# From Announcement to Adoption: The Real State of AI in Banking

## Abstract

AI is now a strategic priority across banking, but the distance between public announcements and scaled production remains significant. As institutions look to deploy AI across customer experience, fraud, AML, risk, operations, and advisory workflows, many are reassessing the foundations required to support meaningful transformation. Join Evident Insights and NVIDIA for a conversation on how banks are approaching the next phase of AI adoption: building flexible platform architectures, retaining control over proprietary data and models, and identifying the use cases with the greatest potential for enterprise-wide impact. In this webinar, you’ll learn: How leading banks are rethinking AI platform architecture Strategies to optimize the role of frontier and open source models in the enterprise Where agentic AI may create meaningful value across banking How transaction foundation models could reshape fraud, AML, risk, and customer intelligence What separates AI announcements from meaningful enterprise adoption How banks can better control their models, data, and infrastructure

## AI Summary

_No AI summary available._

## Transcript

**[00:00:02 – 00:00:04]** Good morning, good afternoon, everyone.

**[00:00:04 – 00:00:07]** I know we have people from all over the world.

**[00:00:07 – 00:00:09]** Welcome to another NVIDIA webinar.

**[00:00:09 – 00:00:11]** We're going to be talking about the banking industry,

**[00:00:11 – 00:00:15]** what is happening with AI, and I couldn't be more excited about

**[00:00:15 – 00:00:18]** the guests we have with us today.

**[00:00:18 – 00:00:23]** We have Alexandra, co-founder, co-CEO of Evident AI, which

**[00:00:23 – 00:00:24]** has a lot of things to say.

**[00:00:24 – 00:00:27]** They just released their new index a couple of days ago.

**[00:00:27 – 00:00:32]** And Kevin, which is a senior editor at Evident, and he

**[00:00:32 – 00:00:34]** has the most difficult job ever.

**[00:00:34 – 00:00:37]** He has to make sure that Alexandra and I don't run with this

**[00:00:37 – 00:00:41]** for one hour of a monologue without letting anybody else talk.

**[00:00:41 – 00:00:45]** With that, and because we really want to get through

**[00:00:45 – 00:00:47]** all this, let me introduce myself.

**[00:00:47 – 00:00:48]** I'm Sergio Blanco.

**[00:00:49 – 00:00:51]** I have the privilege of leading the banking and insurance

**[00:00:51 – 00:00:53]** verticals for NVIDIA globally.

**[00:00:53 – 00:00:55]** And, you know, looking forward to this conversation.

**[00:00:55 – 00:01:02]** So, Alexandra, you guys just released the Evident AI Index

**[00:01:02 – 00:01:03]** for the banking industry.

**[00:01:03 – 00:01:06]** There's a lot of talk about it.

**[00:01:06 – 00:01:08]** What can you tell us about this?

**[00:01:08 – 00:01:11]** Yeah, well, I mean, and thank you for organizing this.

**[00:01:11 – 00:01:14]** And it's so great to be spending an hour with you

**[00:01:14 – 00:01:18]** in a deep conversation about AI deployment and financial

**[00:01:18 – 00:01:22]** services, because that is where it's all happening, of course.

**[00:01:23 – 00:01:28]** As you mentioned, we've just released our 2026 ranking

**[00:01:28 – 00:01:33]** results of the biggest banks in North America and Europe,

**[00:01:33 – 00:01:37]** so the Canadians and the US and across all of Europe.

**[00:01:37 – 00:01:41]** And, uh, it's, it's, it's been very interesting to follow, obviously,

**[00:01:41 – 00:01:44]** throughout the year and then sort of see where it lands here.

**[00:01:44 – 00:01:47]** You know, it was on october 6 that it was released, but

**[00:01:47 – 00:01:51]** one thing i would point out is that since the, you know,

**[00:01:51 – 00:01:54]** we released the index the first time, which was back in january

**[00:01:54 – 00:01:58]** 23, this is the fastest pace

**[00:01:58 – 00:02:04]** of, you know, accelerating of the spend, of deployment on innovation

**[00:02:04 – 00:02:07]** levels that are pulled, you know, starting to really see sort

**[00:02:07 – 00:02:11]** of scale and speed, and it feels like there's this, you know, this

**[00:02:11 – 00:02:14]** hockey stick, and we're just sort of on our way, you know, finally

**[00:02:15 – 00:02:18]** sort of up that sort of that, uh, end, end of the hockey stick.

**[00:02:18 – 00:02:24]** So a lot of activity; um, the leading banks are moving very fast.

**[00:02:24 – 00:02:27]** We've found that the top three remained the same as last

**[00:02:27 – 00:02:31]** year, so that is led by J.P. Morgan, second is Capital One, then

**[00:02:31 – 00:02:38]** came RBC, and it is a, well, the ranking hasn't changed, the pace

**[00:02:38 – 00:02:40]** has changed, but they've all three.

**[00:02:40 – 00:02:45]** We've been, you know, moving as fast, you know, and faster

**[00:02:45 – 00:02:49]** than last year, keeping pace with each other, which, you

**[00:02:49 – 00:02:53]** know, we'll get into sort of what that means and what they're doing,

**[00:02:53 – 00:02:55]** but that pace has been incredible.

**[00:02:55 – 00:02:58]** And then you've had, we've seen some movements, you know,

**[00:02:58 – 00:03:01]** some have gone a little bit up, some have gone down, and

**[00:03:01 – 00:03:04]** for a number of reasons, but one thing that I thought was

**[00:03:04 – 00:03:07]** really interesting was that the top 10 are now really dominated

**[00:03:07 – 00:03:13]** by the North American banks, And with UBS the only one remaining

**[00:03:13 – 00:03:17]** in the top 10 now from Europe.

**[00:03:17 – 00:03:19]** And there are a lot of reasons for that, and we'll get into

**[00:03:19 – 00:03:22]** that, but it really does, I think, is sort of testament

**[00:03:22 – 00:03:28]** to the fact that if you've got close proximity to the, you know,

**[00:03:28 – 00:03:33]** the tech talent, to the startups that are focused on financial

**[00:03:33 – 00:03:38]** services, and you have, you know, more access to that kind of talent

**[00:03:38 – 00:03:42]** and also the mindset, et cetera, et cetera, you are, you know, you're

**[00:03:42 – 00:03:44]** able to move a little bit faster, right, because it's right there.

**[00:03:44 – 00:03:47]** And as we know, there's musical chairs between, you know,

**[00:03:47 – 00:03:50]** Frontier Labs and the big tech companies and the banks,

**[00:03:50 – 00:03:53]** in and out, you know, and so you see a lot of the hires taking place

**[00:03:53 – 00:03:57]** in the North American and the U.S. banks that are coming from the tech

**[00:03:57 – 00:03:59]** sector and maybe a little less so.

**[00:04:00 – 00:04:04]** And when we know that technology is, you know, very powerful

**[00:04:04 – 00:04:07]** and it's different from other tech transformations that we've

**[00:04:07 – 00:04:12]** seen over the decades, you really do need, you know, scientific

**[00:04:12 – 00:04:16]** talent, you know, specialist talent moving, you know, AI from ideation

**[00:04:16 – 00:04:20]** into production and building the platforms that are needed for

**[00:04:20 – 00:04:22]** scale and building the harnesses.

**[00:04:22 – 00:04:23]** That needs particular type of talent.

**[00:04:23 – 00:04:25]** So that proximity is probably what's helping

**[00:04:25 – 00:04:26]** the North American banks.

**[00:04:26 – 00:04:29]** So we're seeing, we're seeing a lot of activity.

**[00:04:29 – 00:04:33]** The results have been, have been really, really interesting,

**[00:04:33 – 00:04:35]** and not only in the top, but also looking at the sort of the

**[00:04:36 – 00:04:39]** middle of the pack, sort of who's really moving and leapfrogging,

**[00:04:39 – 00:04:42]** but also from the some of the banks, like really at the,

**[00:04:42 – 00:04:46]** you know, the sort of bottom, you know, end of the index, that are,

**[00:04:46 – 00:04:50]** some are sort of really shifting upwards and making that decision

**[00:04:50 – 00:04:55]** to put in place the strategies, you know, put the money behind it and

**[00:04:55 – 00:04:57]** really get motoring on execution.

**[00:04:58 – 00:05:01]** So that's been really interesting to follow.

**[00:05:01 – 00:05:04]** I want to start, actually, with the conversation on speed.

**[00:05:04 – 00:05:07]** I think, Alexandra, you gave a great overview there.

**[00:05:07 – 00:05:09]** And the headline stat from, or one of the headline stats,

**[00:05:09 – 00:05:14]** I should say, from the report was the average score of the bank's.

**[00:05:15 – 00:05:23]** I'm excited to have this conversation today because

**[00:05:23 – 00:05:26]** Alexandra, Sarah, I think you two are probably two

**[00:05:26 – 00:05:31]** of the best situated to look at multiple banks and see

**[00:05:31 – 00:05:35]** how that pace actually manifests in the way that banks roll out

**[00:05:35 – 00:05:37]** tools, the way they hire talent.

**[00:05:37 – 00:05:40]** I guess I'll kick to you, Sarah, to start with.

**[00:05:40 – 00:05:44]** You just heard some takeaways from the research that Evident has done.

**[00:05:44 – 00:05:47]** You have conversations with banks all the time.

**[00:05:47 – 00:05:51]** Is that pace kind of mirrored in the way that bankers are talking

**[00:05:51 – 00:05:53]** to you and the questions that they're asking and the types of

**[00:05:54 – 00:05:56]** things that they're taking on now?

**[00:05:58 – 00:06:00]** Absolutely. By the way, I was taking notes when Alexandra

**[00:06:00 – 00:06:06]** was talking because the fast-paced piece is definitely a theme.

**[00:06:06 – 00:06:09]** And as I start putting together my predictions for 2027, I

**[00:06:09 – 00:06:13]** was going to say that's going to be one of them.

**[00:06:13 – 00:06:18]** I expect to see a significant acceleration of AI investment and

**[00:06:18 – 00:06:21]** that AI translating into results.

**[00:06:21 – 00:06:26]** So, back to your question, it is palpable in the industry.

**[00:06:26 – 00:06:32]** If you think about this, one year ago we were talking to the industry

**[00:06:32 – 00:06:37]** and saying, your data is the most fundamental asset you have.

**[00:06:37 – 00:06:38]** We need to keep that in mind.

**[00:06:38 – 00:06:41]** Banks are fundamentally data companies.

**[00:06:41 – 00:06:43]** And this revolution, right?

**[00:06:43 – 00:06:46]** Let's forget headlines for one second and about this

**[00:06:46 – 00:06:47]** is about, I don't know what.

**[00:06:47 – 00:06:51]** This revolution is the revolution of being able to take data and

**[00:06:51 – 00:06:54]** turn that data into intelligence at a speed of light.

**[00:06:54 – 00:06:55]** This is what it is.

**[00:06:55 – 00:06:57]** We can create insights, we can create knowledge, we can

**[00:06:57 – 00:07:00]** create intelligence that are really fast-paced.

**[00:07:00 – 00:07:02]** And again, banks are just data.

**[00:07:02 – 00:07:04]** So one year ago, we're talking to the banks and saying, listen, your

**[00:07:04 – 00:07:06]** data is your intellectual property.

**[00:07:06 – 00:07:10]** That data, it should turn into your own models that you can

**[00:07:10 – 00:07:15]** use to build something that really transports your processes, right?

**[00:07:15 – 00:07:19]** And that, connected to the right amount of compute, will create

**[00:07:19 – 00:07:21]** a mode for the future, right?

