Plaid Effects 2026

Indexado: 2026-05-22

[00:00:27]Don't get turned out. Yeah. Yeah.

[00:00:42]Hack back.

[00:01:35]Turn it up.

[00:01:47]every time.

[00:02:04]Yeah.

[00:02:25]Heat. Heat.

[00:02:47]I'm turning it up.

[00:03:02]H Jack again Heat.

[00:03:36]Hey. Hey. Hey.

[00:03:54]Turn it up.

[00:04:08]Get back.

[00:04:11]Get back.

[00:04:25]Heat. Heat.

[00:05:06]Turn it up.

[00:05:18]All right.

[00:05:23]Happy Jack.

[00:05:36]Get back.

[00:06:14]Turn it up.

[00:06:28]Dip crack.

[00:07:15]Hey.

[00:07:20]Hey. Hey.

[00:07:26]Turn it up.

[00:07:37]All right.

[00:07:58]Hey on hey me hey hey Turn it up.

[00:08:45]back.

[00:08:58]Back down.

[00:09:34]Yeah.

[00:09:36]Nah.

[00:09:45]Turn it up.

[00:10:03]Pap.

[00:10:25]back.

[00:10:52]Turn it out.

[00:11:07]Grand X.

[00:11:43]Hey, hey, hey.

[00:12:04]Turn it up.

[00:12:20]happy.

[00:12:24]Jack happy.

[00:12:55]Hey, come on.

[00:13:12]Turn it up.

[00:13:26]Grand Hey, hey, hey.

[00:14:24]Turn it out.

[00:14:38]H.

[00:15:02]Hey. Hey.

[00:15:31]Turn it up.

[00:15:42]Jack cla Heat. Heat.

[00:16:43]Turn it up.

[00:16:58]Every hand doo doo doo doo doo doo doo doo doo doo.

[00:17:22]back.

[00:17:36]Heat.

[00:17:51]Turn it up.

[00:18:05]back.

[00:19:03]Turn it up.

[00:19:22]Jack happy back again.

[00:19:53]hey back again baby.

[00:20:10]Turn it up.

[00:20:21]Black train.

[00:20:49]Heat.

[00:21:00]Heat.

[00:21:29]feel.

[00:21:32]Hey Please welcome to the stage Plaid co-founder and CEO Zack Pereé.

[00:22:01]>> Hello, welcome to effects. Thank you so much for joining us. Um, this is by far the biggest effects that we've ever done and we have so much that we're going to share with you all today. Um, for me, it's really great to be back in New York City. My co-founder and I actually founded Plaid here, uh, in a tiny little office in Union Square almost 13 years ago. Uh, and since that time, it's it's wonderful to reflect on just how far this amazing industry has come. I'm not sure about y'all, but I'm having the most fun of my career right now. Fintech has truly gone mainstream. Um, fintech and financial services, the two industries are continuing to merge and become just one industry. And now almost every bank, every lender, every investment product builder that you talk to is creating their products digital first and mobile first. This is kind of the idea that we had at the very beginning. That's taken us a long time to get here. We also are starting to see more and more large tech companies, retailers, and so many others launch embedded finance products. They're putting wallets in. They're putting rewards in. They're launching BNPL and so many other things. But what's most exciting to me in this moment right now is the huge quality improvements that we're seeing in the AI native financial products that are out there. Over the holidays, I decided to build a personal financial app. The idea was that my wife and I wanted a tool that we could think about the hardest financial questions that we needed to answer. um the kinds of things that we would normally take to an investment adviser or to an accountant. Um I was able to sit down and actually build the core of the app using cloud code in just an afternoon.

[00:23:35]And now it's a product that we use almost every single day. And apparently I'm not alone. There are more than 4,000 developers that sign up for a cloud API key every single week. And when I talk to them, when I talk to all of you, um you're telling me that you're building things just like what I created. You're creating your own personal financial tools to do the analysis that you always wished you could do very easily. Using generative AI has been a huge game changer. Being able to talk with your finances is pretty amazing. So, let me give you an example. This what's up here. When I graduated from college, I had three different student loans. I had a little bit of credit card debt and almost nothing in my checking account.

[00:24:12]Really, almost nothing, like less than $100. Um, when I finally got my first real paycheck, I had no idea what to do with it. Do I pay down debt? Do I build a rainy day fund? Uh, if I do pay down debt, which loan do I pay first and in what order? Um, these were non-trivial financial questions to try to answer.

[00:24:30]And I knew a lot about financial services, but the questions themselves were quite complex. I ended up creating a spreadsheet. It wasn't exactly this one, but it looked something like this.

[00:24:40]Um, I remember it was like a five or a 10 tab model with a payoff waterfall. It took me almost a week to do. I had to go pull all the data from all these different sources. Um, and I and I knew how to do this stuff. Um, I'd taken a lot of math classes. I thought a lot about financial services. Today, all I would need to do is ask the right question. AI can do the rest. It's kind of like having a top tier financial adviser that knows all of my data in great detail in my pocket at all times.

[00:25:06]Every month, more than 200 million people ask ChatGBT questions about their finances. And just last week, ChatGpt launched a new financial experience powered by Plaid. Users can now link their financial accounts and ask questions that they would previously only have been able to answer with spreadsheets like this or by talking to financial adviserss. I highly recommend that you all use it. Um, give it a try.

[00:25:29]It's really good, especially at the questions that are that are quite hard to answer. All right, so let's take a look at a quick demo here.

[00:25:53]So, this is a real world example of uh someone that works at Plaid. I will not tell you who. Um using chatbt to um ask questions about opening a 529 account for their their their their child. Um, and then thinking about how much they could contribute to the 529 account at different points in time and how how it would affect their lifestyle. Now, these are questions that you could answer with spreadsheets, but it's so much easier to be able to just chat with it very quickly. We actually made this demo, I think, like yesterday. Um, so it's pretty amazing to see these are real time, very very uh very live types of products that you can use. Um, so really proud of what we've been able to launch here. But zooming out a little bit, one way that you could look at AI, it's as a disruptor. Um, when I think about it, I see it as an enabler. Uh, many consumers are using AI to become better informed about their financial lives, to become more involved in their financial lives, and they're ready to act. They're more likely to open a new account when they have more data. They're more likely to apply for a loan or to make a payment.

