What is modern ML, and how it affects business and helps fintech and banks worldwide to be a step further than competitors?
Machine learning methods have been around for decades, but only the big data revolution and the sharp decline in the cost of computing power have truly unlocked the potential of trainable models. In particular, the financial and banking sectors are leaders in applying machine learning methods, which are the basis of automated decision-making systems. For example, based on online ADM tests, the system determines the probability of a potential borrower not repaying the loan. Similar systems are used in medicine and jurisprudence. We will explore this as a big step forward in data analysis and show how the banking systems benefit from pinpointing the exact financial indicators with a stronger influence than others on the bank's health.
- What is ML, and why are machine learning banking algorithms an underrated yet invaluable asset for the Fintech landscape?
- Addressing the issues of bias
- The value of user experience
- How to embrace the full force of machine learning and be better for it?
- Why must banks become ML?
- Processing huge amounts of unstructured data
- Automation of routine tasks
- Deepening marketing personalization
- Identifying business trends
- Acceleration of the development cycle
- Improving the banking credit scoring
- Better safe than sorry – the eternal security issue.
- Examples and use cases of machine learning in fintech and banking
- JPMorgan Chase and their COiN
- Bank of America and Erica, their virtual assistant
- Wells Fargo invests in machine-learning Chatbots and Startups
- Citibank and fighting fraud with FeedzAI
- U.S. bank
- Why banks are increasingly using ML: Things to consider when implementing machine learning in Banking
- Final thoughts: Is it worth implementing machine learning in banking?
Pride & Prejudice: What is ML, and why are machine learning algorithms an underrated yet Invaluable Asset for the Fintech Landscape?
During the last two decades, with the advent of powerful computers, the Internet, and the mass digitization of information, machine learning has experienced a real boom. The term has been used in science for over half a century to describe programmed pattern recognition. However, the concept is even older. Mathematicians started talking about such processes at the beginning of the 19th century. McKinsey’s analysts weigh the benefits and risks of using machine learning algorithms in business in general and banking in particular. They discovered that the myth of “perfect intelligence” has somewhat distorted the understanding of what ML is and what it can bring to the table.
Machine learning in artificial intelligence increasingly refers to the decision-making process with the help of a computer based on a statistical algorithm. Among the most obvious applications are predictive models, widely used in popular business applications, for example, to automatically provide recommendations to customers or for the loan approval process. Machine learning algorithms make decisions faster and cheaper in automated business processes than humans. Machine learning also promises to improve the quality of decision-making due to the presumed absence of human biases.
However, let's face it: algorithms and AI, in general, are just as prone to bias as humans. For example, clients with a long history of credit maintenance without delinquencies are usually defined as low-risk clients when assessing creditworthiness. But imagine that those customers' mortgages were paid off for years with substantial tax benefits that are no longer available. Of course, the degree of risk varies. However, if the program "does not know" about it, it cannot give an adequate assessment.
What's more, machine learning can perpetuate and even reinforce people's behavioral biases. We all face this problem on social media: news feed filtering is based on user preferences, thus reinforcing readers' natural biases. The site may even systematically prevent evidence of the opposing point of view from appearing. Therefore, algorithmic bias is one of the biggest risks, as it can cause costly mistakes for businesses, steering projects and organizational goals in the wrong direction. The good news is that with awareness of this problem, biases in algorithms can be detected and corrected.
Fighting “The Prejudice”: Addressing the Issues of Bias (complex three-level approach)
Actions can be taken to eliminate bias or protect against destructive effects. To do this, users of machine learning algorithms must first understand the shortcomings of the algorithm they are using and refer ML n from asking questions, the answers to which will be unequivocally wrong due to the biases of the algorithm. Using machine learning models is more like riding a car than an elevator. To get where you're headed, users can't just push a button; they must first learn about work procedures, traffic rules, and safety practices.
Second, data scientists who develop algorithms must shape data samples to minimize the possibility of bias. This step is an important and difficult part of the process. For the moment, note that available historical data are often insufficient for this purpose, and fresh, objective data must be generated through controlled experimentation.
Third, business leaders need to know when to use machine learning algorithms. They must understand the true values of the tradeoff: algorithms offer speed and convenience, while manually developed models (such as traditional decision trees or logistic regression) are more flexible and transparent.
Don’t Ignore the Humans: The Value of User Experience
Let's take a look at the role of users. From the user's perspective, machine learning algorithms are black boxes. They offer quick and easy solutions for those who know little or nothing about what's inside. They are used at their own discretion. But judgment must be based on knowledge. Business users seeking to avoid harmful applications of algorithms are somewhat similar to consumers seeking to eat healthy food: such consumers need to study nutrition literature and read labels to avoid excess calories, harmful additives, or dangerous allergens. Users should also review the algorithms they use in their business.
