Finance is one of the industries most likely to be disrupted by AI. The field is data intensive and small improvements in process and accuracy offer major opportunities for companies looking to out-innovate their competitors and industry challengers. According to CB Insights, finance was one of the major drivers of AI-related venture capital last year and there doesn’t seem to be any sign of slowing down. Here’s how AI can help improve financial organizations:
Fraudulent transactions are a costly business. Financial institutions unable to detect fraud suffer from not only monetary losses, but also hits to their reputation. The prevalence of false positives, instances where users are wrongly identified as fraudsters, has historically been a thorn in the sides of big banks. In an industry where detection accuracy traditionally hovers around 40%, it is much easier to run a false positive than the other way around. In fact, $118 billion in credit card sales are declined each year, even though real fraud only amounts to $9 billion. Each year, almost 15% of all American cardholders experience at least one wrongfully declined transaction.
There are many reasons fraud is difficult to detect with current systems. The number of actual fraudulent transactions is low, it can be hard to discover patterns and fraudsters are constantly changing their strategies to avoid detection. Analysts often attempt to build up a vast set of rules derived from historical data and industry knowledge, however these processes are limited by the amount of data they can ingest and the ability to handle exceptions. These are all areas where AI can help.
Companies like Stripe are actively innovating in this space. By using machine learning, they have been able to reduce fraud by over 25% without increasing the number of false positives. This was accomplished by integrating hundreds of new signals into the detection algorithm and constantly retraining their model to stay up to date with the newest tricks fraudsters are using.
Money laundering is the process of running illegally obtained money through a series of transactions to give the appearance of legitimacy. Illegal operators attempt to wash their funds of association with criminal activity, similar to how everyday people wash their clothes of association with dirt. Unfortunately, money launderers appear to be quite good at it. The UN reports that money laundering transactions comprise 2% of global GDP, however banks are only able to seize less than 1% of all laundered money.
As a result, regulatory bodies hold financial institutions to a high standard when it comes to money laundering compliance. All parties and transactions are subject to a thorough and intensive due diligence process. This involves developing an understanding of who the senders and receivers are, what their relationship may be, navigating many layers of shell companies and getting a sense of historical transaction history. In the US, companies spend up to $7 billion on anti-money laundering and compliance operations each year. The analysis needed to determine if a customer is engaging in money laundering is long, inefficient, and almost entirely performed by humans.
Banks like HSBC are already getting a move on building more intelligent compliance processes and have invested heavily in research, development and partnerships. However, there are regulatory hurdles that need to be dealt with, as regulators worry that human reviewers will mindlessly follow the recommendations of black box AI systems. Fully automated financial compliance departments are still a thing of the future, but it is encouraging to see the largest financial institutions in the world actively working towards it.
This era of rapid development in artificial intelligence would have been impossible without an accompanying explosion in the availability of data. In the past, banks were able to get by with just using historical data of transactions and payments to determine how creditworthy a potential borrower was; however, with the increased availability of new structured and unstructured databases, novel new techniques are now necessary.
This new constellation of data gives lenders the opportunity to produce more accurate segmentation of their borrowers, as well as produce a more nuanced view of creditworthiness by incorporating qualitative factors such as willingness to pay and consumption behaviour. Since a larger array of data is now consulted, lenders can make more accurate assessments of traditionally difficult to score individuals, such as those without a credit history. Crucially, this clearing process is able to happen almost automatically, resulting in a better customer experience.
There are a slew of fintech startups already employing this model, especially in the developing world where banking histories are more sparse. A prominent example is Ant Financial, an arm of the Alibaba Group. They have developed an application that pulls from a combination of traditional and non-traditional data to produce a comprehensive credit score.
If you’re like me, then you probably use multiple financial products in your day-to-day life. Most financial products offer a high degree of customization, whether it’s for your credit card, your savings account, or your investment portfolio. For example, the relevancy a particular loan or line of credit will differ depending on what life stage you are in. Customers just graduating from university will have vastly different financial needs than those preparing to buy a house, or putting aside money for their first child’s education fund.
These shifts in consumer behaviour aren’t currently accounted for by most banks, despite the fact that 72% of customers used digital channels to open a chequing account in 2016. Banks capable of providing a more personalized experience with offers more relevant to the customer’s current needs would decrease acquisition costs and increase conversion and loyalty. Additionally, better product recommendations would allow financial advisors to optimally serve their most profitable customers, without losing focus on the rest of their client portfolio.
Financial institutions are complex organisations, with many different departments, roles, and lines of communication. The potential points for increased efficiency are almost limitless, and properly identifying these opportunities requires a deep understanding of both the state of the art in AI and domain expertise. As an example, an AI model could analyze documents such as loan or mortgage agreements to measure the bank’s exposure to risk, flagging documents that appear problematic. This would allow analysts to focus their time on more rewarding, and more profitable tasks.
JP Morgan Chase, the biggest bank in the US, recently implemented a program called COIN to do just this. The software analyzed commercial-loan agreements at a faster rate and with less errors than human reviewers. Chief Information Officer Dana Deasy doesn’t see this as a case of labour displacement, but rather as a case of augmentation, framing it as a way to “free people to work on higher-value things”.
Finance is not immune to innovation and disruption, and artificial intelligence has the potential to unlock many layers of value. The list of possible use cases will continue to grow thanks to innovations in machine learning that are increasing the scope of the technology and its potential applications. It will require careful planning, broad executive buy-in, sharp data science expertise and the right partner to make it a reality.
Interested in starting your AI journey? Contact us today.