Qualtrics Inc. today used its X4 Experience Management Summit conference to unveil a set of specialized artificial intelligence agents it says can address most customer interactions automatically and without human intervention.

Experience Agents are trained on the company’s database of over 15 billion questions and answers derived from customer interactions. They are designed to interact directly with customers and employees in personalized, proactive and empathetic ways, the company said.

Experience Agents can be deployed across any channel and respond instantly to address complaints, including offering compensation if approved by company guidelines. They’re currently targeted at what Qualtrics calls “moments of friction” but will become more proactive in the future.

“Historically, customer satisfaction has been addressed post-interaction, and organizations only close the loop on a small fraction of what comes in,” said Brad Anderson, president of product and engineering at Qualtrics. “Our unique value is the ability to understand human emotions and to turn listening systems into action.”

Massive database

Qualtrics claims to be the world’s largest survey platform, with an average of 50,000 survey interactions per minute. In addition to asking for feedback, the software can detect evidence of digital frustration, such as “mouse thrashing,” in which users move a mouse back and forth rapidly, usually in response to frustration. An AI agent can intervene and step the user through the process.

The difference between agentic AI and more common forms of call center automation is that “agentic AI is autonomous,” Anderson said. “It understands emotion and intent and takes the action that is most appropriate.”

Qualtrics codeveloped the agents with about 10 customers. Training typically lasted less than 48 hours. Customers can feed a set of runbooks and policies into the system to guide its behavior and specify which actions can operate autonomously and which require a manager’s approval.

While each Qualtrics customer customizes its surveys to some extent, most ask a basic set of similar questions, Anderson said. Qualtrics used anonymized questions and answers to create demographic and behavioral categories that guide agent responses to the most common complaints. “We’ve had enough customer conversations that we’re comfortable that customers trust us to use that data,” Anderson said.

Most of the agents run on self-hosted Llama models in a Qualtrics data center but customers can choose whichever large language model they prefer. Customer data “is never used to train a general-purpose public LLM,” he said. Model training was reinforced by extensive human review.

Reports on agent activity are aggregated in a dashboard called the Location Experience Hub. Managers can create and review responses for agents to deliver.

Agents are expected to be available early in the second half of this year.

Image: SiliconANGLE/Ideogram