|The Challenge||A large regional bank was looking to improve revenues, and improve the customer experience. As part of their growth efforts, they put increasing emphasis on reducing the occurrence of fraudulent debit card transactions. The bank was losing about $100,000 per month and risked negatively impacting customers through rejected transactions and reissued debit cards. The bank’s existing third-party detection solution was adequate for detecting coarse-grained indicators of fraud (e.g., unusually-large purchases, customers exceeding their credit limit) but other types of fraud remained undetected.|
|MapR Solution||The bank chose the MapR Converged Data Platform and the MapR Professional Services (PS) team to help them develop a more effective fraud detection capability. This bank was interested in the MapR Platform because they were already using it successfully for a data warehouse offload solution and were reaping the benefits of an enterprise-grade, highly reliable platform for customer data. However, they did not have the in-house data scientists who were well-versed in big data modeling and data engineers to quickly develop a custom fraud detection solution.|
|MapR PS developed a fraud detection model that enables the customer to choose a fraud detection threshold that maximizes the savings per account and minimizes the operational costs of detecting the fraud. The table below shows an example of that model in action.|
|Monthly Savings||Many third-party fraud solutions are not sophisticated enough to be customized for the subtleties of a particular business. The bank has a lot of other proprietary customer information that can help make fraud detection much more accurate. A MapR data scientist explained that if all these data sources could be brought together in one place, the bank could gain new insights beyond just the credit card transactions, leading to far more accurate fraud detection.|
Fraud Detection with Precision
A big data approach to fraud detection can determine the precise incremental value to the business of a fraud prevention engagement with MapR Professional Services.
A live test with the bank’s call center was conducted during the 3rd week of the engagement with MapR Professional Services. High-scoring transactions were routed to operators, along with some probable reasons why the account looked suspicious.
This resulted in several new frauds being detected in real time.
The process of a PS engagement will vary depending on the project and customer. This engagement was roughly divided into three phases: discovery, solution building, and the transition. At the beginning, the MapR team came in, described what they could do, what it might look like, and what the deliverables would be. In this example, MapR was hired for a three-week Professional Services Quick Start Solution engagement.
Phase 1: Discovery. In the discovery phase, the MapR data scientist asked a lot of questions to understand the nuances of the bank’s business and its customers. In some cases banks don’t even know the full extent of the problem. The first discovery made was to identify and show the bank its monthly fraud exposure.
Since fraud is slightly different in every organization, it’s important to understand all the nuances of the business and what options a bank has to take steps to prevent fraud. The MapR data scientist talked to the bank’s fraud experts, business leaders, and call center reps to understand the extent of the problem, how they thought the fraud was happening, and what they wanted to accomplish. Collaboration between the data scientist and business stakeholders is key to transitioning ownership and maintenance of the model to those in the customer who must own the process. This customized discovery phase lays the groundwork for the rest of the engagement and the solution building.
Phase 2: Solution Building. Once the MapR data scientist clearly understood the problem and the context, he got to work developing the solution. During this process, domain experts at the bank provided anecdotal information of accounts, bank policies and how their customers interact with the branches to make sure they get the most out of their data. The MapR team then explored the data, started to develop a model, and experimented with adding different data sources and trying different algorithms. Development is an iterative process, where different machine learning models are tested, and performance and value are assessed. The processes is repeated, the model is trained, and the solution is continually customized and improved to produce the optimum value.
Phase 3: Transition. Once the solution is developed, the MapR team did a series of presentations and knowledge transfers to help the client understand what was built, how to use it and the value they can get from it. A live test with this bank’s call center was conducted during the 3rd week of the engagement. High-scoring transactions were routed to operators, along with some probable reasons why the account looked suspicious. This resulted in several new frauds being detected in real time. The MapR team collaborated closely and gave constant updates to the customer throughout the engagement. At the end of the engagement, the MapR team gave the bank’s team what they needed to maintain this process and improve it over time.
Fraud prevention through predictive analytics is fast becoming a permanent part of a bank’s arsenal of tools to combat bad actors and to improve bottom lines. It goes beyond traditional packaged fraud detection approaches because the fraud detection models are custom tuned for the particular data that is available for this bank. For this particular engagement, the MapR solution was able to quickly reclaim an average of $2 per account in fraudulent transactions. Given the bank’s large customer base, this quickly added up to substantial savings.
There are multiple benefits to engaging with MapR Professional Services. It’s challenging and time consuming to find and train people with the right expertise in big data technologies such as Hadoop and Spark. With MapR services, the customer works with world-class data engineers and data scientists who can immediately help them get the most out of their big data investment. It’s a lot less expensive to build something designed by experts right from the start..
Customers also get to see results within days or weeks instead of the typical six-month project cycle. In three weeks, the MapR PS team can help the client see what is possible. The customer gets a huge head start and can save a lot of time, effort and budget.