Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Fintech: Under the Hood 🎙️

Fintech: Under the Hood 🎙️ artwork

🎙️ Plaid CTO, Will Robinson: Shaping the future of financial infrastructure, Plaid’s north star metrics, efficient innovation flywheels, and AI (transaction models, MCP, & Agents)

Apr 10, 202654 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Hey Fintechers and Fintech Newbies 👋🏼 Welcome to episode four of the podcast mini-series which is another one I’m super excited to share with as it’s a rare deep discussion with a fintech CTO! If you’ve used a fintech app in the last decade, whether it’s for payments, investing, lending, budgeting, or embedded finance, there’s a good chance Plaid was working quietly in the background. It’s the connective tissue between banks, fintechs, and consumers, powering account connectivity, identity, risk, payments, and the broader open finance ecosystem. Will Robinson, the CTO of Plaid, has been at the forefront of financial data innovation. With a background in tech giants like Google and startups like Coinbase, Will brings a wealth of experience to Plaid, where he leads the engineering vision for one of the most important financial data networks in the world that connects thousands of financial institutions to thousands of fintech applications, managing massive volumes of sensitive financial data, while balancing reliability, security, privacy, and performance at scale. Here’s a little preview: In the full episode with Will, we get into Plaid’s architecture, how the platform has evolved beyond aggregation, how they think about data quality and intelligence, how AI is reshaping financial infrastructure, as well as: * The 2008 crash sparking curiosity in how FS works and how to make it better * Low-coupling, High-Cohesion computer science principles and their importance to fintech * Why brownfield development is a lot harder than greenfield dev * What the Plaid x Perplexity partnership means for consumers, hobbyists, and innovators * The tech metrics Will obsesses over and that are key to Plaid continuing to scale * The now staple AI question which touches on their new transaction and behavioural models, the importance of their MCP server, and how Will thinks about Agents using Plaid’s data and services * …and so much more. This episode is for you if you…: * You’re curious about a CTO’s journey to a world renowned fintech * Want to understand how a fintech CTO thinks about metrics * Want to learn about how Plaid is using AI and how it thinks about the future References from the pod 👇🏽 * For those not familiar with Lost in Space, here’s the “Danger Will Robinson” reference and here’s his LI URL: https://www.linkedin.com/in/dangerwillrobinson/ * Here’s some additional info about the transaction model we touched on * This is the latest press release which actually covers some additional functionality to the initial Plaid x Perplexity partnership we discussed on the pod * And as Will mentions in the episode, Plaid is hiring. Check out the available roles. Make sure you subscribe by searching for “Fintech: Under the Hood” on Spotify or Apple Podcast and hitting subscribe to get the next one directly in your favourite podcast app. And while you’re there, give it 5 stars, drop a review, and share it with a fintech friend.☺️ If you’re listening here, then drop a like and comment below! Thanks and see you next time! J. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit jasshah.substack.com

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

We have a network's eye view on a consumer financial behavior and activity that lets us make really critical observations about is this a real user request or is

SPEAKER_01this potential fraud? Or if this user is trying to take out this loan, what is the likelihood that they'll be able to pay it back? We are in a unique position given all of the global network signals we see across many different banks, many different fintech apps, many different devices. We can put all that together and offer these higher level signals to our customers, to both fintech apps and banks that nobody else is in a position to offer. I think we are soon going to have a similar model that is trained more on sequences of activity.

And so again, this kind of higher level semantic concepts around what is my usual pattern?

SPEAKER_01Is my usual is this deviating from my usual pattern I always get paid I tend to eat out this much I get spendier after my payday whatever it is and all of those notions can be embedded within these models that we're now building and then turned around and used for things like fraud detection right this user is acting differently than they usually do and being able to make those kinds of classifications at scale is something we're really really excited about and something that we're in a unique position to help power.

SPEAKER_00Welcome to the FinTech Under the Hood podcast, where we go under the hood of FinTech innovation with the builders and leaders shaping modern financial services. We're back with another episode in this series, and this one's a standout, a rare behind-the-scenes conversation with a FinTech CTO. If you use a FinTech app in the last decade, whether it's for payments, investing, lending, budgeting, or embedded finance, there's a good chance Plaid was working quietly in the background. Plat isn't a consumer brand in the traditional sense.

SPEAKER_00It's infrastructure. It's the connective tissue between banks, fintechs, and consumers, powering account connectivity, identity, risk, payments, and the broader open finance ecosystem. And as financial services become increasingly API-driven and AI-enabled, the role of trusted, reliable financial data infrastructure becomes even more critical. Our guest today, Will Robinson, is the CTO of Plaid, leading the engineering and technical vision behind one of the most important financial data networks in the world.

SPEAKER_00Will sits at the heart of a platform that connects thousands of financial institutes to thousands of fintech applications, managing massive volumes of sensitive financial data while balancing reliability, security, privacy, and performance at scale. In this conversation, we'll go properly under the hood into Platt's architecture, how the platform has evolved beyond aggregation, how they think about data quality and intelligence, and how AI is reshaping financial infrastructure. If you want to understand how modern fintech actually works

beneath the surface,

SPEAKER_00this is the episode for you. Now, let's get into it. Welcome, Will.

SPEAKER_01Thank you so much, Jess.

SPEAKER_00So I usually start these with a kind of how did you get to Plaid? So it'd be great to understand a bit about your journey, where you were previously and your kind of journey into Plaid's CTO role.

