Question: We’ve been investigating contact recording, quality monitoring and speech analytics. The team is quite impressed with the capabilities provided by the various vendors. A few of the vendors claim to be able to spot problems by analysing the customer’s and agent’s conversation in real-time. Having identified various sorts of problems, the system sends a notification according to rules so that either automatic guidance is provided on their screen or by someone joining the conversation. This is intriguing. Is it real?
William Durr of Verint Answers:
Almost sounds like science fiction, doesn’t it? And yet, it’s true. The technology has gotten good enough to spot key words in real-time and trigger some action. There have not been many deployments of the technology as far as I can tell. I’ll speculate why.
First, no speech analytic engine is error-free even when they process recordings after-the-fact. Burdened with analysis on the fly in real-time, errors can only go up in numbers. To some degree, interventions will occur that need not have and there will be no intervention when it was warranted.
Second, someone needs to produce a tight list of key words that, when uttered, indicate emotional exchanges or interaction woes. Such a list is challenging to create, refine and maintain.
Third, I am concerned about operationalising the technology at the production floor level. Consider the agents. The typical agent is already juggling numerous applications on their desktop while avoiding dead air time and engaging the customer in a friendly and professional manner. Suddenly, messages start popping up on the agent’s screen, directing the agent to steer the conversation in a particular direction or to make a particular offer. Or, a supervisor “barges” into the conversation and takes control. Now imagine that nothing took place in the conversation that triggered those interventions? It would only take a matter of some days before the agents would begin to chafe and become resentful.
Fourth, all these notifications go where? If you would like to automatically trigger reminders or guides triggered by speech analytics, you need to identify and define responses and guides for quite a large number of cases. Likely so large a number that it becomes unmanageable. So, the answer is a supervisor notification. Then, the supervisor can monitor silently to determine whether to barge in.
This means that supervisors have to be ready and available to handle a stream of notifications throughout the day. Do you require supervisors to man their stations as if they are agents? Do you set up a pool of cross-trained supervisors? How does this work assignment affect supervisor attitudes? All that said, smart people are working to solve these problems and bring the technology mainstream.