IVR Design Lessons From the Fortune 500
IndustryView | 2014
Interactive voice response (IVR) systems are both crucial for call centers and an infamous pain point for consumers.
An IVR is a system that enables callers to complete tasks by navigating menus using their voices or keypads. IVRs can reduce the call volume that call centers handle, but they can also alienate customers if poorly designed.
To identify IVR design best practices, we called the IVRs of 50 customer service-oriented companies in the Fortune 500 and analyzed how they’ve designed their systems. We then spoke with specialists in operations research to get their thoughts on IVR design.
One of the basic purposes of IVR systems is to help customers find answers to their questions and complete routine tasks on their own. When a customer “zeroes out” (i.e., dials “0” or speaks a response that takes them to a human agent), this purpose is defeated.
Zeroing out is a serious problem for call centers. A Purdue University study discovered that 63 percent of consumers say they’ll stop using a company’s products after a bad experience with an IVR system.
Nitzan Carmeli, a master’s student in operations research in the Faculty of Industrial Engineering and Management at the Technion—Israel Institute of Technology, explains that the detrimental impact of callers zeroing out is why “many organizations just count how many customers end a call without speaking to an agent.”
On the other hand, for some businesses, it’s critical to ensure that IVRs allow callers the option to speak to a human.
Bruce Belfiore, CEO and senior research executive at BenchmarkPortal (a leading firm in call center benchmarking, training and certification), explains that “whether it’s a good option or not depends on your business model. If you’re the Ritz-Carlton, you want to put that option right up front. ... If, on the other hand, you’re a low-cost provider and you’re trying to train your caller cohort to utilize those channels that cost you least, then you’re going to bury it a little further down.”
Additionally, he notes that IVR design involves continual experimentation: “How far down [the option] goes is something you can actually play with in a scientific way—try to find out how much putting the option right up front increases your call volume and how much burying it further down increases the number of hangups.”
Eric Tober, principal speech solutions scientist for Interactive Intelligence (a leading supplier of contact center solutions), notes that this sort of experimental approach is especially necessary for systems that depend on speech recognition.
“When deploying a speech recognition system, it’s always necessary to ‘tune’ (or optimize) an application using live caller data,” he says. “Every application is unique, leading to callers responding in unpredictable ways in variable audio conditions. Due to these factors, tuning is required in order to achieve peak performance and maximum [return on investment, or] ROI.”
To help businesses arrive at a rough estimate for when to provide the option to zero out, we repeatedly pressed zero for dial pad IVRs or requested to speak to an agent for voice recognition IVRs until we could reach a human being. Here are the results:
Nearly one-fourth of the IVRs that Software Advice called did indeed provide the option to speak to an agent in the first menu. The majority of companies, however, forced callers to wait until they’d reached the third menu before allowing them to zero out. On average, we had to navigate 2.94 menus before IVRs sent me to an agent.
That said, few IVRs bury the option to speak to an agent deeper than four menus into the system. Not a single IVR that we called forced callers to wait for more than eight menus, though one IVR did repeat the same menu endlessly instead of allowing me to speak to a human when we pressed zero. The fact that only one company out of 50 used this tactic indicates that the decrease in call volume it offers isn’t worth the massive increase in customer frustration, as callers vainly struggle to reach someone who can help them to resolve their problems.
Eric Soderstrom, speech solutions architect for Interactive Intelligence, explains that, as a rule of thumb, “the ability to speak to an agent should be available—but not advertised—on every menu.” This tactic helps keep call volume low, while allowing callers who are having trouble with the system to get help from a human agent as soon as they need it.
Ultimately, Belfiore argues that the “magic metric” for call centers to track is not when customers zero out, but rather a metric known as “first-call resolution.” This term means that a caller is able to resolve a call on the first try, without having to either call back or be called again.
First-call resolution is important, Belfiore explains, because his firm’s research has shown that it is strongly correlated with customer satisfaction. In other words: by tracking this metric, you can understand how effectively your IVR is helping to solve your customers’ problems.
There’s an ongoing debate among specialists in IVR design that concerns both the number of options the initial menu in an IVR should contain as well as the depth of the IVR’s menu structure (i.e., how many menus a caller navigates in order to complete a task).
