Content type
Blog
Category
AI

Machine Learning in Finance: 7 Applications and Their Global Footprint

Iflexion’s latest research looks into selected artificial intelligence applications in finance and their future in different markets.

Introduction

The global financial market is undergoing a structural shift. Various factors are driving this change, such as evolving consumer preferences, new regulations, and technological advances. All these factors influence the exploding volume of financial data1 and the potential amount of work required to catalog, process, and act upon it.

Artificial intelligence in finance and AI consulting as the technology's professional manifestation will be among the biggest enablers for these processes, as standard methodologies, legacy analytics systems, and statistical approaches fail to keep up2.

The new dynamic requires a cardinal shift in thinking and tactics from financial professionals all over the globe. This push should happen on an unprecedented scale, which would promise unimaginable gains for companies able to harness its full potential. That’s why the World Economic Forum pulls no punches when it addresses the transformative effects of AI and machine learning in finance as ‘the new physics of financial services’3 driven by hype, real achievements, and even fear.

tremendous_excitement_about_the_ai_moment

In this report, we’ll review seven major use cases of AI in finance that help propagate this global transformation. We’ll cover some of their benefits and hopefully dispel some myths about artificial intelligence, its purpose, and real-life ROI for the financial sector.

This report tries to cover a broad spectrum of financial problems in order to highlight the massive extent of the AI revolution that’s been happening in the financial sector for the past couple of years. It will be useful for any finance professional, be they an executive in an international bank or a head of analytics in a small fintech startup.

A 2019 overview of select Artificial Intelligence applications in finance and their future in different markets.
Tweet

The Case for AI in Credit Markets

The credit markets are booming. In the United States alone, the amount of consumer credit outstanding has increased by 22% from 2014 to 2019, surpassing $4 trillion4. The total bank credit (US commercial banks) has grown by $3 trillion in five years, reaching $13.3 trillion in March, 20195.

Value of loans of all commercial banks in the United States from January 2016 to November 2018 (in billion U.S. dollars)

In these circumstances, banks and other financial institutions are doing their best to balance their risk appetite with the opportunity. This creates a competitive environment where financial institutions don’t want to miss out. Somewhat loosened regulations also mean that banks are willing to take riskier loans6.

Artificial intelligence in finance is one of the latest and most rapidly adopted additions to the toolboxes of these organizations, as they race to capture the best credit opportunities or understand bad ones that are already in their portfolios. In the 2018 WEF report, 73% of financial services and investment companies surveyed were to adopt machine learning by 20227.

Let’s take a closer look at some of the specific use cases and processes this technology advancement is going to cover.

AI is one of the latest and most rapidly adopted additions to the toolboxes of fintech organizations. 73% of financial services and investment companies surveyed in 2018 were planning to adopt machine learning by 2022.
Tweet

1. Leveraging AI in Finance to Analyze Unstructured Loan Data

Credit risk analysis is an ancient concept. However, with the latest technological advances, it has received a new boost. Just the fact that people fill out loan applications electronically already provides a plethora of unstructured data for the financial institutions. And it’s been proven that credit risk assessment that includes AI insights for unstructured data (loan application text) is better at predicting loan default than just the analysis of the financial data alone8.

When Words Sweat: Identifying Signals for Loan Default in the Text of Loan Applications

Financial institutions are already recruiting AI vendors9 to help them with the task. This rising trend of using machine learning in finance will only pick up further, as banks and fintechs explore other venues for unstructured data, like analyzing complaints and transactional data10.

2. Automating the Credit Pipeline

The majority of banks globally employ relatively manual credit review and allocation processes. Document reviews, credit risk calculations, and form submissions are done manually or semi-manually in many banks. That’s why it may take hours to process credit applications, especially if we’re talking about mortgages or other non-trivial cases. For this reason, mortgages are usually underwritten in no less than three days11.

Automating loan processing can save significant amounts of money and tremendously shorten the time it takes to process a credit application. Many financial institutions are already reaping the benefits of artificial intelligence, which can help with automation.

For example, Citigroup successfully uses artificial intelligence, although for corporate loan processing, to analyze applicants’ financial statements12. Implemented first at their Hong Kong credit analysis department as a trial project, this project has already helped the global banking giant reduce their load processing times from 10-15 days to just 2 days.

Capital Float in India automatically assesses credit risks, creates early warnings for loans that are likely to go bad, and achieves many other benefits with the help of AI systems that process all of this data and build risk predictions13. They’re treating the whole process as something that can be 100% automated if they push long and hard enough.

