YT YouTube2Text

YouTube2Text → Transcripts → Plaid Effects 2026 | Keynote

Plaid Effects 2026 | Keynote — Transcript

by Plaid · 9,662 words · 1,601 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:24Please welcome to the stage Plaid
0:26co-founder and CEO Zack Pereé.
0:31>> [applause]
0:35>> Hello, welcome to effects. Thank you so
0:38much for joining us. Um, this is by far
0:40the biggest effects that we've ever done
0:42and we have so much that we're going to
0:43share with you all today. Um, for me,
0:46it's really great to be back in New York
0:48City. My co-founder and I actually
0:50founded Plaid here uh in a tiny little
0:53office in Union Square almost 13 years
0:55ago. Uh, and since that time, it's it's
0:58wonderful to reflect on just how far
1:00this amazing industry has come. I'm not
1:02sure about y'all, but I'm having the
1:04most fun of my career right now. Fintech
1:06has truly gone mainstream. Um, fintech
1:09and financial services, the two
1:10industries are continuing to merge and
1:12become just one industry. And now almost
1:15every bank, every lender, every
1:17investment product builder that you talk
1:19to is creating their products digital
1:21first and mobile first. This is kind of
1:23the idea that we had at the very
1:24beginning that's taken us a long time to
1:26get here. We also are starting to see
1:28more and more large tech companies,
1:30retailers, and so many others launch
1:32embedded finance products. They're
1:34putting wallets in, they're putting
1:35rewards in, they're launching BNPL, and
1:37so many other things. But what's most
1:39exciting to me in this moment right now
1:41is the huge quality improvements that
1:43we're seeing in the AI native financial
1:45products that are out there. Over the
1:48holidays, I decided to build a personal
1:50financial app. The idea was that my wife
1:53and I wanted a tool that we could think
1:56about the hardest financial questions
1:57that we needed to answer. Um, the kinds
1:59of things that we would normally take to
2:01an investment adviser or to an
2:03accountant. Um, I was able to sit down
2:05and actually build the core of the app
2:06using Cloud Code in just an afternoon.
2:09And now it's a product that we use
2:10almost every single day. And apparently
2:13I'm not alone. There are more than 4,000
2:15developers that sign up for a cloud API
2:17key every single week. And when I talk
2:19to them, when I talk to all of you, um,
2:22you're telling me that you're building
2:23things just like what I created. You're
2:24creating your own personal financial
2:26tools to do the analysis that you always
2:28wished you could do very easily. Using
2:30generative AI has been a huge game
2:33changer. Being able to talk with your
2:34finances is pretty amazing. So, let me
2:37give you an example. That's what's up
2:38here. When I graduated from college, I
2:41had three different student loans. I had
2:43a little bit of credit card debt and
2:45almost nothing in my checking account.
2:46really almost nothing, like less than
2:48$100. Um, when I finally got my first
2:50real paycheck, I had no idea what to do
2:52with it. Do I pay down debt? Do I build
2:55a rainy day fund? Uh, if I do pay down
2:57debt, which loan do I pay first and in
3:00what order? Um, these were non-trivial
3:02financial questions to try to answer.
3:04And I knew a lot about financial
3:06services, but the questions themselves
3:08were quite complex. I ended up creating
3:10a spreadsheet. It wasn't exactly this
3:12one, but it looked something like this.
3:13Um, I remember it was like a five or a
3:15ten tab model um with a payoff
3:17waterfall. It took me almost a week to
3:19do. I had to go pull all the data from
3:20all these different sources. Um, and I
3:22and I knew how to do this stuff. Um, I'd
3:24taken a lot of math classes. I thought a
3:26lot about financial services. Today, all
3:28I would need to do is ask the right
3:30question. AI can do the rest. It's kind
3:32of like having a top tier financial
3:34adviser that knows all of my data in
3:36great detail in my pocket at all times.
3:40Every month, more than 200 million
3:42people ask ChatGBT questions about their
3:44finances. And just last week, ChatGBT
3:46launched a new financial experience
3:48powered by Plaid. Users can now link
3:51their financial accounts and ask
3:53questions that they would previously
3:54only have been able to answer with
3:56spreadsheets like this or by talking to
3:58financial advisers. I highly recommend
4:01that you all use it. Um, give it a try.
4:03It's really good, especially at the
4:04questions that are that are quite hard
4:06to answer, right? So, let's take a look
4:08at a quick demo here.
4:27So this is a real world example of uh
4:30someone that works at Plaid. I will not
4:31tell you who. Um using chatbt to um ask
4:35questions about opening a 529 account
4:37for their their their their child. Um
4:40and then thinking about how much they
4:41could contribute to the 529 account at
4:43different points in time and how how it
4:45would affect their lifestyle. Now, these
4:47are questions that you could answer with
4:49spreadsheets, but it's so much easier to
4:51be able to just chat with it very
4:52quickly. We actually made this demo, I
4:54think, like yesterday. Um, so it's
4:56pretty amazing to see these are real
4:57time, very very uh very live types of
4:59products that you can use. Um, so really
5:02proud of what we've been able to launch
5:03here. But zooming out a little bit, one
5:06way that you could look at AI, it's as a
5:08disruptor. Um, when I think about it, I
5:10see it as an enabler. Uh, many consumers
5:13are using AI to become better informed
5:15about their financial lives, to become
5:17more involved in their financial lives,
5:18and they're ready to act. They're more
5:21likely to open a new account when they
5:22have more data. They're more likely to
5:24apply for a loan or to make a payment.
5:26And I think that they're better
5:27customers for all of you. Just ask a lot
5:30of the companies in this room that are
5:31leaning heavily into AI. Companies like
5:33Perplexity, Rex, Ramp, Copilot, and many
5:36others that are here. They're leaning
5:38into AI, and they're seeing huge gains
5:40in their business as a result.
5:44We're going to talk a lot more about AI
5:46today, but I'd like to shift gears
5:47before we do that. Let me just give you
5:49a quick update on what Plaid strategy,
5:51how it's evolved, and the products that
5:53we are we've launched in the first half
5:55of this year.
5:57We first shared Plaid with the world in
5:592014. It had taken us a while to build
6:01before, but we launched in 2014. At the
6:03time, there was no way to digitally
6:05interface with your bank account. And
6:07that's where we started. We integrated
6:09with 12,000 banks, credit unions,
6:11fintech apps, and wallets to make it
6:13really simple for consumers to link
6:15their digital accounts with the
6:16financial products that they wanted to
6:18use. Today, Flat enables easy, instant
6:21financial connections across pretty much
6:23everything that you would want to
6:24interact with. Consumers can make an
6:27investment with their Coinbase account
6:29or their Robin Hood account. They can
6:30apply for a loan using SoFi or Rocket.
6:33They can even buy a car online in just a
6:35couple of minutes. Oftentimes sitting on
6:37your couch. Believe me, I've tried it. I
6:39actually bought a car from Carvana
6:40online. Sitting on my couch, it took me
6:42about five minutes. More than half of
6:44the people that have bank accounts in
6:45the US have used Plaid to connect with
6:47their financial apps or to power
6:49services um which collectively have
6:51generated billions of financial
6:53interactions every single year. Now
6:56during co something really remarkable
6:58happened. Consumers were stuck at home,
7:00but they still needed to use financial
7:02products. And in that moment, all of you
7:04in this room and so far beyond stepped
7:07up to deliver the quality of products
7:08that consumers really needed. Fintech
7:10grew at a very rapid rate. And this
7:13industry itself really reached mass
7:15adoption. As an industry in that moment,
7:18I was able to reflect and say that we we
7:20basically solved financial access.
