Latest / The Payments Shed / Ep. 36 - Nvidia on Payments Foundation Models, Connected Intelligence and the Next Wave of Financial Innovation, with Georgios Kolovos, Payments & Fintech Leader @ Nvidia

Transcript

Ep. 36 - Nvidia on Payments Foundation Models, Connected Intelligence and the Next Wave of Financial Innovation, with Georgios Kolovos, Payments & Fintech Leader @ Nvidia · The Payments Shed

• 0:00
  Welcome to the Payment Shed Podcast, the weekly podcast that
• 0:03
  dives into the big topics, trends and people shaping the
• 0:06
  world's of payments, fintech and business leadership with your
• 0:09
  two Co hosts, myself, Grant Evans and Justin Hannah.
• 0:13
  Welcome back to another episode of the Payment Shed Podcast.
• 0:15
  Today we're going deep into Ain financial services from black
• 0:19
  box LLMS to sovereign agentic commerce.
• 0:22
  We are joined by Georgios Kolovas, payments and fintech
• 0:26
  leader at NVIDIA. Georgius, welcome to the show.
• 0:28
  Thank you and you said my name correctly.
• 0:31
  Well, that's a rarity for myself so.
• 0:34
  Great to be here. Can you tell us a little bit
• 0:36
  about your experience, how you got into financial services and
• 0:39
  and what you do at NVIDIA today? Yeah.
• 0:41
  I mean, I've, I've been lucky enough to see financial service
• 0:45
  and payments in general from all sides of the industry have been
• 0:50
  with GE Money when GE was very active in the banking and the
• 0:53
  FINCO space. I've have the pleasure to work
• 0:56
  with some of the most interesting banks in Central and
• 0:59
  Eastern Europe, one of them being Guarantee Bank in Turkey,
• 1:01
  which was extremely innovative. So I've seen it from the issuing
• 1:05
  side and spent about seven years in PayPal.
• 1:08
  So I was driving a lot of the partnership, a lot of the
• 1:12
  inorganic growth of PayPal, expanding into different
• 1:14
  services, so in wallets, mobile wallets and crypto payments and
• 1:19
  you name it. Then for a short period of time
• 1:21
  worked on a merchant site with one of the divisions of Booking
• 1:24
  holding OpenTable, which was quite interesting.
• 1:27
  So having this complete view of the industry end to end, it was
• 1:32
  very obvious to me that data is still quite in abundance but not
• 1:37
  use as much as it can be. So I tried experimenting with a
• 1:41
  few companies around how do we start harvesting some of the
• 1:45
  insights that comes from this data.
• 1:48
  And last year started, was it not last year, two years ago
• 1:51
  started at NVIDIA and it was the perfect time.
• 1:54
  So there was a lot of interest in the market, a lot of very
• 1:57
  good capabilities that develop, but still quite a lot of
• 2:01
  uncertainty around how people using accelerating computing in
• 2:04
  general, but also AI, particularly in payments.
• 2:07
  So the last year and a half, two years have been looking with
• 2:11
  together with a lot of the partners with Nvidia's ecosystem
• 2:15
  around what we can be doing, how we can be driving some of this
• 2:19
  engagement. How we can kind of even like
• 2:21
  take it to the next stage of evolution of payments businesses
• 2:25
  across the side from issuers to networks to acquirers and along
• 2:29
  the way of a lot of very interesting fintech
• 2:31
  conversations. And NVIDIA, undoubtedly one of
• 2:34
  the most talks about companies in the world at the moment I
• 2:36
  would say as well. Correct.
• 2:37
  There's a lot of interest, a lot of interest.
• 2:39
  Then we'll talk a little bit further.
• 2:41
  We we're hosting our biggest developer event in about 3 weeks
• 2:46
  in San Jose and just the interest is phenomenal.
• 2:49
  We already sold out the in person views and kind of now the
• 2:53
  only way people want to join that is online, but the interest
• 2:56
  is there. Interest hasn't stopped over the
• 2:59
  last 3-4 years. There's still so much more to be
• 3:02
  explored. We talk about kind of data
• 3:04
  points in the ecosystem. The new NVIDIA report tells a
• 3:08
  little bit about that and kind of your findings from that.
• 3:11
  We do every year we do State of Financial Services and AI
• 3:16
  report. This year there are probably two
• 3:20
  things that were kind of, not that it was a surprise, but they
• 3:25
  were most interesting to me. First, almost 100% of all the
• 3:28
  people that responded saying that their budget is growing,
• 3:32
  right. It was a good validation that
• 3:33
  this is a technology that will be here to stay and people
• 3:36
  continue investing. The other one, which was also
• 3:39
  very interesting, we are getting to company, getting most of the
• 3:43
  payments company, a financial services company have already
• 3:45
  some sort of experience in working with AI and now they
• 3:48
  start looking into, OK, how do we get it to scale?
