Read “The State of Fraud and Abuse” report
Over the past six months, we have been laser-focused on developing and testing new ways to deploy AI in the fight against fraud and abuse. Today, we are excited to reveal what we have been working on in our first ever Signifyd Spark – a biannual multimedia release of our newest technology and feature updates, including what they mean for your business and how you can make the most of them.
In this first installment, we dive into enhancements that allow our customers to:
- Investigate fraud and abuse patterns with always-on transaction oversight.
- Act quickly on insights using the very same AI risk signals leveraged by our fraud model.
- Convert more good transactions than ever before with market-leading pre-authorization protection.
Team up with AI to investigate and respond to novel threats
Signifyd tackles novel threats from two angles. The first is our Anomaly Detection system, a machine-learning model trained to recognize new fraud attacks before chargeback data becomes available. This is important because chargeback data can take up to 90 days to earn out and in the meantime, fraudsters exploit new vulnerabilities quickly and at scale.
Instead of relying on outcome data, Signifyd’s Anomaly Detection system analyzes hundreds of early warning signals, such as drops in approval rates, spikes in risky identity data, or shifts in behavioral patterns. Over the last six months, we’ve significantly enhanced our system to act faster and smarter, automatically adjusting thresholds in real time during adversarial actions. We have reduced time-to-detect by over 60% and increased accuracy of detection by 25%.
The second method for detecting novel threats is user facing, through our Custom Insights module that allows you to build customized dashboards to track anomalies across the metrics that matter most to your business.
Users can now build custom reports using unique AI features, such as a model output that scores the likelihood of address manipulation, and they can leverage a variety of visualizations to represent that data. With stronger risk signals and more customizability, Custom Insights can alert your team to anomalies in order traffic that indicate emerging abuse threats. Insights in hand, users are empowered to intervene quickly, leveraging some of the same risk signals to build dynamic abuse policies in Decision Center.
Act quickly on insights with model-derived risk signals
As fraudsters continue to adopt new AI methods to infiltrate anti-fraud systems, we too have deployed AI in new and innovative ways to anticipate and respond to evolving threats faster.
One such method is what we call chained models — specialized models that are “chained” to our global consortium model such that their outputs become inputs for our consortium model’s decisioning. These models evaluate specific risk signals, such as address manipulation, IP risk, product SKU risk and more — adding a depth of analysis to the consortium’s breadth.
These same model outputs are now available as policy variables in Decision Center. Because these variables are model-derived, they continually adapt to new risk patterns and are regularly refreshed by our human team of risk experts and data scientists. This allows you to build policies that shift dynamically in response to changes in the risk landscape.
Convert every good order into a successful sale
When fraud screening occurs after payment authorization, merchants have no influence over or visibility into the orders turned away by issuers. Issuers screen all merchant traffic, regardless of it being good or bad, and are forced to make sub-optimal decisions — leading to falsely declined orders and higher processing fees for merchants.
Though the tradeoffs are clear, post-authorization has long been the de facto method of deployment as it requires less data and model sophistication to accurately distinguish between bad actors and legitimate customers, while also giving merchants the flexibility to manually review orders before fulfillment.
To screen for fraud prior to bank authorization requires a much higher level of technological sophistication. Models need to be faster and much more precise — capable of delivering real-time decisions while the shopper is still in-cart. Though this is clearly the better experience for customers, it’s a CX differentiator that’s been largely untapped by ecommerce retailers given the limited capabilities of many fraud solutions.
Deploy pre-auth to increase conversion
Not so for Signifyd merchants. Pre-authorization protection has long been a focus of ours, and in the last year, we have added more partnerships with issuers and made it a lot easier for merchants to move from post- to pre-auth. Today, hundreds of merchants leverage our pre-auth capabilities and the results are pretty compelling.
For example, one of the world’s largest electronics retailers has seen a conversion rate increase of over 200 basis points by switching to a pre-auth fraud screening setup. Another retailer in the fast fashion space has seen a roughly 300 basis points increase in conversion rates. Our issuing partners are reporting a decline reversal rate of over 50% meaning they are overturning their internal fraud assessments over half of the time, thanks to Signifyd’s network data.
The cumulative impact of these advancements not only helps you stay ahead of evolving fraud tactics but also enables you to deliver a seamless and rewarding customer experience that sets your business apart. Today’s Winter 2025 Release of Signifyd Spark marks the first of many announcements detailing exciting product releases as we continue to push the boundaries of AI in fraud and abuse detection.
Photo by Getty Images
Want to talk through your fraud and abuse challenges? We’re here for you.