7:22anyone anywhere that needs access to a
7:25financial product was able to get it
7:26quite quickly through the products that
7:27you all built. But when I talk to
7:30consumers these days, they're very happy
7:32to have more financial access, but you
7:34still hear frustrations about the
7:36quality of the underlying products and
7:37the quality of the financial system that
7:38sits behind it. Some of the fundamentals
7:41of our financial system are harder than
7:42they need to be. Let's take an example.
7:45Let's look at credit scoring. Let's say
7:47that you were um let's say you were on
7:48the subway and you went to talk to
7:50someone about their credit score. Well,
7:52I understand they would probably just
7:53walk away and roll their eyes and think
7:55it's weird because no one talks to
7:56people on the subway, but let's say they
7:58did talk to you. Um, most people don't
8:00understand their credit scores. Or if
8:02they do, they'll tell you a story about
8:04how the credit score went down because
8:06they finally paid off that loan that
8:08they've been working so hard to pay off,
8:09which seems completely backwards.
8:11Consumers don't get it. It doesn't make
8:12sense. The fact is consumers don't
8:15understand our credit system and lenders
8:17themselves don't always get the
8:18predictive value that they want from
8:20traditional credit scores. So being a
8:22company that had a lot of access to data
8:24and a lot of access to customers that
8:26were giving us really great feedback, we
8:27knew that we could build something
8:29better. Given the size of our data
8:31network, the hundreds of millions of
8:33financial accounts that have been
8:34linked, the trillions of data points
8:36that we see um and all the data that we
8:38see we see through user identities, user
8:40actions, devices and so much more. We
8:43decided to build something bigger. We
8:45were thinking in this vein and that led
8:48us to a set of products that we call
8:49plaid intelligence. The first one we've
8:51talked about a little bit, we'll talk
8:52about more today. It's called Lens
8:54Score. It's a consumercentric credit
8:56score that pulls in all of your real
8:57world data. We also brought plaid
8:59intelligence to our anti-fraud product
9:01suite called protect, which we'll talk a
9:03little bit more about very shortly. And
9:05of course, we brought intelligence to
9:06our payments products. We were able to
9:08use intelligence to improve payment
9:09certainty and settlement. On the other
9:11side,
9:13when I reflect on the state of financial
9:14services today, I've never been more
9:16optimistic. The quality of AI enabled
9:19financial products that all of you are
9:20building is incredible. So, I just want
9:22to say thank you for your partnership.
9:24Thank you for the feedback. Thank you
9:26for all of you do all you do to serve
9:28your customers and to make financial
9:29services better for everyone. Okay,
9:32let's dive in. Next, I'm going to hand
9:34it off to our CTO, Will Robinson. Please
9:37welcome to the stage Plaid's Chief
9:39Technology Officer, Will Robinson.
9:47>> Hi, thank you very much.
9:50Over the last decade plus, we that's all
9:53of us in this room. We've quietly
9:55revolutionized consumer finance and it's
9:58happening again now, but this time is
10:00different because AI is reshaping how
10:03things go from idea to reality. And it's
10:06happening, frankly, faster than anything
10:08I've ever seen.
10:10Over just the last weekend, I was
10:12playing around with the Plaid CLI and a
10:14coding agent to link a few of my bank
10:16accounts, making a simple tool to help
10:18teach my 10-year-old son the basics of
10:20managing money. It's nothing polished,
10:22of course, but what would have taken a
10:25team and a week or two just a few years
10:26ago, I did in a couple of days. For
10:30someone who spent his life building
10:32products of all shapes and sizes, this
10:34felt different. And it's not because I'm
10:37a better engineer than I was. It's
10:39because AI is changing how we build.
10:43Whether you're a solo founder with a
10:45coding agent or a Fortune 500 company
10:48with a team launching a new product, the
10:50speed with which ideas become real has
10:52changed permanently. And the financial
10:55experiences that your users expect have
10:57changed as well. They want things that
10:59are more personalized, more responsive
11:01to what's actually happening in their
11:02lives. We call this the start of
11:05intelligent finance. But intelligent
11:08finance only works if the infrastructure
11:10underneath is built for it. And that's
11:12what I want to share with you today. How
11:14at Plaid, we're making our
11:16infrastructure even better so you can
11:18build faster and reach more users.
11:22Reliability,
11:24coverage, and conversion. As this
11:27industry innovates faster and faster,
11:29these fundamentals just get more and
11:32more important. Today, an unstable
11:34connection doesn't just mean a
11:36frustrated user. Instead, in an AI
11:39powered product, that can mean your
11:41model losing context entirely. And that
11:43separates a product from works, a
11:45product that works from one that
11:47doesn't.
11:49Reliability.
11:51API uptime is not just a metric to us.
11:54It is a commitment. I'm really proud to
11:56say that we've achieved more than four
11:58nines of API uptime over the last year,
12:00and we're going to keep pushing that
12:01number higher.
12:03But Plaid being up is only part of the
12:06picture. The connections to your users
12:09specific accounts need to work too.
12:11That's why we build granular diagnostics
12:14allowing you to see what's actually
12:15happening with a given bank connection.
12:17Right there in the Plaid dashboard, you
12:18can see live success rates and drill
12:20into a breakdown of errors so
12:22troubleshooting happens in real time. Of
12:25course, the best bank breakage is one
12:28that Plaid fixes too quickly for you to
12:30even notice. That's why we've also built
12:32AI agents that scan bank accounts
12:34continuously for breaking changes and
12:37autogenerate gross fixes running in
12:39parallel across thousands of bank
12:41integrations. Our engineers can then
12:44review and apply these fixes more than
12:4620 times faster than before,
12:48dramatically improving API reliability
12:50for you and your users.
12:53Coverage. We have industryleading
12:55coverage of more than 12,000 financial
12:57institutions, including community banks
12:59and credit unions. So, wherever your
13:01users bank, they're well covered. But to
13:04deliver on personalized experiences, you
13:07need even more data and a better
13:08understanding of your users's holistic
13:10financial lives. That's why we've
13:12expanded coverage to support over a 100
13:14top requested institutions in just the
13:16last 12 months, including RAMP, PennFed
13:19Credit Union, Gemini, and Gusto. And
13:22we're continuing to expand our coverage
13:24to support more data types, including
13:26business, identity, investment, and
13:29mortgage. Let's take mortgage data as an
13:32example.
13:33We recently quadrupled the number of
13:36mortgage accounts that are on the Plaid
13:37network. But coverage is still only part
13:39of the story. When a user connects their
13:42account for a refinancing offer, you
13:44need more than a loan balance. You need
13:46origination date, escrow balance, next
13:48monthly payment. That's why we've
13:50increased the fill rate on that sort of
13:51data by 20% in just the last 90 days.
13:56Last but not least, conversion. We've
13:59made a bunch of improvements to boost
14:01conversion, including a redesigned link
14:02experience, phone number prefill, and a
14:06new progress bar that guides users
14:07through the flow. Frankly, these kinds
14:10of upgrades might look small in
14:12isolation, but those three together with
14:14dozens more have increased conversion by
14:165% across the network. And at Plaid
14:19scale, that means millions more users
14:21successfully connecting their accounts
14:23to all of your apps. We're not done,
14:26though. We've also cut latency
14:28significantly. Link is up to five times
14:30faster when you preload it through our
14:32SDKs.
