Blog article
See all stories »

Banking on machine learning

New customer channels are changing the face of traditional banks and disrupting existing banking models. Rising mobile penetration has transformed the way consumers bank. Previously, routine tasks like getting a new checkbook, transferring money, viewing account balances, etc., required multiple visits to physical bank branches. Today, all of this can be accomplished with a few taps and clicks on one’s mobile banking app.

Not surprisingly, this has far-reaching consequences for traditional banking players. In 2016, Bank of America, Citigroup and JPMorgan closed 389 branches within one year. Interestingly, this trend does not signify a drop in business growth but a fresh understanding of the emerging banking habits of the digital consumer.

With such understanding, there is added pressure on banks to transform their service strategy and deliver game-changing products in innovative ways. Without constant innovation, banks will quickly lose their edge as competitors whisk away consumers with new offerings and customized products. One technology that can enable such transformation in front as well as back-end operations is machine learning (ML).

ML for financial services

Machine learning refers to the use of mathematical and statistical models to teach machines about new phenomena. It involves ingesting raw information in large datasets, understanding patterns and correlations and drawing inferences. While this may seem similar to how humans learn, machine learning algorithms ‘learn’ at much faster speeds with the ability to adapt from mistakes and course-correct. Needless to say, there are numerous applications of ML in any banking field that requires repetitive work (like back-office functions), high-accuracy tasks (like loan underwriting) or even informed decision-making (like providing financial advice).

Take data security, which is a key concern for banks. Deep Instinct, a cyber security company that leverages deep learning for enterprise security, states that new malware often contains code that is similar to previous versions. With this in mind, machine learning programs can easily identify typical user patterns and detect anomalous network behavior, making it easier to siphon resources into investigating legitimate cyber attacks.

Even in areas such as compliance, machine learning has the potential to infuse automation and run tasks that adhere to changing regulatory protocols with relative ease.

Applications of ML in banking

  • Customer service – In 2014, CGI conducted a study to understand what banking customers want. According to their findings, 52% of customers wanted banks to show them how to save money and 55% wanted access to wealth-building advice by leveraging their financial data. Today, three years later, these demands are becoming a reality. Banking chatbots use machine learning to understand customer behavior, track spending patterns and tailor recommendations on how to manage finances. Erica, a Bank of America chatbot, helps customers perform routine banking transactions while offering simple insights on improving finance management. Machine learning also uses insights from customer data to curate targeted offers and promote relevant products and services, thereby increasing customer satisfaction. 
  • Fraud detection – As mentioned earlier, ML programs can identify anomalous actions for near real-time fraud detection. The value of this in cost alone is significant considering that in 2016, customers lost nearly US $16 billion in identity theft and online fraud. In banks, machine learning can establish patterns based on the historical behavior of account owners. When uncharacteristic transactions occur, an alert is generated indicating the possibility of fraud. Such predictive analytics can also be used as an anti-money laundering (AML) tool to trace the true source of money by identifying disguised illegal cash flow – a tactic commonly used when laundering money.
  • Risk assessment – Banks are always trying to improve loan approval processes – this is another area where machine learning can play a key role. Through intelligence gained from various data sources such as credit scores, financial data, spending patterns, etc., ML algorithms can prescribe accurate risk scores and predict the possibility of a user defaulting on a loan. Armed with such information, banks are better positioned to tailor loans, craft relevant terms and conditions and customize services to suit different customer profiles.  
  • Trading – With the ability to process large datasets at high speeds, machine learning algorithms can make instant predictions for stock market trading based on volatility patterns, risk, latest news, etc. Investment banks such as JPMorgan Chase, United Bank of Switzerland (UBS) and Goldman Sachs, to name a few, have gone beyond mere automation to leverage machine learning algorithms to track trading volatility and support equities trades as well as asset and wealth management. In fact, UBS uses a system that can scan directions in client emails on how to allocate funds across different trading blocks and then processes and executes them. This reduces the time taken to execute custom trades from 45 minutes to 2 minutes, achieving over 95% time savings and freeing up investment bankers for more complex tasks.