For example, when assessing creditworthiness, a built-in stability bias prevents machine learning algorithms from considering certain rapid behavioral changes in loan applicants. An item that often becomes a sign of risk in this context is the credit term. Customers with a higher degree of risk, as a rule, prefer loans with longer periods, taking into account possible difficulties in returning funds. On the contrary, many customers with a low level of risk seek to minimize interest costs by choosing loans with shorter repayment periods. A machine learning algorithm will pick up on such a pattern, giving loan applications with a longer term a higher risk rating. However, if the client, precisely to avoid receiving a high-risk assessment, chooses a loan with a shorter period, and then cannot fulfill his obligations due to a high amount of monthly payment (due to a short loan term), the system will not be able to react to such behavior that will cause credit losses to grow.
An institution considering the possibility of using an algorithm to solve a business problem must be guided by the desire to reach a tradeoff between costs and benefits. Therefore, when choosing machine learning algorithms, you should consider the following questions:
- How quickly do we need a solution? The time factor is often of paramount importance in solving business problems. An optimal statistical model can be out of date by the time it is completed. When the business environment changes rapidly, a machine learning algorithm developed overnight can significantly outperform traditional models that take months to build. For this reason, machine learning algorithms are better for fighting fraud.
- What is our analytical picture? Algorithms work depends on the data. If there is insufficient input, it is often better to bring in a consultant to help the organization develop it.
3. What problems should be solved? One of the promises of machine learning is that it can solve problems that were once unrecognized or considered too expensive to solve with manual models. When solving such tasks, institutions should identify those that significantly affect economic activity and attract the best data processing specialists.
Addressing “The Pride”: How to Embrace the Full Force of ML and be Better for it?
Can a business entity implement the newest, freshest tech and avoid the risks? Yes. It takes paining attention to a couple of critically important aspects:
- Business standards for validating machine learning: It is necessary to develop a template for the documentation of models that standardizes the process of accepting modeling applications. It should include business context and prompt questions with specific questions about business impact, data, and cost and expense tradeoffs. Such a process involves the active participation of users to find the most appropriate solution to a business problem (note that passive checklists or guidelines are generally ignored). The model's key parameters must be defined, including the standard set of analyses to be performed on the input data, the processed sample, and the simulation results. Finally, the model must be discussed with business users.
- Professional verification of machine learning algorithms: A clear process for validating and approving machine learning algorithms is needed. Depending on the industry and business context, especially the economic consequences of errors, it may not be as rigorous as a formal review of banks' risk models. However, the process should establish review standards and an ongoing program to monitor the new model. The criteria should consider the characteristics of machine learning models, for example, automatic algorithm updates when new data appears. If the algorithms are updated, for example, every week, the validation procedures should be completed within hours or days, not weeks or months.
- A culture of continuous knowledge development: Banks and other financial entities must invest in developing and disseminating knowledge in the field of information technology and business applications. There is a need to constantly monitor new ideas and best practices in machine learning applications to create a culture of improving knowledge and raising awareness of the challenges and benefits of using such programs.
Embrace the “Perfect Intelligence” as is: Why must banks become ML?
Machine learning can bring five benefits to your business. ML can bring benefits not only to large companies but to almost any organization, regardless of its size and direction of activity. As a result, the demand for machine learning technologies is rising. It may be premature to try to look for some kind of Moore's Law-type patterns, according to which the computing power of computers doubles exponentially every two years (formulated 50 years ago, it has only recently begun to fail), but there is no doubt that the machine learning industry is growing rapidly.
As algorithms get smarter and more banks and financial entities integrate this technology into their processes, almost every business entity should consider how to benefit from Machine Learning (ML). Here are key areas where this technology can help your business.
Processing huge amounts of unstructured data
One of the most popular arguments in favor of machine learning is that this technology allows you to process masses of information that cannot be realistically handled using more traditional approaches. This is especially true for small firms, which often have more transaction and customer data than they can process. How ML is used depends on the goals you have in mind. Perhaps the most important thing for you is to make more informed decisions about new product development. Or is the priority to attract new customers? Is the most important thing — to analyze and improve internal processes?
Automation of routine tasks
In the beginning, the biggest expectations associated with machine learning were related to increased efficiency. Even now, when this technology is far from being limited to automation, it remains one of its key functions. Using machine learning to automate routine tasks allows you to save time, manage resources more efficiently, and ultimately reduce expenses and increase revenues. The list of functions that can be automated with ML is very long. In particular, using machine learning, you can automate the process of classifying data, creating reports, monitoring I.T. threats, preventing abuse, and conducting internal audits.
