Your Guide to Advanced Fraud Prevention with AI

Artificial intelligence (AI) and machine learning (ML) technologies derive patterns from large quantities and diverse sources of data to generate relevant insights that influence decision-making, business operations and processes, and the customer experience. Finance is one of the most data-rich industries, which is ideal for the strategic application of AI and ML in areas like fraud prevention.

As the financial services industry is currently experiencing acute disruption, change, accelerated digitalization, data breaches, an uptick in contactless payments and a growing threat landscape, financial services organizations (FSOs) like banks and credit unions are positioned to leverage AI and ML to capitalize on the potential benefits of autonomous intelligence and strengthen fraud prevention strategies.

From April to May 2020, consumer bank traffic decreased by over 30 percent compared to the previous year, and evidence indicates that consumers will continue to shift to digital banking regardless of when they begin to feel confident about revisiting businesses in person. Physical bank branches are rapidly being replaced by digital banking channels, whether it’s for checking account status and balance or money transfer, and 85 percent of consumers who have used digital platforms for financial services will continue to champion this form of interaction post-pandemic.

This shifting behavior is an indication of the digital model of the future and a catalyst for digital growth across the financial services industry, as well as an opportunity for fraudsters looking to profit from escalating online activity and weaknesses in existing fraud prevention systems. Fraudsters are adapting their tactics and evolving, and FSOs must emulate this mindset and act now to improve their technologies and digital services, and continuously address diversifying, complex and well-orchestrated fraud schemes.

2020 is a Transformative Year for Fraud

While total combined fraud losses reached $56 billion in 2020, identity fraud scams comprised $43 billion of that cost. Fraudsters are adapting their approaches as consumers increasingly transact and interact digitally, seizing on new opportunities provided by ubiquitous digitization to directly target FSOs via identity fraud scams. One-third of identity fraud victims felt unsatisfied with the lack of resolution offered by the FSOs where the fraud occurred, leading to 38 percent of victims closing their accounts.

The fraud culture, which also encompasses a surge in fraud losses related to peer-to-peer (P2P) payments, digital wallets, phishing, authorized push payment (APP) fraud, friendly-fraud and the gamut of identity fraud scams has resulted in growing vulnerabilities for FSOs.

Additionally, within this complicated threat landscape, FSOs are striving to push urgent business initiatives forward, including:

• Digital transformation acceleration

• Improving the customer experience

• Modernization of payment systems aligned with transformation efforts

• Supporting the shift to faster and real-time payments

• Preparing for and supporting ISO 20022 migration and optimization

• Merging fraud and cybersecurity/information security

• Maintaining regulatory compliance (GDPR, California Privacy Rights Act)

Executing effective advanced fraud solutions that coincide with diverse digital transformation initiatives in a post-pandemic world, all while maintaining high quality, seamless digital experiences is a multifaceted challenge. Robust fraud prevention and management must not come at the expense of consumer convenience. Intrusive efforts that potentially flag and decline authentic transactions by legitimate customers raises the likelihood of a customer abandoning the platform, service or bank.

Lack of real-time data insights to rapidly detect and respond to fraud may also impede contributing financial institutions from leveraging a proactive approach to fraud prevention. Instead, they must rely on historical information that does not accurately reflect emerging and new fraudster behaviors. This can further slow down time-to-insights and resolution and ultimately leads to more substantial fraud losses.

Additionally, relying solely on human analysts for fraud investigations can be highly problematic, as FSOs report false positive rates upwards of 90%, along with an excess of false declines, investigations that take overly long to conclude and frustrating customer experiences. This demonstrates the need for smarter automation and advanced analytics tools to accelerate accurate decisioning on the fraud operations side to equip fraud analysts to work more effectively and optimize customer friction according to risk levels.

As the stakes get higher, FSOs are pursuing the promise of advanced AI and ML to combat and check fraud effectively and address the breadth of challenges surrounding fraud prevention.