• 3:50
  How do you start looking into a little bit more holistically?
• 3:54
  Most importantly, how do we start thinking on a, if we
• 3:57
  deploy this scale across our entire organization where you
• 4:00
  get that? And about 80% of the company
• 4:03
  that responded start looking in open source and open source is a
• 4:07
  quite an interesting area that NVIDIA has been investing quite
• 4:10
  heavily over. I mean, NVIDIA, I don't know if
• 4:13
  you're aware of CUDA or developer library.
• 4:15
  So a lot of that, a lot of this library is open source.
• 4:19
  So NVIDIA is investing heavily in this space.
• 4:21
  I. Think that's a bit of a
• 4:23
  transition. We've seen in general that
• 4:24
  people are kind of going down that open source for, you know,
• 4:27
  it's like sort of saying to the competition, there you go.
• 4:31
  How do you how do you follow suit as well, right?
• 4:33
  It's not competition only, it's about, and this is what I can
• 4:36
  find out, I love about payments. I mean, we do work in some other
• 4:40
  financial services industries, but payments by far is the more
• 4:44
  collaborative part of the the financial services industry,
• 4:47
  right? Even if everyone competes with
• 4:50
  everyone at the same time, the way the industry work, we all
• 4:53
  work together, right? So there's no people are not
• 4:57
  afraid to share. There's a lot of very
• 4:59
  interesting technical papers. So a lot of very interesting
• 5:01
  events where people openly discuss their their progress.
• 5:05
  And I think this is one of the the biggest trends of the
• 5:07
  industry. On the other side, you can say
• 5:09
  it's also a bit of a curse. There's a over lot of
• 5:11
  information. I mean, you look into LinkedIn
• 5:13
  and all the reports that are coming, OK, AI, changing
• 5:16
  payments, banking and all of that.
• 5:18
  So it's very difficult to stay on top of that.
• 5:21
  But the interest they are in the collaboration is looking.
• 5:23
  So this is where open source come quite strongly, right?
• 5:26
  The more companies are going to experiment and work with open
• 5:30
  source, the more they share, the more the industry's elevated for
• 5:34
  them. And I think when you look at the
• 5:36
  financial services industry and there is a real onus on, you
• 5:41
  know, owning the AI stack, you know, and that's from from
• 5:43
  hardware to, you know, regulation as well.
• 5:46
  Why do you think that we look at it that way in financial
• 5:48
  services? So there's a, there's another
• 5:51
  report that we published early kind of meet last year, which
• 5:54
  was quite interesting. We did it with ServiceNow report
• 5:58
  about the maturity of financial services.
• 6:02
  This was very focused on banking and their adoption of AI.
• 6:06
  And actually in 2025, the maturity has dropped because the
• 6:10
  industry AI, it's progressing with such a speed and kind of
• 6:16
  companies in financial services and payment in particular not
• 6:19
  always move with the same speed. And it's for good reasons.
• 6:23
  We're handling quite a lot of very, a lot of money, a lot of
• 6:26
  importance. So we need to stay on top of
• 6:29
  that. To answer your question, the
• 6:30
  reason why we, we see that the more control you have them
• 6:34
  across the entire stack, the better you'll be able to evolve.
• 6:37
  It's not about doing a short term fix on a particular case.
• 6:42
  It's more about how do we change the structure of, of our
• 6:46
  operations and there's tons of service stance of reports and
• 6:50
  companies that are taking this more sustained, more more
• 6:55
  holistic approach delivering a much more sustainable results.
• 6:59
  There's a lot of some of companies are doing whatever 100
• 7:03
  to 100 AI pilots, very difficult to to gain access onto that.
• 7:07
  It was very good in the beginning of when this
• 7:09
  technology was starting to get into the financial services
• 7:12
  space. You want to get your hands and
• 7:14
  you want to experiment it, but to really get a meaningful
• 7:17
  impact, you need to be able to, I don't know what I mean,
• 7:21
  control is the hard work, but you want to have a very clear
• 7:25
  strategy around that. So every time I talk to boards
• 7:29
  in payments but also in the broader financial services, you
• 7:32
  typically need 3 strategies which are very fundamental for
• 7:37
  you to be successful of using this technology.
• 7:39
  First is you need to have your compute strategy and it's
• 7:44
  respectively what it is, is you decided to be on Prem because of
• 7:48
  sovereigns questions. So you decided to go for on on
• 7:53
  cloud or some sort of a hybrid. It really doesn't matter.
• 7:56
  The important thing is that you have a very clear strategy that
• 7:59
  you understand what's computer requirements are needed and you
• 8:03
  stay and you invest consistently onto this one.
