Skip to main content

Unlock customer credit insights with machine learning

Executive summary

The credit card industry is becoming more complex. Advanced loyalty, targeted offerings, unclear rate conditions, and many other factors can often make it difficult for banks to identify the right customer.

Ultimately, the financial services firms that will succeed in this environment will engage the right customers with the right message at the right time.

Market leaders will be those who can accurately forecast the revenue and risk for each prospective and existing customer.

While the credit card environment has changed, the analytics and modeling techniques have largely remained the same. These models are highly valuable, but do not offer flexibility to evaluate granular and complex customer behaviors incumbent in a financial services firm’s data and other public and private data sets.

Machine learning and deep learning (collectively, machine learning) change the paradigm for predictive analytics.

In lieu of complex, expensive, and difficult to maintain traditional models, machine learning relies on statistical and artificial intelligence approaches to infer patterns in data, spanning potentially billions of available patterns.

These insights, not discoverable with traditional analytics, may empower the financial industry to make higher-value, lower-risk decisions. In this brief article, we discuss three potential opportunities that Fusion expects should add high value to the financial services industry.

Advanced analytics for banking

Machine learning uncovers patterns in complex data to drive a predictive outcome. This is a natural fit for the banking industry as firms are often working with imperfect information to determine the value of incoming customers.

How it works: Traditional models vs. machine learning

Credit scorecards represent the basis of most credit card issuance decision making. Whether a firm leverages off-the-shelf models or applies bespoke modeling, Fusion expects the following is representative of a credit scorecard:

FICO Score Distribution: Comparison by Age from 2014 Fair Isaac Corporation FICO Banking Analytics Blog

In the aggregate, these models are highly valuable. But on a per-applicant basis, patterns and details are lost.

In machine learning, we can explore detailed and expansive public and private data about segmented applicants for marketing purposes in real time. For example, we can supplement our existing models with data that can be used to segment potential customers such as:

  • Regional FICO trends
  • Educational attainment
  • Social media sentiment analysis
  • Mortgage and equity analysis
  • Much, much more

Machine learning can apply artificial neural networks to uncover patterns in your applicants’ history across millions of data points and hundreds of model statistical training generations. When detecting these patterns, machine learning models can uncover risk in approved applicants and value in sub-prime applications.

For example, by exploring existing customers, machine learning could potentially reveal that applicants with low FICOs but high educational attainment for a specific city suburb have historically resulted in minimal write-offs.

Conversely, a potentially high FICO applicant may have recently moved into a higher-net-worth neighborhood, requiring a high expenditure on a financial institution’s credit lines, resulting in repayment risk.

Ultimately, your customer data can tell a far richer story about your customers’ behavior than simple payment history.

Machine learning opportunities

Financial services firms can gain more insight and capitalize on the benefits of machine learning by applying their marketing dollars towards customers who are more likely to fit within their desired financial portfolio.

Lifetime customer value for customer with limited credit data

Currently, credit score is determined based on traditional data methods. Traditional data typically means data from a credit bureau, a credit application, or a lender’s own files on an existing customer.

One in 10 American consumers has no credit history, according to a 2015 study by the Consumer Financial Protection Bureau (Data Point: Credit Invisibles). The research found that about 26 million American adults have no history with national credit reporting agencies, such as Equifax, Experian and TransUnion.

In addition to those so-called credit invisibles, another 19 million have credit reports so limited or out-of-date that they are unscorable. In other words, 45 million American consumers do not have credit scores.

Through machine learning models and alternative data (any data that is not directly related to the consumer’s credit behavior), lenders can now directly implement algorithms that assess whether a banking firm should market to the customer segment, thereby assigning customer risk and scores, even to credit invisibles (thin-file or no-file customers).

Let’s look at a few sources of alternative data and how useful they are for credit decisions.

  • Telecom/utility/rental data
  • Survey/questionnaire data
  • School transcript data
  • Transaction data – This is typically data on how customers use their credit or debit cards. It can be used to generate a wide range of predictive characteristics
  • Clickstream data – How a customer moves through your website, where they click and how long they take on a page
  • Social network analysis – New technology enables us to map a consumer’s network in two important ways. First, this technology can be used to identify all the files and accounts for a single customer, even if the files have slightly different names or different addresses. This gives you a better understanding of the consumer and their risk.

