before they even told us what they wanted."
Lost in Translation
Further issues arise when a company tries to translate an existing DTMF system directly into a speech IVR. Morphing a system into a flat, hierarchical menu structure means more than making all the touchtone prompts accessible through speech; it means an entire restructuring of the system.
"Design requirements are very important, and not just as simple as converting DTMF into speech," says Roberto Pieraccini, chief technology officer at SpeechCycle. "[Companies] don’t want to spend a lot on their new IVR, so they just try to convert DTMF into speech IVR. It just doesn’t work."
For example, when switching from a five-menu IVR to a speech-enabled IVR with natural language capabilities and the same number of menus, a company will receive a greater variety of responses. Instead of choosing from fixed menu choices, customers now have the power to say whatever comes to mind. A company must closely examine its end user statistics and typical call path flows, and then build in the grammars applicable to those transactions.
But the heart and soul of an IVR system, arguably, centers around dialogue design. Deborah Dahl, president and cofounder of Conversational Technologies, and Nancy Jamison, an independent market analyst for Jamison Consulting, both cite this as the make-or-break of a successful IVR. Jamison states that a system’s technology—DTMF or speech—does not matter if the prompts and menus are well-structured.
"If you design the application so that it works well and has the information the caller needs, it doesn’t matter," Jamison says. "If people have incentive to use [IVR], and it gets them to their destination, it doesn’t matter if it’s speech or touchtone."
It also becomes vitally important to fill an IVR with prompts and menus that make sense to customers, something Dahl says can be done by analyzing calls prior to design. "Analyze your calls, pick out the top reasons customers call, and streamline with three choices presented from the queries," she states. "Organize the menus, taking into account frequency of different call paths people follow. People don’t get too mad if they have to spend a little bit more time to get across an unusual question, because they know that they’re doing something that doesn’t happen very much."
In the case of natural language, developing prompts conducive to certain responses requires careful planning. Rather than focusing on accuracy in a speech recognizer, Dahl notes that phrasing a question in a way that leads to a select number of responses is more important. She gives the "either/or" example as a typical scenario and easy-to-avoid mistake in VUI design.
"If you use, Would you like your checking account balance or your savings account balance?, you get people saying ‘yes’ because that’s like asking, ‘Would you like coffee or tea?’" she says. "With a speech recognizer, you have to reword the question so it’s not yes or no, but checking balance, so you would use, Which balance would you like, checking or savings?"
For most designers, the best way to determine if a dialogue is well-crafted is through usability testing. One of the most popular testing methods, Wizard of Oz (WOZ), tests various caller scenarios using prerecorded prompts and untrained callers. Camack says that WOZ testing, which runs on .wav files created on a computer, is both inexpensive and quick, but can still replicate the customer call experience. "They hear exactly what they would hear if it was a fully robust prototype, but there is none."
A company’s work, however, should not end at WOZ. Knowing how easily customers can navigate and understand a system is step one, but from there, an enterprise must take action. "Do a pilot. Test it out on a smaller set of your customers before you take it to a broad deployment," Gartner’s Kraus advises. "From that set, do a combination of customer feedback surveys and analytics."
After that, he says, monitoring and tuning should be done "fairly intensively" every day during early deployment stages, and scaled back to monthly reports as the system matures. To protect the IVR investment, he says an organization cannot afford to "take their eye off the ball; it’s a big mistake to not tend to the care of it."
And it’s not just VUI designers championing a careful, calculated approach to dialogue design. Michael Zirngibl, CEO of Angel.com, a company that provides hosted IVR solutions, says implementations can fall flat if a budget is constructed solely around technology. "In traditional IVR deployments, a lot of the resources go into the plumbing—the aspect of the application that doesn’t have immediate caller value, [such as] writing VoiceXML code and grammars," he says. "And there is often not much resource left at the end to focus on user design."
Some Can’t Be Swayed
Blame it on a lack of faith. It’s one of the top reasons customers choose to opt out of an IVR, and it’s practically unavoidable. There will always be a group of end users who dial into a customer service line with a preconceived bias against IVRs, and slam the zero key before the first prompt begins.