• 8:06
  So this is the foundation. Then the the second layer and
• 8:10
  the second strategy is around data insights and models.
• 8:14
  We've been talking about data for long time, right?
• 8:18
  Put all the data in one place and there's particularly in
• 8:22
  payments. People have grown through
• 8:23
  acquisition, through different things.
• 8:25
  Data is a little bit scattered, but in now it's becoming less
• 8:28
  about, yes, the data can need to be in one place, need to be in
• 8:31
  control. But it's all about about how do
• 8:33
  you transfer this data into insights and how do you leverage
• 8:37
  this insights to add value to your customers?
• 8:39
  Is it the merchants on the acquiring side is into the
• 8:41
  consumers on the left? And this is also converting not
• 8:44
  only on the tradition, the models that are necessary to run
• 8:48
  a payments operations, but also to the LLMS.
• 8:51
  How do you use the LLMS? What is the data that you fine
• 8:54
  tune the LLMS, et cetera. So when you think strategy for
• 8:57
  compute strategy for your data and then the the final piece is
• 9:01
  strategy for your applications, right.
• 9:03
  So we have started upside down. We've started from application.
• 9:06
  There was a lot of rush. We need very quickly a lot of
• 9:08
  application. If you don't have the other two
• 9:11
  in place, it's very it's either will become very expensive very
• 9:14
  quickly or you not be able to get a lot of the the benefits
• 9:19
  because your data and insights are not are not in place.
• 9:23
  When you talk kind of to other leaders in the industry, what do
• 9:27
  you kind of hear the opportunities that are available
• 9:30
  to them by kind of this AI infrastructure kind of of the
• 9:33
  next generation I? Mean the conversation is
• 9:36
  changing, right? The conversation is changing in
• 9:40
  the sense that people realise that this technology is here to
• 9:43
  stay, while before kind of even if you look into the beginning
• 9:48
  of last year, people were still in this space.
• 9:50
  I'm not sure if this is not another great tech term that
• 9:53
  people are trying to to use, but now companies start thinking
• 10:00
  holistically. We've seen a lot of the early
• 10:02
  progress being on how do you take cost out, right?
• 10:05
  So how do you drive efficiency from this technology?
• 10:10
  Now it's all about, OK, this technology can potentially
• 10:13
  change the way the business model operates.
• 10:15
  This can start generating some new revenue.
• 10:17
  So I'm seeing more and more conversation moving that
• 10:21
  direction. Nevertheless, there's, there's a
• 10:23
  massive gap. So on one side you have the
• 10:27
  technology organization that is quite, very, quite well
• 10:30
  understanding what is required and what is it on.
• 10:33
  Sometimes the leadership, the business side of the story not
• 10:36
  ever is catching into that understanding.
• 10:41
  And it's not surprised. A lot of the boards that I
• 10:43
  talked today, everyday someone comes to a knock on their door
• 10:48
  and say, Hey, we can sort all your AI needs.
• 10:51
  So we have everything of that and and it's going back to the
• 10:54
  strategies, right? If you don't have a very clear
• 10:56
  strategy, if AI is not fully embedded into your strategy,
• 11:00
  very difficult to be able to cut the noise right and be focused
• 11:04
  on the on the real thing. That's quite funny saying in
• 11:06
  itself though, because a lot of the time these companies don't
• 11:08
  know what their AI needs actually are yet, right?
• 11:10
  Because they haven't started playing around with it.
• 11:14
  Yes, it's yeah, it, it really depends The way to look for me,
• 11:21
  very simple metrics around when you want to understand the
• 11:24
  maturity of a particular organization about their
• 11:27
  understanding is when you look into what is their data science
• 11:30
  and AI teams in place. And this, the difference is
• 11:35
  quite stark. It's not a complicated thing to
• 11:38
  do particularly, but we have so many tools and LinkedIn is
• 11:40
  pretty good, a good indicator of this.
• 11:45
  But when you look into the people that are progressing
• 11:48
  well, these are people that have a very, very strong data science
• 11:52
  teams. And it's not about the number of
• 11:54
  people, it's the quality and the ability to execute.
• 11:58
  And that's a nice segue into connected intelligence,
• 12:00
  something that NVIDIA talk about.
• 12:02
  So, so we see connected intelligence as the the next
• 12:06
  frontier is kind of one of those those statements that NVIDIA
• 12:09
  would would make in this area. So if we start broadly on this,
• 12:12
  what do you guys actually mean by connected intelligence
• 12:14
  specifically? In payments, so let's let's
• 12:17
  start from one of my favourite and most of the topics that most
• 12:20
  of the companies that I talk to in the payment space with a
• 12:22
  conversation typically start from the fraud outside.