SPEAKER_01Absolutely. So for me, I spent a long time working at Google and then I was at a few startups, most recently at Coinbase. And then I've been at Plaid for just about a year now. And when I think back on it, I think the seeds of my being here at Plaid were actually planted in the 2008 financial crisis. A lot of people in sort of my generation, I came up toward the tail end of the first dot-com bubble and then was somewhat early in my career in a formative state when the financial

crisis happened. And although I was deep in technology and infrastructure and that was really

SPEAKER_01filling my vision in my work and my career at the time, when you lived through those times, they left a mark. You couldn't help but notice that there was this whole massive, maybe inscrutable or mysterious financial system out there affecting all of us, part of all of our lives. And I think a lot of people that lived through that experience took away a curiosity, maybe a fear, maybe an excitement around, well, what's just happened here? Let me understand it. Let me know it better. And that was a background interest for me for many years as I

worked in many

SPEAKER_01different places, led me to Coinbase ultimately and pretty deep down the crypto rabbit hole. But the overriding interest was always, how does the system work? How did it go wrong in 2008 in ways that affected Main Street very visibly? How could we make it better in people's lives? And I think one thing that really excites me about Plaid is we're really at the heart of that story of how do we make it better for everyday people? I'm really proud of what we've done there.

SPEAKER_00Well, that's really cool. Yeah, I mean, 2008 crash. I graduated in 2008 and it was the very same reason I wanted to get into financial services is because... I had my computer science degree and I thought, well, where are there some challenges? You know, is there stuff that I could build in an industry that there's some regulatory oversight, but there's actually lots of innovation and lots of things are changing. And it sounds like you had the same experience. Was there anything in particular at that time that you were hyper curious about in

SPEAKER_00financial services? At the time, I think like a lot of people have been engineering bent, sounds like you as well.

SPEAKER_01I wanted to understand the mechanism, right? Like what the heck just happened? What are these collateralized debt obligations? What are these credit default swaps? How do they work? What are these tranches I'm hearing about? And it's this interesting thing where it's not rocket science, but it's also not trivial. And you do have to pull on it. You have to pull on the thread a fair bit to get you know, even just a kind of high-level end-to-end understanding of what was happening, where in the economy,

SPEAKER_01and how did that ultimately mean people were struggling financially, losing their homes, et cetera, et cetera. And this research journey is easier now that the Big Short moves movie is out and a lot has been packaged up for more widespread popular consumption because it did affect us all and aroused so much interest but trying to figure that out yourself reading articles listening to podcasts and so forth that was that was really the focus for me initially and was that was that your the core way you learned was just going off and doing like um self-imposed

research to really pull

SPEAKER_00up the thread and and go deeper on that topic

SPEAKER_01Yeah. Self-imposed research. That sounds more like a punishment. Yeah. Self-directed research. I call it self-imposed research because sometimes it does feel like a punishment. Self-directed research. And, you know, I had a few friends who'd been working in finance in New York and, you know, boy, boy, were they tired at that time of getting these questions, but they were very patient and, you know, gamely answered. And, and I do think that they there are parallels that sort of rhymes with a lot of the technical thinking that you get as you're doing a computer science degree,

SPEAKER_01as you're building big distributed computer systems as well, right? Things like pressure building up in the system, robustness constraints, weird independence assumptions that then bite you in the tail when they turn out to be just a little bit wrong. Yeah. So once I started learning, it was kind of natural in a way to keep learning.

SPEAKER_00Yeah. I mean, one of the things I learned in my degree and that always resonates with in pretty much every role I've been in is low coupling, high cohesion. Core principles are when you're building, you build things that are lowly coupled and highly cohesive. And then my second job at Citibank, I was like, these things are really highly coupled. All of this stuff is interwoven and you change one thing, you can only do four releases a year because everything is so tightly coupled that

SPEAKER_00one change here affects 50 systems. So that was an interesting learning for me. I don't know. Did you have anything? I guess your experience and journey has been almost very tech first companies who were born out of the era of new innovative technologies with those computer science principles baked in, whereas banks and FIs haven't had that. They've been traditional brick and mortar banks now. then the tech has kind of been thrust in the middle of this era

and then they've had to adjust.

SPEAKER_00Is that, what's been your experience working at Google Cornbus? I guess it's a bit more modern.

SPEAKER_01A bit more modern. I mean, everywhere wants to be better, of course. And it's funny. So two things that you just sparked. First of all, I thought you were going to talk about the coupling in the financial system itself at that time, because that was insane, right? Obvious in hindsight, but the powder keg of all the credit default swaps, wiring everyone to everyone else. this sort of global financial suicide pact, right? That is a classic pathology that you look out for as you're designing a distributed computer system, right? We've gotten over coupled and thus like all this brittleness

has snuck in around the edges.

SPEAKER_01When it comes to, so I haven't worked at a bank or a big FI myself. I've talked with a lot of people who do and who have, of course. And, you know, the thing I'll say is I think modern banks who have turned the corner, I bet if you went back to Citi now, and I could name a bunch of other banks, I think they look and they strive to be more and more like tech companies that have a bank attached, right? I do think that that is at least an aspiration, that that is just how seriously they're taking tech.

And we see this in the financial institutions that we work with in almost every case.

SPEAKER_01here at Plaid. And so I do want to sort of give them the credit where it's due. And the other thing I'll say that's sort of changed for me over the course of my life and career, I think like a lot of people coming out of school, it was the thing I loved most was building in a very pure way, often with a big green field in front of me, like an artist with an empty canvas. I can finally express myself. And I do, I still think that's cool, but My experience over the few years I've been at this game is that

there's almost

SPEAKER_01always where something really matters, where there's real value and real important work to be done. The field is very brown, right? Like banks are very mature and they've taken a really, really tough, windy, like path dependent road to get to where they are now. And so somewhere hiding in the bowels of your favorite bank might be some cobalt, running on some ancient mainframe and the purist engineer in me wants to turn my nose up and say, oh, gosh,

SPEAKER_01like let's wipe the slate clean. The pragmatic business value enjoyer in me says, yeah, of course, there's some cobalt there. That's a sign that they've been doing something important and real. And, you know, we accrete these things over time. And part of what tells you you're doing something that matters is you got to deal with a bit of a mess along the way. And so I got a lot of respect for people who are dealing with that mess too.

