It looks like you’re a Fintech Takes Network member but haven’t been receiving the newsletter. We’ve added you to make sure you get the full experience (and if you’d rather not get it, you can always update your preferences at the bottom of this email).

Happy Wednesday, Fintech Listeners!

I hope your week is proceeding well. I’m on my way back home, after a productive few days in Salt Lake (my new favorite fintech conference location?)

Many thoughts are bouncing around my brain after yesterday’s AI-Native Banking and Fintech conference (interviewing Raj Date tends to have that effect). Many of these thoughts will eventually manifest in (hopefully) smart insights and ideas in the newsletter.

And today? Today I have a wonderful podcast for you!

— Alex 

---

3 BIG IDEAS FROM THE PODCAST

Jason Mikula and I are back with a new episode of Fintech Recap (because fintech doesn’t recap itself).

In this round, we continue two of our never-ending conversations (open banking and BaaS Island): Plaid’s decision to pay JPMorgan Chase for API access (what happens when facts on the ground move faster than the CFPB) and the FBI’s widening probe into Evolve (BaaS Island has transformed from metaphor to case file).

Plus, something new! FICO is attempting something new-ish with small, domain-specific AI models.

And at the end, a very Dr. Strangelove reality: some risks don’t get fixed; we just learn to live with them … but will we? Will I?

Tune in for the full conversation here

And read below for my three big ideas... 

#1: Model Size Is a Risk DecisionCopy anchor linkCopied

Every language model starts with the same underlying architecture. That part stays fairly consistent. 

What does change (and where the essential choice lies) is deceptively simple: how much data do you want to train it on?

If you take the first path in this Choose Your Own Adventure (and pour in tens of billions of parameters like OpenAI, Google, or Anthropic do), you’ll end up with a model that’s very smart and very broadly capable.

The kind that can summarize CFPB’s 1033 rule and then, for no reason other than your amusement and command, rewrite it in the voice of Dolly Parton (highly recommended, by the way). 

But the trade-off is steep: LLMs are ridiculously expensive (trillions of dollars expensive) and time-consuming to train. They’re slow, costly to run, and prone to hallucinate in ways that are difficult to predict or explain.

And as I wrote about in last week’s newsletter, hallucinations aren’t glitches, but a symptom of how transformer models are probabilistic by design. In other words, a certain amount of hallucinating is par for the course.

There is, of course, another path (because what would Choose Your Own Adventure be without choice?). 

This is the path of smaller, domain-specific models. These models are trained on less data. They’re faster and less expensive to build. Cheaper and faster to use. And less likely (though not immune) to act erratically. 

They’re narrower and honed for the tasks they’re built to do. Bloomberg tried this with BloombergGPT. Stripe did this, too — training a model only on payments data, treating payments as its own language. Because payments have syntax and structure, Stripe’s model could find patterns traditional machine learning never could.

So that’s the trade-off: big models can do everything, but nothing reliably. 

Small models can’t do as much, but as for what they can do, they do better (generally speaking).

#2: FICO Tries to Thread the AI NeedleCopy anchor linkCopied

The dialectic between big, general-purpose models and smaller, domain-specific ones sets the stage for FICO’s latest announcement. 

FICO has launched a Focused Foundation Model for Financial Services.

Underneath the branding, it’s really two things: a focused language model and a focused sequence model. 

And unlike OpenAI, the model’s not built on all the information on the internet, but on the proprietary datasets FICO already controls.

The bet is that smaller, domain-specific models (which, again, are a thousand times less expensive to build) are more reliable for targeted use cases like fraud detection, real-time risk assessment, and next best action recommendations. 

FICO isn’t leaping into the LLM deep end; they’re trying to layer a bit of predictive value on top of classic machine learning in places where it already has an edge.

It’s a prudent strategy, even sophisticated in how it leverages proprietary data. 

The problem is adoption. 

FICO is trying to carve out a middle lane: an AI not too big to trust, but not so narrow that it becomes irrelevant. Big banks like Chase and AmEx already believe they are better at building analytic models than FICO is (debatable, but that’s what they believe), while small lenders may not be ready or interested in adopting generative AI at all. 

Which brings us to the question: Who is this for?

And does a “middle” market segment even exist?

