Low Latency Fraud Detection & Prevention
Digital transformation’s acceleration has increased exposure to online frauds and scams, and according to a new report by Grand View Research, the rise of incidences—whether it’s mobile payment fraud, email phishing, or card fraud—increases the downstream impact on businesses and subsequent financial losses.
A PwC survey showed that in 2020 about 49% of companies had experienced fraud amounting to $42B in losses, and the cost of fraud for banks grew by 17% since 2019. This reality is increasing the investment companies are making in fraud detection and prevention (FDP).
Consumers interact with businesses via multiple digital touchpoints, and the digital platforms and online banking systems that support them are complex, causing the platforms to suffer from being slow, inefficient, and expensive. These fraud detection and prevention platforms require a system architecture that’s designed with sound-first principles including speed, efficiency, and cost effectiveness. Traditional models cannot detect fraud in line with the growing rate of transactions. Time is money and milliseconds matter significantly when trying to detect and prevent fraud—speed is a high-order bit, and an important first principle.
What do we mean by first principles? First principles are foundational propositions or assumptions that stand alone. Therefore, one cannot deduce first principles from any other proposition or assumption. These principles establish a foundational baseline, and from there one organization can build new solutions to address complex problems. In the context of fraud detection and prevention, speed is the first principle. There are other principles as well such as data consistency, model accuracy, end-to-end security, and architectural simplicity, but detecting and preventing fraud requires fast computational processing.
Thanks to modern software platforms, transactions are executed nearly instantly, but the same processing speed that creates a great experience for customers leaves banks and payment processors with less time to identify and react to fraud. Companies lose tens of billions of dollars to fraud each year in the form of fines, settlements, and erosion of the trust and customer loyalty that underpins the financial services industry. And the increased complexity, volume, and speed of today’s online transactions mean organizations will need more advanced methods of fraud detection to keep up with malicious actors.
Across the industry, there are many examples of companies successfully innovating their digital platforms with accurate and fast fraud detection and prevention systems. For example, PayPal created a data service that efficiently leverages data pipelines in real-time to make accurate decisions in detecting fraud. The underlying system handles billions of events per day and maintains dozens of petabytes of data. Similarly, DoorDash’s risk team is using machine learning models for payment protection, preventing thousands of dollars a day in fraud losses, with increased efficiencies amounting to two to three times.
Taking these examples into account, along with so many others across the industry, the ability to enable fraud detection and prevention in real-time is a key differentiator. What matters most in architecting a real-time platform to address fraud? In working with several customers, we identified three specific areas that matter a great deal. First is the ability to update digital identities in real-time. Secondly, using AI models to increase accuracy and speed of detection. And in third, leveraging high-speed statistical analysis to reduce cost. There needs to be an underlying technology that can meet these specific needs, while also offering the right developer experience and flexibility in activating these capabilities. This is where Redis Enterprise can be considered. Let’s expand each of these three capabilities.
Keep digital identities updated in real-time
Since consumers expect a responsive digital experience, it’s important to update digital identities in real-time to provide a seamless consumer experience. A user’s identity can change based on a variety of inputs including geolocation for example. A consumer’s score can be influenced by suspicious connections they may have or connections to fake identities. Therefore, the underlying platform needs to search very fast, and then deduce a risk-score based on a consumer’s connections. Both RedisSearch and RedisGraph can enable real-time updates of identities to arrive at an accurate point-in-time profile of a consumer.
Increase accuracy and detection of speed with AI/ML models
The processing of ML models slows down significantly because of the separation of ML data (e.g. features) in relation to ML models. This slowness can affect the accuracy of models for instant fraud detection. While most ML models leverage pre-calculated features, new user inputs require real-time calculations and updates that need to be delivered to the online models. Since the time window for these live updates is short, the faster these can be calculated, stored, and delivered to meet the requirements of real-time predictions, the more accurate the models.
ML and deep learning models need to be closer to the data. For real-time predictions, Redis can be used as an online feature, which serves feature vectors to RedisAI, a specialized module that brings inferencing capability where the data resides within Redis to decrease the latency. Using RedisGears, teams can write and execute functions that implement data flows in Redis, while abstracting away the data’s distribution and deployment. And since rule-based systems are based on ML models, it’s important to seamlessly orchestrate data flows across ML model inference and data residing in memory.
Reduce detection costs with high-speed statistical analysis
As companies experience increased volumes of transactions and need additional infrastructure footprints to support higher traffic, cost becomes a key challenge. Being able to detect anomalies from observing historical trends and comparing transactions against known patterns can guide teams on how to optimize the underlying platform. To activate these capabilities, RedisTimeSeries and RedisBloom provide value-added capabilities to drive this level of operational efficiency, and therefore cost savings.
Bringing it all together
A technology like Redis Enterprise can bring all of these capabilities together, and be used by risk management professionals to support real-time fraud detection and prevention across digital identities, statistical analysis, AI transaction risk-scoring, anomaly detection, and more. Organizations can leverage Redis’ multi-model in-memory database to expedite frictionless online transactions and reduce false positives without overcomplicating your enterprise architecture or external vendor solutions.
In addition to Redis, as new and emerging technologies continue to impact and transform many industries, organizations must be laser-focused on enabling real-time platforms to meet low-latency usage scenarios such as fraud, product recommendations, inventory management, and preventive maintenance. Real-time capabilities offer organizations competitive differentiation, while also both the top and bottom-line business objectives.
You can see Taimur's full video presentation below:
About Taimur Rashid
As Chief Business Development Officer, Taimur is responsible for developing emerging businesses at Redis Labs, and strategic business & corporate development. He is currently leading initiatives related to AI/ML. Prior to Redis Labs, Taimur led Worldwide Customer Success for Microsoft’s Azure Data Platform, Analytics & AI business. And before that, he was Managing Director for Amazon Web Services (AWS) Platform Technology and Applications where he led business development from 2008 (near its inception) to 2018 when the business reached $25B in ARR. Taimur helped forge some key partnerships and customers including Airbnb, CapitalOne, Dropbox, Liberty Mutual, NASA JPL, Nasdaq, Netflix, Nintendo, Intuit, SAP, Samsung, and Societe Generale.