In the past few decades, we’ve seen a number of significant disruptions across the economy propelled by breakthroughs in computer technology. One of the most significant has been artificial intelligence, the use of advanced algorithms to emulate human intellect. AI algorithms are excellent at uncovering subtle relationships in massive data sets and automating processes that conventional computer technology can’t manage.
The banking industry hasn’t been as quick as others to adopt AI, but there are signs that this is beginning to change. Bank of America has started to pioneer the use of artificial intelligence in banking. Their use of AI so far may set the tone for how the broader industry adopts the technology.
A Powerful Communication Tool
One of Bank of America’s most widely used AI innovations is Erica, their AI-powered customer service chatbot. The bot — first piloted in late 2017 and made available to the public in 2018 — helps more than seven million customers. Around half a million new clients start using the bot every month.
Customer service chatbots have become a popular tool for businesses wanting to reduce the workload on customer service representatives and streamline communication. These bots can field early customer questions and concerns and, depending on how sophisticated they are, resolve more straightforward requests. Erica, for example, can search for past transactions, call up credit score info and lock or unlock credit cards.
Chatbots are useful even with conventional programming, but AI can make them more powerful. AI enables tech like natural language processing (NLP), which chatbots can use to parse customer requests and respond in a way that feels natural. NLP can also gather behavioral data on customers — what they need and how they feel — which the software can pass on to customer service reps, giving them the information required to provide the best service possible.
A Way to Comb Through Data
Bank of America is also experimenting with using AI to improve automated fraud detection. Traditionally, fraud detection algorithms employ hard-and-fast rules to detect fraud. Preprogrammed conditions can flag transactions that are likely fraudulent — a customer using a card in a state they don’t live in, for instance — and block them.
AI allows for a more sophisticated approach. Bank of America can leverage the data they have on every fraudulent transaction they’ve caught and train AI to intelligently flag potential fraud.
The tech can also apply to billing disputes in the same way. Typically, a bank associate may spend hours combing through data to find who is at fault. An AI algorithm with the right training could scan the same information in minutes and render a verdict.
The Bank of America also hopes to leverage AI for real-time data analysis in the near future. Recent regulations, like the EU’s Markets in Financial Instruments Directive 2, which went into law in 2018, mandates additional reporting on trades — meaning banks have a lot more publicly-available trading to work with. An AI algorithm can parse through this data in real time, providing better info on current market conditions for the bank’s clients.
Potential Drawbacks of AI in Banking
While AI may offer significant benefits for the banking industry, there are also some risks associated with the technology. A sufficiently-trained AI will only recreate reality — meaning it will replicate any bias found in the information it was trained with. If a bank typically sides with vendors instead of customers in billing disputes, for example, there’s a good chance the AI will behave similarly, even if it seems to be making decisions objectively.
Some researchers are worried that AI can’t overcome bias, and that there’s no way to control for bias in training data.
AI-based automation can also be somewhat difficult to troubleshoot. Because an AI isn’t directly programmed and is instead trained on massive datasets it uses to “learn,” these algorithms can be somewhat opaque. Developers may not always know how or why the AI came to a particular conclusion.
Conventional, rule-based algorithms can make less intelligent decisions, but developers have more fine-grain control over how they process information and make data. If traditional algorithms start making unusual or ineffective decisions, they can typically be tweaked. AI algorithms are often harder to fix.
How the Public Has Responded
Despite the possible drawbacks, public response to Bank of America’s artificial intelligence projects has been positive so far. Erica’s overall user satisfaction rate has tracked around 82 to 83 percent, according to Christian Kitchell, the bank’s AI Solutions and Erica Executive.
While Bank of America’s artificial intelligence approach has been one of the most ambitious, most other financial institutions are experimenting with similar applications of artificial intelligence. It’s likely that AI, for better or worse, will play a significant role in the future of banking.
Follow Us On
Get the latest tech stories and news in seconds!
Sign up for our newsletter below to receive updates about technology trends