SPEAKER_00I agree. The Brownfield stuff is a lot harder, but it's a lot harder because it's, like you said, so interconnected and interwoven with what people are actually using, what consumers are using every day. And it's important to so high that that's why it's difficult.

SPEAKER_01There you go.

SPEAKER_00Yeah. Nice. Yeah. And you touched on the big short. I won't go into the big short in too much detail, but your linkedin url is one of the more interesting ones that i've seen danger will robinson of course is that that a lost in space the movie or with matt leblanc or is that does that go further back to the tv it goes further back i mean i've always been will robinson and when i was born there was just the tv show which was already

SPEAKER_01old when i was born just to be clear but but all of my life i've had uh had excited people hear my name and start waving their arms frantically and saying danger will robinson danger and so it's just one of those fun things you gotta you gotta embrace uh yeah i mean if if there's any faux pas i'll start swinging my arms around and shouting danger will robinson totally fine let's get into let's get under the hood of plaid um at its core for anyone who doesn't know what problem does plaid solve today and how's that problem or the challenges

that it solves for

SPEAKER_00evolved over time

SPEAKER_01Absolutely. So the core business that Plaid is best known for is aggregating connections between financial institutions, banks and others. You can just think of banks and a bunch of fintech apps and other applications that consumers want to use. So we all know if you're playing in the financial realm that financial data is historically like very hard to use hard to get your hands on very fragmented across institutions surprisingly heterogeneous and less standardized

than you might expect if you were starting from scratch in a greenfield with a bunch of computers right they've grown up in this path dependent way and so it's a pretty messy world to try to make sense of and then on the other side of that aggregation picture you've got

SPEAKER_01a bunch of Big financial technology apps, if you're using a Chime, SoFi, Robinhood, Venmo, most of the big fintech apps that you could name or think of, their customers are Plaid, they're powered by Plaid, and we make it simple as a sort of one-stop shop. for developers that are building apps like that to come to Plaid and let their end users, the ultimate consumer, link up their bank accounts, their financial institutions of choice and share transaction information,

SPEAKER_01account information. facilitate ACH transfers and payments, all these kinds of things. So there's a big, messy, long tail dominated set of banks on one side. And there's a bunch of fintech applications being built on the other. And we make it easy for the two to connect without... without each of them having to do a ton of redundant work. So that is definitely the thing that Plaid is most known for today. We've got a bunch of other really exciting stuff going on that

kind of builds on top of that foundation that I think we'll probably get into a bit later today.

SPEAKER_01But that is Plaid in a nutshell.

SPEAKER_00Yeah, it's a very clear description. I think some people don't quite understand what Plaid does and the value it adds. And I think that's kind of nailed it. Yeah, so early plaid was more about aggregation. I think the founders, Will and Zach, from what I can remember, it was from them trying to build their own PFM app and realizing that actually the data connectivity is not quite there. So they decided we'll build the aggregation,

we'll build the data connectivity. I'll just jump in and to be crystal clear, probably obvious, but different Will.

SPEAKER_01I think he goes by William. was a co-founder of Plaid 13 years ago, along with Zach, our current CEO. And the lore as I have it is that they set out to build their own personal financial management application. They very self-deprecatingly say that in hindsight, they were really bad at that. But along the way of doing that is just an implementation detail they needed to get right. There's this really annoying problem of all these people have these different bank accounts and we just want one clear way to connect to all of them.

SPEAKER_01They were well-networked with the fintech scene, I think, in New York at the time where they were living. And as they were showing their doomed-to-failure finance app to all their peers in the room, everyone said some form of, that looks nice, and the app's crap, basically. But how are you connecting all the banks? Could I have just that part? And it is a really critical moment in the life of an entrepreneur to notice when the world is telling you, I want to buy this thing from you that you haven't even yet conceived of as a

SPEAKER_01product. And credit to Zach and William. They twigged that pretty quickly and really pivoted and started what is Plaid today.

SPEAKER_00Today, there's a bunch of stuff in the Plaid stack, identity, payments, risk, credit, data analytics. So from your perspective, and obviously you haven't been there for the entire time, but how has the architecture evolved now? to support the expansion across a bunch of different use cases?

SPEAKER_01Great question. And so this is partly the story that I mentioned briefly earlier, where we've still got this deep critical foundation of core connectivity. And that is a mission that we can never fall asleep on. And also having done things we're really proud of in that space, we've connected the majority of Americans with bank accounts, for example, to at least one fintech app at this point. So the breadth and depth of our financial data connectivity of

our network is how

SPEAKER_01we like to talk about it is really unrivaled in the industry. And that presents a natural opportunity to say, in addition to simply connecting from points ABC to points DEF, there are a bunch of things to do with this data that we're in a unique position to do with appropriate consumer consent and education and permission. But we have a network's eye view on consumer financial behavior and activity that lets us make really critical observations about

SPEAKER_01is this a real user request or is this potential fraud? Or if this user is trying to take out this loan, what is the likelihood that they'll be able to pay it back? Or this user is trying to transfer money from their bank account to this crypto trading app, what is the likelihood that that transfer will actually go through and not be reversed? We are in a unique position given all of the global network signals we We see your activity across many different banks, many different fintech apps, many different devices.

