Improving Customer Service with Speech

Customer Service by Phone For the average customer, across the vast majority of industries, telephone based customer service today usually involves interaction with an IVR. Most call centers seek to reduce human-to-human interaction to a minimum, typically and reluctantly providing it only upon failure of their IVRs. This isn't a problem for those customers who are willing to interact with a machine. But many people do not like IVRs and will actively seek to avoid or circumvent their use.   Why is this? The answer is simple: for most users, IVR still means DTMF. As such, IVR has come to mean, for far too many people, poorly designed, illogical and annoying systems that are often incapable of providing the service that these people sought when they picked up their phones.   None of this needs be. DTMF systems can be well designed, logical and a pleasure to use. They can also provide fast, efficient and accurate information for the majority of their users in many application domains.   While all that is certainly true, I do not believe that a well-designed DTMF system can in any way compete with a well-designed speech system.  

But Can Speech Improve Customer Service by Phone? Can speech improve customer service by phone? Simply put, yes. But what does "improved customer service" really mean? The simple answer "yes" actually belies a number of technological advancements and voice user interface design values.  

Facts and Values First of all, speech recognition has improved so dramatically in the last few years, that I now recommend its use even for numerically intensive and even numeric-only applications that were heretofore thought to be better suited for DTMF. Up until fairly recently, speech was thought to heighten the probability of errors during numeric data entry, relative to the incidence observed in DTMF. The basic argument was that errors only occur in DTMF entry when the caller actually gets "fat-fingers," (hits the wrong key).   However, I can attest from personal observation that speech recognition accuracy rates are currently essentially equal to DTMF "recognition accuracy" rates. For example, in a dictation test using one of the industry's leading ASR engines, I submitted approximately 100 ten-digit numbers to an un-tuned application and I did not encounter a single error. The active grammar was not a finite set of ten-digit numbers. The grammar would permit any digit in any ten-digit sequence, a huge search space of 1010 possibilities. These tests were performed using landlines, cell phones and even speakerphones.  

So What Does "Improved Customer Service" Really Mean? In the context of speech-based IVRs, it means solving the caller's problem more efficiently, more conveniently and more effectively than any other alternative that is readily available to the caller. But this cannot be accomplished with technology alone. An IVR affording improved customer service as such would have to be designed to minimally possess all of the following properties:  

• Ease of use: Ideally a design should model or simulate the way the task would be done with people.
• Speed: Turn taking and chatter should be reduced to a minimum.
• Choice: Users should have various ways to perform their tasks and applications should support DTMF as an alternative and backup.
• Professional production quality: The IVR should appear polished and smooth.
• Efficiency: IVRs should never ask a user to enter something more than once and they should continuously signal the user that he is moving toward his task completion.
• "Benevolent Punt Policy": If speech is not working for a particular caller, get the fellow to a CSR before he has a chance to get mad! As a general rule: 2-3 strikes and you're out. ("You're saved" is actually more accurate).

Back to Basics

Too often designers forget that nobody seeks customer service unless they have a problem. They want something or they need something. They need some information; they want to order a product; their product has failed; their bill is fouled up; their claim isn't paid; or their account isn't balanced. What's more, most people would prefer to speak to a person who will quickly and politely solve their problem.   The advantage of speech, now being evinced in many well-designed VUIs, is that it can solve people's problems as fast or faster than humanto- human interaction. And that can translate into a considerable improvement in customer service.