Quick summary:
- Improper payment rates ranging from 5% to 20% cost enterprises billions annually through both errors and fraud
- AI models analyzing 200+ feature points verify business identities, payment accounts, and relationships in real time without requiring enterprises to change existing workflows
- The fraud arms race is already AI versus AI, with bad actors reactivating dead business registrations and using deepfakes while verification systems learn and adapt continuously
Network Solutions had a problem. The company was one of the first to accept credit cards online, and fraud was eating them alive. More than 30% of transactions were fraudulent.
Ben Turner, then at the company, realized that if they could understand the identity behind an IP address, they could decide whether a payment was safe. That insight laid the foundation for what became Verituity, where Turner now serves as President and CEO.
On this episode of The Disruption Is Now, Turner explains to host Greg Matusky how AI-powered payment verification works, why the fraud problem is bigger than most people realize, and how enterprises can cut improper payments by 70% starting with their first transaction.
Watch now:
Key takeaways:
Errors cost as much as fraud but get half the attention
Most people think enterprise payment verification is about stopping fraud. Turner says that misses half the problem.
“It’s actually about errors and fraud,” Turner explains. “A term that people have heard in the news is improper payments. The U.S. government making improper payments. Well, that’s both errors and fraud.”
The improper payment rate varies depending on industry, use case, enterprise size, and technology age. Some companies see rates around 5% to 6%. Others hit the low 20s.
Errors come from legacy workflows and old mainframe systems running COBOL. Mistakes just happen. Fraud spans everything from insider threats, where bad actors change supplier payment information, to email compromise attacks that target insured individuals expecting claim payments.
The range of problems makes human detection nearly impossible, which is why Turner started with the premise that this is fundamentally a data problem requiring the right models to solve it.
Relationships matter more than checklists
When you think about payment verification, you probably imagine a system checking boxes. Does the business have a physical address? Does it have banking history? Turner’s approach goes deeper.
Verituity’s models look at relationships between data points rather than isolated factors. The system verifies the payer-payee relationship, matches session behavior to identity and geography, confirms the connection between payment accounts and businesses, and monitors other activity across the internet through what Turner calls a consumer index.
“The reason we take that approach, it’s really hard to beat relationships,” Turner says. “And if you understand those relationships, then you can build a high level of accuracy or precision in your models.”
Humans can’t process 200 features simultaneously — but AI can
When Verituity verifies a business as a payment recipient, the system analyzes more than 200 feature points to make its decision. Contributing to those features are thousands of additional data elements.
The verification process happens in layers. The system verifies the business entity’s identity. It authenticates the email address. It confirms the relationship between the administrator and the business. It verifies the admin’s identity. It validates the payment account and its relationship to the business.
This complexity explains why AI became necessary. Humans can’t process hundreds of features across thousands of data points in real time. Rules-based systems require constant updates and manual teaching. AI learns continuously and handles genuinely complex data sets that would overwhelm traditional approaches.
The technology also detects emerging fraud patterns like the reactivation of dead business registrations. Bad actors revive dormant business entities, run transactions that appear legitimate, then let the registration sunset again. That pattern only becomes visible when AI tracks registration status changes over time and correlates them with transaction behavior.
Historical data isn’t necessary for the system to improve
Most AI implementations require training on historical data before they deliver value. Turner’s approach works differently.
“From day one, that first transaction, we’re able to start to identify improper payments or errors and fraud,” Turner says. The system doesn’t need to analyze a client’s past payment files unless specifically requested to identify existing problems.
This immediate effectiveness comes from AI’s learning nature. The models optimize themselves as they process transactions. The trick is implementing the right controls to prevent the learning from drifting off track.
Verituity follows a philosophy of “just send us your check file.” When a payment file emerges from an ERP or legacy mainframe, the system accepts it as-is. AI transforms that data into something useful, typically the ISO 20022 standard, then begins looking for anomalies and indicators that a payee might not be legitimate. The process continues through settlement, examining the payment file, the payee and payment account, and the final payment instruction.
The result is a 70% improvement in both accuracy and fraud reduction without requiring the enterprise to modify existing systems or processes.
Bad actors already weaponized AI, so defenders must too
When asked if fraud will soon become AI versus AI, Turner says that it already is.
AI provides an advantage in this environment. A rules-based system would require armies of analysts updating detection logic, examining edge cases, and responding to new attack patterns. AI handles that adaptation automatically while a specialized team focuses on improving the underlying models and addressing novel threats.
The technology also allows Verituity to detect emerging fraud techniques that require pattern recognition across massive data sets and the ability to identify subtle anomalies that would slip past human review.
Key moments:
- How Turner’s Network Solutions experience laid the groundwork for Verituity (3:59)
- Why AI multiplies knowledge by hundreds of thousands of data points (6:08)
- The five verification steps for business payment recipients (7:08)
- Taking payment files as-is and transforming them with AI (8:44)
- How the system achieves 70% improvement from day one (9:57)
- Using AI to analyze 200+ features per business (11:34)
- How relationship analysis beats factor checklists (13:12)
- Why U.S. banking structure creates verification gaps (14:28)
- The incentive structure that keeps bad actors in the game (15:23)
- The fraud arms race is already AI versus AI (16:07)
Q&A with Ben Turner, President and CEO of Verituity
Q: How does AI handle complex business verification?
A: We look at over 200 feature points to make a decision. We’re verifying the identity of that business entity. We’re authenticating the email or maybe text. We’re verifying the relationship between the admin and the business. We’re verifying the admin’s identity, and then we’re verifying the payment account and the payments account relationship to the business. So, all that’s driven by a sophisticated set of AI models that are interlinked so that, at the end of the day, you’re able to make a decision.
Q: Why do relationships between data points matter more than individual factors when analyzing payments?
A: One of the largest issues in payments in the U.S. is a lack of coverage of information on bank accounts. This is a result of how the banking industry got built in the U.S. Therefore, you can’t rely on just looking up against a third-party data set to say, is that a good bank account? A lot of times, you’re not going to get an answer. However, AI can solve this by examining all these features and their relationships with one another.
Can you trust the identity? Can you trust the session? Do they have a history of having used that account? What’s the risk around the email or mobile number they’re using? All those features, plus more, come together to allow you to build a high level of accuracy or precision in your models, which gives you the trust that you have the right bank account.
Q: With AI becoming more accessible, fraudsters are evolving quickly. What does the next phase of this battle look like?
A: It’s already AI versus AI. Fraudsters have low costs and high upside, so they’re constantly experimenting with new tactics. Defenders need systems that learn just as fast. That’s why we invest heavily in data science. It’s an arms race that won’t end, but AI helps level the field by allowing small teams to protect enormous transaction volumes.