We can put all that together and offer these higher level signals to our customers, to both fintech apps and banks that nobody else is in a position to offer.

SPEAKER_01offer. And so in order to do that, we've really had to up our game building on top of that infrastructure foundation we had for connectivity, building a lot of sophistication around data itself, right? When you go from being a data pipeline purely to also being a data refinery, we're doing a bunch more analytics, a bunch more ML, a bunch more AI with that data to build these high order products around payments risk, credit worthiness, fraud risk. And that has been the journey that we've been on in the last couple of years adding those creating those value-add services

and increasing product lines and increasing use cases what does that mean from from a systems design perspective on reliability for plaid scale because obviously there's more use cases there are more customers therefore there's more reliance on plaid how do you as a cto as a tech team look at reliability and what are those what are the key

SPEAKER_00things to to look at when you scale when you continue to scale

SPEAKER_01Yeah, it's a great question. Keeps me up at night in a good way, right? So one of the things that we look at, of course, is just what is our uptime brute metrics around is the service working? But I think where it gets really interesting is as you expand, cover more banks, cover more financial institutions, cover more fintech apps, the natural thing would be for average quality to kind of degrade, right? Your network is getting bigger and bigger. At the margins, maybe you're adding things that are more flaky or your your attention is being spread over a wider

and wider surface

SPEAKER_01area and things are just degrading over time. And so the standard that we hold ourselves to and the thing that we're always working to build our systems in support of is we want to simultaneously have a bigger broader and deeper network and also a higher and higher average level of reliability and predictability across that network as we go. And so far over my year or so here, I'm really proud of how we've achieved that. And I think that that credit goes back many years before I was here, all the work that the team's been doing, the big investments we've made in our platform,

SPEAKER_01in our infrastructure. And frankly, the muscle memory and experience in working with banks and working with fintech apps that's just encoded in the DNA of our company. It is such a complex space. The stakeholders involved, even, you know, arm's length regulators through the banks are so specific in how you need to work with them, how you need to talk with them. And that is, that is experience that you just can't

SPEAKER_01get overnight, right? And so Plaid has been through that school of hard knocks and I think come out on the other side now supporting the majority of financial institutions, supporting the vast majority of fintech volume in a place where Now that complex, complicated mass is actually working for us in a sense, like we've mastered that domain. And so it is very, very likely that the 12,000 and first bank that we add is super high overlap

SPEAKER_01with some subset of the 12,000 we've already added. So at first, it is a real headwind against you just how gross and heterogeneous the consumer finance space is in America. But once you've mounted that curve, then it sort of becomes part of your moat.

SPEAKER_00Yeah, it's a headwind. And then when you kind of get shifted in the right direction, it becomes very much a wind that takes you in the right direction.

SPEAKER_01Yeah, absolutely.

SPEAKER_00Yeah. And you've touched on some interesting metrics there. Which ones are you really obsessive over? I guess if we exclude the data ones, because I guess we could get into a 30-minute discussion just on data metrics. But if we talk about things like connectivity, uptime, latency, which are the ones that are your North Stars as a CTO? And then which are the other ones that you're looking to improve?

SPEAKER_01Yeah. You know, ask a technologist his or her favorite metric is like asking, you know, who's your favorite child, right? There are a bunch that we care about really, really deeply. I'll just give them sort of a few in no particular order. So obviously there's coverage, right? Coverage of financial institutions one of our selling points like the value proposition to a fintech app customer of ours is we can connect to any bank any financial institution that your end user want has works with or wants to connect to we can do it really reliably and the conversion is going to be really

fast and seamless right and a lot of the metrics that we look at most as north stars flow very naturally from that so there's a gross coverage metric of banks and financial institutions

SPEAKER_01but A half-click deeper than that, there's also the sort of quality of that coverage. It's not just were we able on one day to... get connected to this small credit union in Tulsa, but is that connection live? Is it active? Are we sure? How do we know? Are they using modern OAuth and API techniques if possible versus like an older technique where we're more manually getting data from their website or their application, pushing, pushing, pushing, not just can we check a box and say,

SPEAKER_01yes, we support 12,000, but is the level of support for each of those as high as it possibly could be. And that's really important to our customers. It's a huge selling point for Plaid. Another one we look at is the latency and freshness of that data. So one of the many things that we're trying to manage is what's our interaction pattern with these banks and FIs, financial institutions. You know, 13 years ago, nobody knew what Plaid was. It was just getting started. And we've been on a really long journey of getting more awareness of who we are

and becoming better and better partners and collaborators with the banks and financial institutions that we interact with and so being really good stewards of that and of those relationships I think is another really important thing to us we have a lot of metrics around that and then the error rate and how quickly we can recover from errors right so we are dealing with such a large and messy world here

SPEAKER_01It is one of these backdrops as a technologist where the goal is not, just cannot be, all banks are up all the time with instantaneous freshness and latency. We can only be as good as the world we're sitting on top of or interacting with. to some extent and so it's more about agility and speed as banks are going up and down having problems having data issues right are we twigging those as quickly as anyone possibly could are we nimbly recovering have we plumbed

the right things through to our customers the larsie fintech apps so that they've got all of the tools and like the best freshest state that they can possibly have to serve their

SPEAKER_01customers

SPEAKER_00Nice. And on the error rate, is that partly self-regulated based on an all data view of the market and maybe spotting some of those data points being off because you can see similar data patterns from other banks? Is it customers saying, we don't think this is right, and then you go through a review process? What is that?

