SpeechCycle’s Persistent Optimization Process (POP) regularly updates deployed applications to ensure that caller satisfaction and call resolution metrics are maintained. Data transcribed from actual calls is analyzed on a regular basis for the life of the application. Content modification within the scope of the application as well as voice recognition performance metrics are considered within 3 areas:
- Speech Performance Analysis – Is recognition as accurate as possible?
- Dialog Evaluation – Is caller experience optimal?
- Call Driver Analysis – Are there additional call drivers that should be automated?
Analyze – Logs are mined, using SpeechCycle’s Call Explorer, to identify problematic areas of a call flow. These are targeted as potential opportunities for improvement, both for completion rate and caller satisfaction. This analysis also identifies calls that are symptomatic of other possible issues.
Listen – Listening to a selected number of full call recordings helps in understanding the potential issues that affect the caller experience.
Transcribe – A large number of selected utterances (a collection of things people say in answering a question) generates a wealth of annotated speech used to improve speech recognition rates as they relate to speech grammars and statistical language understanding models. New trends and problems are also detected through a detailed analysis of the transcriptions and annotations.
Improve – The process of analysis, listening and transcription combines to produce continuously improved versions of the automated agents. Once an improved version is deployed, the cycle of continuous improvement begins all over again – expanding the breadth, depth and quality of the caller experience, call after call after call.