🎬 DIRECTOR'S COMMENTARY

I wrote more about FICO’s choice to pursue small language models, rather than building on top of large language models, in Monday’s newsletter, in case you missed it.

TL;DR (although it’s honestly not that long and you should just read the whole thing): I think FICO is making a smart choice that prioritizes customer value over hype. 

#3: BaaS Island is a Federal InvestigationCopy anchor linkCopied

The FBI has now entered the chat. 

What began as bankruptcy court drama at Synapse and operational and compliance chaos at Evolve has widened into a full-fledged federal probe into Evolve and its sprawling network of fintech programs.

Investigators are digging through Synapse’s Google Workspace (emails, calendar, Drive) and asking the basic questions: when it came to the shortfall in funds, who knew what, when?

The focus isn’t just about the shortfall in funds, though. It’s also about international money movement. 

Evolve is known to be linked to at least one $15 million pig butchering scheme, and there are numerous other examples of risk control failures that would, understandably, be interesting to the FBI (Jason outlines them all in extreme detail in this post). 

The bigger tension here is accountability. Regulators usually put 100% of the responsibility on the company that holds the bank charter (and understandably so). 

However, these failures always take two or more to tango. Middleware, platforms, fintechs, investors — they’re all part of the chain. We may need to start thinking about how we can rebalance incentives across the entire chain, rather than relying solely on the bank. 

And when incentives are broken, failure isn’t an accident. It’s inevitable.

---

WHAT I'M LISTENING TO

#1: Cash Flow Conversations (Fintech Takes x Nova Credit) 🎧Copy anchor linkCopied

A new podcast series, which kicked off with a panel that I moderated at the Nova Credit Cash Flow Underwriting Summit (episode 1) and then a series of fascinating conversations with PayPal (episode 2), Atlanticus (episode 3), and Nova Credit (episode 4).

I deeply enjoyed these conversations (I’m a cash flow nerd) and I have a hunch you will too! 

#2: Vertical SaaS: Fintech Disruption by a Thousand Cuts (Fintech Takes x Pipe) 🎧Copy anchor linkCopied

Luke Voiles (CEO of Pipe) and I are on a mission to learn as much about vertical SaaS and embedded finance over the course of this podcast series. Episode 1 (a crash course on vertical SaaS) and episode 2 (the importance of customer centricity) are out now.

Please take a listen. I’m very proud of this series and I’m excited to for the next batch of episodes to drop!

---

MONEY20/20 SPOTLIGHT

It’s officially Money 20/20 season, which means I’ll be highlighting a handful of (rotating) sessions, meetups, and happenings in every newsletter. 

Some I’ll be at, some just look too good to miss. You’re welcome!

🏀 Fintech Takes The Court @ Money20/20 | 10/26 | 10am–2pm PT

Sunday morning: we’re playing basketball, baby! 

Join us for a high-energy 3x3 pickup game, hosted by Fintech Takes & SOLO. All are invited - female, male, young, old, seasoned, out of shape. Bring your A-game (or just your sneakers) and come play or hang courtside with fellow fintech enthusiasts. RSVP here.

💰 Deepfakes, Real Risk: Fighting Fraud in an Age of Synthetic Identity | 10/26 | 3:00pm–3:30pm PT 

I’m thrilled to be moderating this discussion at Money20/20, featuring the head of fraud at Varo and the co-founders and CEOs of SentiLink and Oscilar. 

☕ Nova Credit Coffee + Conversation | 10/27 | 8:15am–10:30am PT

Start the AM with lending leaders unpacking the real-world journey of cash flow analytics (where to begin, how to apply it, and what it takes to make it work). Breakfast, networking, and discussion included! RSVP here.

🍸 MX Happy Hour Panel: Data into Action | 10/27 |  3:30pm–6pm PT

Small panel conversation featuring Jane Barratt (Chief Advocacy Offer, MX) and yours truly (among others!): 3:30pm–4pm panel, followed by drinks and hors d’oeuvres at The Grand Lux Cafe, Venetian (5-6). RSVP here.

🍸 Fundbox After Hours | 10/27 | 7:30pm–9:30pm PT

Come for the conversation on the future of embedded finance and small business lending. Stay for the one-on-one conversations (over drinks and appetizers, of course!)

---

Thanks for the read! Let me know what you thought by replying back to this email. 

— Alex