**[00:07:21 – 00:07:23]** That was one year ago.

**[00:07:23 – 00:07:25]** And it was a little bit of, at that point, it was still

**[00:07:25 – 00:07:29]** a lot of, oh no, but you know, chatbots is the thing, this is all

**[00:07:29 – 00:07:31]** about Gen-AI and all these things.

**[00:07:31 – 00:07:32]** Okay, let's pause this.

**[00:07:32 – 00:07:37]** About your data transforming into intelligence, turn that

**[00:07:37 – 00:07:40]** into transformation of the bank.

**[00:07:40 – 00:07:44]** Look at today, just in the last month, how many banks

**[00:07:44 – 00:07:46]** have announced they are building their own foundation model.

**[00:07:47 – 00:07:51]** Some call it a banking foundation model like Bragman, Revolut,

**[00:07:51 – 00:07:54]** others call it the foundation model of the bank, like Lloyd's,

**[00:07:54 – 00:07:56]** and call it Autumn.

**[00:07:56 – 00:08:01]** You have a full spectrum of neo-banks and traditional

**[00:08:01 – 00:08:04]** banks, so there is no longer that excuse of like, oh, this is

**[00:08:04 – 00:08:07]** something only a new bank can do.

**[00:08:07 – 00:08:09]** The full spectrum of banks, you look at the leaders in

**[00:08:09 – 00:08:12]** your score, all of them are building things along these lines.

**[00:08:13 – 00:08:18]** And that is the principle, and today every main bank is doing that

**[00:08:18 – 00:08:22]** because they realize that you build that intelligence and you can start

**[00:08:22 – 00:08:25]** activating in so many ways, right?

**[00:08:25 – 00:08:29]** And as you know in your index, right, it is very easy to

**[00:08:29 – 00:08:33]** make an announcement and say we're doing a lot in AI.

**[00:08:33 – 00:08:36]** It is very hard to justify and prove that with the outcomes

**[00:08:36 – 00:08:38]** and all these things, right?

**[00:08:38 – 00:08:41]** And so the outcomes come from this, through intelligence connected

**[00:08:41 – 00:08:44]** to the real processes, right?

**[00:08:44 – 00:08:47]** And you see how fraud transforms, marketing transforms, lifetime

**[00:08:47 – 00:08:51]** value calculations transform, credit assessment transforms.

**[00:08:51 – 00:08:52]** That's the exciting piece here.

**[00:08:53 – 00:08:56]** But, Alexander, what are you hearing from the industry

**[00:08:56 – 00:08:59]** about that, about the pace?

**[00:08:59 – 00:09:01]** The pace? Yeah, I think it is, and I'm also getting a

**[00:09:01 – 00:09:04]** bit distracted by the really good questions in the chat because they

**[00:09:04 – 00:09:07]** are so good about, you know, the ratio of open versus closed source

**[00:09:07 – 00:09:10]** models and weight models and so on.

**[00:09:10 – 00:09:14]** But the pace has been, I couldn't agree with your assessment more.

**[00:09:14 – 00:09:16]** I think, you know, we're very aligned in what we're seeing

**[00:09:16 – 00:09:22]** and, obviously, you know, the sort of technical capabilities

**[00:09:22 – 00:09:24]** of the leading banks and what they're able to do.

**[00:09:24 – 00:09:26]** And I know we'd probably get on to, you know, what has

**[00:09:26 – 00:09:31]** been built in these banks, on the neobank side, but we're seeing that

**[00:09:31 – 00:09:33]** merit in the leading banks as well.

**[00:09:33 – 00:09:39]** They definitely are keeping up very much with the

**[00:09:39 – 00:09:40]** capabilities of the neobanks.

**[00:09:40 – 00:09:44]** There is a particular question in the chat about the traditional

**[00:09:44 – 00:09:45]** banks that we measure.

**[00:09:45 – 00:09:51]** We keep in mind that the reason we built the index was to look at the

**[00:09:52 – 00:09:57]** pace of transformation in terms of AI deployment in traditional banks.

**[00:09:57 – 00:10:02]** Hence, we don't include a Nubank or Revolut in the index.

**[00:10:02 – 00:10:03]** Hopefully, we'll do that.

**[00:10:03 – 00:10:05]** Next year, we'll set up something on challenger banks.

**[00:10:05 – 00:10:08]** Next year, we do have Nubank in the Latin America index,

**[00:10:08 – 00:10:11]** but not in this what we call the global ranking.

**[00:10:11 – 00:10:15]** But we try to mainly look at the traditional banks because

**[00:10:15 – 00:10:19]** they are the ones comparing, you know, to each other, to

**[00:10:19 – 00:10:23]** their peers, and how quickly are we able to go from legacy

**[00:10:23 – 00:10:27]** to AI first, and that journey is what we measure in the index.

**[00:10:27 – 00:10:29]** But of course, when looking at the neobanks and looking at

**[00:10:29 – 00:10:33]** those capabilities, you know, this, they're born AI native, right?

**[00:10:33 – 00:10:37]** They have a completely different starting point, and they don't

**[00:10:37 – 00:10:39]** have legacy data systems, to your point, Asa.

**[00:10:39 – 00:10:41]** That is one of the things that you really got to focus

**[00:10:41 – 00:10:43]** on if you're a traditional bank.

**[00:10:43 – 00:10:46]** Your data is fragmented, you've got to knit it together, you've got

**[00:10:46 – 00:10:49]** to figure out what platform you're building that can sit across all

**[00:10:49 – 00:10:53]** of that, and then all of the things you can do on, on that platform.

**[00:10:53 – 00:10:58]** And neobanks, they can set it up really well from the get-go, right?

**[00:10:58 – 00:11:05]** So, and to sort of the amount of growth in customers that you're

**[00:11:05 – 00:11:11]** seeing both NewBank and Revolut achieve is really astonishing.

**[00:11:11 – 00:11:14]** But, you know, so when you're looking at the traditional

**[00:11:14 – 00:11:16]** banks and you're looking at the challenges that are coming

**[00:11:16 – 00:11:20]** at them, it's not just looking at their peers as in, you

**[00:11:20 – 00:11:23]** know, what are my neighbors doing that are also traditional

**[00:11:23 – 00:11:24]** banks and I need to move as fast.

**[00:11:24 – 00:11:26]** NVIDIA, IndeX, AI.

**[00:11:26 – 00:11:29]** But I also have the NEO or challenger banks coming at me

**[00:11:29 – 00:11:34]** because they might not necessarily take people and become, you know,

**[00:11:34 – 00:11:38]** the primary bank account, but they can take, definitely what we're

**[00:11:38 – 00:11:41]** seeing is that they will become the secondary bank account and

**[00:11:41 – 00:11:44]** moving into the territory of what the primary bank account is doing.

**[00:11:44 – 00:11:48]** And then you've got the frontier labs that are also thinking

**[00:11:48 – 00:11:52]** about, well, you know, what if people can start transacting

**[00:11:52 – 00:11:54]** through our platform?

**[00:11:54 – 00:11:57]** And then you, and even if that might take some time, you

**[00:11:57 – 00:12:02]** and I, and Kevin, right, we might be asking Claude or chat, like,

**[00:12:02 – 00:12:05]** how do I optimize more portfolio, even if there's no transactions

**[00:12:05 – 00:12:08]** that are being taken, but we're walking into the traditional

**[00:12:08 – 00:12:13]** banks with a complete behavioral change going on with different

**[00:12:13 – 00:12:16]** expectations, and the banks are needing to keep up with that.

**[00:12:16 – 00:12:18]** And the leading banks, when we're looking at the pace of change,

**[00:12:18 – 00:12:21]** you know, you, you know, obviously all of that is happening from

**[00:12:21 – 00:12:25]** an engineering of infrastructure, you know, perspective, but

**[00:12:25 – 00:12:29]** they've also got to work out, I've got to make my products

**[00:12:29 – 00:12:33]** work for the way that our customers are changing their behavior.

**[00:12:33 – 00:12:36]** So I've got to look two or three years out and see, you know,

**[00:12:36 – 00:12:40]** what is it that that interaction might need to look like?

**[00:12:40 – 00:12:42]** So you've got a lot of blue sky thinking needing to take

**[00:12:42 – 00:12:43]** place at the same time.

**[00:12:43 – 00:12:45]** And that is something that we're seeing in the leading

**[00:12:45 – 00:12:48]** banks, that they're looking at that, they're experimenting

**[00:12:48 – 00:12:52]** with that, in addition to still dealing with huge legacy systems

**[00:12:52 – 00:12:55]** that they need to transform.

**[00:12:55 – 00:13:00]** Sarah, that kind of blue sky thinking, is that really what

**[00:13:00 – 00:13:04]** was behind the development of the Pragma model with Revolut?

**[00:13:04 – 00:13:10]** I mean, how did the bank kind of go about crafting that

**[00:13:10 – 00:13:14]** kind of a foundation model and thinking about the customer

**[00:13:14 – 00:13:17]** a couple of years ahead of time?

**[00:13:17 – 00:13:21]** We are going through the most fundamental transformation

**[00:13:21 – 00:13:24]** of our industry in our lifetime.

**[00:13:24 – 00:13:27]** None of us remember anything like this.

**[00:13:27 – 00:13:32]** And that's what was behind the idea, right?

**[00:13:32 – 00:13:36]** When you are a player and really want to grow fast,

**[00:13:36 – 00:13:40]** what you want to do is go beyond what is possible, right?

**[00:13:40 – 00:13:44]** If you look at the last four or five years, right?

**[00:13:44 – 00:13:47]** We've been helping banks build foundation models.

**[00:13:47 – 00:13:49]** Which is a foundation model in finance?

**[00:13:49 – 00:13:51]** It is a model, right?

**[00:13:51 – 00:13:54]** Think about something similar to a transformer that is used

**[00:13:54 – 00:13:57]** to predict language in the Netherlands, but that is focused

**[00:13:57 – 00:13:59]** on financial data, right?

**[00:13:59 – 00:14:03]** So it's a model that is helping you predict financial events.

**[00:14:03 – 00:14:06]** And I think the aha moment of Revolut was saying, Okay,

**[00:14:06 – 00:14:09]** but why are we limiting this to the transactional data

**[00:14:09 – 00:14:10]** the bank is doing?

**[00:14:11 – 00:14:13]** Why don't we actually build into the model everything

**[00:14:13 – 00:14:15]** we know about the customer?

**[00:14:15 – 00:14:19]** Because the suspicion they had was the level of intelligence

**[00:14:19 – 00:14:22]** we're going to be able to create is exponentially larger.

**[00:14:22 – 00:14:24]** It's not only going to be about what, is this transaction

**[00:14:24 – 00:14:27]** fraud or not, but it's going to be what is the client supposed to do,

**[00:14:27 – 00:14:30]** how it's going to be buying things, what is going to be the next action

**[00:14:30 – 00:14:34]** to take, how it's going to be, you know, anything about the consumer.

**[00:14:34 – 00:14:36]** You're going to have that great view.

**[00:14:37 – 00:14:40]** So I want to use this, actually, as the blue sky comment, the

**[00:14:40 – 00:14:43]** comment about what they were thinking, to use something

**[00:14:43 – 00:14:45]** to make it very real for people.

**[00:14:46 – 00:14:48]** Think about the previous big revolution that happened in

**[00:14:48 – 00:14:49]** our industry, right?

**[00:14:49 – 00:14:51]** It was the 1960s when the industry moved from being

**[00:14:51 – 00:14:54]** paper-based to be computer-based.

**[00:14:54 – 00:14:59]** In 1960, a bank had a checking account, which was the same

**[00:14:59 – 00:15:00]** in the U.S., was the same for every single.

**[00:15:00 – 00:15:01]** It was mandated by the government.

