How predictive models can drive competitive advantage in retail banking 🏆

Innovative applications of machine learning are transforming many different industries, and retail banks that provide deposit and lending service to consumers have a significant opportunity to capitalize on this disruptive technology. Incumbent banks face increasing competition from financial technology companies, however in the arms race to deliver better customer experiences they have a critical advantage – enormous amounts of historical data.

Machine learning empowers businesses to leverage their historical data, and retail banks and credit unions have an imperative to capitalize on this opportunity. This article discusses a few of the ways that machine learning can empower retail banks to drive better customer experiences and improved bottom-line results. For a primer on machine learning generally, see this post.

Machine Learning Use Cases

1. Bank Account Opening

The account opening process is an opportunity to make a great first impression with a customer and an area that many banks have struggled to automate due to process requirements, fraud risks and increasingly complex regulatory frameworks.

Machine learning and predictive analytics can have a profound impact on account opening. By automatically analyzing historical data – which typically includes information about historical applications and whether the application violated any requirements – predictive models can be trained that streamline the process by automatically flagging applications for approval, manual review, or rejection.

In certain cases, due to compliance requirements a machine learning model’s “decision” may not be sufficient. In these situations, banks can still drive significant process improvements by using predictive models as decision support tools that assess risk and provide recommended actions. This assessment becomes part of a broader review, reducing the manual effort required to make account opening decisions.

2. Transaction Processing (Fraud Detection)

Retail banks can also use machine learning to identify suspicious transactions. By analyzing historical transactions, models learn when to flag applications for review. When live transactions occur, the program prevents the transaction from processing without additional review.

This quick response can dramatically reduce the occurrence of wire fraud and money laundering. However, it has a potential downside – in some cases, it creates worse customer experiences by flagging legitimate transactions for review. With the development of increasingly powerful machine learning algorithms and banks’ growing data sets, automated machine learning platforms such as DigiFi’s are increasingly able to reduce this risk by building highly accurate models that flag real issues without preventing legitimate transactions.

3. Credit Underwriting Decisions

Most retail banks have significant lending operations, including consumer loans (personal loans, credit cards, auto loans, mortgage, etc.) and business loans (small business and commercial). Credit underwriting – the process of determining which applicants to lend to, and at which rates – is therefore a critical analytical process that retail banks and credit unions devote significant resources to optimizing.

Machine learning can greatly strengthen credit underwriting in a number of ways:

  • Estimating the probability that a borrower will default if a loan is issued
  • Predicting whether an applicant will prepay their loan
  • Recommending the ideal interest rate (which the customer will accept but will also generate a profitable return)

Credit underwriting is typically one of the easiest areas to implement powerful machine learning because banks and credit unions have large historical data sets on the loans they’ve issued. Basic forms of machine learning, including logistic and linear regression, have been used for many years by the largest banks, and the industry is now beginning to use forms of “deep learning”, such as artificial neural networks, which can further strengthen predictive models when applied to large data sets.

4. Workflow Prioritization

A lesser discussed opportunity for machine learning at banks lies within workflow management. Predictive models can drive significant operational efficiencies by eliminating key areas of manual judgement from business processes and properly prioritizing personnel.

A good example of this is application queue prioritization – which refers to the order in which a loan officer reviews a set of applications. In many cases, loan officers make their own decisions regarding the order to review applications, resulting in inconsistencies and inefficiencies within banks. Machine learning can be leveraged to predict things like “Which application is most likely to close?”, “Which application will be most profitable” or “Which customer is most likely to leave if I don’t respond immediately?”. The answers to these questions can be used to automatically prioritize workflows, resulting in better customer experiences and better business results.

5. Predicting NSF/No-Pay on Loans

Machine learning can also be used within loan servicing to predict which customers will miss their payments if action is not taken.

Predictive models analyze large volumes of data regarding historical payments and then accurately determine which customers are unlikely to make their next payment. This valuable information can be used to prioritize bank staff to focus on high-risk customers, prepare the bank for upcoming defaults and as an early-indicator for identifying potential issues with the underwriting process.

6. Product Recommendations

Knowing which product a customer is most likely to be interested in is highly valuable information that can have a major impact on marketing decisions. Machine learning can also assist with this, by analyzing customers’ prior history and recommending which product they’re most likely to want in the future (specific credit cards, different bank account features, etc.).

The recommendations can take a number of forms, including:

  • Recommendation models, which produce a single result (i.e. the most likely product)
  • Probability prediction models, which predict the probability that a customer will want a given product (i.e. each product is assigned a probability)

Recommendation engines typically require significant amounts of data, because consumer behavior can change quickly. Therefore, this type of machine learning is usually only available to large institutions with excellent data management practices.

Potential Compliance Concerns

The reality of modern-day machine learning is that it can produce extremely accurate results, but typically in a “black box” fashion. Specific decisions are hard to explain because the models used to make generate the decisions are exceptionally large and complex.

In most cases this is fine – we often don’t understand why humans make certain decisions either! However, in the highly regulated banking industry, this can be a cause for concern in areas such as credit underwriting and account opening.

So, how do you benefit from machine learning while staying in compliance? The answer differs based on the institution and specific use-case, but a good option can be to use predictive models as a decision support tool as opposed to a decision making tool, especially when rejecting transactions or applications. For example, if a model predicts a low fraud risk you may be able to automatically approve the transaction, whereas if a model suggests a high fraud risk, you may need to proceed with additional review to confirm or reject the model’s decision.


We continue to see more and more retail banks and credit unions find ways to use machine learning across their organization to drive powerful improvements. Increasingly large data sets, continued enhancement of deep learning models and ongoing compliance acceptance will accelerate the pace of adoption, which will benefit both retail banks and their customers.

About DigiFi

DigiFi is a technology company that helps businesses make better automated decisions.

Our platform lets businesses easily use automated machine learning and rules management to optimize critical decisions with no coding or technical expertise required. Repetitive work that used to take hours can now be completed in minutes, letting your team focus on what matters most.

Learn more at