[00:26:52]And I think that they're better customers for all of you. Just ask a lot of the companies in this room that are leaning heavily into AI. Companies like Perplexity, Rex, Ramp, Copilot, and many others that are here. They're leaning into AI and they're seeing huge gains in their business as a result.

[00:27:10]We're going to talk a lot more about AI today, but I'd like to shift gears before we do that. Let me just give you a quick update on what Plaid strategy, how it's evolved, and the products that we are we've launched in the first half of this year.

[00:27:23]We first shared Plaid with the world in 2014. It had taken us a while to build before, but we launched in 2014. At the time, there was no way to digitally interface with your bank account. And that's where we started. We integrated with 12,000 banks, credit unions, fintech apps, and wallets to make it really simple for consumers to link their digital accounts with the financial products that they wanted to use. Today, Flat enables easy, instant financial connections across pretty much everything that you would want to interact with. Consumers can make an investment with their Coinbase account or their Robin Hood account. They can apply for a loan using SoFi or Rocket.

[00:27:59]They can even buy a car online in just a couple of minutes. Often times sitting on your couch, believe me, I've tried it. I actually bought a car from Carvana online. Sitting on my couch, it took me about 5 minutes. More than half of the people that have bank accounts in the US have used Plaid to connect with their financial apps or to power services um which collectively have generated billions of financial interactions every single year. Now during co something really remarkable happened. Consumers were stuck at home, but they still needed to use financial products. And in that moment, all of you in this room and so far beyond stepped up to deliver the quality of products that consumers really needed. Fintech grew at a very rapid rate. And this industry itself really reach reached mass adoption. As an industry in that moment, I was able to reflect and say that we we basically solved financial access. anyone anywhere that needs access to a financial product was able to get it quite quickly through the products that you all built. But when I talk to consumers these days, they're very happy to have more financial access, but you still hear frustrations about the quality of the underlying products and the quality of the financial system that sits behind it. Some of the fundamentals of our financial system are harder than they need to be. Let's take an example. Let's look at credit scoring. Let's say that you were um let's say you were on the subway and you went to talk to someone about their credit score. Well, I understand they would probably just walk away and roll their eyes and think it's weird because no one talks to people on the subway. But let's say they did talk to you. Um, most people don't understand their credit scores. Or if they do, they'll tell you a story about how their credit score went down because they finally paid off that loan that they've been working so hard to pay off, which seems completely backwards. Consumers don't get it. It doesn't make sense. The fact is consumers don't understand our credit system and lenders themselves don't always get the predictive value that they want from traditional credit scores. So being a company that had a lot of access to data and a lot of access to customers that were giving us really great feedback, we knew that we could build something better. Given the size of our data network, the hundreds of millions of financial accounts that have been linked, the trillions of data points that we see um and all the data that we see we see through user identities, user actions, devices and so much more. We decided to build something bigger. We were thinking in this vein and that led us to a set of products that we call cloud intelligence. The first one we've talked about a little bit, we'll talk about more today. It's called Lensore. It's a consumer-centric credit score that pulls in all of your real world data. We also brought cloud intelligence to our anti-fraud product suite called Protect, which we'll talk a little bit more about very shortly. And of course, we brought intelligence to our payments products. We were able to use intelligence to improve payment certainty and settlement on the other side.

[00:30:39]When I reflect on the state of financial services today, I've never been more optimistic. The quality of AI enabled financial products that all of you are building is incredible. So, I just want to say thank you for your partnership.

[00:30:50]Thank you for the feedback. Thank you for all of you do all you do to serve your customers and to make financial services better for everyone. Okay, let's dive in. Next, I'm going to hand it off to our CTO, Will Robinson.

[00:31:03]>> Please welcome to the stage Plaid's Chief Technology Officer, Will Robinson.

[00:31:17]Hi, thank you very much.

[00:31:20]Over the last decade plus, we that's all of us in this room. We've quietly revolutionized consumer finance. And it's happening again now, but this time is different because AI is reshaping how things go from idea to reality. And it's happening, frankly, faster than anything I've ever seen.

[00:31:40]Over just the last weekend, I was playing around with the Plaid CLI and a coding agent to link a few of my bank accounts, making a simple tool to help teach my 10-year-old son the basics of managing money. It's nothing polished, of course, but what would have taken a team and a week or two just a few years ago, I did in a couple of days. For someone who spent his life building products of all shapes and sizes, this felt different. And it's not because I'm a better engineer than I was. It's because AI is changing how we build.

[00:32:13]Whether you're a solo founder with a coding agent or a Fortune 500 company with a team launching a new product, the speed with which ideas become real has changed permanently. And the financial experiences that your users expect have changed as well. They want things that are more personalized, more responsive to what's actually happening in their lives. We call this the start of intelligent finance. But intelligent finance only works if the infrastructure underneath is built for it. And that's what I want to share with you today. How at Plaid, we're making our infrastructure even better so you can build faster and reach more users.

[00:32:52]Reliability, coverage, and conversion. As this industry innovates faster and faster, these fundamentals just get more and more important. Today, an unstable connection doesn't just mean a frustrated user. Instead, in an AI powered product, that could mean your model losing context entirely, and that separates a product from works, the product that works from one that doesn't.

[00:33:19]Reliability.

[00:33:21]API uptime is not just a metric to us.

[00:33:24]It is a commitment. I'm really proud to say that we've achieved more than four nines of API uptime over the last year, and we're going to keep pushing that number higher.

[00:33:33]But Plaid being up is only part of the picture. The connections to your users specific accounts need to work too.

[00:33:42]That's why we built granular diagnostics allowing you to see what's actually happening with a given bank connection.

[00:33:47]Right there in the Plaid dashboard, you can see live success rates and drill into a breakdown of errors so troubleshooting happens in real time. Of course, the best bank breakage is one that Plaid fixes too quickly for you to even notice. That's why we've also built AI agents that scan bank accounts continuously for breaking changes and autogenerate gross fixes, running in parallel across thousands of bank integrations. Our engineers can then review and apply these fixes more than 20 times faster than before, dramatically improving API reliability for you and your users.