**[00:15:01 – 00:15:03]** The checking account is this much, it pays this.

**[00:15:03 – 00:15:05]** There was one checking account.

**[00:15:05 – 00:15:07]** And if you wanted any kind of lending, like a mortgage,

**[00:15:07 – 00:15:10]** you need to go to the bank and negotiate with them and see

**[00:15:10 – 00:15:13]** what the hell is going to happen.

**[00:15:13 – 00:15:18]** You fast forward to 1980, and then every bank has multiple type of

**[00:15:19 – 00:15:22]** checking accounts, saving accounts, 401k, investment accounts,

**[00:15:22 – 00:15:26]** credit cards, debit cards, and you keep naming it, right?

**[00:15:26 – 00:15:29]** Think about the amount of innovation that happened

**[00:15:29 – 00:15:30]** in those 20 years.

**[00:15:30 – 00:15:33]** If you're in 1916, you tell someone you're gonna be paying

**[00:15:33 – 00:15:35]** with a piece of plastic, they would probably put you in

**[00:15:35 – 00:15:40]** an asylum or something and saying you're absolutely mental, right?

**[00:15:40 – 00:15:44]** So what we have ahead of us is gonna be something like that.

**[00:15:44 – 00:15:47]** I was, the other day, pushing; I was at Cybers and I was

**[00:15:47 – 00:15:51]** talking to the industry and said, listen, you know, we have today in

**[00:15:51 – 00:15:54]** a bank 20-something different type of accounts, just for checking,

**[00:15:54 – 00:15:56]** because we're going to try to put the client in the one that

**[00:15:56 – 00:15:58]** actually fits better their needs.

**[00:15:58 – 00:16:01]** I can see a future where actually a client, a bank has as many

**[00:16:01 – 00:16:03]** checking accounts as clients they have, because you're

**[00:16:03 – 00:16:06]** really going to create one product for every client that is

**[00:16:06 – 00:16:08]** perfectly set up for that client.

**[00:16:08 – 00:16:10]** Again, it may seem crazy today, like having a piece of plastic

**[00:16:10 – 00:16:13]** to pay things in 1960.

**[00:16:13 – 00:16:14]** Probably we're gonna get there.

**[00:16:15 – 00:16:17]** We're gonna have products that are truly, truly unique

**[00:16:17 – 00:16:18]** products for everybody.

**[00:16:18 – 00:16:21]** Think about the implications in wealth management and asset

**[00:16:21 – 00:16:22]** management, right?

**[00:16:22 – 00:16:25]** Breathing advice to a client that is truly unique to them,

**[00:16:26 – 00:16:28]** which we are not there yet.

**[00:16:28 – 00:16:32]** So blue sky thinking, I'm very excited about this.

**[00:16:33 – 00:16:36]** Alexandra and Sarah, what you just kind of described is something

**[00:16:36 – 00:16:40]** that I think we hear from banks all the time, right, the reimagining

**[00:16:40 – 00:16:45]** of processes, the rewiring of the bank, you know, the reorchestrating

**[00:16:45 – 00:16:49]** of kind of how all the workflows around it work and, you know,

**[00:16:49 – 00:16:54]** kind of rebuilding it up from the ground around this technology.

**[00:16:54 – 00:16:58]** I think those types of words are blue sky thinking in themselves

**[00:16:58 – 00:17:01]** and sometimes mean different things to different people.

**[00:17:01 – 00:17:05]** Alexandra, I want to go over to you because, again, you're

**[00:17:05 – 00:17:07]** talking to banks all the time.

**[00:17:07 – 00:17:10]** When someone says that to you, that we are rewiring

**[00:17:10 – 00:17:15]** our process or we are reimagining our process, what is the tell

**[00:17:15 – 00:17:19]** that says they're actually serious about that versus something that

**[00:17:19 – 00:17:23]** feels like it's just lip service to what is supposed to be said?

**[00:17:23 – 00:17:26]** Yeah, I think, I mean, I think that takes us into this question

**[00:17:26 – 00:17:29]** about what's, like, what is table stakes, as we define

**[00:17:29 – 00:17:35]** it, versus what is frontier, and you get to the frontier if you're

**[00:17:35 – 00:17:36]** actually going into the workflows.

**[00:17:36 – 00:17:41]** And, but even, even just sort of having the operating model

**[00:17:41 – 00:17:47]** and the strategy and the game plan to push AI capabilities down into

**[00:17:47 – 00:17:51]** the lines of business is obviously a first necessary starting point.

**[00:17:51 – 00:17:54]** So when you're looking at how the banks are organized, you

**[00:17:54 – 00:17:58]** know, to your point, Asa, like over the years, like pendulum swings

**[00:17:58 – 00:18:01]** on whether things are centralized or decentralized, and you go

**[00:18:01 – 00:18:05]** back and forth, you centralize, you decentralize, and then, and right

**[00:18:05 – 00:18:11]** now, we're looking when we look at mature banks, they will have gone

**[00:18:11 – 00:18:15]** through a phase of centralization, and then they decentralize, and

**[00:18:15 – 00:18:20]** they're saying, okay, all of my 10 to 18, whatever it might be, lines

**[00:18:20 – 00:18:24]** of business, depending on what bank you are, the responsibility

**[00:18:24 – 00:18:28]** for you, head of capital markets, or you, head of retail, or

**[00:18:28 – 00:18:35]** you, head of wealth management, to implement and figure out

**[00:18:35 – 00:18:43]** the KPIs for, for impact, when I am using AI, that's on you, right?

**[00:18:43 – 00:18:46]** And typically what that looks like is that you've got some

**[00:18:46 – 00:18:50]** kind of a two-in-a-box system where you've got maybe an

**[00:18:50 – 00:18:53]** applied AI team, you've got your engineers, but they're, they're

**[00:18:53 – 00:18:57]** in your line of business because you bring them down close to

**[00:18:57 – 00:18:59]** the actual work that is happening.

**[00:18:59 – 00:19:01]** And so you've got to map, you know, and I'll get to the workflow

**[00:19:02 – 00:19:04]** a bit in a minute, but you've got to then map out, you know, what's

**[00:19:04 – 00:19:07]** back office, what's middle office, what's front office, what's the

**[00:19:07 – 00:19:11]** day in the life of a wealth manager or an investment banker, and

**[00:19:11 – 00:19:18]** then make sure that the data that's in the bank, is it at the disposal

**[00:19:18 – 00:19:21]** to that person that is needed?

**[00:19:21 – 00:19:23]** How can we be more seamless?

**[00:19:23 – 00:19:28]** And so that is where, when we're looking at workflow

**[00:19:28 – 00:19:32]** changes, is someone going in and saying, okay, for this

**[00:19:32 – 00:19:38]** banker, there's 42 steps, and, you know, 12 of that can be automated,

**[00:19:38 – 00:19:43]** or we can drive that information much faster so, you know,

**[00:19:43 – 00:19:46]** banker Lucy doesn't have to go to Bob and, you know, Christian to get

**[00:19:46 – 00:19:47]** all of this, but it's all there.

**[00:19:47 – 00:19:49]** This is obviously stating the obvious, but when we're

**[00:19:49 – 00:19:52]** looking at table stakes, you know, you're looking at banks

**[00:19:52 – 00:19:56]** that are saying, yes, we've got AI productivity tools rolled out

**[00:19:56 – 00:19:59]** to sort of part of the bank, and we've got developer productivity

**[00:19:59 – 00:20:01]** tools, and that's great, and we're seeing 20% efficiency

**[00:20:01 – 00:20:03]** gains, or what it might be.

**[00:20:03 – 00:20:05]** That is what we would probably consider, you know, Kevin

**[00:20:05 – 00:20:08]** and I, you and I spend a lot of time parsing every single

**[00:20:08 – 00:20:10]** use case, right, in the banks.

**[00:20:10 – 00:20:12]** We would say that probably belongs to table stakes.

**[00:20:12 – 00:20:16]** Now it's a question of, is that tooling, you know, being designed

**[00:20:16 – 00:20:18]** specifically for a particular team?

**[00:20:18 – 00:20:20]** And then we look at, like, are you getting into the

**[00:20:20 – 00:20:24]** trenches of the workflow?

**[00:20:24 – 00:20:26]** And that's where workflow transformation, I mean, it's

**[00:20:26 – 00:20:30]** such a buzzword, but it actually really matters that you pull

**[00:20:30 – 00:20:31]** your team out, right?

**[00:20:31 – 00:20:35]** Think about how this is happening right now, and, like, how

**[00:20:35 – 00:20:37]** can I completely re-imagine this?

**[00:20:37 – 00:20:39]** Or can I? With some of it, you can't, right?

**[00:20:39 – 00:20:42]** Some of it, you just have to say, no, that's good.

**[00:20:42 – 00:20:47]** But of the 42 steps, maybe I can get to 20 that are rejigged

**[00:20:47 – 00:20:50]** or collapsed or, you know, we see, you know, and it's,

**[00:20:50 – 00:20:54]** it's like the boring use cases or areas of the bank are actually

**[00:20:54 – 00:20:56]** the most exciting.

**[00:20:56 – 00:20:59]** You know, get your KYC stuff fixed, right?

**[00:20:59 – 00:21:06]** If you've got $10 billion or euros every year, which the big banks do,

**[00:21:06 – 00:21:10]** waiting to come into the bank, and it takes you six months to onboard

**[00:21:10 – 00:21:12]** the clients, I mean, you've just got money sitting on the outside.

**[00:21:12 – 00:21:15]** If you can actually get that onboarding time down to six

**[00:21:15 – 00:21:19]** days, that's an immediate revenue uplift for any bank.

**[00:21:19 – 00:21:22]** So those are the sort of obvious, and it's hard to do.

**[00:21:22 – 00:21:26]** KYC processes sound easy; it's really hard to do, right?

**[00:21:26 – 00:21:27]** And every line of business will have different

**[00:21:27 – 00:21:29]** onboarding systems, right?

**[00:21:29 – 00:21:33]** So it's not KYC for a bank; there's lots of different kyc.

**[00:21:33 – 00:21:36]** So, so I would say, you know, to answer your question, back

**[00:21:36 – 00:21:39]** to the workflow, when you see, you know, that, that

**[00:21:39 – 00:21:43]** when you're sort of really seeing workflow changes and the type of

**[00:21:43 – 00:21:49]** kpis that are being tracked, as in how much more can I handle, right?

**[00:21:49 – 00:21:52]** And so we will probably get into the, you know, impact

**[00:21:52 – 00:21:56]** on head count, but often that, what, what we see is that there's

**[00:21:56 – 00:21:58]** so much backlog in banks, right?

**[00:21:58 – 00:22:01]** So the sort of transformation is not only just, like, the

**[00:22:01 – 00:22:05]** workflow transformation, but it's also, like, I can deal

**[00:22:05 – 00:22:07]** with backlog that's sitting there and I can actually get

**[00:22:07 – 00:22:10]** to it, or there's a problem over there that I just couldn't

**[00:22:10 – 00:22:13]** tackle, but now, because I've got the tools and I've probably

**[00:22:13 – 00:22:16]** also got the people, because we've onboarded a lot of clever talent

**[00:22:17 – 00:22:20]** here, I can tackle a problem that I just never thought I could tackle.

**[00:22:20 – 00:22:24]** So there's a lot of that we're seeing going on too.

**[00:22:24 – 00:22:28]** Let me add one thing here, Kevin, if you don't mind,

**[00:22:28 – 00:22:32]** because this is such a fundamental piece, and I'm reading some

**[00:22:32 – 00:22:36]** of the questions on the Q&A, and I feel that the audience really wants

**[00:22:36 – 00:22:40]** to hear about what is different, what the leaders are doing,

**[00:22:40 – 00:22:42]** and what is the path to success.

**[00:22:42 – 00:22:44]** And so I think it's fundamental that we'll really get

**[00:22:45 – 00:22:46]** deeper into this piece.

