Even with well-designed telephone scripts and call flows, personalization from caller history, more accurate ASR, and better grammars, most speech applications are still static and don’t adjust in real time to caller dynamics.
Try as we might to design great systems, each individual has his own pattern of speaking, level of comprehension, and hand-eye coordination skills (used for entering touchtone responses).
Each person also has his own level of concentration and willingness to use automated systems. Even the speed at which we speak varies widely, as the average English-speaking rate ranges from 130 to 200 words per minute.
Comprehension rates similarly vary, with speed decreasing as complexity rises. The calling environment varies considerably also; a power user can exhibit novice caller behavior if they are distracted or using a mobile phone with a poor signal. The result is that, although we can really improve individual caller interaction, there still exists the opportunity to greatly improve results by having the system adapt based on these real-time variables.
Adaptive Technology (such as Interactive Digital’s VUI Cloud service) is technology that takes into account the environmental variables of the call, the individuality of the caller, and her in-call behavior to tune interactions with callers. Just as a good CSR naturally adjusts to call conditions and the type of caller with whom she is presented, adaptive technology listens to how callers behave and adjusts the audio responses accordingly, thereby emulating effective human-to-human communication.
Using adaptive technology, the system can adjust the playback speed of a speech system and the caller response time thresholds that trigger a change in speed or message content. Thus, with experienced callers, playback speed can be increased gradually throughout the call. Similarly, inexperienced callers can trigger the system to slow down the audio playback rate and provide a greater level of detail in the spoken dialogue. In essence, adaptive technology listens for signs from the caller that she understands what is said and is comfortable with the speed of the dialogue.
The software works with both VoiceXML and proprietary IVR solutions, and doesn’t rely on information gleaned from customer databases, web-based profiles, or caller automatic number identification (ANI). Adaptive technology continuously monitors caller behavior, including speech and DTMF input speed and accuracy and the length of time it takes for the caller to reply to a prompt.
Initially, the software listens to callers to gather a calibration sample to profile caller behavior. After gathering sufficient data, it applies that data to automatically adjust the number of words per minute or the voice message content delivery to callers of particular skill levels, in real time. Profiles are built for different types of callers against which caller behavior can be matched.
For typical B2C Retail, Financial, Travel, Medical Insurance and Government applications, analysis using 95% confidence intervals indicated improvements in IVR Utilization of about 17.24-20.44%, a reduction of First-Attempt Caller Input Errors of about 1.02-1.75% (relative reduction ranging from 4.7-8.0%), an increase in Average Handle Rate of about .5-3% and reductions in Average Handle Time of about 6-16% when incorporating adaptive functionality.