In Norway, Kraft Bank is also trying to expand automation throughout its credit management pipeline with the help of Tieto14. It is not a surprise, since there are plenty of other use cases for machine learning in finance, and specifically in credit processing:

Additional Opportunities for AI in Credit Processing

Lead Scoring
Automatically identify the best potential borrowers to give them a processing priority.
Collections
Automatically identify people more likely to repay, the best contact channel and even time to reach out to them—all with the goal of maximizing repayment rates.
Reserving
Automatically calculate an accurate estimation of the potential losses on your credit portfolio to adjust reserving capabilities.

These applications are just a part of the overall trend across many markets and many types of financial institutions. Competition is one of the biggest driving factors in this process.

  • “…aggressive fintechs, some prominent nonbank lenders, and early-adopting incumbents have enhanced their customer offerings, largely automated their processes, and made their risk models more precise. As a result, they can undercut traditional banks on price…”
    —The future of risk management in the digital era, McKinsey&Company
    15
Opportunities for AI in credit processing: applicants’ financial statements analysis, credit risks assessment, lead scoring, collection process, and more.
Tweet

Untapped Trading Opportunities

Just like in credit risk, using math, statistics, and algorithms for trading is also not a new idea. Algorithmic trading is a concept that’s been utilized in markets for almost five decades now. However, applying AI to solve trading problems is a relatively new approach that’s been gaining traction over the past three to five years16.

Google Trends - Analysis for “AI in trading,” “machine learning in trading,” “AI for trading,” and “machine learning for trading” search terms over the last eight years

That’s been happening for the same reasons AI is gaining ground in other industries: a cheaper processing power and an incredible availability of niche data.

This transformation will completely change the reality of the trading industry for some of its insiders. For the financial sector, it is expected that in five years from 2018 to 2022, AI investment will boost employment by 9%17. By 2020, the estimated AI spending by financial institutions will reach nearly $10 billion18. These AI innovations are projected to push the algorithmic trading market over the $20-billion mark by 202619.

AI investment can boost revenues and employment

Market leaders have been shifting towards trading AI systems for years now, with companies like BlackRock, one of the biggest financial management companies in the world, actively cutting their workforce in favor of machines20. Why are all these companies making a push for machine learning in finance? There’s a host of reasons, including the following:

  • AI is available 24/7
  • AI is more predictable than a human (provided it’s an expert system made by experts), which is important in a volatile industry like trading
  • Once the initial investment has paid off, the ROI is potentially endless
  • A lot of the currently manual trading processes are routine and can be automated with RPA
By 2020, the estimated AI spending by financial institutions will reach $10B. By 2026, AI innovations are expected to push the algorithmic trading market over the $20B mark.
Tweet

3. The Intrigue of Reinforcement Learning in Trading

We’d like to point out that reinforcement learning is a relatively new practical field of AI in finance, with a long history of theoretical discoveries but a small period of real-world implementations21. The essential difference from other machine learning domains is the approach to analysis and results.

Reinforcement learning seeks to reward the system for the right decisions, whereas the more traditional machine learning approaches are trying to infer the right decision based on the provided data22. Reinforcement learning systems change with each new data point, while traditional machine learning requires all of those data points to be available in advance.

Data Science Foundation - AI in Financial Trading

The long-term benefits of the reinforcement learning applications in trading aren’t yet completely obvious. However, that doesn’t stop companies and institutions from exploring this potential. Reinforcement learning is being tested as a prime tool for creating trading policies, as it doesn’t have any set rules to follow and can develop its own trading “game plan.” As a result, reinforcement learning can develop its own action policies and respond to real-time market changes in a more sophisticated manner.

Reinforcement learning is a promising development in AI that can completely change how trading is done. However, at this point, it is pioneered only by a handful of companies and a few Kaggle enthusiasts23. Luckily, there are plenty of applications of “traditional” AI/ML systems in trading.

A relatively new practical field of AI in fintech, reinforcement learning is a promising development in ML that can completely change how trading is done.
Tweet

4. “Traditional” AI and Markets

It’s impossible to predict stock prices. Otherwise, companies working with AI in the field would simply earn all of the money by themselves. The market is too volatile and affected by too many factors to be automatically predictable. Instead, what these systems do is improve your likelihood of success.

For example, the previously mentioned unstructured data can be used to analyze stock prices. Items like Facebook posts, tweets, SEC filings, and other pieces of public information can be processed and used for stock predictions24. Imagine a company experiencing a social media backlash. It would take humans much time to get a grasp of the situation, but an AI system that continually monitors online sentiment can identify these risks in advance. That’s what a company called Trade Ideas claims to do. They say that their trading engine, Holly, actually was able to deliver 54.3% of winning trades, which made a profit at an overall profit factor of 1.5425.

