Why It’s So Hard to Put Bank Smarts in Voice Assistants

October 24, 2018
Written by: Nathan DiCamillo
Originally published on: American Banker

Voice assistants are proliferating throughout banking, but consumer demand hasn’t firmed up for virtually assisted banking just yet, partly because the experience is still deemed too buggy.

So for artificial intelligence startups such as Finn AI, the focus right now is on getting voice assistant technology working well inside banks, which is harder than it sounds.

“There is a disconnect in the market around expectations [about] how difficult building a natural language assistant is,” said Jake Tyler, chief executive of Finn AI. “It’s not easy for us. We’ve aimed to deliver great experiences and are managing expectations in not just building deep-learning algorithms and data science but also in supervised training and making sure we’re labeling data in the right way and that we’re managing our conversation design.”

The Vancouver, British Columbia, startup will devote more resources to developing a data science team after closing on US $11 million in Series A funding on Tuesday. The round was led by the venture capital firms Yaletown Partners and Flying Fish Partners and also included BDC Capital’s Women in Technology Fund and 1843 Capital.

Finn AI is doing business with BMO Financial and ATB Financial in Canada as well as Banpro in Nicaragua. By the end of the year, it expects to launch a product with Fidor bank, a digital bank in Europe, and by early next year with TymeDigital, a digital bank soon to launch in South Africa.

Finn AI can already deploy its conversational agent into a bank’s existing web and mobile channels as well as into third-party platforms including Facebook Messenger, WhatsApp, Alexa and Google Assistant. The company powers its service through a central model and engine, which centralizes how its AI learns and responds across banking channels.

With its US $11 million in new funding, Finn AI CEO Jake Tyler said it would grow its data science team. Finn AI is also developing customer coaching tools for its banking clients, so that the virtual assistants that banks deploy through Finn AI will do more than have a conversation about banking services and sales, and begin to encourage customers to be more fiscally responsible.

For one bank customer, Finn AI has developed a credit score coach. In the first half of next year, the company will have developed a budget coach and savings coach.

 

“It’s our goal for everyone to have a personal financial adviser assistant in their pocket,” Tyler said.

 

There are some hurdles to expansion. Bank customers have expressed little desire for virtual assistants, said Emmett Higdon Director of Digital Banking at Javelin Strategy & Research. But banks are very interested in giving their customers a tool they will want to interact with daily (someday) in their smart homes and cars.

Building the required scale to meet client demands is among the more difficult challenges for the company’s team of 50 employees. Every bank needs a tailored product, so that their customers get a personalized experience that is relevant. Otherwise, these clients may more easily turn to Finn AI’s competitors such as Personetics or Kasistos. To achieve this means gathering as much data about customers as possible.

“The more data you bring to the party the better it gets,” said Frank Chang, managing partner at Flying Fish Partners. “And when you already have a few customers signed up, you can offer a great baseline tool that work right out of the box.”

Others in the world of virtual assistant startups have different attitudes toward bank data. “It’s not a technology issue, it’s a business decision and a philosophical decision about how their customers are going to use the data,” said Daniel Latimore, Head of Banking at Celent. “There are others out there who are taking a different approach. They believe proprietary data about bank customers is a liability, and they don’t want to hold it.”

Finn AI insisted, however, that it does not take actions that would compromise customers’ data.

 

“We treat each customer as their own separate instance,” Tyler said. “We do share learnings about how people talk about banking, common intent definitions and ways of asking for things — but not individual customer data. All shared data is cleansed and scrubbed of [personal identifying information] and other sensitive data. This shared learning means we don’t encounter cold start problems, so our customers can get started faster.”

 

Chang, who was formerly managing director of the voice and natural language business at Microsoft and is joining Finn AI’s board of directors, said the company has a unique way of building customer trust.

With each new bank client, Finn AI starts by building trust with customers through frequently asked questions, moves on to doing transaction requests, and then moves into predictive banking.

“As users get more and more comfortable with interacting with an agent like this they will ask it new things,” Chang said. “That also helps train the system.”

Finn AI took a similar trust-building approach with BMO Financial. In the first phase of its relationship with the Montreal bank, Finn AI’s virtual assistants have been answering frequently asked questions in Facebook Messenger. BMO decided to work with Finn AI because of what it had seen with the startup’s work at ATB Financial, and the bank has seen a one-point bump in the five-point customer satisfaction rating it has on Facebook since the bot’s launch in March.

 

Read BMO’s customer story: Why Bank of Montreal Launched BMO Bolt™ to Transform Customer Experience

 

“It’s not a tried-and-tested model where you can say, ‘This is how many mortgages I’m going to sell,’” said Sumit Sarkar, BMO’s Head of Customer Experience and Strategy within personal and business banking marketing. “But we didn’t want to step in and go transactional right away. We wanted to test and learn.”

BMO has also found that users engage with the platform at times when the bank is closed, and often use it for finding out information about tasks they want to do themselves, such as changing their passwords.

Humans are still necessary for the bot’s learning: Whenever Finn AI cannot answer a customer’s question, it hands the customer off to a human agent, and the Finn AI engineering team takes a look at why the bot couldn’t solve the problem.

 

“We don’t expect AI to answer everything,” Sarkar said. “We want a healthy balance between the human behind the scene and the tech behind the assistant. The more value-added conversations should be with the human because that works best for the bank in how we meet the customer needs.”

 

As smartphone sales increase dramatically globally, Tyler predicts that there will be universal adoption of virtual assistants in the banking world.

“We think the next 10 years will be about banks having an AI digital assistant in sales, then core banking and then personal finance,” Tyler said.

For banks, virtual assistants provide the opportunity to cut costs on call center volume and also produce a better customer experience.

“Virtual assistants make it simple to tell me how much I spent on home improvement the last three weeks — that’s cut and dried for a virtual assistant,” Javelin Strategy’s Higdon said. “But it’s difficult to do on your own within the mobile banking app today.”