Eric Stratman is senior director of analytics and insights at ValidiFI, where he builds machine learning systems that help businesses verify bank accounts and catch fraud. His team analyzes data across 171 million payment records, 83 million bank accounts, and 1.4 billion inquiries to identify risk patterns missed by traditional verification methods.
ValidiFI’s AI can increase account verification coverage from 85% to 96% by learning routing and account number structure patterns across thousands of financial institutions. The model also flags suspicious connections, such as when multiple Social Security numbers associated with the same bank account appear within a short time frame.
Here, Stratman explains how ValidiFI trains its models while complying with financial regulations, why quality data is more important than complex algorithms, and what it will take to keep up with scammers getting smarter.
ValidiFI positions itself as a leader in “Predictive Intelligence for Bank Accounts and Payments.” Can you explain how AI and machine learning form the basis of this predictive capability and how your approach differs from traditional account verification methods?
in Valid FIthe foundation of predictive ability is data. Our data team is very passionate about incorporating data into our solutions. This diligence allows us to create solutions centered around bank account and payments intelligence, providing analytical insights into bank account status, ownership, behavior, and performance. We believe that with high-quality data, we can take full advantage of various AI and machine learning techniques to deliver predictive capabilities to our customers. This is done by further expanding coverage through account verification pattern matching, or by analyzing account velocity and behavioral data to uncover seemingly legitimate fraudsters.
You mentioned that AI technology can extend account coverage from 85% to 96% by analyzing patterns in bank account and routing numbers. Can you explain how your machine learning models identify these patterns and whether they are detecting specific signals that human analysts might miss?
Our AI and machine learning pattern matching allows us to analyze the structure of account numbers based on routing numbers. Human analysts are naturally pretty good at identifying patterns, so it’s not necessarily something they can miss. As one example of a trend our machine learning process shows, almost all ACH-enabled account numbers at Bank A are 8 digits and start with 3. If a human analyst were given a list of successful trades from Bank A, they would likely reach the same conclusion. AI and machine learning are great at scale. There are thousands of routing numbers, each with different structural behavior. Using AI and machine learning, we can process this at scale and identify even finer details within account structures. Add to that the fact that real-time data is coming in that can change what was previously thought to be true. AI and machine learning also allow us to adapt as more data and information becomes available. In addition to structural analysis, our pattern matching also incorporates historical data, ensuring that we provide our customers with powerful predictive indicators of account validity, even when direct verification is not available. This increases validation coverage from 85% to 96% when using pattern matching.
Fraud detection systems analyze the connections between bank accounts, consumers, and payment behavior. What types of AI algorithms are you using to detect these relationships, and how do you address the challenge of identifying sophisticated fraud that may appear legitimate on the surface?
in Valid FIour fraud detection capabilities are built on the ability to analyze billions of data elements across a consortium network that connects bank accounts, consumer identity attributes, and consumer velocity behaviors in real time. We utilize various AI algorithms to identify fraud and high-risk indicators. We often leverage supervised learning techniques to identify the most effective indicators when identifying high-risk fraud patterns. When trying to identify the general patterns we are seeing, we may consider other unsupervised learning methods that can reveal relationships and anomalies. For example, you can identify when a bank account is linked to multiple SSNs, but is also typically associated with a VOIP phone. These signals may not individually raise red flags, but when analyzed in the larger context, they can reveal coordinated or new fraud tactics.
With over 171 million payment records and 83 million bank accounts in your database, how are you leveraging AI to turn this vast dataset into actionable insights? What is your approach to training models on financial data while maintaining privacy and regulatory compliance?
Valid FI‘s data network continues to grow every month, and with it our ability to provide deeper and more predictive insights. In addition to 171 million payment records and 83 million bank accounts, our database also covers over 1.4 billion inquiries and over 61 million unique consumers, providing the most comprehensive financial view in the industry.
We use AI as a strategic layer on top of this large dataset, transforming raw data into real-time, actionable intelligence. Our machine learning models analyze patterns across bank account status, ownership, payment behavior, and identity attributes to predict trustworthiness and performance.
