This feature is the first in a series of
CRM magazine articles examining the role of interactive voice response (IVR) systems in CRM and contact centers. The next piece -- "Listen Up!,"
magazine, January 2008 -- suggests proper ways to design the actual scripts of an IVR.
Few customer-care technologies have a more compelling ability to trigger a radical shift in both customer experience and contact center operating costs than interactive voice response (IVR) systems. IVRs with the ability to recognize human speech -- commonly known as speech-enabled IVRs -- pick up where their touch-tone or dual-tone multifrequency (DTMF) counterparts leave off. But just because customers can talk to the application doesn't mean it's a panacea to prevent their frustration. In fact, the design of the system is even more important when speech technologies come into play, and voice user interface (VUI) design continues to serve as a main hindrance to stronger satisfaction rates.
Designing a speech-enabled IVR system is inherently complex and requires in-depth preparation. What follows, though, are 11 process-oriented strategies to make your road to speech-recognition nirvana a little less bumpy.
1. Go From Zero to Hero
A huge chunk of customer frustration with speech-enabled IVRs stems from feeling trapped in the system. Keep aggravation at bay by providing callers with the ability to opt out of the application and to get connected to a live rep -- that's how you align with customer-centric service.
"What we're looking for," says Dan Miller, senior analyst and founder of Opus Research, "is a closer integration of the speech application with the call-routing application and linkage to the business logic that supports the goals of CRM."
But don't think you're finished just because you've added a "zero-out" option. If an agent isn't immediately available, update callers on expected hold times. Once a caller is connected to a rep, make sure the caller doesn't have to repeat any information already provided to the automated system. "The whole integration of the back-end infrastructure [has] to make sure the agent who receives the call has access to what the caller has already done using the speech application," says Vijai Shankar, senior product marketing manager at Genesys Telecommunications Laboratories, an Alcatel-Lucent company.
Robby Kilgore, creative director of professional services at Nuance Communications, stresses the importance of what he describes as letting callers "see around the corner." "Context is king," Kilgore says, "and that's really about maintaining the context around the caller so that callers understand that they are, for example, on step three on a five-step process." Kilgore adds that context contributes to "lower call friction."
to view full-size image.
2. Connect with Customers...
You may think you can put yourself in your clients' shoes, but there is no substitute for talking with your customers for first-hand insight into the customer service experience; you'll learn valuable information like why they're calling, what they're trying to accomplish, likes and dislikes about the application, and what they'd like to see incorporated into the system. "If you don't understand your own customers you're never going to build a speech application that will meet their needs," says Tom Hanson, senior manager of the Avaya speech self-service and IVR solutions.
3. ...and with Reps
Customer service reps are typically the employees who interact most often with your customers. So why not leverage their expertise? Obtaining feedback from the agents will provide tips for improving the application.
In addition, Erin Smith, senior VUI designer at Convergys, suggests getting a recording of actual calls between reps and customers -- and listening to the interactions. Agents, she says, "provide great insight [into] how customers currently feel and how we can continue to [build] strong relationships within an IVR with their callers."
Dave Pelland, director of the Design Collaborative at Intervoice, recalls an instance where his firm worked with a mortgage company handling at-risk loans. Initially, the plan was to authenticate each caller and then ask what that caller wanted to do. Intervoice, however, has a process it refers to as "discovery," which includes going to the client's location for a handful of days, talking to executives and service reps, and listening to calls. Pelland, listening in, noticed that several callers were not account holders but instead were real estate agents calling for account holders, in order to get a payoff quote. If not for the discovery process, Pelland says, Intervoice would never have identified that kind of user, one the mortgage company never prepared for. As a result, he says, "we ended up actually changing some of the early requirements to handle this additional user."
4. Remember That Speech Does Not Equal DTMF
Thanks to standards such as VoiceXML (Voice eXtensible Markup Language) and SALT (Speech Application Language Tags), speech recognition applications go beyond the boundaries of touch-tone systems, enabling customers to carry out tasks too complicated to handle within a DTMF environment. Improved vocabulary recognition also plays an important role. With directed dialogue, callers are guided with fact-based, closed-ended questions like "What is your ticket number?" Natural language dialogue, however, allows for more open-ended responses such as "I want to book a flight from Philadelphia to Chicago on January 1 and fly back to Philadelphia on January 24."