**[00:22:46 – 00:22:51]** The way I will describe it is, the tech stack or the

**[00:22:51 – 00:22:53]** banking stack is changing, right?

**[00:22:53 – 00:22:56]** It used to be, we have the hardware at the bottom that creates

**[00:22:57 – 00:22:58]** from data sets those data sets.

**[00:22:58 – 00:23:01]** Now, a lot of them in a lot of silos are connected to

**[00:23:01 – 00:23:05]** some software applications, and then we have our teams on top

**[00:23:05 – 00:23:08]** using those software applications, and it's all very vertical,

**[00:23:08 – 00:23:11]** very narrow, which creates most of the challenges we have within

**[00:23:11 – 00:23:15]** a bank, is that reason, because application one and application

**[00:23:15 – 00:23:17]** two don't talk to each other because they are connected to data

**[00:23:17 – 00:23:19]** sets that don't talk to each other.

**[00:23:19 – 00:23:22]** And therefore, our people in the call centers, every time they

**[00:23:22 – 00:23:25]** need to do anything, they need to use 18 different tools and screens

**[00:23:25 – 00:23:27]** and copy and paste 27 times, right?

**[00:23:27 – 00:23:30]** So that is the today.

**[00:23:30 – 00:23:33]** The stack of the future, you're still gonna have your underwriting

**[00:23:33 – 00:23:37]** capability, you're gonna have your data, but what we are saying

**[00:23:37 – 00:23:41]** is you need to start thinking about building an intelligence layer that

**[00:23:41 – 00:23:43]** is gonna sit on top of the data.

**[00:23:43 – 00:23:47]** And that intelligence layer is your bank intelligence, right?

**[00:23:47 – 00:23:49]** And when we talk about these banking foundation models,

**[00:23:49 – 00:23:50]** that is one of them.

**[00:23:50 – 00:23:52]** You will have many. You will not have just a single one of them.

**[00:23:52 – 00:23:54]** You will have multiple intelligence, right?

**[00:23:54 – 00:23:59]** You don't want your intelligence that is focused on customer

**[00:23:59 – 00:24:01]** actions, right, to be the same that you have to focus on risk

**[00:24:01 – 00:24:04]** compliance and internal processes.

**[00:24:04 – 00:24:06]** But you need that layer.

**[00:24:06 – 00:24:08]** And then what you're going to be doing is interacting with that

**[00:24:08 – 00:24:12]** layer and rethinking your processes by bringing that intelligence

**[00:24:12 – 00:24:14]** into what you're doing.

**[00:24:14 – 00:24:20]** If you follow that analogy, what happens is, today, the temptation

**[00:24:20 – 00:24:23]** is very high to go top-down.

**[00:24:23 – 00:24:26]** By the way, in the industry, we have a long tradition

**[00:24:26 – 00:24:27]** of going top-down.

**[00:24:27 – 00:24:30]** A new technology comes, we don't want to fix anything.

**[00:24:30 – 00:24:34]** What we do is we put a layer on top, we make it very pretty,

**[00:24:34 – 00:24:36]** and then we figure it out how to make it work with the

**[00:24:36 – 00:24:39]** mess we have underneath, right?

**[00:24:39 – 00:24:42]** And a lot of the websites of banks are based on that.

**[00:24:43 – 00:24:46]** A lot of, and you know, you're being in the industry as long

**[00:24:46 – 00:24:47]** as we're big, right?

**[00:24:47 – 00:24:50]** Did you know that, yes, the website looks great, but actually

**[00:24:50 – 00:24:53]** that website is communicating to 27 different systems and,

**[00:24:53 – 00:24:57]** you know, half of the data is delaying all those challenges.

**[00:24:58 – 00:25:00]** The other way to thinking about this is, no, no, I'm actually

**[00:25:00 – 00:25:03]** going to really start bottom-up.

**[00:25:03 – 00:25:05]** I'm going to build those intelligence layers and transform

**[00:25:05 – 00:25:10]** the processes, like going to the workflows, as Alexandra is saying.

**[00:25:10 – 00:25:11]** You need to do that.

**[00:25:11 – 00:25:12]** I'm going to give you one example.

**[00:25:12 – 00:25:16]** Someone was asking me less than one week ago, you know,

**[00:25:16 – 00:25:18]** because our system is a mess, what we're thinking about doing is we're

**[00:25:18 – 00:25:21]** going to create an agent that is a generalist agent, and we're going

**[00:25:21 – 00:25:22]** to put it on top of everything.

**[00:25:22 – 00:25:25]** We're going to give them access to all the systems, and then

**[00:25:25 – 00:25:27]** we ask the agent a question, and the agent actually gives

**[00:25:27 – 00:25:30]** me an answer, and problem solved.

**[00:25:30 – 00:25:34]** And I was looking and saying, do you realize that you will

**[00:25:34 – 00:25:36]** have no way of verifying that agent actually knows what

**[00:25:36 – 00:25:38]** he's answering about?

**[00:25:38 – 00:25:40]** Because right now in your systems, the same data is

**[00:25:40 – 00:25:43]** in 27 places, and those data will be different, right?

**[00:25:43 – 00:25:46]** You have no way of auditing what is happening, and the

**[00:25:46 – 00:25:47]** only thing you're doing again, you're creating that layer

**[00:25:48 – 00:25:51]** on top, but you're not fixing the underlying problem, right?

**[00:25:51 – 00:25:55]** While if you start the other way around, you create the intelligence

**[00:25:55 – 00:25:59]** layer, and now you create the agent, you can have full certainty,

**[00:25:59 – 00:26:02]** the answers are right, you know, you're getting to the right place

**[00:26:02 – 00:26:04]** and all these different pieces.

**[00:26:04 – 00:26:07]** It is work, yes, it is work, but it's absolutely work, right,

**[00:26:07 – 00:26:09]** because the returns are massive.

**[00:26:09 – 00:26:11]** If you start bottoms up and you build that intelligence layer and

**[00:26:11 – 00:26:16]** you fix fraud or you dramatically improve fraud, as many use cases

**[00:26:16 – 00:26:23]** have come up to the public over the last weeks, it pays for itself.

**[00:26:23 – 00:26:26]** Whatever investment you need to make to do that, it pays back

**[00:26:26 – 00:26:28]** in two months, in three months.

**[00:26:28 – 00:26:30]** It's incredibly fast.

**[00:26:30 – 00:26:35]** So, suggesting for the audience, for all of you in the industry

**[00:26:35 – 00:26:38]** going through this, don't try to solve everything.

**[00:26:38 – 00:26:42]** Just pick that workflow, pick that use case that really

**[00:26:42 – 00:26:44]** matters to you, and think about what intelligence layer

**[00:26:44 – 00:26:49]** you're gonna need to be able to do it, and start there.

**[00:26:50 – 00:26:53]** You don't need massive teams to do these things, right?

**[00:26:53 – 00:26:56]** In Revolut Metapublic, they work with three or four data scientists

**[00:26:56 – 00:26:59]** to build this intelligence layer.

**[00:26:59 – 00:27:00]** You don't need massive teams to do this.

**[00:27:00 – 00:27:02]** That's the other piece, right? Like in banking, we have this

**[00:27:02 – 00:27:05]** tendency to believe that everything is a three-year project that

**[00:27:05 – 00:27:07]** requires 100 people.

**[00:27:07 – 00:27:10]** You can do these things in a matter of months, but, like,

**[00:27:10 – 00:27:14]** three, four months with a small team of talented people.

**[00:27:14 – 00:27:19]** So carrying back to you, I think this is the key aspect of it.

**[00:27:19 – 00:27:22]** Yeah, to play the skeptic for a minute, I mean, Revolut is

**[00:27:22 – 00:27:25]** a bank that's 11 years old, right?

**[00:27:25 – 00:27:27]** It was founded in 2015.

**[00:27:27 – 00:27:31]** A lot of the banks out there don't have the benefit of having

**[00:27:31 – 00:27:34]** data that's probably quite as clean as Revolut was when it started.

**[00:27:34 – 00:27:37]** I guess you're saying it doesn't take a large team to do this,

**[00:27:37 – 00:27:41]** but if you are a bank that has siloed data or has those kind

**[00:27:41 – 00:27:45]** of places that exist, you know, 27 different places, I mean, that

**[00:27:45 – 00:27:49]** does feel like a bigger kind of undertaking that you would need to.

**[00:27:50 – 00:27:52]** Seriously, we couldn't have done this better if I had

**[00:27:52 – 00:27:55]** planted the question from you, because I don't know if

**[00:27:55 – 00:27:59]** you saw it, like three months after Revolut announced the model, and I

**[00:27:59 – 00:28:03]** can share this because they made it public, Lloyd's, which is 261 years

**[00:28:03 – 00:28:08]** old, and I know because I check it so I can answer this question

**[00:28:08 – 00:28:12]** when people ask this question, they did the same thing in four

**[00:28:12 – 00:28:14]** months, and they put it out there.

**[00:28:14 – 00:28:17]** And, and I think in the blog, the tablets, right?

**[00:28:17 – 00:28:21]** I think they call it autumn, uh the model they launched.

**[00:28:22 – 00:28:25]** Did they use as much data as Revolut did?

**[00:28:25 – 00:28:26]** I don't think they did.

**[00:28:26 – 00:28:29]** Revolut used, difference that I know are true, Revolut used

**[00:28:29 – 00:28:33]** three encoders to build the model, Lloyd's used two.

**[00:28:33 – 00:28:36]** But I think the big learning is, even if things are not

**[00:28:37 – 00:28:40]** perfect, start and do it, right?

**[00:28:40 – 00:28:44]** Because even a model that is just 80% of what it could

**[00:28:44 – 00:28:48]** be possible is gonna be a lot better than half nothing.

**[00:28:48 – 00:28:52]** And that is, again, you're talking about an incredible

**[00:28:52 – 00:28:54]** institution, 261 years of history.

**[00:28:54 – 00:28:57]** It's been around for a long time, and I'm pretty sure

**[00:28:57 – 00:29:00]** they have a good legacy set of elements out there, because

**[00:29:00 – 00:29:04]** every bank does, as long as you're more than five years old.

**[00:29:04 – 00:29:07]** And all of them are multi-hundred years old, right?

**[00:29:07 – 00:29:08]** And they figured it out.

**[00:29:08 – 00:29:10]** So you have both sides of the coin.

**[00:29:10 – 00:29:12]** And by the way, you know JP Morgan is doing these things,

**[00:29:12 – 00:29:15]** and Capital One is doing these things, and they are all more

**[00:29:15 – 00:29:16]** traditional players, right?

**[00:29:16 – 00:29:18]** So it's not just two examples.

**[00:29:18 – 00:29:21]** It's two examples that are very public, and they've been very

**[00:29:21 – 00:29:26]** vocal about what they're doing, and they create the perfect comparison.

**[00:29:26 – 00:29:29]** Alexandra, I want to go back to something that Aseer said a couple

**[00:29:29 – 00:29:33]** of minutes ago with the kind of top-down versus bottom-up approach.

**[00:29:33 – 00:29:39]** And I think the cutaway from that is, you know, the bottom-up

**[00:29:39 – 00:29:42]** approach takes a bit more work, takes a bit more time.

**[00:29:42 – 00:29:44]** It sets you up in the long run.

**[00:29:44 – 00:29:47]** The top-down approach at the moment probably looks like it's

**[00:29:47 – 00:29:51]** working to some extent because, quite frankly, the models are, you

**[00:29:51 – 00:29:54]** know, advancing in what they can do, and, you know, you're able to

**[00:29:54 – 00:29:59]** get some sort of a, I guess, when you think about it, you see banks

**[00:30:00 – 00:30:03]** take both approaches, whether, you know, how much they're investing

**[00:30:03 – 00:30:05]** versus how much they're not.