• 12:25
  And fraud is a very nice area because the impact that you can
• 12:31
  see in the ROI that you can generate from fraud is pretty,
• 12:34
  pretty easy to articulate. So the faster you process the
• 12:37
  data, the faster you train the models, the better the
• 12:40
  performance is. So the bottom line impact, the
• 12:43
  savings are much better. So if you move from a
• 12:46
  traditional rules based models, there's a very simple
• 12:50
  acceleration, which is you move the processing from CPU to GPU.
• 12:54
  So CPU is sequential processing. You complete once transactions
• 12:59
  review, then you do the next. The GPU's potential is that you
• 13:03
  can do that in parallel, so you can do multiple transactions at
• 13:06
  the same time. So just moving that, that's
• 13:08
  simple acceleration provide a lot of, a lot of benefits.
• 13:13
  And then you start looking into where the interconnectivity
• 13:16
  works. The the first stage of
• 13:19
  interconnectivity is when you start looking into transaction
• 13:22
  in relations to other transactions.
• 13:24
  So we are leveraging graph neural networks.
• 13:27
  We published recently a blueprint recently, last year, a
• 13:31
  blueprint around how companies can use graph neural networks to
• 13:34
  prevent fraud. But it's still relatively
• 13:37
  concise. You're still looking in the
• 13:39
  fraud use case in isolation. Now we are seeing the emergence
• 13:44
  of payment foundation models which leverage a lot.
• 13:49
  You go deeper into payments, you use the same methodology that is
• 13:52
  used to train large language models, but rather than use this
• 13:56
  methodology on language, you apply to payment transactions.
• 13:59
  So what happened underneath is this this each individual
• 14:02
  transaction is tokenized and then they are put into time
• 14:05
  series. So it's very easy to use to
• 14:07
  start predicting what is the next transaction in this
• 14:10
  particular time series. And the interconnectivity comes
• 14:14
  into that is that you don't necessarily use it only for
• 14:17
  fraud, but the embeddings that come from these models.
• 14:19
  Now you can use it to start empowering other models across
• 14:23
  the value chain of payments, maybe for personalization or
• 14:26
  maybe for for routing or all of that.
• 14:29
  So you start seeing this interconnectivity between
• 14:32
  transactions now in the connectivity about individual
• 14:34
  steps in the value chains of payments.
• 14:36
  And there's a first step, which is the interconnectivity between
• 14:40
  different payments organization. So there's a lot of talk and
• 14:43
  particularly here in the UK where there's a fraud fraud
• 14:47
  prevention strategy coming to be published.
• 14:50
  A lot of that is how do you share data between organization
• 14:53
  to prevent fraud. This is interconnectivity that
• 14:55
  you're bringing between the organizations.
• 14:58
  But the one that we are very excited Biden, is this payment
• 15:01
  foundation model side because this provides the next step of
• 15:05
  evolution of payments. And we see examples.
• 15:09
  Actually, there are six companies in the world that
• 15:11
  today, I can argue are way ahead of anyone else that I've seen in
• 15:15
  in this space. Has network tokenisation been,
• 15:18
  you know, a big factory in kind of moving us forward with with
• 15:21
  that capability as well then? There there's so depending on
• 15:25
  which part of the the equation you look into that.
• 15:28
  So if you look into the acquiring side, yes, there's a
• 15:30
  lot of benefits. Well, but it still looks into
• 15:33
  one portion of the equation, right.
• 15:35
  So when you look into acquiring, you have a lot of individuals
• 15:39
  models, right? You have checkout model, you
• 15:41
  have your fraud models, your authorization, authentication,
• 15:44
  routing settlement in Yuda. In a lot of organization, these
• 15:48
  models will work independently. The, the models that are having
• 15:51
  this payment foundation models connects those models together
• 15:55
  and ultimately they optimize for, for the benefits of the
• 15:59
  merchant, right? So, and you've been in this
• 16:02
  space for, for, for quite some time.
• 16:03
  So you know, there's only three factors that matter to
• 16:06
  merchants, right? The cost of the transaction, the
• 16:09
  risk appetite of the merchants and then the end to end
• 16:12
  conversion. So imagine if you for each
• 16:15
  transaction you can optimize it based on the needs of the
• 16:17
  customers for this particular 3, the mix of this 33 metrics for
• 16:22
  them. So then each models we optimize
• 16:25
  it for all of that. We spoke briefly before around
• 16:29
  Revolute from the an issue in perspective and Adian from the
• 16:33
  quarrying perspective. Kind of what's your your
• 16:35
  thoughts on the way that they're they're doing it from from a
• 16:38
  payments foundation model? Perspective, I don't want to
• 16:41
  stay steal their Thunder. Both of these companies are
• 16:43
  coming to GTC next month to come and talk.
• 16:47
  So they'll share a lot of very interesting insight and kind of
• 16:51
  having a sneak peek of how they are developing.
