1 hr 6 min
The Future is Spoken presents Phillip Hunter as this week’s guest. Phillip is an expert in strategy, design, and AI-powered optimization. He is the founder of CCAI (Conversational Collaborative AI) Services and is part of the team behind Alexa. In this episode, Phillip discusses voice analytics and performance optimization techniques. Before anything is coded for conversational AI systems, voice analytics need to occur. “Many of these systems are built to achieve something specific,” Phillip explains, and analytics—such as user engagement and monthly usage—help measure their success in achieving those goals. A common issue identified through analytics relates to recognition errors. These errors can result in users getting confused or stuck while using an application. Even when a product is well-designed and thought out, Phillip notes that there will always be unpredictable issues. Phillip talks about the goal of reaching a resolution for users. He gives the example of a customer who accesses a bank’s call centre to confirm that a specific deposit has occurred. This kind of request is relatively straight-forward to automate, but it requires AI to gather information from the user and access a back-end information system to see what is happening in a specific account. In an ideal case, the request can be fulfilled independently, using only AI. Phillip explains the importance of anticipating issues through gathering and studying data before an application goes into production. As soon as a product is live, there is the added pressure of tight deadlines and upset users to consider. Once an application is in production, the next step is to validate whether the system’s performance is on or off-target in achieving set goals.
If the numbers are not matching target goals (which is common), a diagnosis is needed to determine the cause. A solution can then be proposed and implemented. After the issue has been “fixed” with the solution, metrics are again examined to ensure that it is working. A lot of teams are tempted to focus on the symptom and assume it’s a “recognition event”