[00:34:23]Coverage. We have industryleading coverage of more than 12,000 financial institutions, including community banks and credit unions. So, wherever your users bank, they're well covered. But to deliver on personalized experiences, you need even more data and a better understanding of your users's holistic financial lives. That's why we've expanded coverage to support over a 100 top requested institutions in just the last 12 months, including RAMP, PenFed Credit Union, Gemini, and Gusto. And we're continuing to expand our coverage to support more data types, including business, identity, investment, and mortgage. Let's take mortgage data as an example.

[00:35:04]We recently quadrupled the number of mortgage accounts that are on the Plaid network. But coverage is still only part of the story. When a user connects their account for a refinancing offer, you need more than a loan balance. You need origination date, escrow balance, next monthly payment. That's why we've increased the fill rate on that sort of data by 20% in just the last 90 days.

[00:35:26]Last but not least, conversion. We've made a bunch of improvements to boost conversion, including a redesigned link experience, phone number prefill, and a new progress bar that guides users through the flow. Frankly, these kinds of upgrades might look small in isolation, but those three together with dozens more have increased conversion by 5% across the network. And at Plaid scale, that means millions more users successfully connecting their accounts to all of your apps. We're not done, though. We've also cut latency significantly. Link is up to five times faster when you preloaded through our SDKs. That's the infrastructure getting better every year. So, your business keeps growing in an AI first world.

[00:36:11]Now, the way you and all developers work has also changed, and Plaid has to keep up. That's why we launched Sandbox Studio so you can configure the perfect test environment for your new ideas without the setup overhead. Sandbox Studio is all about collapsing that multi-tool dance. Postman for flows, dashboard for keys, handedited JSON for test users into a single experience right on the Plaid dashboard. You can spin up custom users via a form or via plain language prompt. Then hit run on any scenario or endpoint to see live responses.

[00:36:44]Now, for developers just starting to build on Plaid, the CLI that I mentioned earlier and our MCP servers are how you work in an agentic environment. Going from an idea to a working prototype shouldn't be the hard part. With our MCP servers, your AI agents can interact directly with Plaid Sandbox environment, generate test users, trigger web hooks, pull sandbox tokens without writing a single line of code.

[00:37:09]And for banks and financial institutions, later this year, we're launching an additional MCP server that lets you manage your Plaid integration using AI tools like Claude and Cursor.

[00:37:19]You'll be able to run automated validations of your API endpoints, debug failing requests by ID, and monitor integration health with LLM suggested code fixes. All of this will be possible without you even needing to open a browser.

[00:37:35]Now, we've talked about how you can build faster on an ever more reliable base, but the real secret sauce is the thing I'm always most excited to talk about, and that is the data foundation we've been building for a decade.

[00:37:52]Again, AI has raised expectations across finance for your users and for your business. Your users want more than a record of just what happened. And you need to make more better decisions faster for them. Faster fraud calls, more accurate credit decisions. To help deliver on that, you need the right data foundation. That's where the Plaid network really comes in. The connections that form that network give Plaid something no other platform has.

[00:38:21]Financial data at scale across more than 12,000 institutions, across millions of users, and over time.

[00:38:30]We're now using this data to train models purpose-built for finance. That is the foundation of the next consumer finance revolution. What you can build on top of it, Sudu is about to show you.

[00:38:41]Thank you very much.

[00:38:43]>> Please welcome to the stage Plaid's head of data and AI, Sudu Sashadri.

[00:39:00]Will is right. What you can build with AI is only as powerful as the data foundation underneath it. But better data alone isn't enough. You need models purpose-built to understand it. Most AI and finance is focused on prediction.

[00:39:18]Every financial product is trying to make a better decision under uncertainty.

[00:39:23]Can this person afford a home? Are they ready to invest? Is this transaction fraudulent? Those are the right questions. But to answer them, you need to first focus on representation.

[00:39:36]Does your model actually understand what it's seeing? Because if it doesn't, the prediction doesn't matter. That's the problem I came to solve at Plaid.

[00:39:45]There's no other place where you can see how accounts, institutions, and devices are connected across the financial network. That data set is one of a kind.

[00:39:55]a data nerd's dream to work with. And now we're using it to build an intelligence layer that can actually make better decisions and drive better outcomes for you and for your customers.

[00:40:07]But before we get to the outcomes, let me show you what's underneath it.

[00:40:13]Most general purpose models pattern match on text. A generic model would label this transaction as a deposit. But is it severance, salary, or a reimbursement?

[00:40:27]A financial model has to know the context because underwriting, fraud, and cash flow decisions all change from there. To get this context, we built a transaction foundation model that was trained on deidentified data across the plat network.

[00:40:44]It starts with a transaction interpreter. We extract entities from raw bank text. When merchant or payment signals are ambiguous, we add more context. Then we generate plain English representations of what a transaction likely means.

[00:41:00]Next, we train the model with contrastive learning. Instead of memorizing labels, the model learns by comparison.

[00:41:10]For each transaction, we generate two positive interpretations and one hard negative. We run those through a pre-trained encoder specifically adapted for financial transactions and pull true economic matches together and push the false ones apart.

[00:41:26]That's how dozens of messy merchant strings resolve to one merchant identity. How a payroll deposit separates from income. how subscriptions get recognized even when naming conventions vary.

[00:41:41]When we apply this model to our existing capabilities, the performance difference was meaningful. Primary categorization is up 13%.

[00:41:51]Loan payment detection is up 14%.

[00:41:56]And we're delivering 89% precision on income classification.

[00:42:04]All of these improve signals to help you understand how your users and customers manage their money.

[00:42:11]To be clear, this isn't building a better classifier for transactions. This is actually a better identification of a financial event.

[00:42:20]But even the most perfect transaction is still a snapshot.

[00:42:25]Because financial life isn't a bag of transactions, it's a sequence. That's where a sequential model comes in. The architecture here is different. Each event first passes through a fusion layer that combines transaction with time, amount, merchant, institution, and account context. Those events then flow through a transformer backbone that can read a financial history like one continuous narrative.

[00:42:55]And the real technical leap here is how we teach the model time.

