- Chatbots in financial services are expected to save companies over $7 billion annually by 2023
- 64% of companies in the sector report that chatbots are also boosting their Net Promoter Scores
- The most advanced assistants can make proactive recommendations and understand intent
“Speak to agent.”
If this is what you say most often when you talk to your bank through an 800 number, you’re not alone. Just 46% of bank customers today say they’re happy with their bank’s overall ease of use, quality of service, and customer experience, according to a recent Deloitte survey of more than 17,000 consumers.
That script may soon flip, thanks to increasingly sophisticated chatbots powered by machine learning and natural language processing (NLP).
The newest generation of customer-service chatbots aren’t just trained to understand complex human communication. They’re learning from the data generated by each interaction and getting better at the contextual and emotional understanding that more customers increasingly expect—but frequently don’t get—from banks and insurance companies.
Customer service chatbots showed early promise in sectors like retail and healthcare. Now they’re becoming a major factor in financial services. Chatbots are expected to save banks $7.3 billion per year by 2023, up from an estimated $209 million in annual savings in 2019, according to Juniper Research. That translates to more than 862 million man-hours saved in operations.
More than 4 in 10 consumers say they are more comfortable sharing personal details with a chatbot than with a friend.
Customer engagement with chatbots in financial services is also on the rise. In 2017, just 28% of consumers reported interacting with a retail banking or financial services bot, according to a Capgemini report. In 2019, 50% did so. The same report shows that, in addition to time and cost savings, 64% of companies in the sector report that chatbots are boosting their Net Promoter Scores.
As consumers get more comfortable with chatbots and they become a mainstay of support, developers are working hard to raise their functional and emotional IQ. Advances in NLP are allowing them to grasp sentiment and context and communicate more fluidly and personally.
From chat to voice
In 1966, Joseph Weizenbaum created ELIZA, the world’s first chatbot. It paired keyword recognition with a script that mimicked human conversation. More recently, chatbots like Siri and Alexa have become ubiquitous virtual assistants on mobile and home devices. Low-code software development tools and open source APIs have put chatbot design and deployment within reach of almost anyone.
However, chatbot communication skills have mostly been confined to simple requests and FAQ-like searches. As anyone who has endured a dead-end “conversation” with Alexa knows, even the most sophisticated bots can fail to grasp linguistic subtleties like context, emotion, irony, and idiom.
NLP and machine learning are changing the game. NLP has advanced significantly since 2013, when the introduction of Word2Vec, a neural-network model, allowed algorithms to grasp context despite missing words. Since then, NLP has incorporated neural-network models to give chatbots better memory retention and the ability to handle sentiment detection and analysis.
For consumers, these innovations mean that bots are starting to respond appropriately to messy, unstructured human communication as it occurs—with midstream digressions, contextual haziness, and subtle emotional nuance.
“Three years ago, converting unstructured data to assess topic, intent, discourse, quality, and sentiment was unheard of,” says Amy Matsuo, national leader for regulatory insights at KPMG. Today, she adds, banks are showing how these new capabilities hold promise.
In 2018, Bank of America launched its banking bot Erica. Within a year, Erica had handled 50 million customer requests and queries. Through iterative learning from those engagements, Erica’s machine-learning algorithms have effectively doubled the number of ways customers can ask her questions, from 200,000 at launch to 400,000 today.
Getting personal with NLP
The next frontier of bot development in financial services is likely to focus on even more personalized interactions. “The amount of data available to banks is immense, but turning that data into useful information has always been a struggle,” says John Ing, principal product lead in customer experience at tech consulting firm ECS. “Machine learning is making that a reality.”
Banks are building new capabilities that enable bots to act as virtual financial advisors, not just support agents.
Today, BofA’s Erica doesn’t just check your balance. She can proactively make suggestions to improve your finances. Erica can recommend steps to boost a low credit rating, for example, or suggest optimal payment amounts on a hefty credit card balance to lessen monthly interest accumulation.
Banking and insurance company USAA has discovered that customers are even willing to broach sensitive financial topics with a new virtual assistant that integrates with Amazon’s Alexa. Through Alexa, customers might ask questions like “I have no money, what can I do?” and “I don’t know how to save for college. How can I start?”
This isn’t unusual. According to the Capgemini study, more than 4 in 10 consumers say they are more comfortable discussing personal details and issues with a chatbot than with a friend.
SEB Bank’s customer chatbot, Amelia, is also taking on a more personal role to improve customer experience. Based on conversational context, Amelia can offer advice on account management or recommend action when she detects fraudulent transactions.
Amelia has also been trained to upsell: using customer data and credit ratings, she can suggest you apply for a new credit card with better terms.