The Simility Blog
Omnichannel Fraud Prevention for the Digital Age Banks
Swastik BihaniDecember 26, 2017
Consumers are engaging with banks more each year using multiple channels, products, and devices. As banks increase their communication and product channels to provide greater customer convenience, their fraud management challenges are increasing multifold. This is because a customer can go back and forth between these various channels and create newer scenarios that may not exist in the more uni-channel model. As a result, the focus must shift from detecting fraud in isolated channels to identifying anomalous behavior across channels.
Banks that want to provide a frictionless consumer experience while still detecting and preventing fraud need a comprehensive view of users and transactions across products and channels. Here are three top considerations when evaluating a fraud prevention solution for your omnichannel business:
Ingest any type of data
Since the goal is to create a 360-degree view of the users across products and channels, it’s imperative that the fraud solution be able to ingest data from any channel or end-user device. Not all data is created in the same way – hence the solution should be able to ingest structured, unstructured data, in real time as well as batch mode, along with data like biometric data, device fingerprint, IP, location, etc. Customers should also be able to integrate other feeds that they’ve invested in over time – both homegrown ones as well as external third-party feeds. The solution should provide smart-ingest capabilities that can significantly simplify the incorporation of data from any source or format. These enhanced fraud prevention capabilities can further streamline processes with greater flexibility, leading to faster analysis while driving deeper insights.
Run concurrent machine learning models
Machine learning (ML) is a natural fit for fraud detection as it is great at identifying patterns spread across a large number of rows and columns of data. However, a bank has multiple areas where ML can be used – e.g. ATO detection, fraudulent transactions, chargebacks, etc. – each requiring a unique ML model potentially running on the same set of data. Hence a superior fraud detection solution should be able to run multiple ML models from the data lake’s myriad of data types in parallel. It should provide different, better-suited models for different scenarios through its rich library of both supervised and unsupervised ML models.
Provide a seamless workflow with insightful dashboards and reports
Effective, efficient case analysis requires intuitive visualization tools that can assist fraud analysts to review their cases more comprehensively with relevant data. Further, quickly adapting to evolving scenarios requires understanding the schemas of these varied data sources and profiling various entities—be they users, events, devices, etc. By using these visual tools and the overall workflow element, analysts can quickly drill down to the most relevant information and make a decision on the transaction.
Finally, the solution should give omnichannel businesses the opportunity to set their own rules and configurations to manage fraud movement across multiple channels, evolving with the business or as threats mutate.
Case Study: How Simility’s adaptive fraud detection solution helped a bank prevent omnichannel fraud
Effectively stopping the fraud that is spread over multiple channels is a journey. Working with a leading Fortune 1000 bank, we planned it in various phases to ensure that we’re headed in the right direction with incremental steps while enabling real measurable business value.
STEP 1: Successful deployment
To ensure a smooth on-premise deployment, we spent time with the bank to ensure that the deployment was well orchestrated as it had a lot of moving parts. These included procuring the infrastructure, firewall rules, network interconnects, connecting to other sources – e.g. IBM MQ, etc. Once the infrastructure was up and running we set up a rules-based model to detect transactional wire fraud while ensuring in-depth customized reporting and dashboarding. In the process, we were not only able to deliver a more efficient solution but also get them comfortable with Simility’s tool and workflows.
With this initial deployment completed, we took on the next piece, mobile banking.
STEP 2: Enabling an additional channel via the mobile app
As we planned to integrate the mobile channel the key goals were to:
- Provide better fraud protection for the mobile channel
- Create a more holistic user view via the multi-channel enablement
To do so, we ensured that the data entities were created from a multi-channel perspective. In addition to the mobile-specific rules, we also added user specific behavioral rules that incorporated various aspects of our device fingerprinting solution. With this technology in place, we were able to analyze events like funds transfer, beneficiary list updates, unusual spending patterns, etc. By adding more capacity to the initial deployment, we clearly demonstrated that the system was linearly scalable as loads increased. We could also now showcase the user 360-degree view of events (forgot user id, forgot password, adding beneficiary, making a transaction, etc.). This allowed us to investigate many cases, including account takeovers, check kiting, and more.
The next milestone was to start incrementally enabling additional channels.
STEP 3: Add additional channels
Having completed the other first two steps successfully, the next steps were to add additional channels and products to start getting a more holistic view of the user across various products and channels. These channels and products included IVR, teller based transactions, web transactions, loan origination, card transactions, etc.
Simility helped the bank continually explore and leverage new data sources to identify fraudsters, while providing a seamless experience to good customers.
Are you still using rule-based fraud prevention solution to detect fraud for your evolving business? Are you getting complete customer (OR fraudster) insight from your data lake?