14:33That's the infrastructure getting better
14:35every year. So, your business keeps
14:37growing in an AI first world.
14:40Now, the way you and all developers work
14:43has also changed and plaid has to keep
14:45up. That's why we launched Sandbox
14:47Studio. So you can configure the perfect
14:49test environment for your new ideas
14:51without the setup overhead. Sandbox
14:53Studio is all about collapsing that
14:54multi-tool dance postman for flows,
14:57dashboard for keys, handedited JSON for
15:00test users into a single experience
15:02right on the plaid dashboard. You can
15:04spin up custom users via a form or via
15:07plain language prompt. Then hit run on
15:09any scenario or endpoint to see live
15:11responses.
15:13Now, for developers just starting to
15:15build on Plaid, the CLI that I mentioned
15:17earlier and our MCP servers are how you
15:20work in an agentic environment. Going
15:22from an idea to a working prototype
15:24shouldn't be the hard part. With our MCP
15:27servers, your AI agents can interact
15:29directly with Plaid Sandbox environment,
15:31generate test users, trigger web hooks,
15:33pull sandbox tokens without writing a
15:36single line of code.
15:39And for banks and financial
15:40institutions, later this year, we're
15:42launching an additional MCP server that
15:44lets you manage your Plaid integration
15:46using AI tools like Claude and Cursor.
15:49You'll be able to run automated
15:51validations of your API endpoints, debug
15:54failing requests by ID, and monitor
15:56integration health with LLM suggested
15:58code fixes. All of this will be possible
16:01without you even needing to open a
16:03browser.
16:05Now, we've talked about how you can
16:07build faster on an ever more reliable
16:09base, but the real secret sauce is the
16:13thing I'm always most excited to talk
16:15about, and that is the data foundation
16:18we've been building for a decade.
16:21Again, AI has raised expectations across
16:25finance for your users and for your
16:27business. Your users want more than a
16:30record of just what happened. And you
16:32need to make more better decisions
16:34faster for them. Faster fraud calls,
16:37more accurate credit decisions. To help
16:39deliver on that, you need the right data
16:42foundation. That's where the Plaid
16:44network really comes in. The connections
16:46that form that network give Plaid
16:48something no other platform has.
16:51Financial data at scale across more than
16:5312,000 institutions, across millions of
16:56users, and over time.
17:00We're now using this data to train
17:01models purpose-built for finance. That
17:04is the foundation of the next consumer
17:06finance revolution. What you can build
17:09on top of it, Sudu is about to show you.
17:11Thank you very much.
17:13>> Please welcome to the stage Plaid's head
17:15of data and AI, Sudu Sashadri.
17:29Will is right. What you can build with
17:32AI is only as powerful as the data
17:35foundation underneath it. But better
17:38data alone isn't enough. You need models
17:42purpose-built to understand it. Most AI
17:45and finance is focused on prediction.
17:48Every financial product is trying to
17:49make a better decision under
17:51uncertainty.
17:53Can this person afford a home? Are they
17:55ready to invest? Is this transaction
17:58fraudulent? Those are the right
18:00questions. But to answer them, you need
18:02to first focus on representation.
18:06Does your model actually understand what
18:07it's seeing? Because if it doesn't, the
18:10prediction doesn't matter. That's the
18:12problem I came to solve at Plaque.
18:14There's no other place where you can see
18:17how accounts, institutions, and devices
18:19are connected across the financial
18:21network. That data set is one of a kind.
18:25a data nerd's dream to work with. And
18:28now we're using it to build an
18:30intelligence layer that can actually
18:32make better decisions and drive better
18:34outcomes for you and for your customers.
18:37But before we get to the outcomes, let
18:39me show you what's underneath it.
18:42Most general purpose models pattern
18:45match on text. A generic model would
18:48label this transaction as a deposit. But
18:51is it severance, salary, or a
18:55reimbursement?
18:57A financial model has to know the
18:59context because underwriting, fraud, and
19:02cash flow decisions all changed from
19:04there. To get this context, we built a
19:07transaction foundation model that was
19:09trained on deidentified data across the
19:11plat network.
19:14It starts with a transaction
19:15interpreter. We extract entities from
19:17raw bank text. When merchant of payment
19:20signals are ambiguous, we add more
19:22context.
19:24Then we generate plain English
19:26representations of what a transaction
19:28likely means.
19:30Next, we train the model with
19:32contrastive learning. Instead of
19:34memorizing labels, the model learns by
19:37comparison.
19:39For each transaction, we generate two
19:42positive interpretations and one hard
19:44negative. We run those through a
19:46pre-trained encoder specifically adapted
19:48for financial transactions and pull true
19:51economic matches together and push the
19:53false ones apart.
19:56That's how dozens of messy merchant
19:57strings resolve to one merchant
19:59identity. How a payroll deposit
20:02separates from income. How subscriptions
20:06get recognized even when naming
20:08conventions vary.
20:11When we apply this model to our existing
20:13capabilities, the performance difference
20:15was meaningful. Primary categorization
20:18is up 13%.
20:21Loan payment detection is up 14%.
20:26And we're delivering 89% precision on
20:31income classification.
20:34All of these improve signals to help you
20:36understand how your users and customers
20:39manage their money.
20:41To be clear, this isn't building a
20:43better classifier for transactions. This
20:46is actually a better identification of a
20:48financial event.
20:50But even the most perfect transaction is
20:53still a snapshot.
20:55Because financial life isn't a bag of
20:58transactions, it's a sequence. That's
21:01where our sequential model comes in. The
21:04architecture here is different. Each
21:06event first passes through a fusion
21:08layer that combines transaction
21:11with time, amount, merchant,
21:14institution, and account context. Those
21:18events then flow through a transformer
21:20backbone that can read a financial
21:22history like one continuous narrative.
21:25And the real technical leap here is how
21:27we teach the model time. Our model
21:31encodes cyclical patterns like hour of
21:34day, day of week, and payday effects
21:38while still learning continuous time
21:40deltas. The difference between 5 minutes
21:43apart and 5 weeks apart. And we train
21:48this pre-trained this with
21:49self-supervised objectives like
21:51corruption detection and future
21:53observation prediction and temporal
21:55consistency.
21:57And before we ever fine-tune it for a
21:59product, we train the model to
22:01understand the behavior of financial
22:03system itself, what belongs in a
22:06sequence, what should come next, and
22:09when the story stops making sense. In
22:12other words, we are teaching the model
22:14the grammar of money.
22:17That is because most important
22:19transactions often live between these
22:22signals.
22:24Imagine Alex and Jake.
22:27two friends from New York City who both
22:29earn $4,000 a month. Alec gets a direct
22:32direct deposit every two weeks, pays
22:35rent on the first and keeps his savings
22:38cushion.
22:39He shows responsible financial behavior.
22:43While Jake has irregular inflows,
22:46surprise expenses, frequent overdraws, a
22:49sign of financial strain.
22:52Same monthly income, completely
22:54different temporal structure.
22:57This is the difference between a
22:58snapshot and a story.
23:00And when we take those learned sequence
23:02representations and apply them in real
23:05systems, the impact shows up where it
23:07matters.
23:09Our early testing of a sequential model
23:11shows that we've been able to detect 26%
23:14more high-risk AC transfers without
23:17increasing how many transactions get
23:19flagged.