**[00:30:05 – 00:30:09]** And I guess when you think about the trade-off there of taking one

**[00:30:09 – 00:30:14]** approach versus the other, do you think the bigger risk is kind of

**[00:30:14 – 00:30:20]** missing out on the ultimate benefit that this could have if you take

**[00:30:20 – 00:30:22]** the longer road and go bottom up?

**[00:30:22 – 00:30:27]** Or is it getting left behind by not doing enough, even from the

**[00:30:27 – 00:30:31]** top-down perspective immediately?

**[00:30:32 – 00:30:36]** I mean, I think the issue is, if you don't, to Asa's point,

**[00:30:36 – 00:30:42]** I think, unfortunately, we agree a lot, but if you don't do the

**[00:30:42 – 00:30:46]** work, it's going to be very hard to get the full-on benefits long term.

**[00:30:46 – 00:30:50]** When we're looking at, for example, there's a lot of questions

**[00:30:50 – 00:30:54]** uh about, you know, the cost and token cost and overspending

**[00:30:54 – 00:30:58]** inadvertently on, because there's someone somewhere in the bank

**[00:30:58 – 00:31:02]** that, that sort of got overexcited and used, you know, a system and

**[00:31:02 – 00:31:05]** then ended up with a massive bill.

**[00:31:05 – 00:31:08]** But if you've built your platform, right, and you've got controls

**[00:31:08 – 00:31:12]** in, and you've got dynamic model routing, and you've got those

**[00:31:12 – 00:31:15]** systems, which is no easy thing to build, by the way, but, like,

**[00:31:15 – 00:31:20]** taking the time to actually do that, you know, you're optimizing,

**[00:31:20 – 00:31:24]** you know, the choice of model in terms of both cost and capability.

**[00:31:24 – 00:31:28]** And what we find is that it's very often not necessarily

**[00:31:28 – 00:31:32]** the newest and most capable model that is necessary.

**[00:31:32 – 00:31:34]** You know, more often than not, it's, like, an old model,

**[00:31:34 – 00:31:38]** an open weight model, or something like that that you, or small

**[00:31:38 – 00:31:40]** language, for example.

**[00:31:40 – 00:31:43]** So there's a lot of that, like, get the work done so one can do that.

**[00:31:43 – 00:31:47]** And then it is also, it's very hard to go up that hockey

**[00:31:47 – 00:31:52]** stick if one hasn't got the controls around the models,

**[00:31:52 – 00:31:55]** Gen-AI and the agents, because then you actually can't let it, let

**[00:31:55 – 00:32:00]** it go and work because if you don't build the harness or the controls,

**[00:32:00 – 00:32:02]** you know, it's just, you're sort of kind of standing still.

**[00:32:02 – 00:32:06]** So you also need to do the hard graft of doing, you know, you

**[00:32:06 – 00:32:10]** know, you know, try and automate the governance, and, and so on.

**[00:32:10 – 00:32:15]** So, so that work has to be done in order to speed up,

**[00:32:15 – 00:32:19]** and so that there is, there is, there is no way around it.

**[00:32:19 – 00:32:22]** And sort of one of the other things that you see is common

**[00:32:22 – 00:32:25]** denominators of the leading banks, not all of them, but

**[00:32:25 – 00:32:29]** mostly, is, is that they started a long time ago, right?

**[00:32:29 – 00:32:32]** Jamie Dimon went out in 2017 famously and said, we're an

**[00:32:32 – 00:32:35]** AI organization, we're AI first.

**[00:32:35 – 00:32:41]** And then he hired Manuela Veloso to run that AI lab that was central.

**[00:32:41 – 00:32:42]** It was a centralized brain.

**[00:32:42 – 00:32:47]** Like, what are all of the things that might be things

**[00:32:47 – 00:32:49]** that we need to tackle two to three years from now, is why

**[00:32:49 – 00:32:52]** he set it up in 2017 or 18, right?

**[00:32:52 – 00:32:57]** Super prepared for really having done a lot of the hard work and

**[00:32:57 – 00:33:00]** a lot of the thinking and a lot of the thinking about the technology.

**[00:33:00 – 00:33:04]** So when you have chat GPT suddenly come online in November

**[00:33:04 – 00:33:07]** 22, you're sort of better prepared to think about this

**[00:33:08 – 00:33:12]** technology that, prior to that, was maybe touching 30% of the bank.

**[00:33:12 – 00:33:15]** Now you've got this technology touching 100% of the bank.

**[00:33:15 – 00:33:17]** How do we think about that?

**[00:33:17 – 00:33:20]** And you're much more prepared for that and the work that

**[00:33:20 – 00:33:23]** needs to be done versus those that maybe started a lot later.

**[00:33:23 – 00:33:25]** And not to say that one can't catch up.

**[00:33:26 – 00:33:27]** Of course one can catch up, right?

**[00:33:27 – 00:33:31]** But it is, you just have that advantage from doing that

**[00:33:31 – 00:33:33]** thinking and the hard work.

**[00:33:33 – 00:33:37]** NVIDIA AI, NVIDIA AI, NVIDIA AI, NVIDIA AI, NVIDIA

**[00:33:57 – 00:34:01]** I actually am not in the proximity of deep talent pools, and

**[00:34:01 – 00:34:04]** so, in order to move fast, I'm going to have to partner.

**[00:34:04 – 00:34:06]** And it makes a ton of sense for a lot of banks to partner,

**[00:34:06 – 00:34:09]** right, on certain things.

**[00:34:09 – 00:34:12]** Anyway, but that system of build, partner, like what

**[00:34:12 – 00:34:16]** do we do, you know, diversify, we don't want to have vendor lock-in,

**[00:34:17 – 00:34:21]** all of that is constantly being debated at a group level, but also

**[00:34:21 – 00:34:24]** down in the lines of business, and it's a little hard to say exactly

**[00:34:24 – 00:34:28]** where it lands, you know, entirely because different banks will

**[00:34:28 – 00:34:32]** pursue different routes, for sure.

**[00:34:32 – 00:34:35]** But yeah, but I also see that there is that question, and

**[00:34:35 – 00:34:37]** I might actually want to throw that over to you, Asa, about

**[00:34:37 – 00:34:39]** the model selection.

**[00:34:39 – 00:34:43]** Sorry, I'm stepping in on your territory, Kevin, but

**[00:34:44 – 00:34:48]** we hear that like maybe a year ago, 18 months ago, there's

**[00:34:48 – 00:34:52]** like 80% of the models that were, you know, chosen for, you know,

**[00:34:52 – 00:34:57]** problem solving and implementation were commercial closed models

**[00:34:58 – 00:35:01]** and 20% were open weight models, and now it seems to flip.

**[00:35:01 – 00:35:04]** Some banks still don't really want to use open weight models

**[00:35:05 – 00:35:08]** as much, but generally speaking for the sector, that's more

**[00:35:08 – 00:35:12]** now 2080, i.e., many more are using open-weight models

**[00:35:12 – 00:35:15]** and fewer are using the commercial.

**[00:35:15 – 00:35:18]** Is that what you see, too?

**[00:35:18 – 00:35:19]** Yes, I mean, 100%.

**[00:35:19 – 00:35:24]** And within, at NVIDIA level, we've been incredibly vocal

**[00:35:24 – 00:35:27]** since the beginning, right, that is, the future is hybrid

**[00:35:27 – 00:35:32]** and you're always going to need to use a combination of things.

**[00:35:32 – 00:35:36]** My personal way of explaining this is always the same, that

**[00:35:36 – 00:35:39]** is, listen, you don't want to have a garage that actually has just

**[00:35:39 – 00:35:43]** a Ferrari, right, because as fun as they are, and you could argue the

**[00:35:43 – 00:35:46]** best car you can have, a thousand horsepowers and does everything

**[00:35:46 – 00:35:51]** incredibly great, it is not the right tool to go to buy things

**[00:35:51 – 00:35:55]** from the store or to take your four kids to the school, right,

**[00:35:55 – 00:35:58]** you need to do like four trips to take four kids to the school, and

**[00:35:58 – 00:36:02]** it would be a lot slower than just taking it on the Odyssey, right?

**[00:36:02 – 00:36:04]** So you need the right tool for the right job.

**[00:36:04 – 00:36:07]** And with the model, it's exactly the same thing.

**[00:36:07 – 00:36:10]** And, you know, there's plenty of research that will tell you

**[00:36:10 – 00:36:14]** for very specific tasks; go back to my comment at the beginning.

**[00:36:14 – 00:36:18]** The banks have data that no frontier model ever had.

**[00:36:18 – 00:36:22]** The banks know how credit markets in the U.S., in Canada,

**[00:36:22 – 00:36:27]** in Europe have acted and performed over the last 30 years.

**[00:36:27 – 00:36:28]** That is not available information.

**[00:36:28 – 00:36:31]** They are the only ones that understand how their consumers,

**[00:36:31 – 00:36:36]** how their businesses represent risk and what is possible or not.

**[00:36:36 – 00:36:39]** Models you create with that information, they will outperform

**[00:36:39 – 00:36:41]** anything that is out there, right?

**[00:36:41 – 00:36:44]** Always, right? Because again, the data is the secret sauce.

**[00:36:44 – 00:36:47]** The model is a little bit of a commodity to some point.

**[00:36:47 – 00:36:50]** And you listen to the voices of most of the true experts

**[00:36:51 – 00:36:52]** in AI, they're really the big, big names, they will

**[00:36:52 – 00:36:56]** tell you, you know, the model's great, and they keep apping each

**[00:36:56 – 00:36:59]** other a little bit better at each other, but that is a path where

**[00:36:59 – 00:37:00]** it's a little bit of a commodity.

**[00:37:00 – 00:37:04]** They are very similar, all of that; the data is what

**[00:37:04 – 00:37:05]** is going to bring the richness.

**[00:37:05 – 00:37:09]** So we are seeing it the same path you see every, I think

**[00:37:09 – 00:37:11]** the number is just from the beginning of the year till

**[00:37:11 – 00:37:16]** now, it has flipped; it was mostly closed models and tokens and then

**[00:37:16 – 00:37:20]** very few open, and now the model is the opposite; it's mostly open way

**[00:37:20 – 00:37:24]** models, closed models a bit less; again, the overall growth has been

**[00:37:24 – 00:37:26]** massive in both, which is great.

**[00:37:26 – 00:37:27]** I think it's demonstrating that point.

**[00:37:27 – 00:37:30]** You need to have the right tool for the job, right?

**[00:37:30 – 00:37:34]** Yeah. There are banks in Europe now launching incredibly

**[00:37:34 – 00:37:37]** small models that actually can live in your phone.

**[00:37:37 – 00:37:39]** So you can have your phone and you have the financial

**[00:37:39 – 00:37:43]** app in your phone, running in your phone, connected to the bank.

**[00:37:43 – 00:37:45]** Yeah. That is an incredible idea.

**[00:37:45 – 00:37:48]** So we are very excited about all this that's happening.

**[00:37:48 – 00:37:51]** There's a question from Ram Kumar.

**[00:37:51 – 00:37:55]** I don't know, sorry, Kevin, keep going, but there's a

**[00:37:55 – 00:37:59]** very interesting one on exactly that in the chat.

**[00:37:59 – 00:38:01]** But back to you, Kevin, in the studio.

**[00:38:01 – 00:38:03]** We'll kick it to that one next.

**[00:38:03 – 00:38:06]** I was just gonna ask you, Sarah, I mean, you said the

**[00:38:06 – 00:38:07]** future is hybrid.

**[00:38:07 – 00:38:09]** You talk to a lot of banks.

**[00:38:09 – 00:38:12]** What percentage of them is still kind of driving that

**[00:38:12 – 00:38:14]** Ferrari more often than not?

**[00:38:14 – 00:38:18]** You know, how far along are banks in getting to that hybrid future?

