Nova KBM’s AI assistant is more successful than anyone dreamed

Nova KBM’s AI assistant is more successful than anyone dreamed 1120 840 Ben McCulloch

Here’s how Nova KBM, a market-leading Slovenian bank, automated its customer engagement, and the results it saw. 

Not only had the pandemic pushed many of Nova KBM’s customers towards ‘sofa banking,’ but also many of its Gen-Z and Millennial customers were keen to bank via digital channels.

Aleksandra Brdar Turk, Director Banking Operations, Nova KBM, and her team set out to build a digital assistant to meet those customer needs.

Supporting the Slovenian language

Here’s where it gets tricky. It’s assistant had to speak Slovenian – a language with roughly 2 million speakers and no NLU support at all.

“We are a small nation, but we are very proud of our culture, and it would be inconceivable to have a chatbot that couldn’t speak Slovenian to a person. So that was a must. It was like our basic requirement. Without that we were not moving forward.” Aleksandra told Kane Simms in a recent VUX World webinar.

Starting with a partner

So before doing anything, Nova KBM had to find a vendor that could build an entire NLU system for a new language from scratch!

First they brainstormed; Aleksandra and her team checked out their options throughout Europe, and tested platforms. They played with demos, and eventually narrowed down to a shortlist of providers.

But the Slovenian language hasn’t had any support from conversational AI providers so far. How could they support the language, which is a non-negotiable part of the project?

It turns out, Kore AI could do it.

Serving a language with 2 million speakers

Kore AI actually built Nova KBM’s Slovenian NLU from scratch. Slovenian is a Slavic language with its own nuances and a complicated grammar. To get this wrong would have undermined the assistant right from the start.

Alexandra was sceptical that anyone could pull it off. Kore AI built the NLU while Nova KBM acted as subject matter experts on the language.

How long did it take?

6 weeks!

That’s incredible! 6 weeks to build an NLU from scratch for a language that hasn’t received any support from conversational AI companies to date.

A month and a half was all it took to get a working prototype that worked around 85% of the time.

Once they had proven that the assistant would be able to support the Slovenian language, the project could move forward.

Building the team

To bring a conversational AI initiative to life, senior leadership support is crucial, and is one of the biggest challenges most organisations face.

Nova KBM’s CEO and management were keen from the start. They wanted a strong digital presence – to be closer to their customers and to understand their needs. This enabled Aleksandra to put together a core team to work on the project.

The project team consistent of people throughout Nova KBM. IT portfolio managers, architecture and infrastructure specialists, business analysts, subject matter experts and many more were all helping to make it happen.

“A motivated team is what I think pushed us very quickly towards the finish line. And we were able to roll out the finished product and go into production in five months, which was unheard of for a project this size. So remember, six weeks just for the language, then we had to build all of the conversational flows and all of the automated flows and build the whole knowledge base behind the NLP and machine learning engine, which then have to go through a couple of cycles of machine learning… But it was done in five months.”

Starting with live chat

Nova KBM had live agents handling chats with customers before they began automation. This is a great place to start and is common, as it allows you to gather real data on customer needs and utterances so that you can prioritise your use cases and build your language model.

Automating the simple things first

Nova KBM took the same approach and picked the simplest, most common conversations to start automating first.

Aleksandra and her team also put some effort into the personality design of the assistant. They didn’t want to appear like a stiff, conservative old bank. They wrote the assistants dialogue with a little wit and a relaxed feel, but still ensured that the answers were succinct and useful.

Once the assistant went live, they optimised the language model based on what and how people conversed with it, as well as iterating on new requirements as additional queries came in. Since then, and even today, the assistant receives regular updates based on customer questions. Another best practice.

Better impact than dreamed

“We were hoping to get our NPS scores increased, which we did, but with such a huge impact we’d never dreamed of.”

It’s no exaggeration. NPS scores jumped by 25%. 95% of chats are automatically serviced.

They went from 5000 chats a month, to volumes of 50,000 chats a month after deploying the assistant into the Nova KBM mobile app.

Those are amazing results, and there’s more.

One of the biggest benefits of AI assistants is that they free live agents up to do more complicated tasks. Another benefit is that the team have more time for training.

Nova KBM’s assistant, Niko, has enabled them to improve customer service across the board because live agents now have time for more coaching. They can improve their skills, widen their knowledge or learn how to better deal with unhappy clients.

“So we’re trying to make [automation] work for us, as well as for the customer” says Aleksandra.

Listening to Aleksandra talk during the webinar, you can hear her enthusiasm for this project. That enthusiasm is reflected in the result. Aleksandra was intrigued by the potential of conversational AI but also grounded in reality. She knew the project had to handle her mother tongue properly or it would fail from the start.

It’s great to see that Niko has been an incredible success, and Nova KBM expect it to have a brighter future:

“We’re now educating Niko. I would say he’s now a high school graduate. We want to push him to university – we want him to become an expert in banking, online banking, support the clients, give them as many ‘to the point’ answers as we can.”

If Nova KBM can do it, you can, too

Regardless of the challenges you face on your conversational automation journey, few can be greater than the lack of availability of the underlying core technology.

If Nova KBM was able to solve that problem, with the help of Kore AI, then it’s unlikely that the challenges you face can’t be solved with the right team, the right attitude and the right partner.

    The world's most loved conversational AI event is back
    This is default text for notification bar
    Share via
    Copy link
    Powered by Social Snap