23:20That's meaningful reduction in losses at
23:23scale. And with credit risk scoring at
23:27the same approval rate of 70%, we've
23:30we've been able to help lenders borrow
23:33borrowers approve 13% more in terms of
23:37like who are less likely to default.
23:40That's significantly less loss on your
23:42books. Same operating point, better
23:46decisions. That is the bigger point of
23:48all of this. Plaid is not building
23:51one-off models. We are building a
23:54financial intelligence layer. And that
23:57intelligence layer is shared across all
23:58of our products. So every improvement to
24:02a financial model would mean improvement
24:04across risk, payments, fraud, and
24:07financial management without every team
24:10solving the same problem from scratch.
24:13Here's what this means for you. Smarter
24:16fraud detection without new rules.
24:19better risk decisions without rebuilding
24:21your stack. Products that just don't
24:24react. They understand what's changing.
24:3016 years ago, I came to the US from
24:32Bangalore as a student.
24:35I did all the right things. Studied
24:38hard, saved, paid bills on time. But
24:42with no local credit history, progress
24:44felt slow. Financial freedom felt far
24:47away.
24:49My story isn't unusual. For many out
24:52there, there's often a gap between what
24:55shows up in their actual financial
24:56behavior and what's in a credit file.
24:59That's because the system was designed
25:01to measure history.
25:03But people can't always wait for that
25:05history to build. That's why this work
25:07matters to me. We are helping you see
25:10who your users actually are and what's
25:12happening in their financial lives right
25:14now.
25:16And by doing that, we designing a more
25:18open and fair financial system.
25:23Finance often looks like a ledger, but
25:26it behaves more like a language.
25:28Transactions have semantics, sequences
25:31have grammar, and when you can read
25:34both, you can build products that
25:36actually understand people and make
25:38better decisions for them. That's
25:41intelligent finance.
25:43And throughout effects, you will see
25:45where it's already showing up and where
25:47it goes next. Thank you.
25:51[applause]
25:58[music]
26:00Hello everyone. We are so excited that
26:03you are here. You are here at the Plaid
26:05Shopping Network at Plaid Time TV. My
26:08name is Alyssa and I'm here to tell you
26:10some exciting things that we have going
26:12on. And today with me I have two new
26:14friends. This is
26:17>> Hi everyone. I'm Benjamin Franklin. You
26:19probably know me best from the $100
26:21bill.
26:22>> And I'm President Abraham Lincoln. I
26:26don't know what this is or how I got
26:28here. [laughter]
26:29>> Well, welcome. Today we're going to talk
26:31about the financial future.
26:34>> Okay.
26:34>> Okay. I'm still trying to catch up to
26:36the financial present, but that sounds
26:38interesting.
26:38>> There's a big chunk of the past I'm
26:39figuring out, too.
26:40>> Yeah, actually. Same.
26:41>> And you get to be on television. Are you
26:43excited about that?
26:45>> I don't really understand what it is,
26:46but
26:48>> the moving picture
26:50>> doesn't make any more sense when you say
26:51it that way.
26:52>> Kind of like [laughter] a
26:56>> Yes. We're just so excited to show you
26:58all the different things that we have
26:59here at Plaid. There we are. One, two,
27:01three. Our three products that we're
27:03going to take a look at today. I can't
27:05wait to to do this with you, too.
27:07>> It's going to be fun.
27:08>> Yeah, we're going to have a great time.
27:09>> Stay tuned for more.
27:12>> Yes. And we'll be right back.
27:19>> Welcome to the stage product marketing
27:21lead at Plaid, Katherine Schuger.
27:26[music]
27:34Fraud never stops. Everyone in this room
27:37knows that. But AI has fundamentally
27:41changed the game. What's different now
27:43is the speed. AI is dramatically
27:46accelerating how quickly fraudsters can
27:49build, test, and scale attacks. What
27:52used to require real operational effort,
27:56creating believable identities,
27:57generating convincing application data,
28:00rotating communication patterns, can now
28:02be done faster, cheaper, and at a much
28:05greater scale. We're seeing it play out
28:08in real time across the Plaid network.
28:11And three attack vectors keep coming up
28:14again and again.
28:17Account takeover. Last year, ATO losses
28:20exceeded $15 billion,
28:23making it the costliest fraud type on
28:25record. AI assisted scams, fishing, and
28:30social engineering have become
28:31increasingly more effective at not just
28:35access at stealing not just access
28:37credentials, but the so-called full
28:39logs, full identities and access
28:41contexts, including location data and
28:43device snapshots.
28:45First party fraud is becoming much
28:48harder to distinguish from legitimate
28:50behavior. It now accounts for 36% of all
28:54reported fraud incidents globally, up
28:57from 15% last year. Users look
29:00trustworthy during onboarding, establish
29:02transaction history, build account
29:04tenure, and behave normally for weeks or
29:07months before engaging in fraudulent
29:09activity.
29:11Synthetic identities with aged accounts
29:14now allow fraudsters to hide in plain
29:16sight, moving through normal traffic in
29:18ways that were impossible just a few
29:20years ago. Last year, 73% of financial
29:24institutions reported a rise in
29:27synthetic identity fraud. All of these
29:30attack vectors have one thing in common.
29:33If you can only see what happens behind
29:36your four walls, you will get
29:38blindsided.
29:40Last year at Effects, we made a simple
29:42argument. Fraud is a network problem.
29:46Today, I'll bring that to life with a
29:49real attack we saw earlier this year.
29:52A few months ago, a major payments
29:55platform saw their account takeover rate
29:57jump from roughly 30 basis points to 80
30:00in just a few days. When they looked
30:03more closely, they saw that the
30:04invasions were coming from users
30:07connecting accounts to one specific
30:09bank. In fact, the ATO rate to that bank
30:13had hit 7%, 20 times more than their
30:16baseline. They had no idea why or what
30:20to do without having to fully shut down
30:22the connection, impacting thousands of
30:25legitimate users.
30:27When we looked at the sessions inside
30:29that spike, the first thing that stood
30:31out was geography.
30:33The IP locations were thousands of miles
30:36from the addresses on the bank accounts
30:38being connected. Neither the app nor the
30:41bank could see that. We could because
30:44we're watching both the session side and
30:46the account side at the same time.
30:48When we pulled the graph linking
30:50historical identities and bank account
30:52connections, we saw that those sessions
30:55were part of a larger scheme. They were
30:58in much denser components than normal
31:01traffic, shared devices, shared IP
31:04addresses, shared identity clusters
31:07across dozens of accounts. It was clear
31:10this was a coordinated attack built on
31:13shared infrastructure.
31:15After our team investigated the traffic,
31:18we designed a rule that could serve as a
31:20short-term solution. It flagged the
31:22majority of those sessions at a low
31:24stepup rate. We stopped the bleeding,
31:26but we still did not have a long-term
31:28fix. So, we had to go back to work. And
31:30the question we kept coming back to was,
31:33if seeing across the network let us
31:36unravel a ring after it happened, what
31:38would it look like to catch it as it's
31:40forming?
31:42A year ago at Effects, we introduced
31:44Protect, our fraud solution that scores
31:47every user in real time across
31:49onboarding, bank linking, and account
31:52activity, powered by our model, the
31:55Trust Index. The results have been
31:57powerful. On average, Protect could have
32:00detected 46%
32:02more firstparty fraud and could have
32:05prevented 52% of fraud dollar losses.