**[00:38:18 – 00:38:24]** I think that is an area where you see, and there was a question

**[00:38:24 – 00:38:27]** in the chat about how we define leaders, let's agree, Alexandra

**[00:38:27 – 00:38:29]** told me you disagree, but like the top 10 in your index,

**[00:38:29 – 00:38:31]** let's call those the leaders.

**[00:38:31 – 00:38:36]** We call everybody below that maybe fast followers, right?

**[00:38:36 – 00:38:40]** But you know, I think the more you think about the leaders,

**[00:38:40 – 00:38:43]** the more you see about they building their own intelligence

**[00:38:43 – 00:38:46]** layer, they building their own models, and again, they use

**[00:38:46 – 00:38:51]** closed models a lot too, right, but that is becoming more prevalent.

**[00:38:51 – 00:38:55]** The more you go down the rank, you still hear, and Alexandra

**[00:38:55 – 00:38:59]** touched on this a little bit at the beginning, that is

**[00:38:59 – 00:39:02]** because talent may be a little bit more of a scarce resource.

**[00:39:02 – 00:39:05]** It may be a faster path to prove a point.

**[00:39:05 – 00:39:08]** You still see that the dependency on closed models is a little bit

**[00:39:08 – 00:39:10]** higher on the bottom of the list.

**[00:39:11 – 00:39:13]** But someone was saying in the forum right now, hey,

**[00:39:13 – 00:39:15]** it seems that the future is open.

**[00:39:15 – 00:39:18]** We strongly believe that, believe that, I mean, the future is

**[00:39:18 – 00:39:23]** both, but open has a massive role to play because I'm now talking

**[00:39:23 – 00:39:27]** just about the financial services industry, the banking industry,

**[00:39:27 – 00:39:31]** and data is the blood that actually makes everything work in a bank.

**[00:39:32 – 00:39:35]** Banks are going to protect that, and the way to protect

**[00:39:35 – 00:39:37]** that is building your intelligence with your own data and keeping

**[00:39:37 – 00:39:43]** that in your own house, right?

**[00:39:43 – 00:39:47]** So can I just ask you a question on that?

**[00:39:47 – 00:39:51]** If the model ends up becoming, you know, it's not what

**[00:39:51 – 00:39:52]** you pay for, right?

**[00:39:52 – 00:39:55]** Your financial services, like,

**[00:39:55 – 00:39:57]** so when you're looking at the frontier labs, and I'm

**[00:39:57 – 00:40:01]** looking at sort of, obviously we're sort of gearing up for,

**[00:40:01 – 00:40:03]** they are gearing up for an IPO.

**[00:40:03 – 00:40:08]** So the revenue stream is still very strong based on sort

**[00:40:08 – 00:40:09]** of a mix of things.

**[00:40:09 – 00:40:13]** But if you sort of look a year out, what is it that

**[00:40:13 – 00:40:18]** We, not we, but being the banks, will be paying for when you're

**[00:40:18 – 00:40:20]** working with open AI on Swapik?

**[00:40:20 – 00:40:23]** If it's not the model itself, it is the service, it is therefore

**[00:40:23 – 00:40:30]** deployed engineers that help organize the rails for which

**[00:40:30 – 00:40:31]** everything runs on.

**[00:40:31 – 00:40:35]** What do you think that looks like a year from now?

**[00:40:36 – 00:40:37]** I don't know what it's going to look like from a revenue

**[00:40:37 – 00:40:41]** point of view, and I don't think we should be commenting on that.

**[00:40:41 – 00:40:46]** What I can tell you, what I'm seeing is, when banks

**[00:40:46 – 00:40:48]** think about AI, they're gonna be thinking about like three,

**[00:40:48 – 00:40:50]** four different elements.

**[00:40:50 – 00:40:57]** One is they're gonna be thinking about they need to secure compute.

**[00:40:57 – 00:40:59]** And by the way, this is not like people always say, oh, you're

**[00:40:59 – 00:41:01]** saying they need to go on-prem.

**[00:41:01 – 00:41:04]** I didn't mean that. I mean, they need to secure compute.

**[00:41:04 – 00:41:08]** We have a lot of, some of the leading banks are fully cloud-based

**[00:41:08 – 00:41:09]** and there's public information.

**[00:41:09 – 00:41:11]** Capital One is fully cloud-based.

**[00:41:11 – 00:41:14]** Capital One has secure compute, has thousands of GPUs; they

**[00:41:14 – 00:41:18]** know how to use them, and they have the arrays, right?

**[00:41:18 – 00:41:19]** So a bank needs to secure compute.

**[00:41:19 – 00:41:22]** Whatever infrastructure they pick is fine, but they need

**[00:41:22 – 00:41:24]** to secure compute.

**[00:41:24 – 00:41:26]** They're going to be building a lot of intelligence, so

**[00:41:26 – 00:41:28]** they're going to be using that compute to run their own

**[00:41:28 – 00:41:31]** intelligence, and that's something that they keep under control.

**[00:41:31 – 00:41:36]** And they're going to be using closed models for very interesting

**[00:41:36 – 00:41:39]** use cases where it actually makes all the sense.

**[00:41:39 – 00:41:43]** One place that is very common knowledge is coding.

**[00:41:43 – 00:41:47]** Coding is not a place where the data of the bank, the

**[00:41:47 – 00:41:49]** IP of the bank is encoding.

**[00:41:49 – 00:41:52]** It's more of a generalist tool.

**[00:41:52 – 00:41:55]** So those models are doing a fantastic job there.

**[00:41:55 – 00:41:59]** And there is a lot of roles within any enterprise where

**[00:41:59 – 00:42:02]** a frontier model does a fantastic job, and it could be an easier

**[00:42:02 – 00:42:02]** path to get there.

**[00:42:02 – 00:42:05]** You need to do marketing copy, you need a lot of things.

**[00:42:05 – 00:42:08]** Again, the IP of a bank is not marketing copy.

**[00:42:08 – 00:42:10]** Even when some banks would believe that the marketing

**[00:42:10 – 00:42:13]** is the best in the planet, it's not the IP of the bank.

**[00:42:13 – 00:42:16]** It's not what is going to make a bank different versus the others.

**[00:42:16 – 00:42:19]** So I think that a little bit of rationalization between,

**[00:42:19 – 00:42:24]** I always use the NVIDIA example, how we internally think about

**[00:42:24 – 00:42:31]** this is if something, it's more general in nature, it's

**[00:42:31 – 00:42:35]** more of a standard thing.

**[00:42:35 – 00:42:41]** Any partner or any closed model is a fantastic tool.

**[00:42:41 – 00:42:43]** For things that come down to the essence of what we

**[00:42:43 – 00:42:48]** do as a company, cheap design, managing our supply chains,

**[00:42:48 – 00:42:53]** for all these things, we build a lot of intelligence in-house.

**[00:42:53 – 00:42:56]** And I think that framework of preference, like what is

**[00:42:56 – 00:42:59]** connected to your IP, what is connected to what makes

**[00:42:59 – 00:43:03]** you who you are, versus what is more of a standard job and made

**[00:43:03 – 00:43:04]** at this distance, it really helps.

**[00:43:04 – 00:43:07]** And building on what you said before, by the way,

**[00:43:07 – 00:43:10]** Alessandro, there's a lot of great companies in this space.

**[00:43:10 – 00:43:12]** I mean, ServiceNow is an accelerated, it's a company

**[00:43:12 – 00:43:15]** that do a great job, they do a lot of roles, they're incredibly big in

**[00:43:15 – 00:43:18]** banking, and it's a great partner of NVIDIA, a fully accelerated

**[00:43:18 – 00:43:21]** company, right, and there are many examples, like CrowdStrike,

**[00:43:21 – 00:43:24]** there are many examples like that; you can see all their

**[00:43:24 – 00:43:26]** logos in all our announcements.

**[00:43:26 – 00:43:28]** There's a lot of great ways to just benefit from this

**[00:43:28 – 00:43:33]** revolution without having to do everything yourself.

**[00:43:33 – 00:43:36]** Yeah, for sure.

**[00:43:36 – 00:43:39]** Sarah, you just mentioned kind of the infrastructure

**[00:43:39 – 00:43:40]** is what leads to the scale.

**[00:43:40 – 00:43:43]** The scale, I think we have some people in the chat asking about,

**[00:43:43 – 00:43:45]** you know, how that leads to ROI.

**[00:43:45 – 00:43:50]** I mean, Alexandra, I know AI, ROI is one of your favorite

**[00:43:50 – 00:43:51]** things to think about.

**[00:43:51 – 00:43:55]** And I'm curious when you think about it in the context of

**[00:43:55 – 00:44:00]** What those enabling factors are, where do you see the infrastructure

**[00:44:00 – 00:44:05]** fit versus everything else that the IndeX measures and tells

**[00:44:05 – 00:44:10]** you what's good and what's leading?

**[00:44:10 – 00:44:14]** Yeah, the ROI, I mean, it's on everyone's mind all the

**[00:44:14 – 00:44:20]** time, and it is, you know, I want to not say it's elusive

**[00:44:20 – 00:44:24]** because I don't actually believe that it is long-term, right?

**[00:44:24 – 00:44:29]** But it is, there's a lot of questions around, you know, heavy

**[00:44:29 – 00:44:35]** investments into everything that is needed in order to deploy AI, and

**[00:44:35 – 00:44:37]** when are those returns coming back?

**[00:44:37 – 00:44:41]** So CFOs are having, you know, very much that are lifting

**[00:44:41 – 00:44:44]** this sort of sticker and saying, okay, i'm not seeing the return

**[00:44:44 – 00:44:47]** you see what i see next year, will i see it in two years time?

**[00:44:47 – 00:44:53]** And uh and this this heavy spending is continuing, and uh you you know

**[00:44:53 – 00:44:57]** again back to the hockey stick, but just in terms of the roi,

**[00:44:57 – 00:45:00]** i think We should just make, you know, it's very clear, and I'm sure

**[00:45:00 – 00:45:06]** everyone on this call knows this as well as we do, in the sense that

**[00:45:06 – 00:45:13]** monitoring and mapping and being able to calculate ROI is very hard.

**[00:45:13 – 00:45:15]** Like, what is the cost?

**[00:45:15 – 00:45:20]** Is the cost an AI cost or is it just a simple, normal,

**[00:45:20 – 00:45:22]** run-the-bank cost?

**[00:45:22 – 00:45:23]** You know, what goes into what bucket?

**[00:45:23 – 00:45:28]** And we have a lot of debates with the banks because

**[00:45:28 – 00:45:31]** behind the scenes, we've got our public ranking

**[00:45:31 – 00:45:32]** that we released on Tuesday.

**[00:45:32 – 00:45:36]** We also work with most of the banks on a private benchmark,

**[00:45:36 – 00:45:41]** which is based on participation, where they give us their data

**[00:45:41 – 00:45:47]** on AI deployment and ROI or impact, and then on an anonymized basis,

**[00:45:47 – 00:45:53]** we send that back and show the banks individually where are they

**[00:45:53 – 00:45:57]** ahead in their deployment, what's the distance to frontier; we break

**[00:45:57 – 00:46:01]** it down in maturity frameworks for each of the lines of business.

**[00:46:01 – 00:46:04]** We look at what is table stakes versus frontier, and we look

**[00:46:04 – 00:46:07]** at the KPIs associated with each of these lines of business and

**[00:46:07 – 00:46:09]** where the banks are stacking up.

**[00:46:09 – 00:46:15]** And in that mapping, you get into, you know, as we know,

**[00:46:15 – 00:46:18]** the complications of what is, you know, an ROI framework and

**[00:46:18 – 00:46:23]** how do you measure that impact and what is, you know, hours saved, how

**[00:46:23 – 00:46:26]** do you translate it into a dollar value, you know, how do you ladder

**[00:46:26 – 00:46:32]** that up and send it up to the board of the CEO on a regular basis.

**[00:46:32 – 00:46:35]** Just getting that monitoring in place is extremely hard.