"Some callers, who are thankfully a shrinking subset, are resistant to using speech technology," admits Thomas Smith, vice president of IVR and speech product marketing at Verizon Business. "Callers feel entitled to speak to a real person."
But some callers may dismiss a near-perfect IVR for other reasons. For example, a caller’s anger with a system could stem from a call made to an organization with a clunky system just moments prior. Faced with another automated voice, the customer immediately becomes upset at the prospect of having to relive another moment in IVR hell.
Camack also notes that generational differences can play a substantial role in who will opt out and who will stick it out. "Older people have more time, and frankly, some of them don’t know the games to play," he explains. "They have a general predisposition [against IVRs], not because they don’t like technology, but because they were born into a different generation that prefers personal contact."
The problem with those older or unwilling callers is that they see little to no value in working with an IVR. They might never realize that automated self-service could help them solve a problem quickly because their psychological make-up prevents that from happening."You’re never going to cure the certain percent of people that really want to talk to a live operator," Jamison says.
But whether their distrust of the system stems from prior experiences with an ill-designed system, this group is the trickiest to coax into using an IVR.
On the other end of the spectrum—and good news for companies with IVRs—Gen X, Gen Y, and New Millennials have shown a strong preference for self-service, having grown up using the Web, text messaging, and chat. The speed of these transactions, ingrained in the mind-sets of a younger demographic, creates a general distaste in speaking with an agent, Camack says.
Inconvenient menu structures, a lack of usability testing, poor DTMF-to-speech transitions, or just a consumer’s general disdain for anything automated—the hurdles in building a strong IVR are plentiful. The good news is that only one of these problem areas is hard to fix. The harder a company works to improve the technical and VUI elements of its IVR, the more incentive it will give customers to work with a system. As enterprises struggle to make updates to and constantly tune their systems, chances are better that customer satisfaction will increase along with the IVR’s reputation.
Three Tips to Take Away
1. Choose the technology that’s right for you. Rather than being swayed by flashy applications, pick the solution that’s best for your IVR goals. "There have been a lot of illusions about the use of IVR; that natural language can solve anything," says Roberto Pieraccini, chief technology officer at SpeechCycle. Keeping directed dialogue and DTMF in mind, decide whether your business goal necessitates natural language. Will callers need to describe complex problems in their own words? Are there too many options to list up front? Natural language may work best, but don’t go overboard. Consider directed dialogue or DTMF as the next step.
2. Make IVR part of your marketing initiative. One of the easiest ways to convince callers to use your new IVR is through strong marketing. Christoph Mosing, vice president of worldwide professional services at Envox, cites a New Zealand telephone company that successfully marketed a new system by positioning it as a way to help customers. The company printed "call codes" on direct mail, showing customers which buttons to press to solve different issues. "You’ve got to explain to your caller population what it is that you want to achieve with the application—more consistency, increased speed, etc."
3. Give callers a way out. Transferring calls to an agent costs a business, but can also help prevent low customer satisfaction ratings. Give customers a way to get to agents by designing a system that instructs them how to reach an operator if needed. "If you enable your IVR so that zero doesn’t get a caller to an agent, customers think you’re hiding your reps," says Mark Camack, a vice president at Marketing Strategies International. "If customers struggled, heard the same menu twice, and had to go back to the beginning of the system, that’s the time to tell your customers how to access an agent." —L.S.
Stats Can Speak Volumes
>> The good news: Most customers don’t hate all IVRs. The bad news: They will remember and avoid companies with difficult systems. In a recent Harris Interactive Customer Experience Survey, 72 percent of participants stated that they stopped doing business with a company due to poor automated self-service applications.
>> In the same study, speech won over DTMF by an average of eight to one, and 86 percent of respondents said they would use an automated speech system if notified of its availability in advance.
>> Ignoring an IVR’s role as the voice of customer service can hurt business—and the majority of companies fail to understand the importance of a positive call center experience. In a 2007 Aberdeen Group study of contact centers, only 36 percent of "Industry Average" firms aligned their contact center goals with overall company goals. —L.S.