The Value of AI in Fraud Prevention in Financial Institutions

Most FSOs are aware that AI provides organizational value, and historically organizations have used this technology as a tool for optimizing efficiency. Now, AI is viewed as a primary driver of competitive industry advantage, product and service differentiation and the ability to use data effectively. The early adoption phase of AI is concluding, and the market is progressing toward mainstream adoption largely due to the growing accessibility and cost-effectiveness of many AI solutions, as well as the minimal need for human expertise and intervention.

FSOs are accelerating adoption of AI-powered systems, with over $217 billion spent on AI applications for use cases such as fraud prevention. Eighty percent of fraud prevention specialists attribute AI to lowering payments fraud, and 63.3 percent of FSOs credit AI with being a valuable tool in stopping fraud.

Contrary to popular assumption, AI will not replace the human workforce. Now and moving into the future, there’s a growing emphasis on human-machine collaboration, and FSOs are looking to augment human-centric investigations with advanced AI-powered autonomous intelligence to boost efficiency, accuracy and shift the focus of human assets to more specialized, value-driven tasks and knowledge work instead of devoting time to manual investigation steps.

Fraud analysts are under extreme pressure to manage growing volumes of complex alerts while maintaining a frictionless customer experience. The growing adoption of faster and instant payments is further adding to the stress on fraud analysts and fraud operations teams because a number of the faster payment networks come with a 24/7 requirement for transactioning. Advanced AI can empower investigators and analysts with rich, trustworthy and contextual insights, and enable them to work more efficiently while the window of investigation is drastically shortened.

The rising use of digital payment channels and the corresponding increase in real-time payments is also leading to the need for more robust authentication and payments fraud detection methods across all aspects of the customer’s digital engagement. Advanced forms of authentication, such as biometrics, are replacing static, traditional methods of customer authentication and enabling safer digital transactions while reducing customer friction and potential abandonment rates. FSOs are acting upon opportunities to coordinate advanced, non-intrusive, omni-channel authentication methods in every channel via a unified approach to fraud and authentication management that provides improved fraud detection rates and cross-channel attack identification.

AI also helps FSOs address weak points in their fraud prevention efforts and strategies, and orchestrate a multi-dimensional, multi-layered approach to fraud management.

Opportunities for Artificial Intelligence and Machine Learning in Fraud Prevention

FSOs are deploying advanced fraud solutions powered by AI and ML to increase profitability, reduce costs, generate new value, strengthen their fraud detection and prevention operations and protect their financial assets, data, organization and customers. These technologies help FSOs streamline their approach to fraud management and resolve the limitations of traditional, rules-based fraud prevention systems and manual-intensive processes.

Advanced AI and ML also helps FSOs detect fraud earlier. For example, fraud identification processes can be predictive by using ML models that are trained on comprehensive data sets to distinguish behavioral patterns. Fraud teams can then quickly understand and identify the types of fraud being inflicted against the FSO and make meaningful, data-driven decisions.

The more targeted, sophisticated and differentiated the ML models, the more ideally positioned an FSO is to mitigate some of the most complex types of fraud. This is why the speed in which high performing models detect deviant or anomalous behaviors is critical; it greatly impacts an FSO’s ability to keep up with the rate and complexity of fraud, as well as the constant changes associated with fraud prevention.

Additionally, advanced AI and ML enables different relevant data types to be captured, contextualized and leveraged to help FSOs comprehend the context of different threats and accurately detect fraud risks. For example, transactional data, channel data, customer data and account level historical data can be captured and precisely contextualized so that rich information can be deployed to efficiently fight first and third-party fraud.

An agile, autonomous, end-to-end fraud prevention and management solution, such as IFM-X from NICE Actimize, can help FSOs take advantage of the numerous opportunities provided by advanced analytics and AI to deploy holistic, cohesive approaches to fighting fraud in a dynamic fraud environment. Benefits include:

• Enable accurate identity evaluations, detect abnormal or suspicious transaction behaviors and administer continuous fraud prevention across every stage of customer lifecycle fraud management for a multi-layered fraud defense.