We also value privacy and compliance. Our models are trained using strict data governance protocols, including anonymization, encryption, and access controls. We ensure that all data usage is compliant with regulatory frameworks such as GLBA and FCRA, so partners can trust that their insights are strong and compliant.
Velocity metrics can detect risk patterns such as “3 or more phone numbers in 30 days,” which increases fraud risk by 70.5%. How does AI determine which behavioral signals are truly predictive or coincidental, and how does it avoid false positives that can harm legitimate customers?
To avoid false positives, our models and insights are built on a large and diverse sample. It also doesn’t rely on a single data point. Examine the combination of velocity metrics, bank account status, and identity mismatch. For example, changing your phone number alone poses no risk, but when combined with a spike in calls, mismatched identity data, and account speed anomalies, the risk increases significantly.
We also continually retrain our models and evaluate patterns using real-world results such as confirmed fraud cases, return rates, and resolution data to improve accuracy and reduce bias. This feedback loop ensures that the model evolves as fraud tactics change, while maintaining fairness and accuracy.
Traditional credit scoring often lags behind a consumer’s current financial situation. How is AI used to analyze real-time bank account data to provide a more up-to-date picture of credit worthiness, and what advantages does this give lenders compared to traditional FICO scores?
We leverage AI as a layer to analyze real-time bank account data such as inquiries, payment history, line performance, and banking behavior patterns to build a dynamic, current view of a consumer’s financial health. Unlike traditional credit scores, which are often static and outdated, our model continuously evaluates your financial behavior to reflect your true creditworthiness at any given time. This allows lenders to make accurate and timely decisions and expand consumer reach to underserved populations.
You work with companies like VeleLa and amounts across various industries. How can I customize AI models for different industries? For example, does fraud detection for auto loans require a different algorithm than a convenience store loyalty program?
When developing solutions for different industries, we like to consider what data we feed into the training datasets for our models. As you rightly pointed out, the challenges and patterns we see in retail, especially in terms of convenience, can be very different than in auto loans. The underlying AI algorithms may be consistent across industries, but ultimately it is the data fed into the model development that shapes the performance of those algorithms.
The financial services industry is highly regulated. How can you ensure that your AI models comply with regulations such as NACHA rules and FCRA requirements while providing the predictive insights your clients need?
As a SOC 2 compliant company, we take data privacy, security, and compliance seriously. We maintain robust data security, privacy, and risk management controls, supported by continuous monitoring and evidence-based control validation. Our models are trained using strict data governance protocols, including encryption and access controls. We ensure all data usage is compliant with regulatory frameworks such as GLBA and FCRA, so our partners can rely on both powerful insight and compliance.
As fraudsters become more sophisticated and potentially use AI themselves, how is ValidiFI evolving its machine learning approach to stay ahead? What is your strategy for this technological arms race?
We recognize that fraud is no longer static; it is adaptive, rapidly changing, and increasingly powered by AI. That’s why our strategy is built first and foremost on quality data. Plus, leverage continuous learning, real-time pattern recognition, and consortium-scale intelligence. We are also investing in model agility to stay ahead of AI-powered fraud. Our infrastructure supports rapid iteration and deployment of updated models, allowing us to respond to new fraud tactics in near real-time. The ultimate goal is to make sure your clients not only react to fraud, but anticipate it. By combining a rich data range and AI, Valid FI We help agencies stay ahead of the curve and protect their customers with confidence.
Where do you think AI will have the biggest impact on financial services in the future beyond what it’s doing now? Are there any new AI technologies or techniques that have the potential to be incorporated into the platform?
We are constantly exploring different machine learning techniques that can assess consumer and bank behavior. This allows you to identify strong predictors and trends in risk and fraud. Looking ahead, we are particularly excited about new AI and machine learning technologies that help identify and optimize strategic development for our clients. By tailoring our solutions to each client’s unique needs, we aim to provide more accurate and predictive account, payments and credit risk intelligence.