AI is applied in the more technical trading domains, like technical analysis26. For example, VantagePoint has patented a deep learning system that takes into consideration the vast amounts of technical trading data to build longer-term trading profiles27.

(Patent) Calculating predictive technical indicators

It’s also crucial to treat results delivered by an AI system with caution. Market conditions represent incredibly fine pieces in the puzzle that AI systems are tasked with solving. Having all of those pieces right is extremely important.

  • “The key aspect to keep in mind is that AI is driven by data and data quality can significantly impact outcomes. Also, if the data has biases, naturally the predictions from AI will also carry over the same biases, and in some cases even exaggerate those. This calls for data scientists [and AI engineers] with deep mathematical and business domain knowledge to model these decision-making platforms.”
    —Aditya Gandhi, Forbes India
    28
AI is driven by data and data quality. If the data is biased, the predictions from AI will carry over the same bias. This calls for data scientists with deep knowledge to model these decision-making platforms.
Tweet

Wealth Management, Personalization, and AI

The wealth management market is another booming financial domain. Here, the market of wealth management platforms alone is projected to grow from roughly $1.5 billion in 2016 to $3+ billion in 202229.

Ernst & Young place technological innovations among the key driving factors that will shape this expanding financial industry30. Of course, AI has its spot among these technologies. Wealth management clients think that applying machine learning in finance will have a profound effect on the market and its players31.

How important is AI in delivering results
  • “93% of wealth managers say AI will play a role in the future of their practice”
    —Temenos, AI & the Modern Wealth Manager Report, 201832
 The Potential Impact of AI on Wealth Management: A Global Perspective
North American (30%) and European (41%) wealth managers are the most likely to consider AI a 'game changer'.
Tweet

5. Cognitive Agents

It’s important to make the distinction between chatbots and cognitive agents. A standard chatbot usually communicates in the form of canned responses, while a cognitive agent can learn from past conversations and customer data. A cognitive agent can use natural language processing and advanced machine learning algorithms to simulate a conversation.

This could potentially offload the already overworked advisors and assist with relatively simple financial queries36. Cognitive agents can also be easily scaled through multiple languages and regions.

Cognitive assistants
Difference between chatbots and cognitive agents: a chatbot is (usually) a canned response dispencer, while a CA can learn from past conversations and customer data, use NLP and advanced machine learning to carry a conversation.
Tweet

6. Personalized Advisory Relationships

Consider a scenario: an advisor has a meeting with a customer who wants to invest in a certain company. Before that initial meeting, the advisor’s job is to acquire, analyze, and structure as much data about the potential investment as possible. We’re talking about dozens of documents. And, as you already know, the advisor might have dozens or even hundreds of other clients to take care of. The quality of services is at stake, as well as the advisor’s job satisfaction.

Artificial intelligence in finance can be used to assist them with this challenge. The agent can use AI to summarize documents37, extract additional insights, like sentiment38, or look for mentions of specific financial KPIs. This data can help them compile a holistic representation of the investment target and tailor financial advice to that knowledge. All of these efforts could save dozens of hours per year, which would have been otherwise spent droning over data instead of performing more creativity-demanding work.

In fintech, AI can be used to summarize documents, extract additional insights, or look for mentions of specific financial KPIs. This can help compile a holistic representation of the investment target and tailor financial advice.
Tweet

7. Personalized Financial Products

With thousands of existing customers, wealth management companies are sitting on a treasure trove of data. This data can be used to identify financial services that will be most suitable for specific categories of customers based on their transaction indicators (income, previous investment history), riskiness, keyword searches, social interaction, etc.39 In more common terms, this is sometimes referred to as product recommendations.

AI-powered financial product recommendations

These financial recommendations could then be processed by a human for an even more tailored solution40. The outcomes of such an approach are obvious. The advisory service looks more professional, and the client is happier about their financial decisions.

Takeaways

Machine learning in finance is emerging as one of the most noteworthy innovations. Here are the takeaways summing up its visible impact across the domains:

  • Credit markets are embracing AI in pursuit of new risk management capabilities. Here, AI stretches to loan data analysis and credit risks review.
  • Trading companies don’t miss out on AI adoption either. Increasingly based on reinforcement learning algorithms, AI trading applications are favored as they can effectively operate with high predictability around the clock.
  • In wealth management, AI is indispensable for creating relevant personalized experiences, in many cases through cognitive agents. It is also finding more applications in uplifting advisory services and serving to-the-point product recommendations.
  • Despite its big promise, AI in finance still needs to be used cautiously. It is highly data-dependent, and its effect is only as good as the data fed into it.

Quick Contact

Have a project in mind but need some help implementing it? Drop us a line:

*— means required fields