AI Appreciation Day: A Q&A With NMI’s CTO

From underwriting and fraud prevention to developer workflows and consumer-facing applications, AI is reshaping how financial services are built, delivered and experienced.  With July recognizing the impact of artificial intelligence with AI Appreciation Day, I sat down with the NMI team to unpack the growing role of artificial intelligence (AI) in the payments and fintech landscape. 

During our conversation, I also shared my perspective on the rise of stablecoins, the promise and risks of agentic AI, and how biometric authentication is paving the way for more secure, seamless transactions. 

Let’s dive in.

Question: Ahead of AI Appreciation Day, what’s the biggest AI advancement you’ve seen that you genuinely appreciate?

Phillip: One of the most impressive advancements I’ve seen with AI in payments is with real-time merchant underwriting. It’s changing how quickly and confidently businesses can start taking payments. 

Merchant onboarding has been a historically slow and manual process; it can take days to validate a business. But now, AI tools are looking at large datasets from a variety of sources like the web, social media, business histories, etc. And they’re able to flag merchants for auto-approval in seconds. We’re still in the early days, and not many platforms offer this yet, but it’s a direction that the entire industry will need to move towards to reduce onboarding friction and help partners grow.

Question: From your perspective as a CTO, are there any AI-driven tools you’re especially grateful for?

Phillip: When it comes to tools that my team and I are using, I’m really excited about generative AI. But from an industry perspective, I’m particularly interested in the rise of agentic AI in development workflows.

So, instead of suggesting code snippets or summarizing a document, agentic AI is taking on multi-step, complex tasks that are guided by a developer. It’s writing the code for entire projects, spinning up environments and debugging applications all on its own. If the code isn’t working as expected, it reruns it, generates test suites and triages incidents end-to-end to find the problem. These tools are acting more like junior engineers who have real gumption and initiative than the auto-complete tools we were using this time last year. 

It’s a paradigm shift that’s moving AI away from assisting us to being autonomous. And I think these agents will become more reliable as they integrate with more enterprise systems. As these tools become more advanced, they’re going to free up our best engineers to focus on architectural decisions, customer experience, performance, and innovation rather than just how to make code work.

Question: NMI deploys AI-driven fraud protection. Can you share what types of patterns an AI-based fraud tool might be analyzing behind the scenes?

Phillip: AI fraud detection engines constantly scan for behaviors that look abnormal. They’re looking at things like unusual purchasing patterns, how quickly a transaction is occurring, geolocation, fingerprints, etc. While none of this is new in and of itself, AI–powered tools will look at a combination of these things together, rather than individually. 

And what’s powerful about AI versus traditional methods is that AI is adaptive in its learning behavior. It doesn’t just watch and follow static rules; it evolves in real-time as fraud tactics evolve. That’s essential, especially when fraudsters are increasingly using AI themselves.

Question: As AI becomes more embedded in payments, what privacy and security safeguards are companies prioritizing to earn or maintain consumer trust?

Phillip: Transparency is a big one. In finance, trust starts with explainability. The paper trail, if you will. And consumers want to know why things were approved or not approved, what was or wasn’t flagged, etc. That’s a fairly tall order for a black box AI model. So instead, we’re seeing a shift towards models that can interpret their own decision-making process. There are major companies out there investing heavily in tools that can document their entire chain of decisioning. That’s critical in areas like fraud and lending, where you need to see exactly why things happened and where.

There’s also a strong push for AI that preserves privacy. That’s another safeguard that companies should be prioritizing. Another consideration would be AI auditing. Compliance tooling is becoming non-negotiable to meet consumer expectations, as well as upcoming regulations like we’re seeing with the EU’s AI Act.

Question: Mastercard said its autonomous platform will eventually be able to find and purchase goods and services on behalf of customers. Do you think autonomous shopping agents like this have truly arrived?

Phillip: I hope they arrive, because I think it’s really cool. But pragmatically, while we’re on the edge of this becoming real, I don’t think we’re quite there yet. And I think that’s probably the truth for a lot of AI, especially AI in payments. 

I see autonomous shopping agents as being similar to autopilots. We didn’t immediately get rid of human pilots and start using autopilot to take over flights. In other words, we started with co-pilot scenarios, not full autonomy. In the short term, I think AI shopping agents will thrive with things like subscription-based services and repeat purchases where consumer preferences are fairly stable. But generally speaking, people still want control over their purchases.

Will consumers set spending limits? I think they will. They’ll also want the ability to set different kinds of parameters. That’s just human nature. We want the ability to make things work the way we want, like setting spending limits or parameters around categories, restricted items, and even sustainability (like how much of a carbon footprint a particular product has).

