AI bots trained to detect customer emotions through vocal inflections
Image credit: DT
A new AI system has been developed that detects the emotions of customers ringing into call centres.
Software company Amdocs has used Microsoft’s natural-language processing system to detect the sentiment of callers by analysing patterns in their voice, including volume and intonation.
“In the next couple of years, customer experience is predicted to overtake product and price as a number one competitive differentiator between organisations,” said Doron Youngerwood, product marketing manager with Amdocs.
Explaining the company’s new platform, Amdocs One, Youngerwood said: “Operators are using AI in the contact centre, they have a dashboard which includes something called sentiment analysis.
“Sentiment analysis basically tracks the emotional state of the customer so throughout the conversation they can see if the customer is angry, frustrated or sad.”
Their mental state is logged by the system and tracked in real time. Operators are then directed to take a certain course of action to prevent angry or upset users from hanging up by offering them, for example, a price reduction.
An operator could also be directed to attempt an ‘upsell’ or ‘cross-sell’ to a customer who sounds suitably receptive.
“In many cases nowadays there’s a huge amount of guesswork by the agent to figure out how to resolve an issue,” Youngerwood said. This system should help to eliminate that.
The AI tracks emotions through voice intonations such as volume and word usage and even accounts for the way people speak across different areas of the world.
“It depends on which country, the nuance and the dialects of the language. It picks up on the accent and the dialect and based on the dialect it will know if you’re British, you’re more polite, so if you’re louder you’re more likely to be angry and then the propensity to churn [hang-up] will be higher.
“If you’re based in parts of the Middle East where speaking loudly is more common, then it will change the understanding. It will also provide a recommendation on what the next course of action should be.
“That’s why sentiment analysis is so key, so critical. It’s more than just the text, it’s more than what you’re saying, it’s how you say it.”