Zendesk integrates AI with intelligent triage to speed up CX responses

Customer service is a tough business, with multiple challenges that businesses need to address quickly in order to maintain the best possible customer experience.

In a bid to help organizations deal with customer service challenges more quickly, San Francisco-based Zendesk today announced a series of new artificial intelligence (AI)-powered features that aim to accelerate issue remediation and workflow. The new AI features have their basis in technologies that Zendesk gained in 2021 via the acquisition of Portugal-based startup Cleverly AI. The Cleverly technologies are now serving as a foundation to enable new incident triage and recommendation capabilities in the Zendesk Suite Enterprise Edition.

“At Cleverly, our mission was to make machine learning scalable within the CX [customer experience] industry and we were able to accomplish that,” Cristina Fonseca, formerly CEO of Cleverly AI and currently VP product at Zendesk, told VentureBeat.

AI is having a strong impact on customer service

In recent years, AI has become an increasingly common aspect of customer service workflows.

At the most basic level is the near-pervasive use of AI powered chatbots that are intended to help consumers with simple questions and requests. AI is also helping call centers, providing automation and guidance for agents.

Multiple vendors in the broader customer experience and support market have embraced AI-driven technologies. Startup Capacity, which raised funding at the beginning of the year, provides technology to help organizations automatically organize information so it can be more easily accessed by customer support. In June, contact center leader Genesys bolstered its platform with AI capabilities it also gained via an acquisition, in an effort to better understand and improve customer experience.

A clever approach to automation at Zendesk

The new AI-powered features that the Cleverly-based technology enables includes an intelligent triage system.

Fonseca explained that the intelligent triage helps enable automation for repetitive tasks that customer service agents often need to complete. Among the most common tasks is the need to actually label what a customer request is all about. 

The new AI-powered intelligent triage system has been trained on a large corpus of customer service requests and is able to infer what the customer actually wants. The system will now help agents label requests so they are properly identified and handled. In the retail industry, which is the first target for the new service, the system can properly identify order status, invoice and refund requests, among other common types of customer requests. 

Fonseca added that the intelligent triage also provides skills-based routing, where requests are sent to the right person, as well as prioritization capabilities. The other key AI-powered service that Zendesk is now integrated with is called smart assist. Fonseca said that smart assist will recommend content and replies for agents to be more productive while handling customer service cases. 

Getting clever with sentiment analysis and responsible AI 

With customer support requests, understanding sentiment analysis is a critical component.

Fonseca said that at Cleverly, the company first tried out using a couple of prebuilt sentiment analysis models, but they didn’t quite work right. For one, she noted that most were not properly trained for the customer service use case where, generally speaking, there are more negative types of requests than positive ones. As such, the existing models were not accurate.

Cleverly ended up building and training its own sentiment analysis model, which is now part of the Zendesk platform. Even with a properly trained model though, Fonseca said that there is still a fair amount of apprehension among some people about the validity of an AI recommendation. One of the ways that Zendesk is looking to overcome that concern is with transparency around the predictions.

“Some people still don’t trust AI, so for each prediction we make we also make the confidence of that prediction visible,” Fonseca said. “We believe this is very important because AI is not always accurate.”