I think the key challenge we’re going to see — and this applies to other AI use cases as well — is around transparency and control. People want convenience and customization, but they’ll also want to know what’s happening behind the scenes and be able to change things if they need to.

Question: Going back to agentic AI, what are some risks of rolling out financial agents too quickly, both within banking and shopping applications?

Phillip: A big risk that I see is the over-delegation of things to agents without proper oversight. If a financial agent mishandles funds, has a model error, hallucinates something, or gets an ambiguous user request and makes a mistake, that could lead to a loss of trust. In finance, trust is everything, so we’ve got to make sure there’s proper oversight.

Another risk would be with the biases that are often inherently baked into AI decisioning — especially in the context of lending or approvals. If we think of large, general-purpose models (like ChatGPT or Claude), those are trained on the internet, which is full of misinformation. If models aren’t properly trained on high-quality, diverse datasets, those agents could reinforce systemic inequality or even lead to more discrimination.

In the future, we’ll need to have ethical guardrails and transparency in place before rolling out these kinds of capabilities at scale. That said, you can solve a lot of these problems with small language models (SLMs) that are trained on very bespoke, high-quality datasets. With SLMs, trainers can make sure the AI is trained on real, accurate data.

Question: Stablecoins are all the rage right now, with companies like Amazon and Walmart recently hopping on board. Are stablecoins overhyped? Or will they reshape fintech the same way generative AI has reshaped other industries?

Phillip: I don’t think stablecoins are overhyped. Frankly, they’re probably a bit underutilized in mainstream fintech. We’re in the early innings right now, and there have been regulatory hurdles and other roadblocks to adoption. So while generative AI did reshape some industries, seemingly overnight, I think stablecoins are quietly and fundamentally reshaping finance.

Ideally, stablecoins should give us things like instant global statement settlements with 24/7 liquidity — things that traditional banks just can’t do. But the challenge, again, is regulatory clarity. Once that stabilizes, I think stablecoins could become the default back-end settlement layer for things like B2B payments and cross-border commerce. For embedded payments platforms, that would be huge; it would give them near-zero foreign exchange risk and faster payouts. That’s a game changer.

Question: What are your thoughts on emerging biometric technologies? Could biometric tools represent the next frontier of payments and fraud prevention?

Phillip: From my perspective, I prefer biometrics over traditional passwords. Things like MFA (multi-factor authentication) are necessary to keep passwords safe, but they’re just annoying. With the rise of biometrics, I think we’re being pushed to think of a world beyond passwords.

Frankly, this is a technology I want to see more of in payments. I think biometric verification could really reduce the friction in payment workflows, especially in cases where you’re having to constantly authenticate between system and endpoint or re-authenticate the user or request. That said, the tech itself is more advanced than the general public’s trust in it. So, adoption is going to hinge on the privacy assurances that I’ve talked about, as well as on clear consumer value. 

Question: What were some payment trends that you were keeping an eye on at the beginning of 2025, and how have those trends evolved now that we’re halfway through the year? Did anything take off or slow down unexpectedly?

Phillip: At the beginning of the year, I was watching things like AI-enhanced onboarding (similar to what I mentioned earlier about underwriting), embedded finance for vertical SaaS (software as a service) and faster payouts. 

Now that we’re midway through 2025, we’ve seen an explosive interest in AI-powered underwriting and DevX tools. Meanwhile, faster payments are still lagging. I think that’s largely due to infrastructure constraints, along with compliance considerations — those are tough challenges to solve. 

One thing that surprised me, though, is how quickly ISVs (independent software vendors) are turning into fintechs. And what I mean is that you see more and more SaaS platforms out there that aren’t just embedding payments, but they’re also embedding lending, subscriptions, fraud prevention and even AI. So that transformation towards adopting more fintech capabilities seems to have really accelerated.

Question: Is that something ISVs and SaaS builders are doing to compete? Or just because the technology is more available to them?

Phillip: I think the biggest driver is the fact that it’s easier than ever to innovate and do these types of things. You’ve got platforms like NMI, where we’re making it easier to embed payment components, and AI, which is allowing them to quickly integrate new technologies.

It’s also a competitive advantage. Embedded payments and other financial services are becoming a given. Think of it like a tool in a toolbox; you wouldn’t buy a toolbox that doesn’t have a screwdriver or hammer or a set of pliers, right? Without tools like embedded payments, you’re less competitive. If you’re a SaaS platform, you can use these tools to augment your user experience and help your business and your customers grow.