• 16:54
  It is very, very interesting. ADN have been very consistent
• 16:59
  and I'm kind of quite impressed about the clarity of vision and
• 17:03
  the focus on execution. ADN has launched their product
• 17:07
  uplift last year, then they follow up with dynamic
• 17:12
  identification. So they're using some of this
• 17:15
  insights from foundation models to start generating new
• 17:17
  products, which is kind of in the in the entity space on the
• 17:21
  relevant sides is a very different perspective.
• 17:23
  So they're using foundational models to get a better
• 17:28
  understanding of their relationship with the customers
• 17:31
  and deepening their, their connectivity.
• 17:34
  The the interesting thing, the, the important thing with both of
• 17:37
  them is that they're doing it at scale, R.E.M. processing
• 17:41
  phenomenal amount of revenue to their large margins.
• 17:44
  The same revenue has a quite a diverse and broad customer base.
• 17:48
  So you're seeing this models being used across a very, very
• 17:53
  mature, mature and large organisations.
• 17:56
  In today's episode is sponsored by ISX Financial.
• 17:59
  Most growing businesses end up stitching together multiple
• 18:02
  providers to run their payments and banking, leaving them
• 18:04
  exposed to cash flow, reconciliation and even security
• 18:08
  issues. ISX Financial delivers a unified
• 18:10
  ecosystem with multi currency accounts, online payments, real
• 18:13
  time UK and EU transfers and global mass payouts.
• 18:16
  Within that platform, Paid By provides pay by bank with real
• 18:20
  time settlement notifications, automated reconciliation and
• 18:23
  integrated refund management. One partner, one integration,
• 18:28
  helping your business transact with confidence.
• 18:30
  You can learn more at ISX dot Financial.
• 18:33
  That's ISX dot Financial. And a big thank you to the team
• 18:36
  at ISX for supporting the show. Yeah, I think both of them are,
• 18:42
  you know, definitely leading the way even with stuff like Agentic
• 18:45
  Commerce, which we wanted to talk to you about and kind of to
• 18:47
  get your feeling on where agentic commerce is AT today.
• 18:50
  And then we apply a payments landscape to that and and are we
• 18:53
  already seeing tangible benefits coming from agentic commerce in
• 18:57
  payments that you're seeing? So I have a very different view
• 19:01
  than most of his share on agentic commerce.
• 19:04
  For me, agentic commerce is nothing more than just data
• 19:07
  challenge and the difference then traditional forms of
• 19:12
  e-commerce that we've seen that in some of the models, not in
• 19:15
  all. In some of the models the
• 19:17
  journey starts from one side or one place and ends up in a
• 19:21
  different, which creates a lot of problems and a lot of open
• 19:24
  questions around payments in particular.
• 19:28
  We're going back to the data problem.
• 19:30
  Magente commerce is a data problem interestingly enough
• 19:34
  that you need to be able to capture the customer insights
• 19:39
  one way or the other. So we see it two ways that this
• 19:43
  is happening today, which kind of is the first step of the
• 19:47
  agentic commerce. And one is with the chat based
• 19:50
  interface. You have all the the ChatGPT,
• 19:53
  the perplexity of the world where you can provide your input
• 19:58
  and this can trigger an agentic commerce experience.
• 20:02
  We've done a very interesting. Partnership with PayPal where
• 20:07
  they use our open model Nemotron to create a similar chat based
• 20:12
  experience on their own surface. But they also packaged it and
• 20:15
  sell it to merchants. So this is the first way you
• 20:18
  need to tell the agents what they what they should be buying.
• 20:22
  OK, I want AI don't know a green dot, let's say a blue shirt with
• 20:26
  a yellow duck on side. So you put it in the chat and
• 20:29
  the agent goes and gives you option the other way and the
• 20:32
  other way. I'll link it to the foundation
• 20:34
  model. This is more relevant for real
• 20:37
  repeat transactions. If you know based on experience
• 20:42
  of the the consumer, what are the their typical buying habits,
• 20:47
  this can potentially trigger you.
• 20:49
  If I know that every two weeks I buy coffee from from this coffee
• 20:53
  shop, not coffee shop, yeah, coffee maker, whatever it is the
• 20:58
  right name, then the agent can potentially authorize the agents
• 21:01
  to go and initiate this part. So This is why for me, this
• 21:04
  payment foundation models and getting payments data into
• 21:08
  insights and then from that you can start driving a lot of value
• 21:11
  added services is important. But this is still give you only
• 21:14
  one part of the equation, right? How do you, you as a consumer,
• 21:18
  give the agents the information the agents need to start the
• 21:22
  payment transaction on the other side?
• 21:25
  And it's very interesting when you start talking with the
• 21:27
  merchants, even some relatively simple products, not always well
• 21:32
  prepared and augmented for Argentic commerce, right?
