Businesses lose billions of dollars per year to online payment fraud. Not only do fraudulent transactions impact revenue but they also compromise user trust and lifetime value. To prevent fraud, financial institutions must balance identifying criminal behavior with minimizing friction for trusted users. They need their fraud detection and prevention systems to consume large volumes of data, analyze and discover patterns, and drive decisions in real time.
Managing Tomorrow’s Fraud
Fraud systems rely on many micro-decisions to make an accurate assessment of potential fraud and to create a risk profile. These decisions are based on knowing the customer through the real-time synthesis of data (social, demographics, purchases, preferences, etc.), monitoring transactions across cyberspace and across every available channel, and analyzing the patterns in real time.
Specifically, these fraud systems:
- Ingest data from customers specific to the channel and enable firms to define policies, rules, and algorithms associated with that channel. For example, if a user signs in via a different mobile device or the user makes a payment every month but the past two transactions happened days apart, the platform should not make quick, rule-based judgments on a single pattern but make many micro-decisions across many elements and calculate a risk score.
- Gather details from mobile devices, including geolocation and device information, and combine those details with the profile of the card, account holder, and device to assess the risk of a transaction in real time.
- Drive risk-decisioning rules based on IP address, browser, fingerprint data, location identifiers, and device fingerprints.
- Monitor card-not-present (CNP) and other digital interactions and transactions, and use intelligence from the devices with transactional data such as shopping cart information, payment information, billing and shipping information, loyalty information, and product details.
- Produce a behavioral profile based on geolocation, device profiling, trust scores based on merchant activity, customer engagement with real-time alerting and notification, social network analysis, and link analysis.
Payment fraud needs to be managed in a continuum – consisting of detection, prevention, and recovery – in a way that allows for integration and customization of products and services based on the needs of each customer.
For digital payments fraud prevention strategies to be successful, the processes must integrate with transaction processing systems. This integration enables real-time interdiction and drives actions automatically. Automated systems can provide a comprehensive view of customer behavior by leveraging analytic calculations and algorithms to detect and flag suspicious payments activity. A core benefit of these new technologies is their delivery of low false positives. False positives impact revenue negatively.
Limitations of Conventional Payment Fraud Analytics Systems
Conventional fraud analytics systems are built on systems of record and designed to analyze large volumes of historical data to produce fraud insights and predictions. But these systems are siloed and not designed to keep up with the wide array of attacks on data and data sources. Many institutions perform manual reviews of transactions before initiation, an approach that is laborious, not scalable, and more error-prone than an automated strategy. The wide range of access points for financial information and activity gives fraudsters options to plan and execute their attack.
To keep up with this growing threat, payment providers must evolve from the traditional, siloed method of fraud detection to a proactive, analytic approach. The traditional models were trained on historical data, frozen, then weighted or adjusted in batches. This led to almost no co-operative learning and decision-making, as well as harmful business outcomes, such as customer abandonment, payment denial, fraud, missed cross-sell, and bad customer experience.
Learning in Milliseconds
Modern systems of engagement are incorporating a new generation of application architecture that eliminates the wall between transaction processing and analytics. Many companies are now building transactional analytics systems for fraud to complement their existing architecture.
Research and advisory firm Gartner refers to this as Hybrid Transaction/Analytical Processing (HTAP). An HTAP architecture is best enabled by in-memory computing technology to allow analytical processing on the same in-memory data store used to perform transaction processing.
By removing the latency with moving data from operational databases to data warehouses and data marts for analytical processing, this architecture enables real-time analytics and situation awareness on live transaction data. The ability to run analytics on live data and provide immediate feedback to the system is key to fraud deterrence.
The amount of data that needs to be processed or learned from can be massive. The data could consist of: billions of historical payment data points; analysis of activity correlated to hundreds of millions of devices; behavior and device mismatch across many locations; user actions, preferences, and interactions; geo-policies, dependencies, and myriad sets of third-party information; social and third-party consumer information; and e-commerce transactions.
It is important to note the efficiency and efficacy of the systems that prevent payment fraud depend on their power to harness data, analyze, learn it, and act upon it – with a high accuracy rate and at near-instant speed.
In the future, enterprises operating in the financial services industry will give even more attention to developing their own real-time, mission-critical fraud detection systems that require instantaneous response times, massive scalability, and the ability to accommodate diverse types of data. Most of these applications will need to be built on top of hybrid memory database architectures, which offer significant advantages over traditional NoSQL and relational database technologies.
Lenley Hensarling is the chief strategy officer of Aerospike, a leader in next-generation, hyperscale data solutions. He has more than 30 years of experience in engineering management, product management, and operational management at both startups and large successful software companies. He previously held executive positions at Novell, Enterworks, JD Edwards, EnterpriseDB, and Oracle. He has extensive experience in delivering value to customers and shareholders in both enterprise applications and infrastructure software. He believes that business is now happening in real time and that the right infrastructure for serving data to new real-time applications is a rapidly accelerating requirement for businesses to succeed.