32:09The companies that have figured this out
32:11are already here. Gemini, Health Equity,
32:15Cash App, and many others.
32:18Let me show you what seeing across the
32:20network actually looks like.
32:23We can see that this device opened
32:25accounts at six different platforms in
32:27the last 72 hours. Each of those apps
32:30saw a clean new user, but when viewed
32:33across the network, the behavioral
32:35pattern becomes clear. And that kind of
32:38visibility is what makes firstparty
32:40fraud at scale much harder to pull off.
32:43We can see that the IP address of the
32:45session is thousands of miles from the
32:48address on every bank account this
32:51device has ever connected to across the
32:53entire network, not just your app. ATO
32:56gets a lot harder when that signal
32:59exists. We can see that this bank
33:01account disconnected immediately after
33:04an a transfer was initiated.
33:07Once that's noise across hundreds of
33:09accounts this month, that's a pattern.
33:12That's the power of seeing across the
33:14network.
33:16Today, we're announcing our latest model
33:19powering Protect Trust Index 3. We've
33:22pushed the model further. Deeper graph
33:25traversal, new data points, and over
33:283,000 new features specifically designed
33:31to close the attack vectors we're seeing
33:33across the network. Our latest model can
33:37now traverse the live graph further up
33:39to nine hops deep in real time.
33:42That means we can follow fraud across
33:45devices, profiles, banks, and identities
33:49in a single pass.
33:51We're also enriching our graph with new
33:53data to make TI3 much more powerful.
33:57First, account age. Synthetic
34:01identities, first party fraud, bust out
34:03schemes, they all depend on making a new
34:06account look established. And fraudsters
34:08have gotten good at this. TI3 combines
34:12age data from banks, transaction
34:14history, and network connection patterns
34:17to tell you whether an account is
34:19genuinely tenured or days old.
34:22Second, bank connection velocity across
34:25more than 12,000 banks.
34:28The pattern we see with the rampid
34:30increase in first party fraud is pretty
34:32consistent. Link multiple accounts to
34:35similar businesses across the network
34:36over a short span of time, exploit them,
34:39disconnect, and move on.
34:42By measuring that connection and
34:44disconnection velocity across the whole
34:46network, TI3 can identify intent to
34:49commute commit abusive behavior before
34:52it happens.
34:54Lastly, enhanced device identification.
34:57ATO in particular has gotten more
34:59sophisticated as fraudsters rotate
35:02devices and spoof fingerprints to stay
35:04invisible.
35:05Enhanced device identification powered
35:08by Plaid Link and strengthened with
35:10secure persistent cookie based signals
35:13is significantly more precise than
35:15traditional device fingerprinting. And
35:18because it runs inside link, fraudsters
35:20can't see it to evade it. Together,
35:23these signals power graph features that
35:26catch up to 41% more fraud than the
35:28previous model at the same false
35:30positive rate. This is live for Protect
35:33customers today. And if you're not on
35:35protect yet, come find us.
35:39But fraud doesn't stop evolving, and
35:40neither do we. Sudo just showed you how
35:43we're teaching models to understand
35:45financial data. Not just what a
35:47transaction looks like, but what it
35:49actually means and how behavior evolves
35:52over time. I'm excited to give you a
35:55sneak peek into what we're working on
35:57next, the fraud foundation model. What
36:00we're building doesn't wait for labeled
36:02data. It trains on the data that shows
36:05up before anyone labels it fraud. The
36:07warning signs in the sequence that
36:09precede the loss. The architectural bet
36:12is simple. The relationship between data
36:16and prediction that made language models
36:18better and better applies to sequential
36:21financial behavior too. XG Boost can't
36:24do that. But a foundation model built on
36:27a decade of financial activity across
36:29the plaid network can. It will be ready
36:32later this year.
36:34Here's the thing. We didn't build a
36:37fraud product and then go looking for
36:39data to power it. For over 10 years, we
36:42have been the connective tissue of fint,
36:45the infrastructure that sits between
36:47thousands of apps and the banking
36:49system. Every bank account linked, every
36:52device fingerprint, every time someone
36:55authenticated with their bank. Today,
36:58nearly a million people connect through
37:00Plaid every single day.
37:03That network powers fraud intelligence
37:06no one else can replicate. Not because
37:09they haven't tried, because you can't
37:11shortcut over a decade of being the
37:14infrastructure.
37:16That atto attack I told you about
37:18earlier, the ring was always there. What
37:21was missing was a system that could see
37:23the whole board. That's the gap plat is
37:26closing and now we have the data, the
37:29graph, and the model to do it. Thank
37:31you.
37:35[music]
37:36>> It was a fluke, honestly, the whole
37:38electricity thing, you know. Let's just
37:40say I like tying metal keys to stuff.
37:42Yeah. You know, it worked out for me.
37:45[snorts] Got my face on the $100.
37:47[laughter]
37:48>> OH, HELLO THERE, BENJAMIN.
37:51>> SHAME. DON'T SNEAK UP ON PEOPLE like
37:52that. didn't mean to surprise you like
37:54that. [laughter]
37:56>> So, Abe, what are we talking about next?
37:58>> Well, of course, we're talking about
37:59Plaid Protect. It protects you against
38:02fraudsters who are trying to put forth a
38:04false identity.
38:06>> Well, I love that. I hate being tricked.
38:08>> Yes. And if I'm being honest, Abe,
38:12I'm not Abraham Lincoln. It's me,
38:14Alyssa. [laughter]
38:16And that's Plaid Protect. It sees
38:18patterns that fraudsters can't fake and
38:20others can't see. I certainly love the
38:22idea of being protected from liars and
38:24huers and fraudsters.
38:25>> You know, this might be a good moment
38:27for you. You know, you got to learn.
38:29>> Yeah, this is a good moment for me. This
38:30is a really good moment for I'm glad
38:32we're capturing this.
38:34>> We got it.
38:34>> We got it. We get it from every angle.
38:37>> You can learn people are going to try to
38:38fake you out. But with Plaid Protect,
38:40you're totally protected. And at the
38:43center of this is the trust index. And
38:46the trust index uses thousands of data
38:48points to figure out if someone is
38:50fraudulent or not. And that's Plaid
38:53Protect. We'll be right back to the
38:55Plaid Shopping Network. Please tune in
38:58for more of me.
39:01[laughter]
39:03>> You and I need to talk about this. This
39:05was unprofessional.
39:07Please welcome to the stage head of
39:09credit go to market at Plaid, Mitch
39:11Cook.
39:15>> [music]
39:20>> So, I work in lending infrastructure and
39:24naturally all my friends and family
39:26treat me like I personally approve every
39:29loan in America.
39:31Super popular at parties, trust me. And
39:34a few months ago, my neighbor reached
39:36out to me about his daughter, Ashlin.
39:38Ashlin is 19. She started her own
39:41aesthetics business at 16. And honestly,
39:45she is crushing it, making about 17
39:49grand a month, but Ashlin's never had a
39:52credit card, never taken out a loan. So,
39:56when she went to go finance her first
39:58car, she got denied. Not because she
40:01couldn't afford it, because she didn't
40:03have enough credit history. So, I
40:06connected her with a local credit union
40:08that uses cash flow data in their
40:10underwriting. They had her linker bank
40:12account and suddenly they could see the
40:16full picture. Consistent income,
40:20disciplined spending, strong savings.
40:23Her ability to repay was obvious.