SPEAKER_01Is it a mix of both? All of the above and more. So as in most of these things, the majority are kind of boring data issues, like fairly mundane. The bank's API went down for 10 minutes and we've been through so many thousands of those reps that at this point, our systems are very resilient to that, take it as sort of a matter of course, no human intervention is required. And that gives a good end-to-end experience through Plaid and

all the way to our

SPEAKER_01customers with automated escalation back to banks and FIs if needed, to our customers if needed, right? Like the maturity of the model really positively surprised me. I had a very positive impression of Plaid before I joined, but after getting here and just seeing just how far along that road we are of investment in terms of We've seen a thousand of these failures. We know what they look like. We built up the kinds of capability and detection automation of response that you would hope to. I am really proud of where we stand there.

SPEAKER_01And then there are you know, a smaller but meaningful subset of things that are more subtle, more long lasting, where we can draw on these larger network insights, look for anomalies in how certain FIs are behaving. And we do do some of that, but I don't want to overstate. I think most problems are thin. And the kind of big problem for us is just that we cover a really wide surface area. So there's always something going wrong out there in the financial network at all times.

SPEAKER_00Yeah. But I think, yeah, as you touched on before, the reps comment, because you've been through so many reps of this, that the muscle is developed enough that you can spot, like, we need to apply this process to this thing because this thing has happened again. It's happened 999 times before. That's right. Continuing with metrics, but more like signals, what signals tell you the platform is scaling healthily? Like what's the one thing that you look at to see, you know, is it?

SPEAKER_00Again, I think you mentioned before growth, actually growth and growth of institutes, but the data is still, the accuracy rate is still there, connectivity is still there. What's that scale metric you're looking at?

SPEAKER_01Yeah, so there's really two. I did mention that first one before. And just to emphasize it again, I think it is simultaneously growing our coverage and also growing the average quality within that coverage set at the same time, not making a trade off a across those, but saying we cover more and the data quality has improved, the freshness has improved, the consistency of categorization or whatever across institutions has improved, even as the network has gotten bigger. And that's job number one. The second thing that we talk about a

SPEAKER_01internally, it's maybe one hop further removed from our customers, but I think it's critical in how we drive our business. I'm constantly talking and my team, we're constantly talking about how quickly can we develop new products, new features, new models, for the data heavy products that we have. As we've entered this new phase of more intelligence on the data, more AI, more ML, the name of the game, we all know that that part of the world is changing so quickly. And the name of the game for us is while staying incredibly

reliable and rock solid

SPEAKER_01in everything that we've already built, How can we make sure that our innovation flywheel is operating like as furiously and as quickly as it possibly can. And so that's a really fun and interesting and sometimes exhausting tension to hold as a company. I think customers of Plaid's current products want us to make sure we keep doing exactly what we do, rock solid, and sort of never have to talk to us again. And it's a great, boring story. And that's Plaid in the morning, let's say.

SPEAKER_01But then Plaid in the afternoon in the evening. People are also looking to us to help define the future of building AI enabled products, doing way more with intelligence around fraud and other signals. And so we need to be great at both. And they're not opposites, but they do require like different muscles, different subcultures within the company. It's been a really interesting thing.

SPEAKER_00Yeah. And there's always the resource allocation challenges. Well, we have to keep the lights on, but we want to renovate the back room or add an extension.

SPEAKER_01Well said.

SPEAKER_00Yeah. The electricity needs to stay on though. So how do we do that? And how do we do it quickly and make sure it doesn't impact any other part of the house? It's a tricky challenge. And if you weren't innovating, customers would say, so-and-so over there is building this new thing. Why aren't you innovating at Pace? So you You can almost never win, but you have to keep moving, keep pushing.

SPEAKER_01The competition's a beautiful thing.

SPEAKER_00Yeah, exactly. Yeah, and touching on innovation, are there any recent product or infrastructure launches you're really excited about, especially from that engineering standpoint?

SPEAKER_01Yeah, absolutely. So one that comes immediately to mind, we've been training this transaction foundational model, which is effectively taking a bunch of anonymized de-identified transaction data from a Again, this big, massive network of financial data that we have and training LLM style model on top of that, that is now outperforming all of our previous iterations on tasks like transaction classification. It's a very common thing.

for folks who are using Plaid to say, okay, well, this consumer is hooking up to my product.

SPEAKER_01They're doing maybe a budget management app or something like that. And they want a high accuracy first pass take of, yes, I've got these lines of transaction information from this bank or credit card, but what are the categories of spending? What does it all mean? What's discretionary? What's income? And the raw starting place for all these transactions is just, Text spat out by some COBOL system at some bank, inconsistent across banks, etc. And so it is a really important fuzzy matching fuzzy logic sort

of challenge to

SPEAKER_01give a best guess transaction categorization. that a human would look at and say, oh yeah, that was a good guess. That Dunkin' Duh is probably Dunkin' Donuts. That probably counts as a restaurant. Okay, great. And we recently invested a lot in training our own in-house foundational model that is doing much, much better versus our previous ML based techniques versus our previous heuristics at doing this kind of transaction classification. I think we are soon going to have a similar model that is trained more on sequences of activity.

SPEAKER_01And so again, this kind of higher level semantic concept around what is my usual pattern? Is this deviating from my usual pattern? I always get paid, I tend to eat out this much, I get spendier after my payday, whatever it is, and all of those notions can be embedded within these models that we're now building and then turned around and used for things like fraud detection right this user is acting differently than they usually do and being able to make those kinds of classifications at scale is something

we're really really excited about and something that we're in a unique position to help power that's yeah that's really cool i mean the transaction categorization

SPEAKER_00thing i think if you haven't worked with an open banking company you might not know, but actually most PFMs traditionally have done some element of categorization themselves because they'll get the data from third party. Every bank, I've done this test before because I worked on a PFM app in 2018, is do a transaction on six different cards, but exactly the same merchant, exactly the same product, so exactly the same amount. All the transactions, the descriptions are completely different.