• 21:36
  How do you know that this iPhone and that phone iPhone from 2
• 21:40
  completely separate merchants of the same model and the same
• 21:44
  specifications? So there's companies that are
• 21:47
  start looking into that problem. How do you help merchants to
• 21:51
  augment their catalogues? I don't know if you follow
• 21:54
  Klarna launched a Gente commerce product protocol which is going
• 21:58
  into that direction. And then you have the kind of
• 22:01
  the other sides once you have the protocols or the the
• 22:04
  catalogues augmented and how do you label agents to agents to
• 22:07
  communication? And obviously the third part is
• 22:10
  all around, how do you allow the, what is the role of the
• 22:13
  agents? What are the guard rails?
• 22:15
  How do you do all that? It's a fascinating area.
• 22:17
  I I think it's still less defined that some of the others.
• 22:22
  There's a lot of news out there right on a genetic commerce and
• 22:24
  what he's doing. But I think similar to what you
• 22:26
  guys said before is we don't know how powerful it can be just
• 22:30
  yet. We have these dreams of how
• 22:31
  powerful we want it to be and how it's going to allow any
• 22:34
  business to get one up on their competition or just on the
• 22:37
  injury to help the end consumer. But we just don't know what we
• 22:41
  don't know yet with that. And it is exciting, but I think
• 22:44
  there's a lot of structural hurdles that will still come up
• 22:47
  against that allows us to effectively kind of share that
• 22:50
  data, right? For sure, I mean, and for me,
• 22:52
  this is where going back to the conversation we had earlier
• 22:56
  around the power and the collaboration in the industry,
• 22:58
  the industry need to come together not only as a as a
• 23:01
  payment, but also with the merchant.
• 23:03
  And we've done that before, right.
• 23:05
  The contactless cards has been a long journey, but now you give
• 23:10
  it a for granted. It can happen there.
• 23:12
  It has its own challenges. Probably like a hype thing in
• 23:16
  payments though, don't we? Like I don't think a gentech
• 23:18
  commerce will be that hype thing.
• 23:20
  I'm just interested to see how quickly we take it from A to B
• 23:24
  to C as you know, something people then end up using
• 23:27
  frequently in their day-to-day lives.
• 23:29
  I think there will be different models that we were much.
• 23:32
  So very simply in my mind, if you put it into a 2 dimension
• 23:37
  chart, on one side you have it where the the agentic commerce
• 23:41
  experience start. Is it from the same merchant or
• 23:44
  is it from another side outside of the merchant and then the
• 23:47
  other side is it is it from as I said, experience that you need
• 23:53
  to tell the agents 1 of transaction or is it based on
• 23:56
  your learnings? So there are four models that
• 23:59
  are emerging. I would expect some of the
• 24:01
  existing large margins, but speak travel travel is a
• 24:05
  fascinating place. I don't know if you try to do
• 24:08
  your itinerary using some of the chats based interval, kind of.
• 24:11
  Painful. It's painful, but it's
• 24:13
  interesting, right And you look into some of the larger OTAs or
• 24:17
  the larger travel much and they have a lot of insights.
• 24:20
  They have a lot of a lot of information.
• 24:23
  If especially if you have been using them for long years, then
• 24:27
  what is stopping them to help you build a better iteration
• 24:30
  based on that. And I would expect this will be
• 24:32
  a form of a gente commerce that will get faster traction because
• 24:37
  you are limited into the the property of a particular
• 24:41
  merchant. So a lot of the the complexity
• 24:44
  around switching from one location to another to to much.
• 24:48
  But this is not what we talk about, right.
• 24:49
  Most of what we talk into is kind of this very complex
• 24:54
  scenario, starting from one place, combining a lot and kind
• 24:58
  of doing an invisible check out. I think travel's been called out
• 25:01
  a lot as well. We've we've got trust my travel.
• 25:04
  We've got an episode coming out with them soon and it's 1 you
• 25:06
  should listen to I think as well because we'll from from trust my
• 25:10
  travel has some interesting perspectives on how quickly
• 25:13
  they'll see a genetic adoption in that industry and and where
• 25:16
  it can actually work. So yeah, I think it's, it's
• 25:17
  definitely kind of the poster child example at the moment,
• 25:20
  isn't it? I.
• 25:20
  Mean the, the way I look, there's 22 specific use cases in
• 25:24
  commerce, which I think we all require.
• 25:26
  And it's why one of them is why travel is so interesting because
• 25:30
  this experience your purchases that you need to do a lot of
• 25:34
  research before you click the button and buy, right.