**[00:46:35 – 00:46:39]** I don't think anyone doubts that one needs to invest in

**[00:46:39 – 00:46:43]** AI, obviously not, right, because you're going to be left,

**[00:46:43 – 00:46:48]** if you don't, with a higher cost to delivery and no pathway to deliver

**[00:46:48 – 00:46:52]** more seamless experience with your clients, right, that you in those

**[00:46:52 – 00:46:57]** two things you have to do because otherwise others that are investing

**[00:46:57 – 00:47:00]** will have those advantages over time that you won't have, and

**[00:47:00 – 00:47:01]** I don't think anyone doubts that.

**[00:47:01 – 00:47:05]** There is the reason why, you know, when we look at the

**[00:47:06 – 00:47:09]** sort of growing gap between the leading and the lagging banks and

**[00:47:09 – 00:47:14]** why that is making us nervous is because that points to eventually a

**[00:47:14 – 00:47:20]** consolidation in the banking sector because over time you will start,

**[00:47:20 – 00:47:25]** you know, reducing your market share and not performing as well.

**[00:47:25 – 00:47:27]** There's no doubt that that is a direction of travel that

**[00:47:27 – 00:47:28]** we're going into.

**[00:47:28 – 00:47:32]** So back to the ROI, which is, it is really important

**[00:47:32 – 00:47:38]** to, you know, to really map that in order for not to have the

**[00:47:38 – 00:47:43]** conversation that sounds like this, we're investing a lot, I can't

**[00:47:43 – 00:47:45]** see tangible returns or impact.

**[00:47:45 – 00:47:49]** Therefore, let's put a sort of, you know, You know, let's

**[00:47:49 – 00:47:51]** stop that investment because that's the worst outcome you can have

**[00:47:52 – 00:47:54]** for an organization, for a bank.

**[00:47:54 – 00:47:58]** So, you know, but they are, you know, is it going to, you know,

**[00:47:58 – 00:48:04]** is it going to end of this year or next year show up in a cost to, you

**[00:48:04 – 00:48:08]** know, cost to income ratio, share prices and full on bottom line?

**[00:48:08 – 00:48:09]** Probably not yet, right?

**[00:48:09 – 00:48:13]** Will it show up in those metrics in some years to come?

**[00:48:13 – 00:48:16]** Most definitely it will.

**[00:48:17 – 00:48:20]** Sarah, what do you think about that provocation?

**[00:48:21 – 00:48:24]** No, I have a couple of thoughts.

**[00:48:24 – 00:48:31]** The first one is, we release every year our own FSI survey

**[00:48:31 – 00:48:33]** just to see how AI is being implemented by the banks,

**[00:48:33 – 00:48:37]** and the new one is going to come out in the first week of December.

**[00:48:37 – 00:48:41]** One of the things we see year after year is the fact that

**[00:48:41 – 00:48:44]** when you talk to hundreds of them;

**[00:48:44 – 00:48:46]** they told you, we've seen returns.

**[00:48:46 – 00:48:49]** Again, are the returns what they want it to be?

**[00:48:49 – 00:48:52]** They always want more, but they've seen returns, and

**[00:48:52 – 00:48:54]** they are investing more.

**[00:48:54 – 00:48:55]** They are planning to invest more.

**[00:48:55 – 00:48:57]** And I always go back to the same principle, right?

**[00:48:57 – 00:49:01]** Kind of like an organs racer, the right answer usually is

**[00:49:01 – 00:49:02]** a very simple one.

**[00:49:02 – 00:49:05]** You have an industry that is just driven by math and numbers,

**[00:49:05 – 00:49:08]** and every year they invest more and they tell you they see returns,

**[00:49:08 – 00:49:10]** like there is something there.

**[00:49:10 – 00:49:14]** The other simple way to look at this, just to give you

**[00:49:14 – 00:49:20]** a sense, is look at evidence ranking, and look at the banks

**[00:49:20 – 00:49:24]** that are leading on the top, and they happen to be also the banks

**[00:49:24 – 00:49:27]** that grow the faster and have the higher profits in their markets.

**[00:49:28 – 00:49:30]** And this comparison is never perfect, right, because for

**[00:49:30 – 00:49:32]** example now you look at Capital One and because they bought this

**[00:49:32 – 00:49:35]** cover and everything is a little bit murky, right, but when you

**[00:49:35 – 00:49:38]** think about the principle of it, I mean, Capital One was not even a

**[00:49:38 – 00:49:42]** bank 35 years ago, and now it's one of the largest banks in the world,

**[00:49:42 – 00:49:45]** and it's all because they're always being one step ahead in technology.

**[00:49:45 – 00:49:50]** So I think for me, that is actually the proof point, right?

**[00:49:50 – 00:49:52]** That is the banks that are actually, they are living

**[00:49:52 – 00:49:56]** in this space and they are working in this space and

**[00:49:56 – 00:49:59]** they are actually investing in the computing technology and building

**[00:49:59 – 00:50:00]** their own intelligence, right?

**[00:50:00 – 00:50:03]** And then you look and, oh, those are the banks that are also living

**[00:50:03 – 00:50:05]** from a revenue point of view, from a profit point of view, from

**[00:50:05 – 00:50:09]** a growth pace point of view, right?

**[00:50:09 – 00:50:11]** Look at other industries that have done an incredible job

**[00:50:11 – 00:50:13]** connected to us, right?

**[00:50:13 – 00:50:16]** It's out of banking space, but the hedge funds.

**[00:50:16 – 00:50:19]** The leading hedge fund firms, which is very public, their

**[00:50:19 – 00:50:22]** investments they are making on our infrastructure, you

**[00:50:22 – 00:50:27]** can Google in one second and you get all the reference to the really

**[00:50:27 – 00:50:29]** multi-billion dollar investments.

**[00:50:29 – 00:50:31]** And they were told that the return is almost immediate.

**[00:50:31 – 00:50:33]** They are doubling their revenues per year.

**[00:50:33 – 00:50:36]** And now suddenly the banking industry is looking and saying,

**[00:50:36 – 00:50:39]** how can it be that Jane Street or Hudson River, now they

**[00:50:39 – 00:50:40]** are bigger than Goldman Sachs?

**[00:50:40 – 00:50:41]** Like, how can they be, right?

**[00:50:42 – 00:50:45]** They were like a small firm like five years ago.

**[00:50:45 – 00:50:48]** So I think those are the proof points.

**[00:50:48 – 00:50:51]** If I can give a little piece of advice to this audience,

**[00:50:51 – 00:50:54]** right, it's of course we need to measure everything.

**[00:50:54 – 00:50:55]** We are banks, we measure everything.

**[00:50:55 – 00:50:58]** I spent 20 something years in the industry, so I know

**[00:50:58 – 00:51:00]** we measure everything.

**[00:51:00 – 00:51:05]** But if you focus on the right areas, if you focus on the

**[00:51:05 – 00:51:10]** right workflows, the returns are going to be massive.

**[00:51:10 – 00:51:12]** You need to be talking about like a 10x kind of return.

**[00:51:12 – 00:51:15]** I'm going to spend $20 million building what I need to do

**[00:51:15 – 00:51:19]** the intelligence; I'm going to save $150 million in fraud

**[00:51:19 – 00:51:21]** every year, right?

**[00:51:21 – 00:51:26]** You see that over and over again, and that's where you

**[00:51:26 – 00:51:27]** call it table of stakes, right?

**[00:51:28 – 00:51:31]** That is, yeah, I gave everybody access to Copilot or ChargePT

**[00:51:31 – 00:51:33]** or whatever it is.

**[00:51:33 – 00:51:34]** It's very hard to measure that.

**[00:51:34 – 00:51:37]** You will see it in pockets.

**[00:51:37 – 00:51:39]** Where are you gonna see that massive?

**[00:51:39 – 00:51:41]** It's gonna be when you pick an area and say, oh, by the

**[00:51:41 – 00:51:46]** way, my onboarding of commercial clients for the commercial

**[00:51:46 – 00:51:53]** bank, it used to be six months or six weeks, now is 24 hours.

**[00:51:53 – 00:51:55]** That is the aspiration you need to have.

**[00:51:55 – 00:51:58]** When you're thinking those workflows, the aspiration

**[00:51:58 – 00:51:59]** needs to be like that.

**[00:51:59 – 00:52:02]** I'm going to take something that takes months into something

**[00:52:02 – 00:52:03]** that takes minutes.

**[00:52:03 – 00:52:06]** By the way, you will fail, and instead of minutes will be hours.

**[00:52:06 – 00:52:10]** But remember, you were working in months, and now you're in hours.

**[00:52:10 – 00:52:13]** Call that failure a big success.

**[00:52:14 – 00:52:19]** When you think about a process like that and having, you know, AI

**[00:52:19 – 00:52:22]** take on a bigger workload and, more importantly, having agents take on

**[00:52:22 – 00:52:25]** a bigger workload, I think one of the things that we haven't touched

**[00:52:25 – 00:52:28]** too much on today is kind of the controls that you need to build

**[00:52:28 – 00:52:32]** in to make sure that it actually can be trusted to do more, frankly.

**[00:52:32 – 00:52:36]** I mean, I guess, Alexandru, you, when you're talking to banks, what

**[00:52:36 – 00:52:41]** feels new on the kind of front of actually governing this technology

**[00:52:41 – 00:52:45]** and controlling it so that it, you know, you know, so that ultimately

**[00:52:45 – 00:52:48]** banks can race ahead and do more?

**[00:52:48 – 00:52:50]** Yeah, I mean, I think one shouldn't forget just what a

**[00:52:51 – 00:52:57]** monumental task it is to move from controls that fit deterministic AI,

**[00:52:57 – 00:53:02]** you know, traditional AI, that have clean pathways to, you know, to

**[00:53:02 – 00:53:07]** the results, to gen AI and agents.

**[00:53:07 – 00:53:13]** And obviously, there's been a massive shift in what one NVIDIA,

**[00:53:13 – 00:53:13]** Thrust, AI; NVIDIA, Thrust, AI.

**[00:53:13 – 00:53:14]** So NVIDIA, Thrust, AI. So what NVIDIA, Thrust, AI. So

**[00:53:14 – 00:53:15]** what we've NVIDIA, Thrust, AI.

**[00:53:15 – 00:53:16]** So what we've seen NVIDIA, thrust, AI. So what we've

**[00:53:16 – 00:53:16]** seen is NVIDIA, Thrust, AI.

**[00:53:16 – 00:53:18]** So what we've seen is that, so what we've seen is that,

**[00:53:18 – 00:53:20]** so what we've seen is that the, so what we've seen is

**[00:53:20 – 00:53:22]** that the leading, so what we've seen is that the leading banks, so

**[00:53:22 – 00:53:25]** what we've seen is that the leading banks can, so what we've seen is

**[00:53:25 – 00:53:29]** that the leading banks can do, so what we've seen is that the lead.

**[00:53:49 – 00:53:51]** Obviously, you know, with deterministic AI, you've got

**[00:53:51 – 00:53:54]** your, you know, lines of defense, your first line, your second

**[00:53:54 – 00:53:57]** line, your third line of defense, and that sort of can still

**[00:53:57 – 00:54:01]** work for some things, and that's like what the regulators

**[00:54:01 – 00:54:01]** are used to working with.

**[00:54:01 – 00:54:07]** So you've got not only to shift a system onto working on a completely

**[00:54:07 – 00:54:09]** different type of technology.

**[00:54:09 – 00:54:13]** That's hard, and you've got to get the regulators with you to be

**[00:54:13 – 00:54:15]** comfortable with what you're doing.

**[00:54:15 – 00:54:19]** And most importantly, you've got to make sure that you don't fuck up,

**[00:54:19 – 00:54:23]** because then, you know, something might happen, and you might

**[00:54:23 – 00:54:27]** inadvertently do something for a customer that wasn't intended.

**[00:54:27 – 00:54:29]** And if you do, I think in this age where there's so

**[00:54:29 – 00:54:35]** much tension around AI and the use of AI, you know, as

**[00:54:35 – 00:54:41]** a bank, the last thing you can have is a mistake in an AI decision

**[00:54:41 – 00:54:45]** or a mistake in some AI that is not controlled appropriately.