[00:43:01]Our model encodes cyclical patterns like hour of day, day of week, and payday effects while still learning continuous time deltas. The difference between 5 minutes apart and 5 weeks apart. And we train this pre-trained this with self-supervised objectives like corruption detection and future observation prediction and temporal consistency.

[00:43:27]And before we ever fine-tune it for a product, we train the model to understand the behavior of financial system itself. What belongs in a sequence? What should come next? And when the story stops making sense. In other words, we are teaching the model the grammar of money.

[00:43:47]That is because most important transactions often live between these signals.

[00:43:54]Imagine Alex and Jake, two friends from New York City who both earn $4,000 a month. Alec gets a direct deposit every two weeks, pays rent on the first and keeps his savings cushion.

[00:44:10]He shows responsible financial behavior while Jake has irregular inflows, surprise expenses, frequent overdrafts, a sign of financial strain.

[00:44:22]Same monthly income, completely different temporal structure. This is the difference between a snapshot and a story.

[00:44:31]And when we take those learned sequence representations and apply them in real systems, the impact shows up where it matters. Our early testing of our sequential model shows that we've been able to detect 26% more high-risk AC transfers without increasing how many transactions get flagged.

[00:44:51]That's meaningful reduction in losses at scale. And with credit risk scoring at the same approval rate of 70% we've we've been able to help lenders borrow borrowers approve 13% more in terms of like who are less likely to default.

[00:45:10]That's significantly less loss on your books. Same operating point better decisions. That is the bigger point of all of this. Plaid is not building one-off models. We're building a financial intelligence layer and that intelligence layer is shared across all of our products. So every improvement to our financial model would mean improvement across risk, payments, fraud, and financial management without every team solving the same problem from scratch.

[00:45:44]Here's what this means for you. Smarter fraud detection without new rules.

[00:45:49]Better risk decisions without rebuilding your stack.

[00:45:53]products that just don't react. They understand what's changing.

[00:46:00]16 years ago, I came to the US from Bangalore as a student.

[00:46:06]I did all the right things. Studied hard, saved, paid bills on time. But with no local credit history, progress felt slow. Financial freedom felt far away.

[00:46:20]My story isn't unusual.

[00:46:22]For many out there, there's often a gap between what shows up in their actual financial behavior and what's in a credit file.

[00:46:30]That's because the system was designed to measure history.

[00:46:33]But people can't always wait for that history to build. That's why this work matters to me. We are helping you see who your users actually are and what's happening in their financial lives right now.

[00:46:46]And by doing that, we are designing a more open and fair financial system.

[00:46:54]Finance often looks like a ledger, but it behaves more like a language.

[00:46:59]Transactions have semantics, sequences have grammar. And when you can read both, you can build products that actually understand people and make better decisions for them. That's intelligent finance. And through our effects, you will see where it's already showing up and where it goes next. Thank you.

[00:47:33]Hello everyone. We are so excited that you are here. You are here at the Plaid Shopping Network at Plaid Time TV. My name is Alyssa and I'm here to tell you some exciting things that we have going on. And today with me I have two new friends. This is >> Hi everyone. I'm Benjamin Franklin. You probably know me best from the $100 bill.

[00:47:55]>> And I'm President Abraham Lincoln. I don't know what this is or how I got here.

[00:48:02]>> Well, welcome. Today we're going to talk about the financial future.

[00:48:07]>> Okay.

[00:48:07]>> Okay. I'm still trying to catch up to the financial present, but that sounds interesting.

[00:48:11]>> There's a big chunk of the past I'm figuring out, too.

[00:48:13]>> Yeah, actually. Same.

[00:48:14]>> And you get to be on television. Are you excited about that?

[00:48:18]>> I don't really understand what it is, but >> either >> the moving picture >> doesn't make any more sense when you say it that way.

[00:48:28]>> Yes.

[00:48:29]>> We're just so excited to show you all the different things that we have here at Plaid. There we are. One, two, three.

[00:48:35]Our three products that we're going to take a look at today. I can't wait to to do this with you, too.

[00:48:40]>> It's going to be fun.

[00:48:41]>> Yeah, we're going to have a great time.

[00:48:42]Stay tuned for more.

[00:48:45]>> Yes. and we'll be right back.

[00:48:52]>> Please welcome to the stage product marketing lead at Plaid, Katherine Schugu.

[00:49:07]Fraud never stops. Everyone in this room knows that. But AI has fundamentally changed the game. What's different now is the speed. AI is dramatically accelerating how quickly fraudsters can build, test, and scale attacks. What used to require real operational effort.

[00:49:28]Creating believable identities, generating convincing application data, rotating communication patterns can now be done faster, cheaper, and at a much greater scale.

[00:49:40]We're seeing it play out in real time across the Plaid network and three attack vectors keep coming up again and again.

[00:49:49]Account takeover. Last year, ATO losses exceeded $15 billion, making it the costliest fraud type on record. AI assisted scams, fishing, and social engineering have become increasingly more effective at not just access at stealing not just access credentials but the so-called full logs, full identities and access context, including location data and device snapshots.

[00:50:18]First party fraud is becoming much harder to distinguish from legitimate behavior. It now accounts for 36% of all reported fraud incidents globally, up from 15% last year. Users look trustworthy during onboarding, establish transaction history, build account tenure, and behave normally for weeks or months before engaging in fraudulent activity.

[00:50:44]Synthetic identities with aged accounts now allow fraudsters to hide in plain sight, moving through normal traffic in ways that were impossible just a few years ago. Last year, 73% of financial institutions reported a rise in synthetic identity fraud.

[00:51:02]All of these attack vectors have one thing in common. If you can only see what happens behind your four walls, you will get blindsided.

[00:51:13]Last year at Effects, we made a simple argument. Fraud is a network problem.

[00:51:19]Today, I'll bring that to life with a real attack we saw earlier this year.

[00:51:25]A few months ago, a major payments platform saw their account takeover rate jump from roughly 30 basis points to 80 in just a few days. When they looked more closely, they saw that the invasions were coming from users connecting accounts to one specific bank. In fact, the ATO rate to that bank had hit 7%, 20 times more than their baseline. They had no idea why or what to do without having to fully shut down the connection, impacting thousands of legitimate users.