• Reduce false positives, accelerate investigation time and enhance decision-making via AI-powered scoring and authentication, which helps FSOs eliminate invasive, inconvenient control standards and identity verification protocols that disrupt the customer experience.

• Make smarter, accurate, faster decisions via entity-driven investigations and visual analytics.

• Address fraud prevention capabilities and system vulnerabilities emerging from outdated infrastructures, legacy systems and lack of unification throughout the fraud management ecosystem with AI capabilities that incorporate changing behaviors, emerging trends and anomalies to generate actionable insights regarding customer risk.

• Leverage automated decisioning for real-time transaction approval or rejection to facilitate more proactive approaches to fraud prevention.

• Mitigate fraud loss by preventing fraudsters from penetrating the organization and perpetrating fraud across numerous channels and product lines.

AI and machine learning in financial services are empowering more responsive, collaborative and sophisticated approaches in the fight against pervasive fraud. With the autonomous intelligence and continuous self-learning provided by comprehensive, advanced analytics-based solutions, organizations can modernize their fraud prevention models and rapidly stop a range of fraud attacks.

Challenges in AI Fraud Prevention Solution Implementation

Only 53 percent of organizations have transitioned their AI POCs into production over the past two years. Lack of a comprehensive AI strategy, fragmented data, outdated operating models and an underdeveloped use case can hinder AI adoption. Furthermore, these factors can prevent FSOs from effectively embedding AI across the enterprise, which is a vital component in developing comprehensive, organization-wide approaches to fraud prevention and management.

Some challenges that FSOs may encounter during AI implementation include:

• Workforce impact: Employees may have to adapt to new ways of working, collaborating with new technologies, and shifts in job roles and complexity as automation augments the workforce. Upskilling or new talent acquisition may be required to fill in skill gaps and develop the expertise necessary to meet AI objectives.

• High costs: Developing and implementing AI solutions and projects can be a substantial financial investment, and cost varies according to the use case, solution complexity, current analytics maturity, vendor expenses and resource deployment.

• System architecture: Lack of a robust system architecture can lead to performance and latency issues. Data collection, processing and storage architectures must have the agility and capacity to support fluctuating computing requirements and large amounts of data composed of diversified sources and types, in addition to real-time streaming data.

• Data quality: Poor quality data can impact compliance and lead to skewed, biased and untrustworthy AI outcomes. The AI system is only as reliable as the data that it’s trained upon, and FSOs must choose the right data, monitor that data and closely connect it to analytics.

• Data integration: AI systems can quickly outpace their data capacities, and as new financial services products are exponentially introducing new data types, the solution needs to scale alongside data growth and intricacy, and seamlessly integrate and validate new data sources and types.

• Agility challenges: To stay accurate over time, models need to be updated. As changes in reality or data availability occur, systems must be agile and quick to adapt. Organizations need to increase the speed in which their model governance teams can allow models to go to production, as well as adopt practices and tools that streamline the delivery of ML models into production, while providing constant monitoring of the models performance with an MLOps team.

The potential to reimagine fraud prevention via AI and ML hinges on strategic solution integration and advancement, and the agility to evolve alongside constantly changing risks and approaches towards fraud prevention. These factors ensure that FSOs can respond to constantly shifting fraud patterns, and better align their data, processes, technologies and fraud risk teams with fraud prevention objectives.

Build a Comprehensive AI-First Fraud Prevention Strategy

To become an AI-first organization in the era of ubiquitous AI, FSOs must have a well-developed AI vision and strategy, align their AI strategy with their overall fraud prevention and business objectives, define AI and data governance frameworks, leverage automated tools for testing AI models and possess the capabilities to effectively integrate data for AI initiatives.