40:27And as you can probably guess, she was
40:30approved almost instantly. treated as a
40:34superp prime customer with the best
40:36rates.
40:38And Ashlin is not alone.
40:41We see this everywhere. People with
40:44strong financial footing getting
40:46overlooked,
40:48or people who hit a rough patch a few
40:50years ago, missed a couple of payments
40:52but have completely recovered since
40:54then.
40:55Traditional credit models only capture a
40:58thin slice of someone's financial life.
41:02But the reality is people's financial
41:04stories are much richer than that. And
41:08that's why here at Plaid, we've spent
41:10the last several years focused on
41:12bringing cash flow data to the forefront
41:15of underwriting
41:18across the Plaid network. With more than
41:20a million financial connections
41:22happening every day, we can see patterns
41:25in income, spending, and financial
41:29behavior long before it appears on a
41:32credit report. We work with thousands of
41:36lenders like Upstart, Lending Club, and
41:39Rocket, powering millions of lending
41:41decisions every single day.
41:45And increasingly, we're seeing demand
41:47from capital markets providers who want
41:49to bring these same cash flow signals
41:52into portfolio and investment decisions.
41:55So today, I want to talk about how we're
41:57bringing intelligence to every stage of
42:01lending using cash flow data and AI to
42:05help lenders better understand
42:07borrower
42:09from verification to underwriting to
42:12servicing. helping lenders make better
42:14decisions while expanding access
42:16responsibly.
42:19So, let's start with income
42:20verification.
42:22Every financial decision, whether it's
42:25applying for a loan, expanded credit
42:27lines, rentals, new accounts, you need
42:30to be able to prove you have a job and
42:33you have income.
42:34And income sounds simple, but
42:37historically, it's been really hard to
42:40get right.
42:41Most lenders still rely on payubs and
42:44W2s.
42:45The problem is they're incredibly easy
42:48to fake.
42:50In fact, one in five submitted are
42:53actually fraudulent.
42:55Recently, I was talking to a large auto
42:57lender and they found an approved loan
43:00application where the pastub literally
43:02still had the watermark. This watermark
43:05will be removed after purchase.
43:08That's the baseline we're working with.
43:11So, we rebuilt Plaid income from the
43:14ground up to better reflect how income
43:17actually shows up in the real world.
43:21At the core is the transformer-based LOM
43:23that Sudu mentioned that understands the
43:27context behind each transaction,
43:30grouping deposits into real income
43:32streams based on patterns like
43:35frequency, amount changes, and earning
43:38behavior over time.
43:41That drove 86%
43:44precision for earned income without
43:47sacrificing recall.
43:49Lenders can also apply configurable
43:51filters to include or exclude specific
43:54income types based on their underwriting
43:57criteria.
43:59So things like recurring transfers from
44:01friends or profits from sports betting
44:04can automatically be excluded, making
44:06decisions more consistent instead of
44:09relying on manual judgment.
44:12Now let's talk underwriting.
44:15Lens score is the next generation credit
44:18risk score built on cash flow and
44:21behavioral data. It looks at what's
44:24actually happening in someone's
44:26financial life. Income, spending, cash
44:30flow stability, and how those patterns
44:33change over time. If income goes up
44:36while spending stays flat, the score
44:39improves.
44:40No penalties for paying off a mortgage.
44:43No weird incentives around closing a
44:45credit card. It's designed to reflect
44:48actual financial behavior.
44:51But we've also taken it a step further.
44:54Lens score is the only credit risk score
44:57today that combines cash flow data with
45:01insights from across the plaid network.
45:04Signals traditional credit models simply
45:06can't see.
45:09And this is where it gets really
45:11interesting because these network
45:13behaviors are highly predictive and the
45:17nuance matters.
45:19For example, connecting to a wealth
45:22management app, it shows 20%
45:26lower delinquency risk.
45:29But connecting to 10 or more apps, that
45:32can actually signal double the default
45:34risk. same category, completely
45:38different behavior.
45:41These are the kinds of signals you just
45:42can't get from a traditional credit file
45:45or even cash flow data alone.
45:48And because of that, we're seeing very
45:51meaningful improvements in performance.
45:54On average, Lens Score delivers a 25%
45:58lift in predictive performance compared
46:01to traditional credit data alone and can
46:04reduce risk by up to 41%
46:08at the same approval rates. All at a
46:11lower cost to lenders.
46:13But one thing we've learned is that risk
46:16is not one sizefits-all.
46:20Lenscore gives you a very powerful
46:22general purpose view of credit risk,
46:26but different products have very
46:28different risk dynamics.
46:30And one of the clearest examples is cash
46:34advance and earned wage access.
46:37So these are two of the fastest growing
46:39categories in consumer finance and also
46:41some of the hardest to get right.
46:44They involve highfrequency decisions,
46:47small dollar amounts, and most providers
46:50are still making decisions based on a
46:52single connected bank account.
46:56So, we decided to create something
46:58purposebuilt for this exact use case,
47:03the cash advance index.
47:07The cash advance index predicts the
47:09likelihood of repayment within 30 days,
47:12giving providers a real time score they
47:15can use to approve, size, and manage
47:19advances.
47:21It brings the same depth of network
47:24intelligence into an environment where
47:26decisions need to happen instantly and
47:30repeatedly across the entire customer
47:33life cycle.
47:35In a randomized AB test with a leading
47:38provider, the cash advance index reduced
47:41delinquency by eight percentage points
47:45with no drop in approval rates. And if
47:48you know this space, that is a huge
47:51improvement.
47:53It means you can approve more of the
47:55right users, extend the right amounts,
47:58and manage risk risk much more
48:00precisely.
48:02Okay, so far we've mostly talked about
48:05origination,
48:07but the reality is that risk changes
48:10over time.
48:12Someone loses a job, takes on new
48:14expenses, has a major life event, or
48:17someone like me that has five hungry
48:20kids at home.
48:22Most lenders still don't have a simple
48:24way to monitor those changes in real
48:26time. That's where servicing comes in.
48:31At origination, you're already using
48:33Plaid's income and underwriting products
48:34to evaluate a borrower.
48:37Now, you can monitor that same borrower
48:40over time through ongoing updates to
48:43income, balances, and transaction
48:46activity in real time or once a full
48:50refresh view across all accounts is
48:53available.
48:55So, take a rent splitting app. After
48:59approving a borrower to split rent into
49:01installments, the app leverages Plaid to
49:05monitor the borrower's income and cash
49:07flow.
49:08That gives lenders the ability to
49:10reassess risk before each advance and
49:14even align repayment timing to when
49:16funds actually arrive.
49:20So instead of falling behind on rent,
49:22renters can align their payments to
49:25their cash flow and stay financially on
49:28track.
49:30Coming back to Ashlin,
49:34nothing about her financial life changed
49:36between getting denied and getting
49:38approved. The only thing that changed
49:42was the lens we used to evaluate her.
49:46And that's what makes this so powerful.
49:50Because cash flow data isn't just about
49:52helping thin file borrowers.
49:54It's about giving lenders a more
49:56complete
49:58realtime understanding of financial
50:00health for every borrower.
50:04That's what we're building at Plaid, an
50:07intelligence layer powered by cash flow
50:09and network insights that helps lenders
50:13make better decisions across the entire
50:16life cycle.
50:18and we're here to help you see the full
50:22picture. Thank you.
50:28[music]
50:29>> Oh, we're back. Yes. Hello and welcome
50:32to the Plaid Shopping Network. I am
50:34Melissa.
