The role of machine learning in customer service software

octomedia

10th Aug 2016

Machine learning should be invisible, argues Mikkel Svane, CEO of cloud-based software provider Zendesk.

“The difficulty is it [machine learning] is really hard to productise properly,” Svane told Internet Retailing during a recent visit to Australia.

“Talking in big terms, one of the issues with machine learning is it can just be more information that you have to relate to.”

Svane said Zendesk is investing heavily in building machine learning functions into its core customer service software.

“I really think the future of machine learning, the way we’re going with it is to almost build it into the core product, so you don’t notice it, you don’t think about it, you don’t see it. It just works for you.”

Zendesk’s Melbourne office is responsible for building the company’s machine learning products. This year the Melbourne team has unveiled two products built on Zendesk’s machine learning platform: Automatic Answers and Satisfaction Prediction.

As the name suggests, automatic answers deciphers the language used in email queries and auto-replies to customers with relevant help centre content.

If the automatic answers process clears everything up customers can mark the ticket as resolved. If the question is not adequately answered then the normal customer service agent workflow occurs.

The tool aims to solve inquiries faster, freeing up agents up to spend more time dealing with more complex customer service issues.

Svane says Zendesk began to see patterns in the masses of data being collected which lead to the company’s machine learning play.

“We have 75,000 businesses using our software so we have billions of interactions,” he said.

“Everything from asking about returns to ‘where the hell is my Uber?’ So of course when we have billions of interactions we start to see patterns.”

Zendesk builds a unique machine learning model for each of its clients based on past interactions with customer service agents, ie the way they have previously answered customer questions is used to teach the machine learning model the “best” answers.

Discussing emerging trends in customer service interactions, Svane described chat bots as “a new conversational interface” that allows customers to serve themselves.

Since April, Facebook has let outside businesses create chat bots that users can send a message as they would a friend and receive an automated reply.

“Things like Siri and Alexa are manifestations of bots and over time they will get more and more sophisticated and help us navigate information to find the relevant information,” Svane said. 

Consumers will increasingly turn to bots and virtual assistants to find information such as return policies, the challenge for retailers and software providers is to make that information available so bots can learn it and serve the relevant information to back to the user.