FSOs often use AI to build upon existing tools and capabilities, while eliminating the need to throw away existing fraud prevention investments or deploy rip-and-replace approaches. However, building on existing solutions is not always viable, as attempting to layer new systems over outdated or obsolete legacy systems that aren’t up to par anymore simply won’t work. Likewise, the data being fed into a new AI system may not possess the right degree of granularity or quality, which will impact data ingestion, and eventually, the certainty of the output.

High quality, voluminous, diverse, governed and relevant data is crucial to preventing skewed results. FSOs must also be able to effectively incorporate data from numerous sources to consistently detect suspicious activities without compromising operational efficiency.

Real-time fraud prevention depends on real-time data, which is a crucial aspect of proactive fraud prevention and the ability to detect fraud before it results in fraud losses or reputational, compliance and financial damage. Reactive approaches to fraud detection are no longer feasible today, and must be replaced with a holistic, AI-driven approach that uses advanced algorithms and behavioral data to identify fraud threats in real time.

Fraudsters are using new technologies and automated tools, along with innovative techniques, to ensure fraudulent transactions stay under the radar. Real-time insights and contextual analysis of transaction activity and customer behaviors helps FSOs stay a step ahead of criminals, while facilitating approval for legitimate transactions.

Additionally, FSOs should consider adopting a cloud model as the foundation for their fraud prevention strategy. The cloud is an accelerant and enabler of transformation, while providing on-demand flexibility and unlimited processing power at scale that helps fraud teams build, train, assess and optimize ML models.

Advanced Artificial Intelligence and Machine Learning in Fraud Prevention

Driving real-time responsiveness to a quickly evolving fraud market is key to identifying and mitigating new forms of fraud before they result in reputational and financial damage. Standard rules-based fraud detection systems were developed for a comparably slower paced fraud landscape, and hinges on the ability to first identify a fraud pattern. It often takes months to test and deploy detection rules and models after a new fraud pattern is identified, which at this point can result in significant fraud losses. By this time, fraudsters may have already adapted their approach to elude detection.

FSOs must transition from rules-based fraud detection and predictive models to next-generation, advanced analytics built upon highly-developed crime indicators, fraud behaviors and holistic data repositories. They also need the agility to enable continuous adaptation to new and emerging fraud patterns and behaviors, and the data-driven context to truly understand customer risk.

Unsupervised learning models are capable of detecting anomalies in behavior, even with limited transaction data. These models can self-learn, constantly analyze new data and self-update by identifying patterns to conclude if they’re an aspect of legitimate or fraudulent transactions. Supervised learning relies on labeled data sets of transactions appropriately tagged as either fraud or non-fraud. Pertinent, high quality, voluminous training data contributes to both model accuracy and learning patterns that correlate to legitimate transaction behaviors. Time must be factored into supervised learning model development so that the model is pertinent to the initial fraud typology.

When supervised and unsupervised learning models are integrated, fraud detection and prevention become more dynamic and FSOs can gain a more robust degree of clarity regarding the respective risk of customer behaviors. A self-learning model can also be developed to detect existing fraud, as well as new and emerging fraud, to allow FSOs to stay ahead of changing fraudster activity.

Federated learning is a distributed ML process that leverages distributed data across siloed applications, cloud environments and data centers. This approach is ideal for industries like finance that must maintain compliance with strict data privacy laws because it enables safe collaboration for model training. Essentially, this process delivers the ML models to the source of the data rather than the other way around. The distributed data sets provide accurate insights, which are then transformed into features that are deployed across the industry. Federated learning and cross-entity data enables FSOs to address emerging fraud trends and threats, and gain a single, holistic view of customer intelligence to drive faster decisions and minimize resolution time and costs. Of course, to enable federated learning, some form of connection to the cloud is required. Industry solutions such as ActimizeWatch, a cloud-based analytics optimization solution, can provide this value while offering a cross-institutional view of fraud.