SPEAKER_00Some of the MCC codes were slightly different. And that MCC code doesn't map directly to a specific category that PFM is wanting to use to make the transaction useful for the consumer. So that's a huge one. And I think the behavioral stuff is really interesting because I think things like Payment velocity, right, you're going to hit your overdraft at a certain point because this is a change to how you usually spend. It's an increase in spending in a way that you haven't spent before,

SPEAKER_00which means based on your previous salary payments, you're going to get hit with a $30, $25 or $28 overdraft fee. Overdraft charge, yep. Yeah. Those insights that, again, traditionally, they've been built by the PFMs or the lenders, the payday lenders sometimes, or the salary advance companies themselves. And it's much better to have a centralized model that, again, has the biggest coverage, the biggest access to institutes and the biggest source of data

to have that model there to access if they want to switch or they could just continue to evolve it

SPEAKER_00themselves.

SPEAKER_01Yeah, it's well said. I mean, we're seeing transaction categorization successes or accuracy approaching 95% in some domains with this new model. And so we're really excited. I think that there is some tipping point there where it's like, if it's not good enough, then it's basically useless. And then once it's good enough, it's incredibly useful. So I think we're going to find ourselves more and more on the correct side of that tipping point for more and more customers. And I'd also say what we see in almost all of our customers is they're really happy to have a higher floor to start from by taking advantage of

these kinds of

SPEAKER_01capabilities from a Plaid. And then of course, like their secret sauce, their business, their value add on top of that, there's still work for them to do to differentiate from their competition. And where I think things get really interesting, like with the sequence model that I was mentioning, is it's not replacing something that our customers could have been doing themselves. Because they see everything that's happening on their app, they can't see everything that's happening network-wide in the way that we can.

SPEAKER_01And so in a lot of these cases, and cases that are most exciting to me, it really is complimentary. Like we're bringing a signal to these customers that they've just never imagined having before couldn't have had in any other way it's not a replacement for their in-house software it's a complement to it that supercharges their efforts in a lot of cases so that's the kind of win-win you always love to find in business yeah that's really cool are there any other um i think i saw an announcement a plaid perplexity announcement are

there any other

SPEAKER_00product or infrastructure launches um that you're kind of buzzed about

SPEAKER_01Yeah. So there's always a lot coming and I always forget the calendar of what we've announced and what I'm allowed to say, but I do want to talk about Perplexity for just a second because I think it's a great example of this kind of partnership. So in Perplexity, we've recently announced our partnership with them. You can today go to Perplexity, connect your financial institutions, your investments, your bank accounts, credit cards, and so forth via Plaid and see right there in Perplexity, this very professional, very comprehensive

um interface to say analyze your investments analyze your spending do the kind of analysis that not that long ago you can only really get in that way by spending a lot more money maybe hiring a human financial advisor and the democratization of these kinds of capabilities right that's something that really animates me around

SPEAKER_01our mission our mission is financial freedom for everyone and Part of that is just spreading around these capabilities that some people have had access to for a long time, but without AI, without this kind of automation and without this data flow, would just not be feasible to scale to a mass consumer product or a free consumer product. And as that sun rises over that landscape, the capabilities that we're going to be putting in people's hands, like the level of freedom that they can then express just goes up,

SPEAKER_01right? And it's a pure win, right? Like taking a bunch of people who weren't drinking coffee before and now they have Starbucks. And I think that's just a great thing for the world.

SPEAKER_00Yeah, I think the one thing it also adds is In the fintech industry, we tend to talk about personalization a lot, but actually there is almost a technological restraint to having a product that's hyper-personalized. Again, hyper-personalized is another phrase that comes up a lot. If you're building an app, what people have realized is you can't have every homepage be totally dynamic. You can't have different account management profile screens

being completely

SPEAKER_00dynamic and personalized. whereas you can do that if you have a gen ai layer and an llm and the data connectivity there and consumers can ask whatever questions they want it's completely open they can learn about their finances in a way that they are accustomed to or they feel comfortable with and in that way it's almost the ultimate personalized product you're just providing the connectivity the interface is is the consumer zone to shape as they see fit. However, they need to use it is how they can use it.

SPEAKER_00And I think it's the ultimate kind of personalized product for me.

SPEAKER_01I agree completely. I don't know what we'll all be doing in five years, right? Like my only dash of humility here is as excited as I am about how all of this works now, it's changing so quickly. I'm sure we haven't seen the final form. And I think that right now, the gains to personalization, the novelty and real utility of being able for the first time to just ask your financial self questions about itself are so huge that we're

doing it in a way that

SPEAKER_01we don't sort of notice any downsides, right? Just because that's so novel, like being able to just chat to a thing and say, oh, did I spend on trips last year and get an answer back and it'd be mostly correct. That is a superpower that we just haven't had before. And we're all just reveling in that new playground for now. And also, I suspect, just as with each paradigm shift in how consumer facing computing has happened in the past. We will see a lot of fast follow innovation and how we're

packaging and delivering

SPEAKER_01these insights as well. Not everyone will care about their finances enough to chat a lot with a perplexity or anyone else and so i think that there's a new design space also opening up saying okay cool for the casual user how do we still meet them where they are give them the best of these new capabilities even if they're not going to be power users or or take the time to figure out what their dream custom personalized ui is so more more more to come i would say watch this space I think you touched earlier on the open finance shift to

intelligent finance.