Deepening Marketing Personalization
Machine learning is a powerful tool for personalizing marketing campaigns because it allows you to create unlimited advertising messages for unlimited users. Small businesses that don't have enough marketing experience can benefit from leading digital platforms (Facebook and Google) that have ML algorithms built into them. You won't need to try ML n your own algorithms to run microtargeting campaigns.
Identifying business trends
Machine learning has proven its effectiveness in identifying trends in large data sets. However, often these trends are expressed too weakly and, therefore, cannot be recognized by a person, or the volumes of data are too large to be efficiently processed by a "non-intelligent" computer program. Thus, small and medium-sized businesses are using ML to predict customer churn, looking for signals indicating customers are considering switching to competitors and launching processes to retain them. Companies are also increasingly integrating machine learning into their hiring processes. If the algorithms of previous generations, which reinforced the biases laid down by the developers, sometimes did more harm than good, then newer models can resist the inherent human cognitive errors and increase the probability of making informed decisions.
Acceleration of the development cycle
A machine learning algorithm in the R&D department is like an army of super-intelligent assistants that enable developers to use their time more productively. By partially eliminating the need for trial and error, which makes the development cycle longer, companies free up time to innovate. But the question remains — how do you get machine learning to work for your business? This raises the following question — what operational and structural changes will this technology bring? You may have to cut employees or even an entire line of business. As with all large-scale innovations that improve operational efficiency, ML will not benefit everyone equally. It depends on you how productively the new technology will be used and whether the transition to it will be orderly and relatively painless.
Better safe than sorry – the eternal security issue
During the pandemic, banking consumers have relied heavily on debt payments as The financial hardship resulting from the lockdown has put many jobs at risk, causing customers to be wary of the prospect of credit card debt. According to a recent study, $100 billion of annual credit card spending is expected to "move over" to debit cards. The debit payment method is based on the funds the consumer already has in their bank account. While a debit payment may provide some relief to a cardholder who is afraid of debt, there is a risk that debit card details could be stolen by scammers. While less than 1% of card purchases are scammed, customers who have been scammed must go through lengthy and stressful procedures before their money is returned.
Machine learning is a powerful and flexible tool in the fight against cybercriminals who try to compromise a debit payment. Advanced training technology can be especially useful in the fight against card fraud from overseas scammers. But according to experts in People's United Bank, it works best with a layered approach. In addition, location tracking allows financial institutions to detect suspicious transactions, such as when a user makes a mobile payment from one city while usually living or registered in another. However, this security technology makes it harder to attract new customers.
The pandemic and quarantine have accelerated payment habits toward cashless payments. As a result, users are more actively switching to non-cash payments and more often using e-commerce services. At the same time, the growing popularity of contactless payment instruments and settlements with them continues.
Hands-on: Examples and Use Cases of Machine Learning in Banking
Banks have always been innovators in safety and efficiency. Here are five examples of how leading U.S. banks invest in machine learning. Machine learning (ML) is currently one of the most important success factors in banking. This has been proven by five leading U.S. banks. Starting with the U.S. Federal Reserve. They recently published an official report on the state of affairs in the U.S. banking system. It describes all U.S. financial institutions with a capital of more than $300 million, and it is not surprising that the top five banks from the list have invested the most in introducing artificial intelligence and machine learning into their services and offerings:
- JPMorgan Chase
- Wells Fargo
- Bank of America
- U.S. bank
The specific areas of application of ML differ significantly (from sorting email using NLP algorithms and automatically updating information in a CRM system - to providing more convenient mobile banking services with an ML assistant.) For example, banks can machine learning models to predict stock price movements and better manage investments and stock trading.
We will consider each of these cases, and to illustrate the effectiveness of this approach, here are some interesting numbers for you - these five banks closed more than 400 local branches in 2022, moving customers to mobile banking with ML elements. And this not only did not lead to a loss of profit but also helped to achieve and exceed the planned profitability indicators and attract a substantial number of new customers. Why? Becanew offerings with ML elements exceeded customer expectations and led to the most effective type of marketing - word of mouth.
JPMorgan Chase and their COiN
To automate daily routine correspondence and reduce the time to analyze incoming business correspondence, JPMorgan Chase developed a licensed ML algorithm called Contract Intelligence or COiN is now used to parse documents and important aggregate data from them. This tool allowed the bank to process 12,000 loan agreements in a few seconds, while it took about 360,000 man-hours earlier. Now the bank is looking for new ways to use this tool to optimize business processes and generate additional profit.