[00:51:59]When we looked at the sessions inside that spike, the first thing that stood out was geography.

[00:52:06]The IP locations were thousands of miles from the addresses on the bank accounts being connected. Neither the app nor the bank could see that. We could because we're watching both the session side and the account side at the same time.

[00:52:21]When we pulled the graph linking historical identities and bank account connections, we saw that those sessions were part of a larger scheme. They were in much denser components than normal traffic. Shared devices, shared IP addresses, shared identity clusters across dozens of accounts. It was clear this was a coordinated attack built on shared infrastructure.

[00:52:48]After our team investigated the traffic, we designed a rule that could serve as a short-term solution. It flagged the majority of those sessions at a low step-up rate. We stopped the bleeding, but we still did not have a long-term fix. So, we had to go back to work. And the question we kept coming back to was, if seeing across the network let us unravel a ring after it happened, what would it look like to catch it as it's forming?

[00:53:14]A year ago at Effects, we introduced Protect, our fraud solution that scores every user in real time across onboarding, bank linking, and account activity, powered by our model, the trust index. The results have been powerful. On average, Protect could have detected 46% more firstparty fraud and could have prevented 52% of fraud dollar losses.

[00:53:42]The companies that have figured this out are already here. Gemini, Health Equity, Cash App, and many others.

[00:53:50]Let me show you what seeing across the network actually looks like. We can see that this device opened accounts at six different platforms in the last 72 hours. Each of those apps saw a clean new user, but when viewed across the network, the behavioral pattern becomes clear. And that kind of visibility is what makes firstparty fraud at scale much harder to pull off. We can see that the IP address of the session is thousands of miles from the address on every bank account this device has ever connected to across the entire network, not just your app. Atto gets a lot harder when that signal exists.

[00:54:33]We can see that this bank account disconnected immediately after an a transfer was initiated.

[00:54:39]Once that's noise across hundreds of accounts this month, that's a pattern.

[00:54:45]That's the power of seeing across the network.

[00:54:49]Today, we're announcing our latest model powering Protect Trust Index 3. We've pushed the model further. Deeper graph traversal, new data points, and over 3,000 new features specifically designed to close the attack vectors we're seeing across the network. Our latest model can now traverse the live graph further up to nine hops deep in real time.

[00:55:15]That means we can follow fraud across devices, profiles, banks, and identities in a single pass.

[00:55:24]We're also enriching our graph with new data to make TI3 much more powerful.

[00:55:30]First, account age. Synthetic identities, first party fraud, bust out schemes, they all depend on making a new account look established. And fraudsters have gotten good at this. TI3 combines age data from banks, transaction history, and network connection patterns to tell you whether an account is genuinely tenured or days old.

[00:55:55]Second, bank connection velocity across more than 12,000 banks.

[00:56:01]The pattern we see with the rampant increase in firstparty fraud is pretty consistent. Link multiple accounts to similar businesses across the network over a short span of time, exploit them, disconnect, and move on.

[00:56:15]By measuring that connection and disconnection velocity across the whole network, TI3 can identify intent to commute commit abusive behavior before it happens. Lastly, enhanced device identification.

[00:56:30]ATO in particular has gotten more sophisticated as fraudsters rotate devices and spoof fingerprints to stay invisible.

[00:56:38]Enhanced device identification powered by Clad Link and strengthened with secure persistent cookiebased signals is significantly more precise than traditional device fingerprinting. And because it runs inside link, fraudsters can't see it to evade it. Together, these signals power graph features that catch up to 41% more fraud than the previous model at the same false positive rate. This is live for Protect customers today. And if you're not on protect yet, come find us.

[00:57:11]But fraud doesn't stop evolving, and neither do we. Sudu just showed you how we're teaching models to understand financial data. Not just what a transaction looks like, but what it actually means and how behavior evolves over time. I'm excited to give you a sneak peek into what we're working on next, the fraud foundation model. What we're building doesn't wait for labeled data. It trains on the data that shows up before anyone labels it fraud. The warning signs in the sequence that precede the loss. The architectural bet is simple. The relationship between data and prediction that made language models better and better applies to sequential financial behavior too. XG Boost can't do that. But a foundation model built on a decade of financial activity across the plaid network can. It will be ready later this year.

[00:58:07]Here's the thing. We didn't build a fraud product and then go looking for data to power it. For over 10 years, we have been the connective tissue of fintech. The infrastructure that sits between thousands of apps and the banking system. Every bank account linked, every device fingerprint, every time someone authenticated with their bank. Today, nearly a million people connect through Plaid every single day.

[00:58:36]That network powers fraud intelligence no one else can replicate. Not because they haven't tried, because you can't shortcut over a decade of being the infrastructure.

[00:58:49]That atto attack I told you about earlier, the ring was always there. What was missing was a system that could see the whole board. That's the gap plot it's closing. And now we have the data, the graph, and the model to do it. Thank you.

[00:59:14]It was a fluke, honestly. The whole electricity thing, you know. Let's just say I like tying metal keys to stuff.

[00:59:20]Yeah. You know, it worked out for me.

[00:59:23]Got my face on the >> Oh, hello. Benjamin, Abe, don't sneak up on people like that.

[00:59:31]>> Didn't mean to surprise you like that.

[00:59:34]>> So, Abe, what are we talking about next?

[00:59:36]>> Well, of course, we're talking about Plaid Protect. It protects you against fraudsters who are trying to put forth a false identity.

[00:59:44]>> Well, I love that. I hate being tricked.

[00:59:46]>> Yes. And if I'm being honest, Abe, I'm not Abraham Lincoln. It's me, Alyssa.

[00:59:54]And that's plaid protect. It sees patterns that fraudsters can't fake and others can't see.

[00:59:59]>> I certainly love the idea of being protected from liars and huers and fraudsters.

[01:00:04]>> You know, this might be a good moment for you. You know, you got to learn.

[01:00:07]>> Yeah, this is a good moment for me. This is a really good moment for I'm glad we're capturing this.

[01:00:12]>> We got it.

[01:00:13]>> We got it. We get it from every angle.

