In early 2016, SEB, one of Sweden’s largest banks with a presence in 20 countries around the globe, started integrating Amelia, an artificial intelligence (AI) platform from IPsoft, into its help desk. Amelia is represented by a blond female avatar and is always referred to as “she” rather than “it.”
The artificial intelligence platform is built on semantic understanding, which enables Amelia to interact with users through natural language to determine what actions to take in order to answer a question, fulfill a request or solve a problem. She is also designed to learn through observation.
At SEB, Amelia serves as a customer interface with automated interactions that can scale up to meet expanded support needs. “The driver is to find a way to improve the experience for our customers,” explains Mikael Andersson, the bank’s IT strategy transformation lead. The virtual agent, which speaks Swedish in this implementation, will ensure that the bank can meet anticipated future demands.
With Amelia, “you need to find a business context and a role,” Andersson says. That role is one that integrates with the human staff in what he describes as “an augmented approach” that draws on “both tech and people.” With Amelia’s automation taking care of more mundane tasks, employees are free to do “more value-adding work.”
Four Cases Form the Core of the Pilot Program
When SEB first put Amelia to work for the IT service desk, the help desk people suggested the tasks that should be assigned to the platform. Andersson recounts that they reviewed different types of cases, which they then narrowed down to the four that formed the core of the pilot program. The cases were chosen based on their volume, complexity and capacity for testing the range of Amelia’s capabilities.
Amelia’s progress is based on machine learning through observed dialog. When Amelia cannot solve a particular problem, she can escalate it to a person and observe the dialog between the customer and the service representative in order to learn how to address the new situation.
Before Amelia applies a new approach on her own, some optimization is done to ascertain that the technology’s understanding is sound. Then that process can be automated. Amelia’s management interface reports on the number of cases solved, and proposes processes and dialogs for things that haven’t been solved but have been observed.
Andersson points out that because Amelia’s technology is new, there are “no best practices,” so SEB has to find out for itself what works and what doesn’t. Like Amelia, the bank will be guided by user feedback as it works to improve the process.