Hi everyone, I’m Julie VerHage-Greenberg with This Week in Fintech, and welcome back to our special series co-created with Flourish Ventures, spotlighting the people and ideas reshaping the U.S. financial ecosystem. 

Today, Emmalyn Shaw and I are diving into fraud and AI, and what happens when the tools that make payments faster and more seamless also make scams faster, cheaper, and more adaptive. 

Our guests are Nicky Goulimis, CEO of Tunic Pay, and Oban McTavish, CEO of Spade. Nicky explains why scams require far more context than “transaction data,” and why prevention isn’t just detection…it’s intervention, customer engagement, and reducing false positives. 

Oban breaks down why banks and merchants often operate antagonistically, how merchant-related fraud is evolving, and why legacy systems make real-time intervention harder than most people realize. 

And then we look forward: if transactions become invisible and instantaneous, how will people, and their AI agents, decide what to trust? 

🎧 Episode Summary: 

[00:00 – 02:54] Behind the Scenes & Getting Set UpThe episode opens with classic pre-show logistics: audio checks, mic/headphone talk, camera/light setup, and the kind of “are we recording?” energy that always makes the final cut feel more human. There’s also a quick cameo moment with my daughter Quinn :)

[02:54 – 05:52] Pre-Brief: Keeping It Conversational (and Real)Before the formal intro, the group aligns on what “good” sounds like: a natural conversation, permission to interrupt, and reassurance that anything awkward can be edited out. It’s a subtle but important setup for a topic that can get technical fast, and it helps the conversation stay accessible. 

[05:52 – 08:54] Why Scams Are a Context Problem (Not Just a Transaction Problem)

Julie and Emmalyn kick off the special series framing, spotlighting the people and ideas reshaping the U.S. financial ecosystem, and tee up the theme: fraud and AI as a fast-evolving battlefield where innovation and exploitation move in parallel. 

From there, the guest intros ground the episode in two complementary vantage points: Nicky Goulimis (Tunic Pay) focused on scam prevention and accelerating legitimate payments, and Oban McTavish (Spade) focused on data infrastructure that helps institutions understand spend, rewards, and fraud. 

Nicky lays out a core thesis: scams are hard to stop because the missing signal is often unstructured context (site quality, messaging patterns, user journey, merchant legitimacy), not just the payment event itself. 

[08:54 – 11:35] Real-Time Payments: Faster Money, Less Recourse, Higher Stakes

Julie asks how fraud differs in markets that adopted real-time rails earlier. Nicky points to the UK rolling out real-time payments in 2008, and then zooms out globally (India, Brazil, etc.) to underline the defining feature: it’s not only instant movement, it’s that victims often have no practical recourse once funds are gone. 

The conversation ties this to U.S. adoption dynamics (including hesitation around being “on the hook”), while also acknowledging a key reality: scams happen on every rail, including cards, so the battle is broader than RTP alone. 

[11:35 – 14:39] Collaboration Is Still Broken: Banks vs. Merchants

Emmalyn presses on collaboration, and Oban offers a blunt diagnosis: banks are structurally incentivized to move slowly because the cost of mistakes is existential, but the current system is also deeply adversarial. He describes a world where authorization-rate optimization can push merchants and intermediaries toward manipulative behavior (e.g., shifting MCCs, rotating merchant IDs), and argues that durable progress requires trusted third parties and a “median outcome” neither side loves but both can live with. 

[14:39 – 17:45] AI Adoption in Banking: Excitement Meets Reality (and Data Debt)

The discussion pivots to whether banks are actually moving fast on AI. Oban’s view: banks are excited, but the gating factor is often basic readiness, turning messy institutional data into something usable is hard, slow, and expensive, and payments data quality is famously poor. 

Nicky largely agrees on direction: banks aren’t debating whether to adopt AI; it’s about pace, governance, and where they’re willing to deploy it (customer-facing use cases draw more scrutiny than back-office ones). 

[17:45 – 20:52] GenAI Guardrails: Back Office vs. Front Office

A practical distinction emerges: banks may push faster on internal workflows, but slow down when AI touches customer interactions. Nicky highlights the friction fintechs face navigating model governance and review cycles, especially where generative AI could create customer-facing risk. 

[20:52 – 24:28] Designing Detection That Evolves as Fast as Fraudsters

Emmalyn frames the crux: AI makes fraud faster, cheaper, and more adaptive, so defenses must evolve at the same speed. Nicky delivers one of the episode’s sharpest lines: fraudsters are adopting generative AI faster than anyone, which means defenders have to “meet the moment.” 

From there, the blueprint becomes clear: go beyond structured data into richer context, treat “prevention” as a behavioral intervention problem (not just a detection score), and adapt tooling continuously, including sharing intelligence across institutions because the only way to win is through a network. 

[24:28 – 27:05] The Intervention Gap: Why Data Alone Won’t Save You

This section gets operational. Oban describes the uncomfortable trade-off: “zero fraud” would mean blocking everything, but the real goal is minimizing bad outcomes while preserving good transactions. He flags a particularly ugly reality: authorization rates down while fraud rises, the “worst possible chart.” 

The deeper issue, though, is intervention: many institutions still can’t easily stop transactions in-flight or add logic quickly, sometimes constrained by processors, authorization windows, and legacy change processes. Without intervention capability, “all the data in the world” won’t fix the problem. 

[27:05 – 29:53] What Fraud Looks Like Now: False Positives + Job and Investment Scams

Nicky shares a sobering performance reality: even strong institutions may catch only about half of scams, while generating massive false positives that create painful customer friction and expensive manual review. The theme becomes “higher quality alerts” paired with “higher quality interventions.” 

Then the conversation turns to where fraudsters are spending time: a spike in job scams (often starting in text and escalating to calls), and investment scams that use grooming and small “returns” to build trust before disappearing at the withdrawal moment, which is why point solutions won’t cut it and multichannel views matter. 

[29:53 – 32:50] Merchant-Related Fraud: The “Faceless Internet” Problem

Oban argues merchant-related fraud has been underweighted historically because older models leaned heavily on “divergent behavior,” but e-commerce broke those assumptions. Consumers increasingly buy from faceless or platform-obscured merchants, which blurs risk signals, and the incentive structure in onboarding (“fastest path to processing”) creates room for abuse. 

He describes modern patterns like enumeration, low-dollar “drip” testing, and rotating identifiers (descriptor/MCC/MID), all of which reinforce the episode’s thesis: it’s an arms race. 

[32:50 – 38:40] Regulation: Who Pays, and Who Should Be at the Table

On regulatory change, Oban suggests a UK-style approach: if financial institutions bear more direct financial responsibility for certain scam losses, incentives shift meaningfully. Nicky expands the lens globally (PSD3 and other markets) and makes a key point: you can’t solve scams without involving major tech platforms where many scams originate. 

[38:40 – End] The Next Battlefield: Agentic Commerce and the Future of Trust

Emmalyn closes with the big question: ten years from now, when transactions are invisible and instantaneous, how do people decide what to trust? 

Nicky and Oban converge on “agentic commerce” as both exciting and terrifying: if agents can shop anywhere on an open network, they can also be steered into fraudulent merchants, manipulated via injection-style attacks, and exploited at machine speed, raising the stakes on merchant legitimacy and trust signals.