[01:00:15]>> You can learn people are going to try to fake you out. But with plaid protect, you're totally protected. And at the center of this is the trust index. And the trust index uses thousands of data points to figure out if someone is fraudulent or not. And that's Plaid Protect. We'll be right back to the Plaid Shopping Network. Please tune in for more of me.

[01:00:41]>> You and I need to talk about this. This was unprofessional.

[01:00:45]Please welcome to the stage head of credit go to market at Plaid, Mitch Cook.

[01:00:58]So, I work in lending infrastructure and naturally all my friends and family treat me like I personally approve every loan in America.

[01:01:09]Super popular at parties, trust me. And a few months ago, my neighbor reached out to me about his daughter, Ashlin.

[01:01:17]Ashlin is 19. She started her own aesthetics business at 16 and honestly she is crushing it. Making about 17 grand a month, but Ashlin's never had a credit card, never taken out a loan. So when she went to go finance her first car, she got denied. Not because she couldn't afford it, because she didn't have enough credit history. So, I connected her with a local credit union that uses cash flow data in their underwriting. They had her linker bank account and suddenly they could see the full picture. Consistent income, disciplined spending, strong savings.

[01:02:01]Her ability to repay was obvious.

[01:02:05]And as you can probably guess, she was approved almost instantly. treated as a super prime customer with the best rates.

[01:02:16]And Ashlin is not alone.

[01:02:19]We see this everywhere. People with strong financial footing getting overlooked, or people who hit a rough patch a few years ago, missed a couple of payments but have completely recovered since then.

[01:02:33]Traditional credit models only capture a thin slice of someone's financial life.

[01:02:40]But the reality is people's financial stories are much richer than that. And that's why here at Plaid, we've spent the last several years focused on bringing cash flow data to the forefront of underwriting across the Plaid network. With more than a million financial connections happening every day, we can see patterns in income, spending, and financial behavior long before it appears on a credit report. We work with thousands of lenders like Upstart, Lending Club, and Rocket, powering millions of lending decisions every single day.

[01:03:23]And increasingly, we're seeing demand from capital markets providers who want to bring these same cash flow signals into portfolio and investment decisions.

[01:03:33]So today, I want to talk about how we're bringing intelligence to every stage of lending using cash flow data and AI to help lenders better understand borrower from verification to underwriting to servicing. helping lenders make better decisions while expanding access responsibly.

[01:03:57]So, let's start with income verification.

[01:04:00]Every financial decision, whether it's applying for a loan, expanded credit lines, rentals, new accounts, you need to be able to prove you have a job and you have income.

[01:04:13]And income sounds simple, but historically, it's been really hard to get right.

[01:04:20]Most lenders still rely on payubs and W2s.

[01:04:24]The problem is they're incredibly easy to fake.

[01:04:29]In fact, one in five submitted are actually fraudulent.

[01:04:33]Recently, I was talking to a large auto lender and they found an approved loan application where the pastub literally still had the watermark. This watermark will be removed after purchase.

[01:04:46]That's the baseline we're working with.

[01:04:50]So, we rebuilt Plaid income from the ground up to better reflect how income actually shows up in the real world.

[01:04:59]At the core is the transformer-based LLM that Sudu mentioned that understands the context behind each transaction, grouping deposits into real income streams based on patterns like frequency, amount changes, and earning behavior over time.

[01:05:19]That drove 86% precision for earned income without sacrificing recall.

[01:05:27]Lenders can also apply configurable filters to include or exclude specific income types based on their underwriting criteria.

[01:05:37]So things like recurring transfers from friends or profits from sports betting can automatically be excluded, making decisions more consistent instead of relying on manual judgment.

[01:05:51]Now let's talk underwriting.

[01:05:54]Lens score is the next generation credit risk score built on cash flow and behavioral data. It looks at what's actually happening in someone's financial life. Income, spending, cash flow, stability, and how those patterns change over time. If income goes up while spending stays flat, the score improves.

[01:06:19]No penalties for paying off a mortgage.

[01:06:21]No weird incentives around closing a credit card. It's designed to reflect actual financial behavior.

[01:06:29]But we've also taken it a step further.

[01:06:32]Lens score is the only credit risk score today that combines cash flow data with insights from across the Plaid network.

[01:06:42]Signals traditional credit models simply can't see.

[01:06:47]And this is where it gets really interesting because these network behaviors are highly predictive and the nuance matters.

[01:06:57]For example, connecting to a wealth management app, it shows 20% lower delinquency risk.

[01:07:07]But connecting to 10 or more apps, that can actually signal double the default risk. same category, completely different behavior.

[01:07:19]These are the kinds of signals you just can't get from a traditional credit file or even cash flow data alone.

[01:07:26]And because of that, we're seeing very meaningful improvements in performance.

[01:07:32]On average, Lens Score delivers a 25% lift in predictive performance compared to traditional credit data alone and can reduce risk by up to 41% at the same approval rates. All at a lower cost to lenders.

[01:07:51]But one thing we've learned is that risk is not onesizefits-all.

[01:07:58]Lens Score gives you a very powerful generalpurpose view of credit risk, but different products have very different risk dynamics.

[01:08:09]And one of the clearest examples is cash advance and earned wage access.

[01:08:15]So these are two of the fastest growing categories in consumer finance and also some of the hardest to get right.

[01:08:22]They involve highfrequency decisions, small dollar amounts, and most providers are still making decisions based on a single connected bank account.

[01:08:34]So, we decided to create something purposebuilt for this exact use case, the cash advance index.

[01:08:45]The cash advance index predicts the likelihood of repayment within 30 days, giving providers a real time score they can use to approve, size, and manage advances.

[01:08:59]It brings the same depth of network intelligence into an environment where decisions need to happen instantly and repeatedly across the entire customer life cycle.

[01:09:13]In a randomized AB test with a leading provider, the cash advance index reduced delinquency by eight percentage points with no drop in approval rates. And if you know this space, that is a huge improvement.

[01:09:31]It means you can approve more of the right users, extend the right amounts, and manage risk risk much more precisely.