Benefits of ML

  • Lower cost – When applied in contact centers, machine learning coupled with natural language processing (NLP) and contextual engines can complement customer relationship management (CRM) systems to resolve customer queries with relevant and real-time responses. This enhances the customer experience by offering personalized service at a lower cost – lower than the cost of training a human agent.
  • Greater revenue – Targeted and relevant communication about new products and services provides greater opportunities for cross/up-sell, thereby increasing customer satisfaction and driving greater business growth.
  • Higher productivity – By providing continuous automation, ML injects new efficiencies into banking processes and operations. It slashes time taken to complete routine tasks, ensures higher accuracy with minimal errors and reduces turnaround time for various bank-related activities. Besides delivering productivity gains, ML also frees up the workforce to focus on value-addition and client-facing roles.
  • Better compliance – As an autonomous and self-learning technology, machine learning can be easily programmed to adhere to various regulatory protocols with minimal human supervision. This not only simplifies the complexity of meeting regulatory standards but also helps avoid penalties due to non-compliance.

Look before you leap – the way forward

Complex IT landscapes riddled with legacy systems pose several challenges when adopting new technologies. A collaborative study by the National Business Research Institute reveals that 12% of traditional financial institutions found AI to be ‘new, untested and risky’ while others cited issues of regulatory compliance and disparate datasets as barriers to adoption of ML.

As a disruptive technology, machine learning and AI-driven solutions can transform banking as we know it. However, it pays to err on the side of caution. Here are some aspects to be considered before adopting ML:

  • Displaced workforce: According to the US Bureau of Labor Statistics, automation will cause an 8% decline in the numbers of bank tellers between 2014 and 2024 and the number of insurance underwriters will drop by 11%. As machine learning replaces manual jobs, banking organizations must institute a plan of action on how to handle the displaced workforce. This may involve re-skilling programs and staffing workers in value-generating areas.
  • The right strategy: In a 2015 introductory article on ML, McKinsey warns organizations that, without the right strategy, ML will be relegated to routine applications that conduct repetitive tasks. The potential of ML transcends these boundaries when used correctly. To achieve this, there must buy-in from senior leadership with a strong vision and strategy on implementing ML. In fact, an article by PricewaterhouseCoopers (PwC) recommends adopting AI in two different streams – one in operational areas to demonstrate the tangible benefits of AI and the other in emerging areas to demonstrate the value of innovative insights that humans cannot uncover. For instance, at an operational level, ML can replace back-office tasks pertaining to account opening demonstrating immediate benefits such as time, effort and cost savings. On an innovation level, ML can help banks discover unique metrics that can attribute accurate credit scores to individuals without a credit history by using other behavioral data.
  • Open information: For ML to succeed, it needs access to large datasets. This means organizations must make data freely available to machine learning software so they can ingest inputs and churn out accurate insights and predictions. However, care must be taken to create the right protocols when dealing with databases that contain sensitive and protected customer data to prevent violations.

Today, new players are rapidly gaining higher market share by attracting customers with digital products and innovative services – many of which are made possible through machine learning and artificial intelligence. Such players can easily meet customer demand as they already have lean and agile digital operations. To maintain leadership, traditional banks must keep pace with the new normal of rapid disruption.


Technology is transforming the way consumers behave and this is most evident in the banking industry. With its capacity to learn from large datasets and establish patterns and correlations, ML can revolutionize banking operations. It can inject new efficiencies into tasks such as risk assessment, fraud detection, anti-money laundering, trading, and customer service by providing instant insights, relevant recommendations, and informed decisions in real-time. Such capabilities will help banks optimize operations to reduce cost, improve compliance and increase productivity, thus leading to higher revenue. However, it is important for organizations to establish a clear vision and strategy, ensure information is openly available and roll out change management programs to ensure successful ML-powered transformation.  


Comments: (2)

A Finextra member
A Finextra member 03 October, 2017, 05:56Be the first to give this comment the thumbs up 0 likes

Real insightful article Ajay, liked all the use cases for ML that you've outlined. I think a major challenge for banks in embracing ML wholeheartedly also lies with finding the right talent with ML skills. 

A Finextra member
A Finextra member 23 July, 2019, 22:52Be the first to give this comment the thumbs up 0 likes

We are seeing banks look hard at startups like Sigma IQ ( who are bringing a machine learning approach to harder problems like account reconciliation as part of a new finance tech stack. Startups with ML skills need to pay more attention to finance problems as there is so much data to work with in order to improve efficiency.

Member since




More from member

This post is from a series of posts in the group:

Innovation in Financial Services

A discussion of trends in innovation management within financial institutions, and the key processes, technology and cultural shifts driving innovation.

See all

Now hiring