The universe of machine learning (ML) and its applications is expanding at an ever-increasing rate and there is hardly any industry that is not leveraging the power of sophisticated and cutting-edge algorithms to foster business growth. Solutions using Artificial Intelligence (AI) have gained even more significance as the world has moved to digital properties with the onset of the pandemic era.

The finance industry is no exception. Organizations are increasingly using ML algorithms to build credit risk models, detect fraud, respond to customer queries, and much more.

The post will apprise you of the widespread applications, challenges and trends of employing AI and machine learning in the finance industry, featuring insights from these top industry experts who will be speaking at Data Science Salon NYC on December 7, 2022:

• Harry Mendell, Senior Researcher at Federal Reserve Bank of NY
• Erin Stanton, Global Head of Analytics Coverage at Virtu Financial
• Michael Kortering, Chief Credit Officer at Reprise Financial

Applications of AI in Finance

Risk Assessment

Banks have been using ML algorithms to assess the creditworthiness of loan applicants. Not only do the algorithms help banks in making the binary decision of whether to issue the loan or not, but they also have the power to suggest the extent of loan issuance amount to qualify the loan as a good risk for the bank. With the ever-increasing need for explainability, the algorithms are now able to ascertain the model outcomes thereby adhering to governance measures.

The algorithms are also capable of highlighting some hidden factors outside the books of a general auditor that enlightens the organizations to think out of the box while collecting and storing data.

Fraud Detection

ML algorithms are working at the forefront of banking solutions preventing users from a multitude of frauds. Such advanced solutions analyze the buying behavior and a host of other attributes to distinguish a regular transaction from fraudulent ones to create alerts. There is a human in the loop i.e., a customer executive who acts on such alerts and verifies the original author of the transaction to approve it.

According to Harry Mendell, Senior Researcher at Federal Reserve Bank of NY, there are some enhanced ML techniques that can be leveraged to detect fraud or measure credit risks:

“Language models have revolutionized the way we analyze documents and look for risk. Graph convolutional networks, transformers and other deep learning techniques are showing great promise for fraud detection, risk and climate change impact analysis.”

Customer Service

A lot of banks have employed chatbots to speed up the turnaround time of queries from customers. Many inquiries such as checking account balances or examining account activity are easily facilitated through the help of such chatbots.

The latest advancements in chatbots have proven to be less dependent on customer executives and are self-sufficient in serving the customers’ needs. The smart virtual assistant is a win-win for both – the business as well as the customers. Such reliance on technology provides an added competitive edge and yields greater productivity returns by sparing the employees from redundant tasks and enabling them to contribute to business value.

Further, it saves customers the pain and time of soliciting responses from bank representatives. Erica is an example of a powerful virtual financial assistant launched by Bank of America (BoA) that alerts users about duplicate charges, merchants posting refunds, reminding bill payments, highlighting increases in spending and recurring charges, etc. Owing to the surge in digital banking activity, BoA reported almost 1 billion Erica interactions in the first four years since its introduction.

Cyberattacks Prevention

Financial institutions, by the very nature of their business, adopt strict guardrails to ensure that their customers’ wealth and personal information are not at risk because of foul play by bad actors. AI models monitor the data patterns and assist organizations in keeping a strict eye on any unusual behavior or activity.

Personalized Recommendations

The predictive power of algorithms does not stop here — it also helps meet the users’ demand for more customized and personalized recommendations that better suit their requirements. AI-powered assistants learn patterns from dynamic user behavior to generate meaningful and actionable insights and recommendations. The key highlight is that such recommendations are tailor-made to suit the specific content, services, or applications relevant to the customers.

As per JPMorgan, “our research platform produces over 10,000 pieces of research a year, but until recently, clients did not always know the reports existed. ML techniques solved the issue, and now each client logs into a customized portal that provides unique and relevant research, personalized to their needs.”

Stock Trading

Stock trading is a field where time is of the essence to take advantage of the stock movement and generate profits. The faster one can analyze the stock price pattern, the timelier decision of whether to buy, sell or hold that stock can be made. That’s precisely the forte of AI algorithms. Its power lies in sifting through humongous amounts of data in no time and presenting trading firms with data-driven insights to call the shots.

Algo trading is a common term used for high-frequency automated trading where specialized algorithms are used for stock identification, candlestick pattern identification as well as stop loss decision.

Robotic Process Automation (RPA)

RPA brings in a lot of operational efficiencies by automating several repeat tasks. As per Gartner, the use of RPA minimizes efforts and improves speed and accuracy. Smart Process Automation (SPA) / Intelligent Process Automation (IPA) are the latest terms used in the automation field using ML and deep learning techniques. Banking and insurance firms have benefitted from natural language processing (NLP) applications like invoice processing, service request generation, credit application processing, contract processing, etc.

Challenges of Using AI in Finance

Insufficient Data 

Advanced AI algorithms are largely black-box by nature and must comply with regulations and ethical principles to ensure that the technology is aimed at the benefit of the end user. Insufficient or biased data leads to skewed models, which is a challenge in smaller organizations in particular:

“Having sufficient data is the main challenge. That is less problematic in larger, established organizations, but can be more problematic in small and medium size organizations”, says Michael Kortering, Chief Credit Officer at Reprise Financial.

Data does not only need to be available but also accessible, adds Harry Mendell: “I believe that the main challenge is making our vast troves of data available to data scientists. Data architectures that allow data to be cataloged and be searchable and available with the right governance throughout the organization, such as the Data Mesh, will create a renaissance in both AI and ML in finance.”

Data Bias

The recent news where Goldman Sachs was accused of issuing a disparate credit limit based on gender bias created a lot of stir in the industry. The financial institutions are responsible to explain every model outcome and prediction to instill trust in the ecosystem. This trustworthy and explainable AI will lead to the widespread adoption of such advanced algorithms and improve the quality of services to the users.

According to Erin Stanton, Global Head of Analytics Coverage at Virtu Financial, the results of an ML model heavily depend on the humans training them:

“While Finance has been capturing data for several years, there are still ongoing data quality issues as well as details and granularity that are not always captured. Machine learning, or machine training as I like to think of it, is also still heavily dependent on the human responsible for the model research. The data we include and exclude as well as the model inputs or features we include and exclude have a huge impact on model performance.”

Trends for AI in Finance

When asked about AI trends for the finance industry in the next 3-4 years, Erin Stanton mentions the continuous adoption of ML among organizations:

“I think we'll continue to see firms deploy more and more models, starting first with simple problems and moving to more sophisticated issues over time. Within Virtu Analytics, we had one ML model deployed in production for a few years and this year we've launched 4 new ones. I would expect this growth to continue over the next 3 to 4 years.”

Michael Kortering agrees with Erin and thinks that the analysis of images will be one of the applications with increased popularity:

“Machine learning will grow in usage across all organizations. Companies that are more heavily regulated will continue to adopt it in more areas. New modeling techniques will develop. One area I expect modeling techniques to develop is converting data to pictures and then using models to recognize pictures.”

In conclusion, we can say that the finance industry believes in the future of AI and understands that if used rightly, it is one of the most powerful tools to increase customer engagement, drive innovation and push newer avenues to transform the business.

This article has originally been published here.