• 25:38
  I don't know how many. If you go and search for flight,
• 25:41
  how many? How many flight options will
• 25:44
  come? I'll just go on Skyscanner, I'm
• 25:46
  lazy with it. But still, even in Skyscanner
• 25:48
  you'll see whatever, a whole 10 pages of different options,
• 25:52
  right? You still need to go and
• 25:53
  research, right? You have your criteria as costs
• 25:56
  or do you want direct flight? Do you have preferred airlines
• 25:59
  because of it still requires some research and all that.
• 26:03
  So these are perfect examples or kind of perfect use cases where
• 26:07
  gente commerce will be very useful.
• 26:09
  Same way if you're buying ATV, right?
• 26:11
  There's a lot of other things to mention.
• 26:13
  If someone can take away some of the hassle for the research, why
• 26:16
  not? So I think the discover of
• 26:18
  abilities are an important base. Like the notion of agentic and
• 26:23
  and loyalty merging, you know, everyone likes something free or
• 26:26
  discounted and a lot of your research you can do around is
• 26:29
  like, what's the best deal I can find?
• 26:31
  So I think in learning your, your buying, but how you talk
• 26:34
  about the coffee, right? But then knowing, OK, well,
• 26:36
  Nespresso, if you reorder your coffee this month for doing a
• 26:39
  promotion offer where you get 4 free sleeves of coffee pods,
• 26:42
  right? So you might be better off
• 26:43
  ordering that in advance, Grant, and we're going to order it in
• 26:45
  June for you because we know they're running this promotion
• 26:47
  because we found it somewhere else on the Internet, right?
• 26:49
  I think that could be a a really really good way of deploying
• 26:53
  agentic. I think that agentic commerce
• 26:55
  and the architecture is kind of an ongoing kind of system in our
• 26:58
  right. It's not necessarily here's how
• 27:01
  it is today, it might change tomorrow.
• 27:02
  It's changing between now and tomorrow and you have to be well
• 27:06
  aware that what is working today will need to change when it
• 27:09
  comes to tomorrow. How does that kind of work and
• 27:12
  look in practice kind of when you're advising businesses on
• 27:15
  that? When it comes to strategy,
• 27:18
  right, So we're on the acquiring side.
• 27:21
  The way we typically talk about is do you see a opportunity to
• 27:26
  generate an additional value added services to your
• 27:29
  merchants? How do you help your merchants
• 27:32
  to prepare for for this for this?
• 27:35
  I don't know if you call it new future.
• 27:37
  I mean, it's still too early to define if you'll change every.
• 27:40
  Mission, the mission is never going to stop though, right?
• 27:43
  I think you need to get some when you get there.
• 27:44
  Need to be where going in five years time, we're going to be in
• 27:46
  five years time, right? Never really going to complete
• 27:48
  it. Correct.
• 27:49
  And being worked, I have worked on the merchant side, it was
• 27:53
  very different experience, right.
• 27:55
  The conversation that you have on the merchant side is very
• 27:58
  difficult. Changing payments provide
• 28:00
  changes, setups takes time. There's a lot of focus on the
• 28:05
  risk on the prevention, on the customer experience and all of
• 28:07
  that. Just kind of maybe to close the
• 28:10
  the agentic, I don't think it will be only limited to the
• 28:13
  online experience. I like your point around
• 28:16
  loyalty. I think loyalty, even if you
• 28:19
  think about coffee shops and the technology is also there the to
• 28:24
  to be embedded in these experiences.
• 28:26
  And it can, depending how far you are ready to to share data,
• 28:32
  it can become extremely interesting even in the offline
• 28:35
  experiences. Yeah.
• 28:37
  And then I think that's where par data is being talked about a
• 28:42
  lot at the moment, right, in terms of, you know, collecting
• 28:44
  that point of sale information and knowing who that customer is
• 28:47
  and can we open up that part? It's so.
• 28:49
  Frustrating that we don't see line item data.
• 28:52
  I mean this is just I don't know how many company have tried to
• 28:55
  solve that but the POS space is so fragmented so hard to do it.
• 29:01
  Imagine what we will be able to do if you have the data of what
• 29:05
  I have just popped down by Pratt and have for lunch.
• 29:09
  Yeah, exactly. I mean, if I know exactly, then
• 29:12
  I can play my diet of my dinner can be sorted, I can know if I
• 29:19
  need to do an extra session in the gym and all of that.
• 29:22
  But still this is a gap in the payments that is still quite
• 29:26
  frustrating that we are lacking. It's got.
• 29:28
  To come at some point though, hasn't it?
• 29:29
  I, I just find a point as well and then on like, you know,
• 29:33
  sovereignty and why that matters so much.
• 29:34
  Do you think that sovereignty would only really matter at an
• 29:36
  enterprise level or is this sort of part of a national regional
• 29:39
  strategy too that we need to consider?