**[00:54:45 – 00:54:49]** So you will find the banks are more concerned around

**[00:54:49 – 00:54:53]** controls even than the regulators are, because you cannot have and

**[00:54:53 – 00:54:58]** you cannot break your trust with your clients by making a mistake.

**[00:54:58 – 00:55:00]** So you look at the banks and the way they're sort of trying

**[00:55:00 – 00:55:03]** to automate the governance system that obviously everyone

**[00:55:03 – 00:55:06]** knows building these harnesses, and there's some great questions in the

**[00:55:06 – 00:55:11]** chat on that, and it's hard, right, but you, you know, I think that

**[00:55:11 – 00:55:18]** the banking system in some ways is the perfect blueprint for how

**[00:55:18 – 00:55:22]** you should think about regulating and putting controls around

**[00:55:22 – 00:55:24]** the outer edges of AI development.

**[00:55:24 – 00:55:25]** It's right there.

**[00:55:26 – 00:55:29]** They're doing checks, balance, looking at output, they're looking

**[00:55:29 – 00:55:33]** at the system, they're making sure that before it goes anywhere, the

**[00:55:33 – 00:55:36]** right controls are put around it.

**[00:55:36 – 00:55:39]** It's like what your boss says, Asa, you know, Jensen Huang

**[00:55:39 – 00:55:40]** is talking about this.

**[00:55:40 – 00:55:43]** It's like, these are problems we've had forever and ever.

**[00:55:43 – 00:55:46]** These are software, software moves, right?

**[00:55:46 – 00:55:49]** You've got to put the controls around it so it doesn't

**[00:55:49 – 00:55:50]** create problems, right?

**[00:55:50 – 00:55:53]** This is what the banks are doing to the nth degree and

**[00:55:53 – 00:55:56]** doing it with a lot of care.

**[00:55:56 – 00:55:58]** So I think that is impressive, right?

**[00:55:58 – 00:56:02]** And everyone's working hard on streamlining the governance

**[00:56:02 – 00:56:06]** process, right, and getting this moving fast.

**[00:56:06 – 00:56:11]** If I can add something here, I had one big regulatory body

**[00:56:11 – 00:56:15]** for banking visiting me at headquarters last week, and

**[00:56:15 – 00:56:18]** we're talking exactly about this.

**[00:56:18 – 00:56:20]** And so, two big messages for the room, right?

**[00:56:20 – 00:56:25]** The first one is don't use regulation as an excuse, right?

**[00:56:25 – 00:56:28]** I hear that a lot, that, oh, no, we cannot do anything

**[00:56:28 – 00:56:29]** because of regulation.

**[00:56:29 – 00:56:32]** It's like, listen, there are leading institutions doing a lot

**[00:56:32 – 00:56:34]** of things, and they are regulated by the same entities that you are.

**[00:56:34 – 00:56:37]** So, don't use regulation as an excuse, right?

**[00:56:37 – 00:56:38]** It is not.

**[00:56:39 – 00:56:42]** Second, work with the regulators.

**[00:56:42 – 00:56:45]** I'm going to give you a couple of examples.

**[00:56:45 – 00:56:49]** If you use AI, and again, AI is not a chatbot, let's be

**[00:56:49 – 00:56:53]** clear about this, but if you use AI capabilities, accelerated computing

**[00:56:53 – 00:56:56]** capabilities, you can reimagine completely credit decisions.

**[00:56:56 – 00:57:01]** When you do that, what happens is that you can provide a

**[00:57:01 – 00:57:05]** lot more credit to a lot more people, because you become

**[00:57:05 – 00:57:08]** a lot more fine-tuned to the reality of the risk of every

**[00:57:08 – 00:57:14]** individual, and by doing that, you actually expand the credit pool.

**[00:57:14 – 00:57:16]** Regulators love that.

**[00:57:16 – 00:57:18]** There's no question about that.

**[00:57:18 – 00:57:20]** You go to any regulator and say, listen, I think we can

**[00:57:20 – 00:57:21]** lend to a lot more people.

**[00:57:21 – 00:57:23]** And by the way, the people that benefit from that are usually

**[00:57:23 – 00:57:25]** at the bottom of the pyramid.

**[00:57:25 – 00:57:27]** Regulators love that.

**[00:57:27 – 00:57:30]** You just need to work with them and show them how it's fair, how it's

**[00:57:30 – 00:57:35]** gonna work, why you're convinced that actually the results are bad.

**[00:57:35 – 00:57:38]** But remember my comment before about like in the 1960s, we

**[00:57:38 – 00:57:40]** have one account, in 1980, we have 20 something, right?

**[00:57:40 – 00:57:42]** We're gonna have one for everybody.

**[00:57:43 – 00:57:44]** With underwriting, it's the same thing.

**[00:57:44 – 00:57:48]** Even today, we are excluding a lot of people of underwriting on risk

**[00:57:48 – 00:57:53]** assessment just because we cannot get to that granular understanding

**[00:57:53 – 00:57:54]** of every situation, right?

**[00:57:54 – 00:57:57]** This technology is giving you that opportunity, right?

**[00:57:57 – 00:58:00]** You're truly gonna be able to understand all the elements.

**[00:58:00 – 00:58:03]** And again, it's the same on the, the credit side is the

**[00:58:03 – 00:58:06]** same on the marketing side, the same on the fraud side.

**[00:58:06 – 00:58:08]** There are so many things you can do.

**[00:58:08 – 00:58:10]** And again, talk to regulators.

**[00:58:10 – 00:58:14]** They are eager to listen to you engaging with them and

**[00:58:14 – 00:58:17]** hearing what is out there, because they know that this

**[00:58:17 – 00:58:18]** is coming, it's unstoppable.

**[00:58:18 – 00:58:23]** And so the best thing you can do is to have an open dialogue.

**[00:58:23 – 00:58:25]** We've got time for, I think, one more question.

**[00:58:25 – 00:58:29]** I'm going to ask it to both of you, and it's one of the

**[00:58:29 – 00:58:30]** more practical ones, I guess.

**[00:58:30 – 00:58:34]** If you're a CEO, if you're a bank executive who's been

**[00:58:34 – 00:58:38]** listening to this today and tuning in, and you are tasked

**[00:58:38 – 00:58:42]** with making one AI decision today that will have the biggest

**[00:58:42 – 00:58:46]** kind of snowball effect positively in a year, what's the decision

**[00:58:46 – 00:58:49]** that you should make today?

**[00:58:51 – 00:58:52]** Asa, you go first on this one.

**[00:58:52 – 00:58:53]** No, no, you go first, you go first.

**[00:58:53 – 00:58:57]** You are the guest. I'm going to, let me close.

**[00:58:57 – 00:59:02]** Yeah, that's such a good question.

**[00:59:03 – 00:59:08]** You know, I think there's so much focus right now on, I mean,

**[00:59:08 – 00:59:11]** it's really back to the question around how you put the controls in.

**[00:59:11 – 00:59:16]** You put the controls in, you can move very fast, very safely.

**[00:59:16 – 00:59:18]** And if you've got the confidence in your controls, right, you

**[00:59:18 – 00:59:22]** can shift a lot of work.

**[00:59:22 – 00:59:26]** And so I think, you know, ask that question again in six months,

**[00:59:26 – 00:59:27]** it will be something else, right?

**[00:59:27 – 00:59:29]** But ask the question right now, I think, you know, that

**[00:59:29 – 00:59:31]** that is where I would focus.

**[00:59:31 – 00:59:33]** But, you know, there's lots, like you asked me to say

**[00:59:33 – 00:59:35]** one thing, right?

**[00:59:35 – 00:59:39]** You know, there's so much that also comes from leadership.

**[00:59:39 – 00:59:42]** Someone the other day, you know, talked about, you know, JP Morgan

**[00:59:42 – 00:59:45]** being number one on the index, you know, it's a really big bank, and

**[00:59:45 – 00:59:49]** they're like, deploying everywhere, and they started early, right?

**[00:59:49 – 00:59:51]** And someone said, yeah, but that's because Jamie Dimon is sort

**[00:59:52 – 00:59:54]** of CEO and CTO at the same time.

**[00:59:55 – 00:59:59]** I don't think he would characterize himself as a CTO, but he gets,

**[00:59:59 – 01:00:00]** he gets in the engine room, right?

**[01:00:00 – 01:00:04]** And I think, you know, it's very hard to direct from the top as

**[01:00:04 – 01:00:10]** a CEO, you don't really understand what the technology can do and what

**[01:00:11 – 01:00:13]** is required to get it to do that.

**[01:00:13 – 01:00:17]** So I think it's also, you know, if you've got that leadership,

**[01:00:17 – 01:00:20]** that's not something obviously you can do anything about.

**[01:00:20 – 01:00:21]** You have the CEO that you have.

**[01:00:21 – 01:00:26]** But it is that leadership and that understanding that

**[01:00:26 – 01:00:29]** this is different and getting into the engine room.

**[01:00:29 – 01:00:33]** I've seen that the banks that have moved the fastest are

**[01:00:33 – 01:00:37]** the ones that have that kind of leadership at the end of the day.

**[01:00:37 – 01:00:38]** I'm going to build on Alexander's comment.

**[01:00:38 – 01:00:41]** I'm going to cheat and give a couple of things.

**[01:00:41 – 01:00:46]** So the first thing you need to do as a bank, connects exactly

**[01:00:46 – 01:00:51]** to that point, is understand this is not a technology revolution.

**[01:00:51 – 01:00:54]** This is a business transformation, right?

**[01:00:54 – 01:00:58]** The technology a bank needs to transform with AI, it's

**[01:00:59 – 01:01:01]** been around now for years, and it's been proven.

**[01:01:01 – 01:01:03]** And by the way, like every single time you do anything in this

**[01:01:03 – 01:01:06]** dimension, the results come, right?

**[01:01:06 – 01:01:07]** Again, this is not new.

**[01:01:07 – 01:01:10]** There are companies that have been using these things for now.

**[01:01:10 – 01:01:14]** Again, Google transformed YouTube using our infrastructure

**[01:01:14 – 01:01:15]** starting in 2012.

**[01:01:15 – 01:01:18]** So there is a lot of tradition here.

**[01:01:18 – 01:01:19]** So this is a business transformation.

**[01:01:19 – 01:01:21]** So if it's a business transformation, what you

**[01:01:21 – 01:01:25]** have to do is, number one, pick the business process, the

**[01:01:25 – 01:01:27]** workflow you want to start from.

**[01:01:27 – 01:01:31]** And again, don't try to pick 75, just pick the one, the

**[01:01:31 – 01:01:34]** two, one per business line kind of thing, and secure

**[01:01:34 – 01:01:36]** you have the resources to do it.

**[01:01:36 – 01:01:39]** And that is talent and compute and the capabilities.

**[01:01:40 – 01:01:41]** Let's start there.

**[01:01:41 – 01:01:43]** And don't forget what Alexandra said, that is control.

**[01:01:43 – 01:01:46]** I don't mention control when I talk about this, because

**[01:01:46 – 01:01:49]** in a bank, everything we do needs to be covered.

**[01:01:50 – 01:01:54]** That is not nice to have, it's mandatory.

**[01:01:54 – 01:01:59]** So that is granted, but I think that this is a business problem.

**[01:01:59 – 01:02:04]** Fight the workflow, secure the talent, and compute what

**[01:02:04 – 01:02:09]** you need to execute on that business thing, and go from there.

**[01:02:09 – 01:02:11]** Sarah, Alexandra, thank you so much.

**[01:02:11 – 01:02:13]** And thanks so much to our very engaged audience.

**[01:02:13 – 01:02:17]** We really appreciate you taking the time and spending it with us today.

**[01:02:17 – 01:02:18]** Hope you have a great rest of your day.

**[01:02:18 – 01:02:20]** Thank you.

**[01:02:20 – 01:02:20]** Thank you.