The “Choose Your Own Adventure” of Machine Learning Implementation for Fraud

Financial institutions are sorting through the generalized hype over machine learning and making real decisions about how to operationalize it. What many don’t realize is that machine learning procurement is not one size fits all, but rather “choose your own adventure.”

The general options are to either supplement existing legacy systems through an additive approach, or to replace existing systems from the ground up. Organizations should be aware of their options so they can work with machine learning vendors who can have the flexibility to step in and add value at any stage, including the very beginning.

Small scale: “additive” to solve narrow problems

Machine learning can be deployed as an “additive” solution that solves specific issues without significant platform disruption. The benefit of this option is retaining existing infrastructure but augmenting it with the most current AI advancements. High fraud detection rates or high rates of customer abandonment are several issues that could be addressed in a small scale machine learning integration.

But before adding a machine learning component to the stack, companies need to answer a few questions:

  • Is the system appropriate for the given use case?
  • Can it scale to address issues that emerge as the company grows?
  • Will it integrate with the existing legacy system?

Though additive machine learning isn’t as comprehensive as other options, companies need to answer the above questions before committing to a solution.

Large scale: transformative integrations

Organizations are trying to anticipate the coming innovation inflection points, and many are making the decision to replace their risk infrastructures with new platforms. These larger integrations can redefine and unify their total risk strategy as part of a larger effort in digital infrastructure. Companies who are seeking meaningful business growth will encounter new waves of fraud that threaten every expansion into new use cases, geographies, and business lines. Meanwhile, landscape changes are coming fast: customer demands for instant payments, regulations like open banking, and determined fraudsters presenting new patterns at high velocities.

When a financial institution needs to reimagine its fraud detection system in the wake of these issues, it’s possible to integrate machine learning in ways that completely transform their workflows. Companies must be ready to perform all necessary due diligence before taking the plunge:

  • Is the new system capable of meeting the increased demand for compliance or regulation?
  • Is the system enterprise-grade and purpose-built for fraud?
  • Is the system adaptable to the specific needs of the organization?

The Need for Machine Learning Evolves Over Time

Machine learning can be applied to solve either narrow problems or broad organizational changes, but regardless of how it’s used, organizations shouldn’t think of it as a short-term solution. Machine learning tools become more efficient and more productive as they ingest more data, meaning that their value increases as companies integrate them more closely with existing data sources.

The long-term value of machine learning tools goes way beyond efficiency. Fraud is always evolving, and companies that think of machine learning as a short-term solution to simple problems will leave their companies exposed over time. As fraudsters find new attack vectors, companies will need established machine learning-based fraud detection systems capable of identifying these emerging threats.

In short, machine learning is more than a Band-Aid for underperforming metrics. It’s a scalable solution that future-proofs an organization from both existing threats and threats not yet seen.

Case study for additive deployment

Feedzai recently deployed a new machine learning application for a leading processor. This processor wanted to incorporate PIN-less transactions into its payment system, but to do so, it needed new ways to score fraud. They desired an on-premise system with capabilities beyond those offered by Visa or MasterCard, and that served both issuers and merchants. This was a tough order to fill.

Feedzai integrated its own machine learning solutions for issuers first, and after other business units saw its advantages, merchants jumped on board. This produced several notable improvements to the processor’s system, including a 42% improvement in false positive rates and a 38% improvement in fraud detection overall.

Today, Feedzai still provides this real-time fraud scoring and lives as an inseparable part of the processor’s network. Before Feedzai’s integration, the processor had no options for meeting its goals. Now, its solutions outperform the internal systems of several major banks and have prepared them with a flexible fraud management system that will support them for years to come.  

Operationalizing Machine Learning for Fraud