Robotic Process Automation (RPA) automates manual, repetitive tasks and processes, and is often used in fraud prevention as a tool across use cases such as account validation, checks validation, alert assignments and communications to improve fraud operation efficiency. Certain RPA solutions in fraud prevention can help direct investigative processes and recommend measures that will lead to faster, better informed decision-making.

AI-First Approach to Fraud Prevention

Fraud programs must be able to understand and adapt to the new normal behaviors that emerged during the pandemic and that are evolving in the post-pandemic era to accurately interdict what is truly anomalous. Always on AI, which empowers IFM-X, addresses this need by leveraging industry and behavioral intelligence to continuously learn, discover and adapt to rapidly detect divergent activity and prevent fraud attacks while ensuring FSOs can better protect their organization and customers.

The Always on AI framework is also facilitated by an AI-first approach to fraud prevention due to the continuous embedding of more AI, ML, automation capabilities and features into IFM-X. This is achieved via supervised and unsupervised learning models with predictive capabilities, cross-enterprise model insights and efficient model operationalization and governance processes.

Constant, dedicated developments and investments in this area have produced best-in-class capabilities that enable holistic, end-to-end, scalable fraud prevention and fraud management, while supporting digital acceleration initiatives.

Optimize data acquisition, processing and operationalization

• X-Sight DataIQ and X-Sight Marketplace: A continuously evolving ecosystem of integrated point solutions and data sources covering an increasing quantity of modalities, including identity verification, biometric behaviors and device intelligence.

• X-Sight Connect: Offers real-time, API-enabled orchestration to connect disparate, siloed data sources and point solutions.

• Agile data integration: Helps FSOs map new data sources and instantly leverage them across IFM-X, including risk models, rules and decisioning, and alerts and cases, to support diverse, complex messages that will minimize dependency on IT.

• Scalable data processing: Supported by a continuously improved technology stack that drives quicker, scalable data processing and a cloud-friendly architecture.

Consortium intelligence

• Fraud Insights: A comprehensive bulletin of payments fraud statistics extrapolated from the NICE Actimize consortium that allows FSOs to benchmark their Enterprise Fraud Management (EFM) program and better understand emerging fraud trends.

• Cross-FI Entity Risk Score API: An entity (IP, device, address) level risk score that can be accessed through an API to help catch more fraud.

End-to-end fraud prevention coverage

• New Account Fraud: By using application data and monitoring all payment channels, early monitoring functionality detects complex fraud arising from stolen and synthetic identities, as well as mule risk for high accuracy account monitoring driven by advanced analytics and AI.

• Dark Web Intelligence: This solution provides intelligence curated from the dark and deep web, malware networks, botnets and other infrastructures used by fraudsters to deliver actionable data regarding compromised customer accounts prior to being attacked.

• Authentication Management: Helps FSOs orchestrate omni-channel authentication, while unifying fraud and authentication management to execute real-time risk decisioning, stop ATO attacks and improve the customer experience.

• Payment Fraud: Adaptive AI and ML allows FSOs to support real-time payments and safeguard any payment type via an omni-channel approach to monitoring the entire payment lifecycle.

• Employee Fraud: Detects fraudulent employee activity and violation of corporate policy across multiple business and enterprise lines and channels to provide protection against monetary loss, risk and reputational damage.

AI – The Future of Innovative Fraud Prevention Tools

AI and ML are already becoming increasingly dominant in technological industries, especially in finance alongside the growth of cryptocurrencies, pervasive automation and digital transformation. AI advancement allows FSOs to use autonomous intelligence to eliminate fragmented approaches to fraud prevention and transform their operations to mitigate fraud efficiently, holistically and proactively.

Online incremental machine learning is one of the newest fronts of ML and forward-looking organizations are investing in building a new generation of systems that will enhance systems capabilities to learn in real-time beyond the existing limitations. A future where privacy preserving techniques for insights sharing in real time, federated learning and online incremental ML are blended into one system is where we expect the market to be in a few short years.