SPEAKER_00I think we've experienced this in the UK where we've got open banking, open finance, open data, but I think the industry is moving open finance to intelligent finance, especially with this AI boom. What does that shift mean in practice?

SPEAKER_01Yeah, it's a great question. One of the things that we've seen, I've seen some early polls on this. I don't know how authoritative they are, but from what I read, I think more than half of American adults have already engaged with AI in some way to interact with their finances, to talk through budgeting, to talk through savings, to get investment advice. The consumer demand And the change in consumer behavior around this crop of tools is like nothing I've ever seen.

It's happening more quickly and more sort of omnidirectionally than any other

SPEAKER_01change I've observed in my lifetime. And given that that is happening so quickly, I think we're moving to a world where if intelligence doesn't feel embedded in the product that you're using, very quickly the product is just going to feel broken. One goofy analogy here, if you watch toddlers now come up to like a big screen, like a plasma screen TV, they will very often reach and try to poke it. Because every screen they've ever seen... has been an iPad, has been a phone, has been a touchscreen, right?

SPEAKER_01Like they just think that's how screens work. And I think as you get over this tipping point where almost every app you use, every interaction you're having with a computer has intelligence embedded within it, that then becomes the baseline default expectation. And anything that is lacking that just feels antiquated. So we're seeing this shift happening in real time. And I think Plaid has the core data aggregation role still matters a lot, hasn't gone away at all. And also these additional layers that we're building on top of

that to let these

SPEAKER_01companies build from a higher foundation of more pre-aggregated signals. That is really, really critical to people getting off to the races. The other big trend that I'm seeing is I think a pendulum swing back toward smaller companies, even like hobbyist users, early, early startups, people creating a whole new raft. of AI native, AI first, fintech or financial applications. And we are doing a lot at Plaid right now, actually, to make sure that as we've gotten bigger and served bigger and

bigger enterprise

SPEAKER_01customers and all of that, that we have not lost sight of this great engine for innovation, the one or two people in the garage that These days are just saying, Claude Code, or pick your favorite coding agent. I'm making this fintech app, you connect to banks, make it happen for me. And we want to make sure that Plaid is the thing those agents reach for, and that that experience is all seamless and works really, really well, so that we are the default and de facto choice for this whole next generation of fintech apps that I'm really, really excited to see coming out.

SPEAKER_00As we're moving into this new intelligent finance era, what's the infrastructure change that's needed to enable this next wave?

SPEAKER_01Infrastructure change, that's an interesting one. I think there will be A lot more to come on agent-to-agent communication. I think MCP, again, a sign of the times is just how quickly this is happening. This came from a sort of standard someone threw up on the internet to this lingua franca that we all have to passe in some ways over the course of six months. And whether it's an evolved MCP or new standards, things like that, I think we've not seen anywhere near the last of that trend.

SPEAKER_01Again, I think that a lot of Plaid's direct users, in some sense, are going to be agents over the next year or two. They're going to be agents where there's still a human being behind them, pushing them and giving them the reason to come use Plaid. But making sure that we are positioned for this agent-heavy future that On both sides of the network that we're aggregating, both the fintech apps that are building against us, and also I think the financial institutions that are going to

be increasingly using

SPEAKER_01AI tools for their internal operations, for anti-fraud, for a lot of interactions with us and with others, that drives a bunch of internal infrastructure work and platform work and switches a bunch of priorities around for us. And just a personal curious question, can you spot the difference between an agent and a...

SPEAKER_00a real FinTech innovator if they connect? Because presumably I could connect to an MCP server through some API layer. I think I've seen some of the API docs and I could probably connect to the MCP server. Can you identify the difference between an agent and a real person? And do you have different set of parameters that control what an agent can do versus what a real human can do?

SPEAKER_01Yeah, we do do this today. It's manual right now. What we do is we ask them to tell a joke and that gets sent to me and we see if I laugh. And then if I laugh, it's a human. No, sorry. Yeah.

SPEAKER_00I was going to say, you've got way too many things on your plate.

SPEAKER_01It's an interesting question. And, you know, it's not the main focus for us right now. The line's also fairly blurry. You know, we see these agents acting. When they're here, they're acting very directly at the behest in near real time of a human. And so it almost doesn't matter to us except making sure that whichever combination, like whatever that mix shift is, They're having an excellent experience across the board. If it is a human writing some artisanal Python code by hand,

we want our human facing API docs to be immaculate and our onboarding flow to work

SPEAKER_01really well for that. And I think we've got a certain set of metrics that can cover that story. and we're still hard at work on that. Humans are not going anywhere. And if it is instead cloud code or some other agent, coding agent acting at a human's direction, we want to make sure that that experience is immaculate, that they've got programmatic access to all the documentation that they need, that is the sort of easiest and first things that they find. It's sort of like SEO for human eyes. There's this agent optimization as well for agent eyes. And that is a trend that we're right on top of.

And sorry, just one more follow-up just because of something you just said there. Are you going to be able to handle the potential amount of increased volume in

SPEAKER_00terms of connectivity? Because obviously, if you're a fintech innovator building a fintech app, you're one person or two people. You can ask a plaid representative to set up an account to, you know, exchange keys and then set up a connection and then connect to banks. If there's agents doing it, there's, you know, There's not really a limit to how many connections can be set up. Do you anticipate an increase in growth in connectivity and volume based off the

SPEAKER_00potential of agents creating their own connections?

