When DNB began talking with conversational AI vendors, the largest financial services company in Norway turned to both the usual players; the big vendors such, as IBM and Microsoft; and relatively unknown small and new startups.
DNB didn't go with an enterprise-level vendor. Instead, it chose Boost.ai, an almost one-year-old startup, for its chatbots in banking needs, a partnership that has proven to be mutually beneficial for Boost.ai and DNB.
"They have been really flexible in meeting our demands," said Carla Padró, senior IT director for emerging technologies at DNB.
In turn, Boost.ai, as a new company, has "learned a lot from working with us," he said.
Banking meets chatbots
DNB is a massive financial services organization, with total assets reported at about $310 billion in U.S. currency in 2015.
According to Padró, Boost.ai, based in Norway, developed a conversational AI system for DNB that has now automated more than half the company's online chat traffic.
She explained that DNB ultimately chose Boost.ai because, unlike some of the large-scale vendors, such as IBM, which have really monolithic architectures, Boost offered flexibility and a custom-built platform for DNB.
Henry Vaage IversenCPO and company co-founder, Boost.ai
It took a while for DNB to create the system, Padró acknowledged; development started around January 2018 and didn't produce a finished product until the end of the year. But the wait was worth it, she said. While conversational AI is fairly common across a number of industries and chatbots in banking are not particularly new, Boost.ai claims to have a product that can understand user intent better than other systems.
"Most of the vendors starting today are using some sort of neural network," said Henry Vaage Iversen, Boost.ai's CPO and a company co-founder.
But, he added, "only using a model score for prediction has some limitations."
Semantics and AI
Boost.ai uses a neural network but runs what it calls automatic semantic understanding technology on top of it.
Essentially, the machine learning algorithm first scans textual input at a word-by-word level, internally adjusting spelling errors and basic grammatical mistakes before reading the sentence as a whole to identify a user's intent.
A deep understanding of the industry and the questions customer typically ask about it helps the algorithm correctly discover intent even when it might not be typed out clearly. If the exact intent is not clear, it tries to find close alternatives.
The results, according to Padró, "are impressive."
"This type of technology gives great results not only for the company and employees, but for the customers, as well," she said.
DNB has an ongoing partnership with Boost.ai, Padró said, and they are working together to offer a conversational AI platform that she claimed is "more efficient and smarter," and to turn the DNB chatbot into something "more like a virtual assistant."
According to Iversen, Boost offers prebuilt models for several industries that can be adapted for personalized results. The products currently work with about 25 or so languages and, recently, the company opened its first office in the U.S., in Los Angeles, in hopes of tapping into the American market.