The conventional wisdom on the subject, Belfiore notes, is that the branching tree structure of an IVR “should be no more than five [options] across [in the top menu] and three [submenus] deep. In other words, you give up to five options to callers, and when they push one of those, they can go down as many as three submenus.”
You don’t want the structure of the IVR to be more than three menus deep, Belfiore continues, because numerous menus can be frustrating for callers to navigate. By sticking to three menus, you can provide a comfortable experience for the caller—a principle Belfiore calls “psychological ergonomics.”
He notes, however, that conventional wisdom may not apply in every case. AARP members, for example, require different psychological ergonomics than customers who are business-to-business (B2B) marketing experts.
Additionally, Carmeli notes that creating a deep menu structure can prevent callers from reaching the services that the IVR offers. Using data sources and analytical tools provided by the Service Enterprise Engineering Center at the Technion, she created mathematical models of “customer flow within an IVR system.”
She explains that “in the system we analyzed, there were about 40 service options, but most were rarely visited: Only three services were reached in more than 5 percent of the calls, and only eight services were reached in more than 2 percent of the calls.”
While, on the one hand, you don’t want callers to get lost in a maze of menus, you also need to avoid overwhelming them with options in the top menu. As the number of options increases, callers can quickly become confused.
Carmeli adds that many IVRs don’t need to provide dozens of options anyway, because most callers end up settling on just a handful of them.
“In the IVR we analyzed, about 70 percent of the customers asked for very simple, general information services,” she says.
While the number of options in your own IVR menu should obviously depend on your business model, we counted the number of options in the top menus of each IVR we called to arrive at some rules of thumb.
Most IVRs provided between two and five options in the top menu. The average number of menu options for my sample (excluding those IVRs with open-ended questions and introductory requests for information) was 4.16. But what if you need to offer callers more options in the top menu?
The businesses that didn’t provide any options also didn’t utilize a top menu structure. Two of the companies we called simply prompted me for an open-ended response instead of offering a menu.
Tober explains that this sort of structure, known as “natural language understanding (NLU),” uses “statistical language models and call routing [to] greatly expand the number of distinct selections (150-plus) a caller can make, while enabling a more natural, open dialog.”
Such structures are ideal for companies that offer numerous services through their IVRs, he adds, since they streamline IVR systems by effecting “greater flattening of the menu structure for many callers.” In other words, callers don’t have to go nearly as deep into IVR submenus to complete a task.
Companies that offered only one option generally asked me to input information such as an account number or phone number before sending me to a deeper menu with service options. Obviously, whether or not you should utilize such a structure depends on your business model and your industry. Beginning an IVR with a request to input information is standard in financial and telecommunications services, for instance.
For longer menus, Tober says, you can “break the dialog into multiple menus, [which] can be activated by giving the caller a ‘more choices’ option at the end of the first menu.” He adds that this tactic “helps reduce the ‘cognitive load’ effect, in which callers get quickly overwhelmed when offered more than five options.”
Four of the IVRs we called featured “more” or “next” options in the top menu.
In order to get better data on the depth of IVR menus, we also counted how many menus we had to navigate before we reached self-service options (i.e., options that allow callers to complete tasks such as booking a flight or paying a bill).
Unsurprisingly, the vast majority of IVR systems placed self-service options in the first two menus (94 percent). Only a tiny fraction forced callers to navigate three menus to reach them, and none of the IVRs we examined buried service options more than three menus deep.
IVRs almost universally include introductory messages identifying the company that the caller has reached. While such messages can be great opportunities for promoting your brand, they can also bore and frustrate callers. Thus, the vast majority of the Fortune 500 companies we called kept their introductory messages under 7.9 seconds.
It’s evident from this chart that even those companies that chose to include longer messages kept them under 13 seconds. The longest introduction we encountered was a full 33.5 seconds, which can seem like an eternity when you’re trying to complete a task over the phone.
An even more important consideration is the length of menu option recordings, since callers may become frustrated if the IVR forces them to wait for the option they’re searching for as it reads out lengthy descriptions of other services. Thus, we recorded the length of each top menu option and took an average for each IVR that we called:
Unsurprisingly, companies chose to keep menu options even shorter than introductions. The average menu option was four seconds long. Only two IVRs reached an average menu option length greater than eight seconds. The shortest menu option we recorded was only 0.8 seconds long, whereas the longest extended to an agonizing 19 seconds.