[01:09:41]Okay, so far we've mostly talked about origination, but the reality is that risk changes over time. Someone loses a job, takes on new expenses, has a major life event, or someone like me that has five hungry kids at home.

[01:10:00]Most lenders still don't have a simple way to monitor those changes in real time. That's where servicing comes in.

[01:10:09]At origination, you're already using Plaid's income and underwriting products to evaluate a borrower.

[01:10:15]Now, you can monitor that same borrower over time through ongoing updates to income, balances, and transaction activity in real time or once a full refresh view across all accounts is available.

[01:10:33]So, take a rent splitting app. After approving a borrower to split rent into installments, the app leverages Plaid to monitor the borrower's income and cash flow.

[01:10:46]That gives lenders the ability to reassess risk before each advance and even align repayment timing to when funds actually arrive.

[01:10:58]So instead of falling behind on rent, renters can align their payments to their cash flow and stay financially on track.

[01:11:09]Coming back to Ashlin, nothing about her financial life changed between getting denied and getting approved. The only thing that changed was the lens we used to evaluate her.

[01:11:24]And that's what makes this so powerful.

[01:11:28]Because cash flow data isn't just about helping thin file borrowers.

[01:11:32]It's about giving lenders a more complete realtime understanding of financial health for every borrower.

[01:11:42]That's what we're building at Plaid, an intelligence layer powered by cash flow and network insights that helps lenders make better decisions across the entire life cycle.

[01:11:57]and we're here to help you see the full picture. Thank you.

[01:12:12]Oh, we're back. Yes. Hello and welcome to the Plaid Shopping Network. I am Melissa.

[01:12:18]>> It's me, Ben. Anast >> and we're here to talk about Plaid Lens score.

[01:12:25]>> Tell us a little bit about Plaid. The people at home want to hear all about Plaid Lens.

[01:12:28]>> Oh, you're going to love it. Plaid Lens Score. A more precise way to make credit decisions.

[01:12:34]>> Now, do you folks know what what credit is?

[01:12:37]>> A notch on a tree every time you get a gallon of milk.

[01:12:40]>> That's how I understand it.

[01:12:42]>> Glad Score is a new way to assess credit risk and offers consumers the chance to share a more complete picture of their financial lives. Credit normally looks back at your past, but Plaid Lend Score likes to look at the future of you as a borrower.

[01:12:58]>> Oh, it's almost like a crystal ball or or like uh minority report.

[01:13:04]>> Maybe using lend score, borrowers get a score from 1 to 99 indicating likelihood to repay a loan. Now, let's think about this. What is your lens score?

[01:13:19]>> 100. Ah, no. It's just from 1 to 99.

[01:13:22]>> Yeah. I'm probably 73.

[01:13:25]>> 73.

[01:13:26]>> Yeah. Sometimes I'm a bit of a scamp.

[01:13:28]>> Oh, >> I wouldn't lend to me.

[01:13:38]>> Please welcome to the stage head of payments at Plaid, Brian Demir.

[01:13:48]Good afternoon. Fintech is all about numbers. So, let's start with a really big one. 93 trillion. 93 trillion flow through a last year in the United States. AC in bank payments more broadly are the backbone of the digital economy and are growing three times faster than credit card payments. These are the transactions that run people's financial lives. from funding investment accounts to repaying loans to paying invoices.

[01:14:19]But here's the reality. AC was simply built for a different era. And the gap between how it works and how consumers expect it to work has only widened over time. We at Plaid are on a mission to apply the power of our network to make bank payments match the expectation of the modern consumer. Our goal is to make bank payments seamless in order to drive the growth of your business.

[01:14:46]Nearly 7,000 companies use Plaid to connect their bank account. That includes leaders in lending, remittance, crypto, banking, gaming, accounting, software, and so much more. And 89% of top fintexs leverage us in some capacity.

[01:15:03]The use cases are different, but the core opportunity remains the same.

[01:15:07]Whether it's investment platforms driving assets under management, buy now pay laterers optimizing repayment, or banks driving privacy, building seamless payment experiences is central to all of their missions.

[01:15:23]In order to do this, there are four things that payment leaders and Plaid cares deeply about. Selection, conversion, risk assessment, and money movement.

[01:15:37]Selection is about making bank payments the preferred choice. Plaid's embedded SDK in link flows are dynamic and adaptive. Whether it's a checkout, a bank linking experience, or a one-off payment, it adapts to optimize for that particular experience.

[01:15:56]Embedded institution search, which you can see here, surfaces the banks that a user is most likely to use using real-time geoloccation data as well as network heristics.

[01:16:08]Additionally, our user experience is optimized to create a trusted familiar experience.

[01:16:13]The aggregate result of all of this work is that Plaid customers see a five times lift in bank payment adoption compared to other solutions.

[01:16:24]After that is conversion. Conversion is about making sure the user can finish what they started. Our off engine seamlessly integrates open banking powered off manual entry and even instant micro deposits into one seamless user journey. Importantly, the user never sees the layers. Whether they're a Zoomer who's perfectly comfortable with an app flow or my grandmother who's going to get a checkbook out of her uh drawer, they see one flow that adapts based on their needs and their preferences.

[01:16:59]And for the 100 million Americans who have a Saved Plat account, we also offer a streamlined user payments experience, account connection in as little as one click.

[01:17:11]Returning users consistently convert 11% higher and the plaid experience has lifted conversion by as much as 54% against the competition.

[01:17:23]Together these flows are the front door of modern bank payments.

[01:17:28]We have powered over 9 billion payment sessions in counting and we are constantly running new experiments focused on selection and conversion.

[01:17:38]So the user is converted but now we have to assess risk in bank payments. There are two types of risk. Settlement risk will the funds actually arrive as well as fraud risk. Is this a legitimate payment or is a bad actor involved? This is where Signal comes in. Signal is our AI plower transaction risk model. It has assessed more than a quarter of a trillion dollars in transactions by evaluating account connection history, identity information, prior AC events, and account behavior to tell you whether or not a payment will succeed.

[01:18:14]Brands like Uphold, a top crypto exchange, have used Signal to reduce return losses by over 80%. And we estimate that Signal has saved the fintech ecosystem more broadly over $135 million in losses in just the last few years alone.