But companies are barely scratching the surface with speech recognition deployments if they rely on the same interface-design principles they used with prompts such as, "Say '1' for account balance, say '2' to transfer money." "That approach doesn't add value," says Steven Pollock, executive vice president and cofounder of TuVox. "You're not taking advantage of the technology."
Approaching application design this way is a sign of "not rethinking your interaction or your business process that you're taking with your application," says Tom Chamberlain, director of business process marketing for Aspect Software.
5. Be Smart about Automation
It makes sense to automate frequent tasks such as obtaining stock quotes and making address changes, but steer clear of high-stress situations and tasks that are more complex or used less frequently, maintains Bern Elliot, a Gartner research vice president.
"If you're talking [about] some sensitive material as in someone's health -- maybe some health problems they're having -- they need some empathy and sympathy from the person on the other end; they don't really want to talk to a machine," says Paul Wirtz, Aspect's senior director of solution services. Sometimes, he adds, the reverse is true: "It might be embarrassing and they want to check the status of something without having to interact with a person. You really have to look at each of the applications and see what the right fit is."
6. Understand the Limitations of Speech
While it's clear that speech technology expands what callers can do through automation, shake the notion of any extreme boost in the percentage of self-service interactions as a result of adding speech. Instead, aim for a modest target. "If today 20 percent of the people are being handled by the IVR and you put in speech, it's not going to jump to 95 percent," says Sheila McGee-Smith, president and principal analyst of McGee-Smith Analytics. "But you'll get [another] 15 percent, 25 percent -- and that's good. Don't expect that speech is dramatically going to get you near to 100 percent because you'll be disappointed."
7. Test, Test, Test
Deploying a speech-enabled IVR may seem likely to bolster the quality of customer interactions, but taking a lackluster approach to testing the application will shred any chance of strengthening satisfaction levels. Andrea Holko, Intervoice's senior vice president of global consulting services, puts the importance this way: "Usability testing, to our designers, is the same as quality assurance testing to the developers. It's the only way we know before we go to deployment that what we designed meets [the] end users' needs."
Additionally, there still must be ongoing governance of the system to address day-to-day issues. "It constantly needs improving and tuning," says Daniel Hong, lead analyst of customer interaction technologies at industry research firm Datamonitor. And, by closely examining the metrics produced by the IVR platform, "companies are able to isolate where the user experience is going south," says Miller, of Opus Research -- and then they can make changes to the application.
Greg Simsar, vice president of speech services at Syntellect, suggests that companies listen to live calls or leverage a call-monitoring application once the system goes into production to get a tighter grasp on how callers are interacting with the system. "If you do quality monitoring on your CSRs, why wouldn't you be doing it on your automated calls as well?" he asks.
Part of the plan must also include adopting a gradual strategy -- rather than a big-bang approach -- to rolling out a speech application. Hanson cites an Avaya customer in the healthcare industry as an example. When the company rolled out its application, it did so to a small subset of its customer base, informing them that they would see a change in their interaction with the voice system. The strategy produced useful feedback on customer sentiments about the application. "It really is rolling it out using a strict change-management methodology to a small group of people, understanding their likes and dislikes, adjusting, and expanding it to a larger portfolio of customers as it's rolled out," Hanson says.
8. Get Personal
Like any component of a customer-centric strategy, using a one-size-fits-all formula will make initiatives stale. Use customer data to deliver a more tailored call flow. "In the best system it's based on their [Automatic Number Identification]/Caller ID, so that even before the system plays the very first prompt it knows who the caller is," says Steven Brown, vice president of client services for Angel.com. "But if the Caller ID is not used and it's some other identifying piece of information -- might be an account number -- then use that information to provide a personalized set of menu options back to the caller."
Consider the path taken by Australia-based media and entertainment company Village Roadshow, which operates multiplexes in various countries. In June 2005, VR launched VoiceWeb Ticketing, a hosted speech-enabled IVR built on Envox Worldwide's Envox Communications Development Platform and Nuance's speech recognition functionality.
The deployment features automatic town recognition based on phone number; customer identification/credit-card clearing; natural language understanding featuring multiple commands; and call logs and statistics. Once a caller has been identified, the system can pose tailored questions such as whether the customer would like to use the credit-card number previously used, or order the same number of tickets as before. Through the application, customers can then purchase tickets and place orders quickly; sidestep lengthy queues by providing movie information; and select seats.
