Identify Fraud Patterns with Machine Learning

Fraud happens every day whether we notice it or not. Merchants may not even be aware that fraud is occurring at their store, and yet it can still happen. Preventing that fraud from happening is important to maintaining business revenue through customer trust and brand security.

And yet, in most cases, the merchant would not be responsible for the fraud. The insurance company would foot the bill if the stolen item or money could not be recovered. Therefore, it is even more important for financial services companies to prevent fraud.

What are some of the current pain points the financial services industry faces that are susceptible to fraud?

• Online scams
• Card not present (CNP)
• Identity fraud
• Multi-channel banking
• Credit card disputes

In fact, fraud attempts are on the rise.

“When analyzing transactions originating from the US, the rate of suspected digital fraud attempts* against financial services companies increased 109% when comparing the last four months of 2020 (Sept. 1–Dec. 31) and the first four months of 2021 (Jan. 1–May 1). Globally, the percentage of suspected financial services digital fraud attempts increased 149%.”[1]

So, what is the best approach to balance fraud risk management while maintaining a great customer experience? Through the accurate predictions and recommendations of machine learning.

Machine learning can help to identify:

• Patterns of potentially compromised merchants responsible for credit card fraud.
• Compromised time window and other credit cards that went through the merchant at that time.
• And much more!

Machine learning is not a fortune teller with a crystal ball. Rather, it uses real data to find past fraud and the likeliness of future fraud to occur.

Here are some real-world use cases that ElectrifAi’s pre-built machine learning model, Point-of-Compromise, has proven to be effective:

• A merchant’s employee or unrelated person at a specific location may skim credit cards or use credit card numbers on file to then make unauthorized purchases. This model uses that data input to identify merchants who may be compromised and prevents credit cards from being used at that merchant.
• Identifying where credit card fraud has taken place can be difficult to pinpoint. This model uses many data points to create a prioritization list of potentially compromised merchants to target the originating merchant and location responsible.
• The compromised time is also used to locate other potentially compromised credit cards. The issuing bank can be notified that these cards may have been compromised to reduce the risk of future instances of fraud.

Machine learning is a complex technology that requires experienced data scientists who can create products to solve real business problems.

For example, these are some technical highlights of the Point-of-Compromise Model:

• Merchant Fraud Risk scores powered by an ensemble of Neural Network models.
• The model can be used to augment (improve) other third-party transaction fraud models.
• Investigate overlaps of all the different credit card users that have reported fraudulent transactions and statistically identify the highest overlap in the merchant and time window.
• Analysis and prioritization tools for Investigation teams.

And a few data sources and features used to train the model:

• Signal Categories
• Pre-Fraud Rate Derived Variables
• First Fraud Similarity Comparisons
• Trended Variables
• Geographic Proximity Evaluation
• Advanced Merchant Statistics

Machine learning can be complicated and requires experienced data scientists to help you proceed. There are one of two ways to achieve the benefits of machine learning for your company:

• Start your own data science team and incur all the overhead costs.
• Partner with an experienced firm who have proven domain knowledge in the financial services industry.

Starting your own data science team is expensive and time-consuming. It requires:

• Recruiting data scientists who have experience in your industry and who understand the business requirements, not just the science.
• The overhead costs for that team can be a strain on your budget.
• The time to build the machine learning model and produce results can take a long time.

Partnering with a firm that has pre-built machine learning models used in the real world, however, can save you a lot of time and money. Founded in 2004, ElectrifAi can provide:

• Revenue uplift, cost reduction as well as profit and performance improvement
• A global team of domain expert data scientists and software engineers
• A proven record of transforming structured and unstructured data at scale

Our mission is to help organizations change the way they work through machine learning. We have a commitment to make artificial intelligence and machine learning more understandable, practical, and profitable for businesses worldwide.

Now is the time to put your data to work to prevent fraud. If you’d like to learn more, contact us for a custom demo!

[1] Transunion. (2021, June 3). Rate of Suspected Financial Services Digital Fraud Attempts Rise Nearly 150% Worldwide as Digital Transactions Increase. https://www.transunion.com/blog/global-fraud-trends-Q2-2021?utm_campaign=q2-quarterly-fraud-report&utm_content=blog&utm_medium=press-release&utm_source=press-release&utmsource=press-release.