SPEAKER_01Yes, I think there will be. And I don't want to be overconfident, but frankly, I think we're ready for that. I think that we won't see a step change overnight and the agentic volume will still cap out at much, much lower than like the population of the world or something like that. And so I'm very confident in our ability to sort of elastically scale down and to elastically scale our ability to handle a higher and higher

volume of self-serve

SPEAKER_01traffic, which is effectively how we would experience that kind of inflow or increase. And so we We're well positioned to go with the market if it goes that way and we're flexibly positioned enough that we don't have to guess right a year in advance to build out the capacity or something like that like we can track the market pretty nimbly as it goes and oil boy jazz like that would be a great problem to have i really hope we're talking in a year and

you say well now that there were a million new fintech apps and they're you know each each serve a hundred incredibly happy customers and they've all got plaid connections you know how how has the last stressful year been for you supporting all of those and i will be taking the call from my yacht and telling you it's been a fine year for us

SPEAKER_00Well, I mean, that's a good place to check in. I'll call you in a year.

SPEAKER_01Perfect.

SPEAKER_00There's one last question before we go. Yeah. This is a standard question. It's a FinTech Under the Hood podcast. So when you think about everything happening in the industry right now, what's one thing you think people still don't understand about how it really works under the hood?

SPEAKER_01That's a great question. And I think it goes back to some of the kind of path dependence and legacy code stuff that we were talking about before. This is an industry that, you know, it's almost like if you're drilling in the soil and you see different layers and like different fossils as you go down, right? And it's all still there. Somewhere within the gleaming skyscraper of JPMorgan Chase's new headquarters, there's still like the old leather bound ledger book that

JPMorgan himself like

SPEAKER_01wrote in cursive in, right? And that stack is technical. That stack is cultural. And that stack like going up and down all through it ultimately is all about coordination and trust. And when you engage with this industry and when you try to help push it forward, there's just no substitute for engaging with it on its own terms, like getting into that deep mess of

of history and of culture and of trust and of things that you might not understand

SPEAKER_01at first, but they were put in that way in 1962 for some reason. And it is not an industry for a purist or for the faint of heart, but the value that you can add to the world by engaging with that and really taking it head on and being a true participant. an innovator within that space is like nothing I've ever seen before. So we are hiring, by the way, if anyone listening is wondering,

SPEAKER_01are there still jobs for engineers in this AI enabled future? There's a ton of jobs for engineers and product people and data people at Plaid. I'm really, really proud of the scrappy and lean team that we've had that's taken us to this point in the journey. And I also believe our business is at an amazing inflection point. And this is a journey that in many ways is just getting started for us. So please reach out.

SPEAKER_00That's awesome to hear. And I'll Try and put the jobs link below. And finally, if there's, you know, if people want to learn a bit more about you, a bit more about Plaid, where should they go?

SPEAKER_01One place that I would recommend is our company blog. You know, that maybe sounds like I'm talking from 2007, but company blogs are still great. And we are investing a lot more in ours. And you can read about our tech stack. You can read about business developments, new partnerships, get a good cross section there. So please give it a look.

SPEAKER_00Awesome. And yourself, LinkedIn forward slash Danger Will Robinson?

SPEAKER_01That will do it for LinkedIn. I've got some terrible URL and Twitter. I'm not very social, but you can find me.

SPEAKER_00A true engineer, not very social.

SPEAKER_01There you go.

SPEAKER_00Awesome. Well, thank you so much for your time, Will. And yeah, great having you on.

SPEAKER_01Truly a pleasure, Jess. Thanks so much. If you enjoyed this episode, don't forget to like and subscribe wherever you're listening to it.

SPEAKER_00It really helps with visibility. And if you think a fintech friend or colleague would enjoy it too, please share it with them. That's it for this episode. Thanks for listening and see you next time.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. 🎙️ What Does It Take to Build an AI-Native Bank? | David Henry, CPO, Mal

    Nov 18, 20261 hr 6 minAudio

    Hey Fintechers and Fintech Newbies 👋🏼 Welcome to the first episode of S2 of the Fintech: Under the Hood podcast🎙️ We’re kicking things off with a fascinating conversation about what it actually means to build an...

    Credit the people on this episode

  2. 🎙️ Allica Bank CEO, Richard Davies: Relationship banking, AI powered efficiencies, success metrics, and global expansion

    May 22, 202646 minAudio

    Hey Fintechers and Fintech Newbies 👋🏼 Welcome to episode five and the last episode of this podcast mini-series which ends with a bang, as it’s an in-depth discussion with the CEO of the one of the fastest growing...

    Credit the people on this episode

  3. BVNK Co-founder & CBO, Chris Harmse: Banking to blockchain, stablecoin utility, the stable spending gap, and the future of money

    Mar 13, 202648 minAudio

    Hey Fintechers and Fintech Newbies 👋🏼 Welcome to episode three of the podcast mini-series which is another great one as it explores one of the defining themes of this era of innovation. Stablecoins are often talked...

    Credit the people on this episode

  4. 🎙️ Lean Technologies CPO, Tewfik Cassis: Innovating in the Middle East, AI in the product-dev lifecycle, and the importance of full-stack infrastructure

    Feb 27, 202654 minAudio

    Hey Fintechers and Fintech Newbies 👋🏼 This second episode of the podcast mini-series is a great one for a multitude of reasons. Over the past few years the Middle East has been a hotbed of innovation with fintech at...

    Credit the people on this episode

  5. Checkout.com CPO, Meron Colbeci: Invisible Payments, Agentic Commerce, and Globalising Products

    Jan 23, 202651 minAudio

    Hey Fintechers and Fintech Newbies 👋🏼 Kicking off the first in this interview mini-series with a banger. Payment Services Providers are the invisible force that allow us to safely purchase goods & services online and...

    Credit the people on this episode