Another JPMorgan Chase initiative, the so-called Emerging Opportunities Engine, was introduced in 2021 and has been steadily evolving throughout 2022 and 2017. This solution helps the bank to analyze transactions and find customers who are most likely to agree to purchase additional services. The answer was first used in equity trading and is now expanding into other markets, including debt capital trading. The bank is also investing heavily in developing its private virtual chat assistant, which is currently piloted by 120,000 customers and will soon be rolled out to all of its 1,700,000 customers. This will help save billions in payroll while providing first-class customer support 24/7.
Bank of America and Erica, their virtual assistant
Bank of America was one of the pioneers among financial companies to bring mobile banking to its customers 10 years ago. Then, a couple of years ago, they introduced Erika, a virtual assistant position, as the world's most cutting-edge innovation in payments and financial services. Using cognitive message analysis algorithms and predictive analytics, Erica is a financial advisor to more than 45 million Bank of America clients. In 2019, mobile banking served 12 million customers, which rose to 22 in 2022, indicating that the financial giant is focusing on technologies created in these four years.
By integrating an ML assistant into its mobile banking solution, Bank of America aims to reduce the burden of routine transactions to free up customer support centers to handle more complex cases faster, thereby greatly improving the overall customer experience. 2022 was the second highest-grossing year for Bank of America, which also reported spending $3 billion on technology innovation this year. The business entity is confidently striving for new heights and is constantly growing in the financial industry services market.
Wells Fargo Invests in ML0 Chatbots and Startups
Not so long ago (February 2017), Wells Fargo created a new AI & ML Enterprise Solutions team. The team focuses on connecting to the business entity's payment solutions, using ML to accelerate growth opportunities, and developing cutting-edge APIs to provide excellent services to enterprise-level customers. Just 2 months later, the team released an ML chatbot for the business entity's Facebook Messenger. This virtual assistant is used to reset your password and provide account information. As a result, what used to require filling out several pages of forms has become a simple dialog that takes only a few minutes. After testing with 700 employees, this handy feature will be rolled out to all customers, many of whom have been using Facebook Messenger to transact with Wells Fargo since 2009.
Citibank and Fighting Fraud with FeedzAI
Citibank has its own startup accelerator that brings together many technology startups worldwide. Most of these companies develop financial services and cybersecurity products. One of their most notable moves in this area was Citibank's significant investment in FeedzAI. It is a global enterprise that uses data science to detect and stop fraudulent transaction attempts across various financial businesses, including online banking and mobile banking. Feeds ML uses machine learning algorithms to analyze huge amounts of Big Data in real time and immediately notifies financial institutions of suspected fraud cases.
U.S. Bank is determined to use modern technology 100% and empower its business processes and services through machine learning and artificial intelligence. A Minnesota-based dedicated ML Innovation Leader group has been established to focus on developing conversational interfaces and chatbots to enhance customer service. For example, a machine learning continuous learning algorithm is expected to be able to help bank partners respond more quickly to rarely asked questions. However, this does not mean a complete rejection of human employees - at the moment. U.S. Bank Chief Innovation Officer Dominic Venturo told American Banker that branch employees should not be afraid of bots, as they are just a tool to help people be more productive, not a super-intelligence to replace them.
Why Banks are Increasingly Using ML: Things To Consider When Implementing ML in Banking
According to a study conducted and published by the World Economic Forum and Deloitte, 76% of executives in the banking industry agree that artificial intelligence is a top priority for the development of the sector. Traditionally, it was used for budgeting programs or digital tools, but now ML is needed for many other things. For example, to various segment payments, provide customer suggestions based on their history, advice, and answers to frequently asked customer queries using chatbots.
According to Business Insider, banks could save $447 billion by 2023 if they use artificial intelligence. The savings of $416 billion of this amount will come from the front office and the middle office. In the front office, biometric technologies and personalized offers based on artificial intelligence will reduce the interaction time between customers and employees. In the middle office, using ML will minimize the risk of cooperation with unscrupulous clients who launder money.
The Business Insider Intelligence report highlights the extent to which banks use artificial intelligence technology and real-life cases in the market. For example, 80% of banks understand that machine learning technology can benefit them, according to a survey of financial services professionals conducted by OpenText. Moreover, many are already actively using ML, for example, in Chatbots or to combat fraud.