During the last two years, the VoiceWeb application has served more than 500,000 unique users and automated about 80 percent of customer calls.
9. Use Professional Talent
All too often, companies miss a key element of the overall feel of a speech system: the particular voice used for the recordings. Use professional talent to add to the finesse of the system. "What touches the caller is the quality of your recordings, so it's very important to have a professional recording done and to use as much as possible the same voice talent across the speech application rather than different voices," Genesys' Shankar says.
10. Create User Profiles
Not all of your customers are alike, and speech recognition initiatives often require an understanding of the demographic makeup of your customer base. Design profiles to describe user segments, attaching characteristics that will provide customer visibility. Convergys' Smith says that, on average, three user profiles are probably enough. Assign identifying traits to make it "a living and breathing person," Smith says. "You create these people and you create a true meaning behind the term user."
Take, as an example, a utility company. "Reggie" may be a consumer with one residence, "Elizabeth" may split her time between two units, and "Marie" may represent commercial property owners. "The whole idea is to try and get the company to come at it from a user point of view," Syntellect's Simsar says. "Going through that process really opens up your mind to the different types of people calling your system."
11. Examine the Multichannel Mix
For most companies, the days of interacting with customers solely through the telephone are a distant memory. With the emergence of channels such as email, Web chat, and online self-service, call centers attempt to cultivate customer relationships while evolving into multichannel, multipurpose contact centers. Rather than looking at a speech-enabled IVR in a vacuum, scrutinize how customers interact with your company through all of the service channels offered. But pump the brakes if you're thinking about applying the same design principles for a graphical user interface to a VUI.
For instance, a Web site can feature a pull-down menu with multiple options, but that structure in a phone self-service environment would cause chaos. Where it makes sense, though, take a page from the playbook of other touch points -- perhaps with something as simple as making sure your speech-enabled IVR uses the same terms as those listed on the Web site, and in the same order. "You get ideas of how the customer is used to being communicated to," Smith says.
Moreover, seamlessly integrating channels, although a daunting task, is a key ingredient to not just the success of a speech application, but to customer satisfaction overall. "There needs to be a bigger emphasis on the consolidation of consumer data," says Ken Landoline, senior analyst for Yankee Group. With a consolidated database, he adds, customer activity ends up in one bucket, as it should, regardless of whether that activity took place on the Web or through an IVR, and regardless of whether that IVR had speech capability or not. "On the front end, that's recorded in the same place, so there's always a 360-degree view of the customer across the channels," he says.
SIDEBAR: Shaking the Magic 8 Ball
Ken Landoline, senior analyst for Yankee Group, is the author of "The Evolution of Voice Self-Service Systems to Web Architectures" (February 2007). Here is an excerpt of his take on some of the important developments on the horizon:
- Broader availability of packaged voice components. Pre-built voice self-service components dramatically shorten the upfront development cycle and have great value. Although the voice self-service industry hyped the concept of packaged speech applications during the last few years, we have now entered a phase with more realistic expectations. Components are used to speed the application design and deployment process, but they don't really reduce the lifecycle management expense, similar to traditional packaged software applications.
- Automated sources for voice component development. Transcribing recorded calls or text-based information (e.g., FAQ files) into voice components not only can reduce the initial development expense, but also can expedite the ongoing tuning. Leading vendors are applying extensive linguistic modeling tools to analyze a large base of spoken or written communications and are automatically generating voice components. Historically, developing voice applications required the "brute force" method, where developers manually mapped out what words are being said and in what order. Although these tools only do part of the job of developing a speech application, they can help reduce the TCO of managing speech.
- Adoption of Voice over Internet Protocol (VoIP). VoIP further simplifies and reduces the hardware costs of voice self-service applications. Along with breaking apart the voice self-service system into a Web-based three-tier architecture, VoIP allows IT organizations increased flexibility in network designs for these systems.
- Increased usage in hosting and on-demand options. The voice self-service industry has a long track record of both enterprise-owned and service-provider-managed options. With the movement toward Web architectures, enterprises can more easily choose which tier to outsource and which to retain behind their corporate firewall.
- Voice self-service becoming another applications server. Voice self-service platforms and applications are moving rapidly in this direction today, and this trend will continue.