50:35>> It's me, Ben. Honest.
50:37>> And we're here to talk about Plaid Lens
50:40Score.
50:41>> Can you tell us a little bit about
50:42Plaid?
50:43>> I'm sure the people at home want to hear
50:44all about Plaid Lens.
50:45>> Oh, you're going to love it. Plaid Lens
50:47Score. a more precise way to make credit
50:50decisions.
50:50>> Oh,
50:51>> now do you folks know what what credit
50:53is?
50:54>> A notch on a tree every time you get a
50:56gallon of milk.
50:57>> That's how I understand it.
50:58>> Plaid lens score is a new way to assess
51:01credit risk and offers consumers the
51:04chance to share a more complete picture
51:06of their financial lives. Credit
51:08normally looks back at your past, but
51:11Plaid Lend Score likes to look at the
51:13future of you as a borrower.
51:14>> Oo. Oh, it's almost like a crystal ball.
51:17>> Yes. Using lend score, borrowers get a
51:21score from 1 to 99, indicating
51:23likelihood to repay a loan. Now, let's
51:27think about this. What is your lend
51:30score?
51:31>> 100.
51:32>> Ah, no. It's just from 1 to 99.
51:35>> Yeah. I'm probably 73.
51:37>> 73.
51:38>> Yeah. Sometimes I'm a bit of a scamp.
51:40>> Oh, [laughter]
51:42>> I wouldn't lend to me. Oh.
51:52Please welcome to the stage head of
51:53payments at Plaid, Brian Demir.
52:02>> Good afternoon. Fintech is all about
52:05numbers. So, let's start with a really
52:06big one. 93 trillion. $93 trillion flow
52:12through a last year in the United
52:13States. AC in bank payments more broadly
52:17are the backbone of the digital economy
52:20and are growing three times faster than
52:22credit card payments. These are the
52:24transactions that run people's financial
52:27lives. From funding investment accounts
52:30to repaying loans to paying invoices.
52:33But here's the reality. AC was simply
52:36built for a different era. And the gap
52:38between how it works and how consumers
52:41expect it to work has only widened over
52:43time. We at Plaid are on a mission to
52:46apply the power of our network to make
52:49bank payments match the expectation of
52:51the modern consumer. Our goal is to make
52:54bank payments seamless in order to drive
52:57the growth of your business. Nearly
53:007,000 companies use Plaid to connect
53:03their bank account. That includes
53:04leaders in lending, remittance, crypto,
53:07banking, gaming, accounting, software,
53:10and so much more. And 89% of top fintexs
53:14leverage us in some capacity.
53:17The use cases are different, but the
53:19core opportunity remains the same.
53:21Whether it's investment platforms
53:23driving assets under management, buy now
53:26pay laterers optimizing repayment, or
53:28banks driving privacy, building seamless
53:32payment experiences is central to all of
53:34their missions.
53:36In order to do this, there are four
53:39things that payment leaders and Plaid
53:41cares deeply about. Selection,
53:44conversion,
53:46risk assessment, and money movement.
53:51Selection is about making bank payments
53:53the preferred choice. Plaid's embedded
53:56SDK and link flows are dynamic and
53:59adaptive. Whether it's a checkout, a
54:02bank linking experience, or a one-off
54:04payment, it adapts to optimize for that
54:08particular experience.
54:10Embedded institution search, which you
54:12can see here, surfaces the banks that a
54:14user is most likely to use using
54:17real-time geoloccation data as well as
54:19network heristics.
54:21Additionally, our user experience is
54:23optimized to create a trusted familiar
54:26experience.
54:27The aggregate result of all of this work
54:30is that Plaid customers see a five times
54:33lift in bank payment adoption compared
54:35to other solutions.
54:38After that is conversion. Conversion is
54:42about making sure the user can finish
54:44what they started. Our o engine
54:47seamlessly integrates open banking
54:50powered off manual entry and even
54:53instant micro deposits into one seamless
54:56user journey. Importantly, the user
54:59never sees the layers. Whether they're a
55:02Zoomer who's perfectly comfortable with
55:03an app flow or my grandmother who's
55:06going to get a checkbook out of her uh
55:07drawer, they see one flow that adapts
55:10based on their needs and their
55:12preferences.
55:13And for the 100 million Americans who
55:16have a Saved Plat account, we also offer
55:18a streamlined user payments experience,
55:21account connection in as little as one
55:23click.
55:25Returning users consistently convert 11%
55:28higher and the plaid experience has
55:31lifted conversion by as much as 54%
55:34against the competition.
55:37Together these flows are the front door
55:39of modern bank payments.
55:42We have powered over 9 billion payment
55:44sessions in counting and we are
55:46constantly running new experiments
55:48focused on selection and conversion.
55:52So the user is converted but now we have
55:54to assess risk in bank payments. There
55:57are two types of risk. Settlement risk
56:00will the funds actually arrive as well
56:03as fraud risk. Is this a legitimate
56:05payment or is a bad actor involved? This
56:08is where signal comes in. Signal is our
56:11AI power plower transaction risk model.
56:13It has assessed more than a quarter of a
56:15trillion dollars in transactions by
56:18evaluating account connection history,
56:20identity information, prior AC events,
56:23and account behavior to tell you whether
56:26or not a payment will succeed.
56:28Brands like Uphold, a top crypto
56:30exchange, have used Signal to reduce
56:32return losses by over 80%. And we
56:36estimate that Signal has saved the
56:38fintech ecosystem more broadly over $135
56:41million in losses in just the last few
56:44years alone.
56:46Finally, most critical to any payment
56:49journey is the actual moving m of money.
56:51That's the whole point, right? Which is
56:53where Plaid Transfer comes in. Transfer
56:56is one API for every bank rail. AC,
57:00wires, RTP, RFP, and Fed now all in one
57:04solution. Dynamic routing, a built-in
57:07ledger, and a featurerich dashboard for
57:10full visibility on every single payment.
57:13We process billions of dollars in bank
57:15payments, and are on a mission to make
57:17realtime payments adoption absolutely
57:20seamless.
57:22Selection, conversion, risk, and money
57:25movement. the full payment life cycle,
57:28each step of which is made better by the
57:30Plaid network.
57:32But if you ask our customers
57:34consistently where they still feel pain,
57:36the answer is remarkably consistent.
57:39It's risk.
57:41Every time you initiate a bank payment,
57:43you face a difficult trade-off. AC takes
57:46days to settle. Returns can come days
57:49after that. But you have to decide right
57:52now. fund the account and absorb the
57:54loss if the payment fails eventually or
57:56hold funds and risk losing a customer
57:58who simply won't wait. We all know that
58:01the modern consumer expects their funds
58:04now. So get it wrong in either direction
58:07and you pay for it either in losses or
58:10in churn. The teams that are getting
58:12this dynamic right are the ones pulling
58:14ahead and the ones that aren't are
58:16leaving money on the table with every
58:18single transaction.
58:20And what we hear consistently is that
58:22while teams care deeply about this
58:23trade-off, not every team wants to
58:26become experts in a risk. They simply
58:29want bank payments to work. So we built
58:32a new product for that. We combined the
58:35power of our network, unrivaled
58:37experience in a risk, and worked with
58:40customers who wanted a radically
58:42simplified. With this, we build
58:45guaranteed payments, the
58:47industry-leading risk solution that
58:49takes the guesswork out of bank
58:51payments. And it is ready today.
