For the past three years, Nordic bank DNB has been supporting its staff with conversational AI-powered virtual assistants. A virtual agent called Juno supports the bank’s customer service departments and in 2022 answered more than two million inquiries, which equates to assisting roughly 1200 users daily.
Juno is one of five virtual agents operating across customer and employee-facing use cases at DNB.
The Juno virtual agent is internal, “but it’s grown to become quite a complex application”, says Jan Thomas Lerstein, head of emerging technology at DNB.
“It’s a support tool, or an advisory support tool, for all the customer agents in the bank. It has actually become the most important tool for helping the advisers give good service to the customers,” he adds.
Launched in 2020, Juno supports customer service agents and advisers in accessing the various routines they need to follow when assisting customers. The bot also enables agents to locate key information they need, accelerating their response times when dealing with an inquiry.
Of the more than two million queries answered in 2022, Juno provided 83% of agents with valid answers, answering an average of seven questions per user each day.
“The accuracy is steadily improving,” says Mr Lerstein. “But the main reason that the accuracy is only in the high eighties is because the questions were considered out of scope for that service.”
Asking questions that are not covered by the knowledge base or that the trainers have defined as out of scope or not relevant “kind of depends a lot on the user”, he says.
“The lower accuracy is usually with completely new agents that don’t know exactly how to use the tool since it’s quite an advanced tool. You need to learn how to use it,” adds Mr Lerstein.
Scandinavian fintech partnership
DNB developed these virtual agents with Norway-based boost.ai, a provider of conversational AI software.
The bank’s Juno agent uses the filtering functionality from boost.ai’s platform, which allows a single virtual agent to provide answers across multiple business units without requiring each unit to have its own standalone bot. Filtering enables Juno to tailor its replies so that agents receive a unique, department-specific response instead of a generic one. The feedback function offers direct feedback to the bot, to improve and enhance functionality.
The feedback function for the virtual agents is crucial for its development comments, says Henry Vaage Iversen (pictured), CCO and co-founder at boost.ai.
Not only are the users and the customers accepting this automation, they’re also wanting more of it
The past 12 months have seen an increase in demand from organisations looking for new use cases and ways to solve problems using conversational AI in different ways, says Mr Iversen.
“Not only are the users and the customers accepting this automation, they’re also wanting more of it because they see how it makes both work life and life as a customer of the bank more efficient,” he adds.
Boost.ai is now looking at how open and large language models can be used to improve content, says Mr Iversen.
“We’ve seen a lot of internal use cases where we can use generative models to create test data and training data that will make even the job for the AI trainers a little bit easier,” he adds.
In supporting the customer service team at DNB, Juno can answer questions on more than 3400 topics, with each further customised towards seven different areas within the bank, split between the corporate banking and private market.
According to Mr Lerstein, the bank started experimenting with AI in 2017 to automate and improve internal services that had customer-facing job functions. In recent years, DNB launched a virtual agent suite based on the boost.ai platform.
Other virtual agents at the bank include the customer-facing virtual agent Aino. Within six months of its lunch, Aino begun automating more than 50% of all incoming chat traffic and has interacted with more than a million customers. In addition to Aino and Juno, the bank also runs Hugo, which assists employees with HR-related queries, and Fix, which handles incoming inquiries to DNB’s IT service desk. Additionally, DNB also has a virtual agent named Justina to assist employees with legal questions.