Second, we can identify the individual’s connections with others, such as people in their household. When evaluating a new credit applicant with no or little credit history, the credit ratings of the applicant’s network provide useful information.

Whether a bank wants to more efficiently manage current credit customers or take a closer look at the millions of consumers considered unscorable, alternative data sources can provide a 360° view that provides far greater value than traditional credit scoring.

Alternate data sets can reveal consumer information that can increase the predictive accuracy of the credit scores of millions of credit prospects. This allows companies to target consumers who may not appear to be desirable because they have been invisible to lenders before, which can lead to a commanding competitive advantage.


ON-DEMAND WEBINAR: Learn how to turn data into insights that drive cross-sell revenue


Optimizing marketing dollars to target customers

Traditional marketing plans for credit card issuers call to onboard as many prime customers that meet the risk profile of the bank. However, new customer acquisition is only one piece of the puzzle. To drive maximum possible profitability, banks can consider not only the volume of customers, but also explore the overall profitability of a customer segment.

Once these high-value customer segments are identified, credit card marketers can tailor specific products to these customer segments to deliver high value. Machine learning can assist both in the prediction of total customer value, as well as the clustering of customers based on patterns and behaviors.

Average return on risk by credit score chart

Identifying high-risk credit card transactions in real time

Payments are the most digitalized part of the financial industry, which makes them particularly vulnerable to digital fraudulent activities. The rise of mobile payments and the competition for the best customer experience push banks to reduce the number of verification stages. This leads to lower efficiency of rule-based approaches.

The machine learning approach to fraud detection has received a lot of publicity in recent years and shifted industry interest from rule-based fraud detection systems to machine-learning-based solutions. However, there are also understated and hidden events in user behavior that may not be evident but still signal possible fraud.

Machine learning allows for creating algorithms that process large datasets with many variables and helps find these hidden correlations between user behavior and the likelihood of fraudulent actions.

Another strength of machine learning systems compared to rule-based ones is faster data processing and less manual work.

Machine learning can be used in few different areas:

  • Data credibility assessment – Gap analytics help identify missing values in sequences of transactions. Machine learning algorithms can reconcile paper documents and system data, eliminating the human factor. This ensures data credibility by finding gaps in it and verifying personal details via public sources and transactions history.
  • Duplicate transactions identification – Rule-based systems that are used currently constantly fail to distinguish errors or unusual transactions from real fraud. For example, a customer can accidentally push a submission button twice or simply decide to buy twice more goods. The system should differentiate suspicious duplicates from human error. While duplicate testing can be implemented by conventional methods, machine learning approaches will increase accuracy in distinguishing erroneous duplicates from fraud attempts.
  • Identification of account theft, unusual transactions – As the rate of commerce is growing, it’s very important to have a lightning-fast solution to identify fraud. Merchants want results immediately, in microseconds. We can leverage machine learning techniques to achieve that goal with the sort of confidence level needed to approve or decline a transaction.

Machine learning can evaluate vast numbers of transactions in real time. It continuously analyzes and processes new data. Moreover, advanced machine learning models, such as neural networks, autonomously update their models to reflect the latest trends, which is much more effective in detecting fraudulent transactions.


Bottom line: machine learning can leverage your data to develop patterns and predictions about your customers and applicants. These machine learning models are typically simpler to develop and deploy and may be more efficacious than traditional financial services modeling.

These models also enable a more detailed forecast about your customers, allowing you to reduce risk while targeting more profitable customers through their lifetime with your credit card services.

Related resources

Fusion Alliance has extensive experience in the financial services industry and serves as a preferred solutions provider for many prominent financial services institutions, including Fortune 500 firms. If you’d like to discuss your organization, let us know.

About the author

John Dages

John Dages is Fusion’s Machine Learning Solution Director. With over 15 years of technology leadership experience, John brings a unique and essential perspective to enterprise customers’ cloud and advanced analytics journeys. By leveraging his data, application development, systems integration, and I&O background, John offers a complete picture of how companies can utilize machine learning to drive a competitive advantage and engineer true intellectual property.

Ready To Talk?

Let us know how we can help you out, and one of our experts will be in touch right away.