Though the term IVR includes the phrase “voice response,” the majority of IVRs that we called in fact rely on the dial pad as the sole input method.
Only 28 percent of the IVRs we called offered voice response as the sole input method. Most IVRs with speech recognition capability also allowed callers the option to switch to dial pad input.
Many consumers tend to like dial pad input, as speech recognition has historically been a bug-prone technology. However, there have been a number of technological advances on this front. If you’re considering implementing voice response in your IVR, then you’ll need to look for a solution with advanced speech recognition features.
Soderstrom explains that applications such as Interactive Intelligence’s ISR (Interactive Speech Recognition) can help with simpler recognition tasks, such as strings of digits (e.g., credit card and phone numbers) and “yes” or “no” responses, while other applications (such as the Nuance Communications Nuance Recognizer) assist with more complicated recognition tasks.
Most consumers have called an IVR, only to be informed that they need to listen closely as options have changed. Consumers hear this message regardless of whether they’ve ever called the system before, which may be why only four of the IVRs we called included this infamous message.
Similarly, few companies included a branding statement in their introductory message (i.e., marketing copy summarizing what the company does). Most callers don’t need to be reminded of which company they’re calling; moreover, branding statements dramatically increase the length of introductions.
Indeed, Tober cautions that “marketing messages and legal jargon overload the caller with too much information, and reduce the chances of successfully completing self-service tasks such as getting an account balance. It’s paramount that businesses minimize the number of barriers between callers and their goals.”
More companies used the introduction as a way to inform customers of recent news. The Home Depot, for instance, included an announcement about the recent payment card data breach in the introduction to its IVR system, and even offered a special menu option for customers to learn more about how to respond.
All of the IVRs that we called used human-sounding instead of “robotic” voices, such as the infamous voices popularized by MacInTalk, Apple’s pioneering speech synthesizer. Most IVRs utilized female rather than male voices (74 percent), most likely because research has shown that the human brain tends to prefer female voices.
That said, more than one-fourth of the IVRs we called employed male voices. Ultimately, the decision about which gender of voice to use should factor in the demographics of your customers.
As we navigated IVR menus, we continually ignored the options with which systems presented me in my quest to zero out (i.e., reach a human being). Instead, we dialed zero or voiced a request for an agent. Four of the IVRs we called simply hung up on us after we failed to input the responses the system presented me with.
While such a tactic can reduce call volume, Belfiore explains that it’s ultimately ill-advised to let your system hang up on a customer—as this frustrating experience may motivate customers to take their business elsewhere.
“If you’re in a customer service industry, you need to have a passion for customer service,” he says.
Soderstrom adds that unpredictable callers need to be routed to an agent rather than kicked out of the system.
“If a caller is having difficulty, he or she should be transferred to an agent after, at most, three errors in any recognition state,” he says.
Historically, IVR has been viewed as a technology that is too rigidly automated to accommodate the needs of human beings. One reason for this perception is that some businesses have been guided by the wrong goals.
Tober explains that businesses should, “first and foremost, design with the caller in mind. A lot of IVRs are focused on business goals instead of customer goals. The focus should always be on what the caller wants to accomplish.”
Belfiore concurs: “Always start with the customer—who is calling, and what are they calling about?” He adds that businesses also need to consider what the “goal” of the IVR is.
“Is it to just route the call, or is it focused on self-service applications? These goals will require different designs,” he says.
Moreover, Belfiore notes that self-service is increasingly popular as demographics shift.
"Whereas there are still lots of older callers who become frustrated if they can't talk to a person, there are also lots of younger callers who would prefer not to talk to a person," he says. "If they're getting on the phone, it's because they haven't been able to resolve their issues through the website or some other way of connecting that doesn't involve person-to-person interactions."
Optimizing your IVR system for the performance of complex self-service tasks will become increasingly crucial as callers shift to viewing the IVR as a convenient way to automate problem-solving rather than a necessary evil.
At the end of the day, your IVR system shouldn’t simply make running your contact center cheaper; it should also make your customers happier.
Software Advice called the IVRs of 50 Fortune 500 companies with business models that focus on customer service (e.g., cable companies, credit card companies, airlines etc.). Whenever possible, we called the company’s customer support number. We then navigated menus using the appropriate input methods, attempting to reach a customer service agent in the fewest possible steps. We recorded all interactions to arrive at accurate results.