By Laura Kennedy and Chris Lotspeich
Your call center has decided to implement a speech application, so where do you begin? Speech recognition applications come in all shapes and sizes, from simple call routers to complex ordering systems. The key factor that designers need to keep in mind is ease-of-use. Even with a complex system, the caller must be able to navigate through the system easily, with the feeling like the system truly “understands” them.
Speech applications on the market today boast recognition accuracy in the high 90%, yet why do so many people feel that “speech doesn’t work?” More often than not, it’s because of poor application design that confuses and frustrates the caller, leaving your company with a fruitless investment.
Getting Started: The bulk of the initial work is in the design phase, which includes building the call flow, creating grammars, recording prompts, and usability testing. Speech application designers will continue to modify each aspect throughout the design and internal testing phases.
Defining the Call Flow: Begin the definition process by gathering call data from the people currently answering the phones. Find out what questions or interactions are being handled that could potentially be automated. Also, find out why these are not being supported by your touch-tone IVR system, if you have one. Not all portions of a phone call match well with speech recognition abilities, so don’t force it. Draw out what you feel is a natural progression of call flows and then share this with others to get “fresh eyes” on it. In addition, try role-playing with another person acting as the speech application and see if what the “system” is asking you really makes sense. Finally, collaborate with a speech partner that has technology and industry experience and provides excellent customer service.
Grammars and Prompts: Prompt design and grammar development go hand-in-hand. Designers should decide how much of a “natural language” system callers need or desire. A “How may I help you?” style prompt is effective if the callers know exactly what they want and the speech designer can accurately predict what the callers will say. Usually, however, callers will need some parameters or clues as to what to say. For example, the “How may I help you?” question involves more extensive grammar development and testing than a question like, “We offer three choices: A, B, or C. Which would you prefer?”
The application designer must also keep in mind that there is a constant balancing act between keeping the system as conversational as possible, and not letting callers interact with the speech recognition system like a human. When callers treat the system as if it actually understands them or the grammar and prompts do not match the callers’ needs, then callers get lost or make requests which are completely outside the system’s capabilities. Problems like these can rapidly compound themselves, which is why it is prudent to focus on these steps during the speech application design process.
Usability Testing, with Real Callers: With the initial development and design finished, the ultimate measure of an application is in the first live deployment of the system. It is critical that this first live deployment be a test version with actual callers of the system, as opposed to the programmers who are intimately familiar with the application’s design. This will be the first time assumptions about caller behavior will be seriously tested. The resulting data is invaluable for adapting the application to meet callers’ exceptions and behaviors.
Conclusion: It is important to remember that the goal of any speech application is to allow your callers to accomplish their requests as quickly and easily as possible – and, of course, to reduce operating costs for your organization.
Laura Kennedy is Director of Communications and Chris Lotspeich is Director of Marketing at LumenVox, a Speech Recognition company. Call 877-977-0707 or email info@LumenVox.com.
[From Connection Magazine – April 2006]