[01:18:32]Finally, most critical to any payment journey is the actual moving moment of money. That's the whole point, right?

[01:18:38]Which is where Plaid Transfer comes in.

[01:18:41]Transfer is one API for every bank rail.

[01:18:45]AC, wires, RTP, RFP, and Fed now all in one solution. Dynamic routing, a built-in ledger, and a featurerich dashboard for full visibility on every single payment. We process billions of dollars in bank payments, and are on a mission to make realtime payments adoption absolutely seamless.

[01:19:08]Selection, conversion, risk, and money movement. the full payment life cycle, each step of which is made better by the Plaid network.

[01:19:18]But if you ask our customers consistently where they still feel pain, the answer is remarkably consistent.

[01:19:25]It's risk.

[01:19:27]Every time you initiate a bank payment, you face a difficult trade-off. AC takes days to settle. Returns can come days after that. But you have to decide right now. fund the account and absorb the loss if the payment fails eventually or hold funds and risk losing a customer who simply won't wait. We all know that the modern consumer expects their funds now. So get it wrong in either direction and you pay for it either in losses or in churn. The teams that are getting this dynamic right are the ones pulling ahead and the ones that aren't are leaving money on the table with every single transaction. And what we hear consistently is that while teams care deeply about this trade-off, not every team wants to become experts in AC risk.

[01:20:14]They simply want bank payments to work.

[01:20:17]So we built a new product for that. We combined the power of our network, unrivaled experience in a risk, and worked with customers who wanted a radically simplified. With this, we build guaranteed payments, the industry-leading risk solution that takes the guesswork out of bank payments. And it is ready today.

[01:20:41]And here's how it works. Plaid guarantees a settlement for every approved transaction. If a payment fails after we've approved it, Plaid covers the loss, not you. It is that simple.

[01:20:54]You process the payment and delight the consumer with real time fulfillment.

[01:20:59]Let's dive a little bit into the consumer journey. Importantly, a guaranteed payment looks no different to the end customer. They go through the same experience that I just showed you.

[01:21:11]In this example, a consumer is funding a break brokerage account. Here it is. The person is already in the Plaid network.

[01:21:19]So, after verifying with an OTP, they confirm with one click. On the back end, guarantee is integrated directly into Plaid Transfer. You make a guarantee request in the transfer authorization endpoint, adding a single field to your implementation.

[01:21:37]Behind the scenes are models, which are powered by signal and protect evaluate the transaction in real time across more than 2,000 attributes.

[01:21:46]Have we seen this user before? Yes. How many times has the account been connected on the Plaid network? Four.

[01:21:53]any past a returns? Yes, there was a single R1, but it was two years ago. And the decision comes back in milliseconds, and in this case, it's been approved.

[01:22:05]And if Plaid declines to guarantee, you can still process the transaction and take on the risk yourself. The user gets immediate access to funds and you get a guaranteed settlement from Plaid. That's it. from an OTP to a guaranteed settlement. Radically simple to implement and invisible to the consumer.

[01:22:26]The end result of this is that teams can focus on growth while more consumers get instant experiences powered by bank payments.

[01:22:36]The results so far are truly remarkable.

[01:22:39]One of our launch partners who did a head-to-head test against their own internal logic managing AC on their own versus guaranteed payments saw a 24% higher approval rate against their baseline logic. This is the power of the Plaid network in action. We've been building guaranteed payments with partners like Kelshi, Poly Market and Solve platforms that needed approval rates at the leading edge of the industry in a partner who could handle their scale. We have seen implementations in as little as two weeks and approval rates as high as 90%.

[01:23:15]When it comes to payment risk, every business has a different appetite for control. Some want to manage risk themselves, working directly with model outputs and signals. Others want to focus on their core product and have experts manage risk on their behalf.

[01:23:31]Plaid now has solutions for both, all built on the same underlying models.

[01:23:36]platform and network intelligence for teams who want risk manage for them.

[01:23:41]The solution is guaranteed payments and I'm proud to say that it's available now. We'd love to help you get started on this. So get in touch with your account team to learn more.

[01:23:53]We're so excited to continue working with all of you on this continued evolution of bank payments. The end result of all of this is higher conversion, fewer losses, instant settlement, and with the power of the Plaid network, bank payments are performing better than ever. Thank you.

[01:24:21]Welcome back to the Plaid Shopping Network. As always, I am Melissa and I'm here with Honest Abe himself and we're here to tell you some more things about our friend Plaid Guaranteed Payments and this is brand new. New is a doorbell.

[01:24:35]>> What? I don't know what a doorbell is.

[01:24:37]>> Get up to 90% payment approval rates without added risk.

[01:24:42]>> 90% without added risk. That's absolutely fabulous.

[01:24:45]>> I'm quite riskaverse myself, you know.

[01:24:47]That's why I spend every night safely seated in a dark theater.

[01:24:52]Great. Plaid guarantee payments is a managed a service that makes bank transfers predictable. If Plaid approves the transaction, the transaction is going through.

[01:25:03]>> Well, that's very exciting. It sounds like confidence is built into every a payment.

[01:25:09]>> And we're going to have more fun when we come back to the Plaid Shopping Network and hear even more about our three amazing products.

[01:25:15]>> Call this number using the telephone.

[01:25:20]Yes.

[01:25:22]>> What is it?

[01:25:30]>> Please welcome back to the stage Plaid co-founder and CEO Zack Pereé.

[01:25:40]Okay, that's it for the intro kickoff. I hope that you had as much fun as I did.

[01:25:45]That was so cool to see all of the things that we've been working on for just a little while. When I look at what's happening in fintech right now, all the products that you saw, all the things that you're building, um I've never been more optimistic nor more excited. Uh I think we're still incredibly early in this phase. Uh AI is continuing to change everything and it's going to be a blast as we all work through it together. Um as for what comes next, please come talk to our teams. We'll be around. There going to be a bunch of sessions that are happening very shortly. Um a lot of people here are working on the same problems or similar problems to all of you. So please meet each other. That's what today is for. Thank you all for being here. And you'll hear a little bit more from us in just a few hours.

[01:26:53]breakout sessions are now beginning. If you signed up for one, please reference your email.