58:55And here's how it works. Plaid
58:57guarantees a settlement for every
58:59approved transaction. If a payment fails
59:02after we've approved it, Plaid covers
59:05the loss, not you. It is that simple.
59:08You process the payment and delight the
59:10consumer with real time fulfillment.
59:13Let's dive a little bit into the
59:15consumer journey. Importantly, a
59:18guaranteed payment looks no different to
59:20the end customer. They go through the
59:23same experience that I just showed you.
59:25In this example, a consumer is funding a
59:28brokerage account. Here it is. The
59:31person is already in the Plaid network.
59:33So, after verifying with an OTP, they
59:36confirm with one click. On the back end,
59:39guarantee is integrated directly into
59:42Plaid Transfer. You make a guarantee
59:44request in the transfer authorization
59:46endpoint, adding a single field to your
59:49implementation.
59:51Behind the scenes are models, which are
59:53powered by signal and protect evaluate
59:55the transaction in real time across more
59:58than 2,000 attributes.
1:00:00Have we seen this user before? Yes. How
1:00:03many times has the account been
1:00:05connected on the Plaid network or any
1:00:08past a returns? Yes, there was a single
1:00:11R1, but it was 2 years ago. And the
1:00:14decision comes back in milliseconds. And
1:00:16in this case, it's been approved. And if
1:00:19Plaid declines to guarantee, you can
1:00:21still process the transaction and take
1:00:23on the risk yourself. The user gets
1:00:26immediate access to funds and you get a
1:00:28guaranteed settlement from Plaid. That's
1:00:31it. from an OTP to a guaranteed
1:00:34settlement. Radically simple to
1:00:36implement and invisible to the consumer.
1:00:40The end result of this is that teams can
1:00:43focus on growth while more consumers get
1:00:46instant experiences powered by bank
1:00:48payments.
1:00:49The results so far are truly remarkable.
1:00:53One of our launch partners who did a
1:00:54head-to-head test against their own
1:00:56internal logic managing a on their own
1:00:59versus guaranteed payments saw a 24%
1:01:02higher approval rate against their
1:01:04baseline logic. This is the power of the
1:01:07plaid network in action. We've been
1:01:09building guaranteed payments with
1:01:11partners like Kelshi, Poly Market and
1:01:13Solve platforms that needed approval
1:01:15rates at the leading edge of the
1:01:17industry and a partner who could handle
1:01:19their scale. We have seen
1:01:22implementations in as little as two
1:01:24weeks and approval rates as high as 90%.
1:01:29When it comes to payment risk, every
1:01:32business has a different appetite for
1:01:34control. Some want to manage risk
1:01:36themselves, working directly with model
1:01:38outputs and signals. Others want to
1:01:41focus on their core product and have
1:01:42experts manage risk on their behalf.
1:01:45Plaid now has solutions for both. all
1:01:48built in the same underlying models,
1:01:50platform, and network intelligence.
1:01:53For teams who want risk manage for them,
1:01:55the solution is guaranteed payments, and
1:01:58I'm proud to say that it's available
1:02:00now. We'd love to help you get started
1:02:02on this. So, get in touch with your
1:02:04account team to learn more.
1:02:07We're so excited to continue working
1:02:09with all of you on this continued
1:02:10evolution of bank payments. The end
1:02:13result of all of this is higher
1:02:15conversion, fewer losses, instant
1:02:19settlement, and with the power of the
1:02:20Plaid network, bank payments are
1:02:22performing better than ever. Thank you.
1:02:30>> Welcome back to the Plaid Shopping
1:02:31Network. As always, I am Melissa and I'm
1:02:34here with Honest Abe himself and we're
1:02:36here to tell you some more things about
1:02:38our friend Plaid Guaranteed Payments.
1:02:40And this is brand new. New is a
1:02:42doorbell.
1:02:43>> What? I don't know what a doorbell is.
1:02:45>> Get up to 90% payment approval rates
1:02:48without added risk.
1:02:50>> 90% without added risk. That's
1:02:52absolutely fabulous.
1:02:54>> I'm quite riskaverse myself. You know,
1:02:56that's why I spend every night safely
1:02:57seated in a dark theater.
1:03:00>> Great. Plaid guaranteed payments is a
1:03:03managed a service that makes bank
1:03:05transfers predictable. If Plaid approves
1:03:08the transaction, the transaction is
1:03:11going through.
1:03:11>> Well, that's very exciting. It sounds
1:03:13like confidence is built into every a
1:03:15payment. [laughter]
1:03:18>> And we're going to have more fun when we
1:03:19come back to the Plaid Shopping Network
1:03:21and hear even more about our three
1:03:22amazing products.
1:03:24>> Call this number using the Now, the
1:03:28telephone.
1:03:29>> Yes.
1:03:31>> What is [music] it?
1:03:38Please welcome back to the stage Plaid
1:03:40co-founder and CEO Zack Pereé.
1:03:48Okay, that's it for the intro kickoff.
1:03:51Uh I hope that you had as much fun as I
1:03:53did. That was so cool to see all of the
1:03:54things that we've been working on for
1:03:56just a little while. When I look at
1:03:57what's happening in fintech right now,
1:03:59all the products that you saw, all the
1:04:01things that you're building, um, I've
1:04:02never been more optimistic nor more
1:04:04excited. Uh, I think we're still
1:04:06incredibly early in this phase. Uh, AI
1:04:08is continuing to change everything and
1:04:10it's going to be a blast as we all work
1:04:12through it together. Um, as for what
1:04:14comes next, please come talk to our
1:04:15teams. We'll be around. There going to
1:04:17be a bunch of sessions that are
1:04:18happening very shortly. Um, a lot of
1:04:19people here are working on the same
1:04:21problems or similar problems to all of
1:04:22you. So, please meet each other. That's
1:04:24what today is for. Thank you all for
1:04:26being here. and you'll hear a little bit
1:04:27more from us in just a few hours.
1:04:29[applause]
1:04:35[music]
1:04:35>> Thank you so much for joining us today.
1:04:37And gosh dang it, we ran out of time.
1:04:40>> Dang.
1:04:41>> Well, I have had the time of my life
1:04:42here with you.
1:04:43>> Did you boys have fun?
1:04:44>> Oh, it was a gas.
1:04:45>> Oh, I had the best time.
1:04:47>> Me, too. You guys are my bestest
1:04:49friends. [laughter]
1:04:52>> Absolutely loved it. Well, you're all a
1:04:53part of the Plaid Network, so we'll see
1:04:55you on plaid.com.
1:04:56>> We'll see you there.
1:04:57>> So long.
1:04:58>> Have a good one, everyone.
1:04:59>> We will miss you.
1:05:01>> Yes.
1:05:02>> Yes.
1:05:02>> All right. Good work, everybody. Great
1:05:04job.
1:05:05>> Great call. We love you. We miss you.
1:05:07>> We could hang out. Do you guys hang out?
1:05:09>> Oh my god. I'm really busy tonight,
1:05:10though.
1:05:11>> Yeah. It's kind of a rough one for me.
1:05:14>> All right. That's great. And this has
1:05:16been Alyssa at the Plaid Shopping
1:05:18Network, and we can't wait to see you
1:05:20here soon.
1:05:23Goodbye.

About this transcript

This page contains the full transcript of Plaid Effects 2026 | Keynote by Plaid, generated from the public captions YouTube serves with the video. The transcript has 9,662 words across 1,601 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.