- Security concerns: U.S. Bank uses artificial intelligence to improve the customer experience, reduce risk, fight fraud and money laundering, and make lending decisions. As ML-based decision-making tools proliferate, account managers can more accurately and consistently guide clients with the most relevant products and services for managing personal finances. So, the bank offers consultations using voice assistants: Google Assistant, Siri from Apple, and Alexa from Amazon. The technology is based on artificial intelligence. In addition, the bank collects data about customers and their transactions. It uses deep machine learning to look for patterns that will help identify unscrupulous customers, reduce fraud and ensure cybersecurity.
- Personalized offers: Citibank has prioritized using ML to interact seamlessly with customers. To this end, the bank has begun launching a new analytical platform that collects and processes customer data in real time. It helps Citibank create relevant and timely personalized offers. The group has developed a comprehensive customer profile showing real-time customer activity, such as how often they use apps or ATMs. API channels, artificial intelligence, and machine learning help the bank better understand customer behavior. This platform lets you see why the client calls the bank, improves service quality, and reduces call time. Citibank uses ML at every level of the organization to help retain customers, reduce unnecessary costs, and increase revenue by understanding customer needs.
- Exceptional User Experience: There are various artificial intelligence applications in the banking industry. For example, a growing number of startups, such as Trussle and Habito in the U.K., are looking to use machine learning algorithms to help clients find the best mortgage product on the market. As a result, banks can now develop more products with greater customer loyalty and lifetime value. Consumers can benefit from the convenience of working with a trusted organization that understands their personal needs. As ML-based decision-making tools proliferate, account managers can more accurately and consistently guide clients with the best unique finance products and services.
Account managers can also analyze customer experience with banking services through existing channels. This will allow banks to determine how well their current processes are performing, such as any bottlenecks. After that, they will be able to model and implement process optimization across all their customer service channels and improve the customer experience. ML is one of the already implemented technologies that is being used to improve operational processes at the bank is the use of machine learning to verify tax payments when processing payroll.
Machine learning is one of the areas of artificial intelligence that allows an algorithm to "learn" and "make conclusions" from past data and produce high-precision results, which, in turn, is not available to conventional software algorithms. The main benefits of using this technology, of course, include improved customer experience. After all, this allows you to automatically process payments without involving a person to perform such a check, which significantly speeds up the transfer of funds.
Shifting the routine work that people used to do to automatic systems, supplemented by machine learning technologies, allows workers to focus on more interesting intellectual tasks, positively affecting operational processes and internal NPS. However, a person is still not a robot or a machine; therefore, it is better to leave the routine work of the same type to work and algorithms.
Is it worth implementing ML in banking - final thoughts
As you can see, the top five U.S. banks have all invested heavily in ML and ML. They focus on providing the next level of customer self-service, fast document processing, using chats, fraud checking, loan portfolio management, etc. However, it is obvious that these are far from all possible applications, so let's take a look at more potential use cases for machine training in the banking industry:
- Personalization of offers: When customers see thousands of marketing offers, wherever they are and through any channel they interact with, only delivering the most personalized and relevant recommendations can grab their attention and push them to buy.
- Proper targeting of the consumer to ensure long-term partnerships. Analyzing big data stored in your CRM helps you better understand the needs and expectations of each customer. Clustering internal and external data points help create a holistic picture of the desires of each customer and form the right targeting for each type of customer.
- Improved UX that leads to better sales and conversions. ML algorithms can analyze the visitor's current device, location, open visit history, and business entity interaction history to provide a personalized user experience, which naturally leads to higher satisfaction and increased sales. For example, in a real-time personalization performance review, over 60% of respondents s ML d that using ML models to predict the best site structure for each visitor and adjust content accordingly helped them improve satisfaction and increase customer numbers.
- Self-service portals, Chatbots, and instant messengers. As millennials become the main target audience for banks and other financial institutions, businesses should stick to their preferences. By the way, 44% of U.S. citizens prefer interacting with chatbots rather than customer support employees. It's faster, less annoying, and much more convenient. This number will only grow with time, so the five popular U.S. banks invest a lot of resources and time in building web portals and mobile applications with ML algorithms instead of live agents. This is the right course for the future.
As you can see, these use cases for machine learning in the banking industry clearly indicate that the top five U.S. banks are taking ML very seriously. The ever-growing earnings of giants like JPMorgan Chase, Wells Fargo, Bank of America, Citibank, and U.S. Bank show that this is the right direction and the implementation of banking services through ML solutions is how the industry should develop in the future. What about other banks and financial institutions? They have smaller budgets. Can they effectively compete with the giants? The answer is YES, as access to machine learning and ML-infused platforms are getting cheaper yearly. After all, the question is not “who invested more?”. The question is, who invested better.
Looking for a hands-on team to imbue your banking system with ML algorithms? Contact our AI and ML experts for a consult.