• 29:42
  On an enterprise level, each company needs to have a clear
• 29:45
  strategy and it's matters for sure.
• 29:51
  We can see a lot of interest. We have been supporting a lot of
• 29:56
  what we call NCPS, new clouds that are building sovereign
• 30:00
  capabilities across Europe and the world.
• 30:04
  So there is a lot of interest and I don't think this trend is
• 30:09
  going to change anytime soon. And this kind of leads to your
• 30:13
  the second point of your question for sure.
• 30:16
  Countries are realizing that this is a critical
• 30:19
  infrastructure and they need to have a saying on all that,
• 30:22
  particularly in the world that we are living today, which is
• 30:25
  extremely volatile. The more you can control, the
• 30:29
  more certainty you can provide to your businesses, the better.
• 30:34
  Now to kind of the word that most people are dreading for
• 30:37
  themselves perspective, when we talk about AI, how does
• 30:40
  governance kind of control that innovation within businesses
• 30:43
  when they look at it? That's a.
• 30:45
  That's a good. It's a long and complex topic.
• 30:53
  The good thing is that we have good track records of changing a
• 30:56
  lot of that. We have examples in the past,
• 30:59
  particularly in the crypto space where some company have decided
• 31:02
  to go against the regulator and decided to.
• 31:06
  It's very difficult to see that as a sustainable strategy in the
• 31:13
  in the long run. Regulators are putting their
• 31:16
  efforts to to support, which is very good and right.
• 31:24
  There's two things. The first thing is people are
• 31:27
  scared because they don't understand it, right?
• 31:30
  And if you start with your kind of first thought is I don't want
• 31:34
  to do anything wrong. It's a very, very challenging
• 31:39
  perspective to move, to move forward.
• 31:41
  Of course, you need to understand it to be able to to
• 31:46
  see what's real, what is not, and where there's risk and where
• 31:49
  there's not. There's a lot that need to be
• 31:53
  done and a lot can be done, but I can't go back to the same
• 31:57
  thing. You need to have a very clear
• 31:58
  strategy. If you have 204 hundred
• 32:01
  different pilots that are running at the moment that work
• 32:07
  out of your data, the chances for it to go wrong is pretty
• 32:12
  bad. What can you?
• 32:15
  The only way to do it is to take a much more structured approach
• 32:18
  and much more coordinated approach to do that.
• 32:21
  That fear factor always exists in payments though, doesn't it?
• 32:23
  I think from from chip and pin to contactless to digital, while
• 32:27
  it's it's like the great unknown, there's always people
• 32:29
  wary of it to start with and then adoption grows, trust grows
• 32:33
  and then just becomes a normal part of life.
• 32:36
  It's a process, right? And rightly so, right?
• 32:38
  It should be with design, but you cannot put your kind of head
• 32:43
  in the ground, the sand and say I don't want to touch it.
• 32:46
  By the time you realise that most of your competitors have
• 32:49
  been working with you, you're already on the back foot, you're
• 32:52
  trying to catch up, which is quite, quite difficult.
• 32:55
  I, I see definitely this changing.
• 32:57
  I mean, definitely see a lot more appetite and we did some
• 33:02
  work also with the FCA to, to help bring some of that
• 33:07
  responsible innovation on the forefront and to be able to work
• 33:11
  with different parties of the, of the ecosystem to, to test
• 33:15
  that. So there's a lot, a lot more and
• 33:18
  I'll say it's each country's here approaching it slightly,
• 33:21
  slightly differently, but going back to that, So we payment
• 33:25
  foundation models. So if I have a payment
• 33:27
  foundation models that I know what's my relationship with the
• 33:31
  customer is and this powers a lot of the different downstream
• 33:34
  use cases. It's much more easier to manage
• 33:37
  than if you have, I don't know, 20-30 different models that each
• 33:42
  feeds from different data source is very difficult to control.
• 33:46
  When we talk about that governance piece that Justin
• 33:48
  mentioned before and where you guys might be looking at
• 33:51
  specific regions or, or areas where you can probably make the
• 33:53
  biggest impact quickly, there's something like the EUAI Act
• 33:57
  where you've got a collective region looking at this as as one
• 34:00
  sort of group of countries. Does that kind of tactically
• 34:04
  make that region more appealing to yourselves with your products
• 34:07
  and services that you offer as well Europe?
• 34:09
  'S I mean, there's a lot of very interesting development that
• 34:13
  happened last year, in particular around Europe
• 34:17
  starting to realise that need to build capacity, need to build
• 34:21
  the right framework. There's I'm very impressed with
• 34:24
  what's happening in France. Mistrau has been doing an
• 34:27
  amazingly to ride the ecosystem to allowing different players to
• 34:33
  to experiment. Even if it's one region, still
• 34:36
  there's a lot of flavours in the local dynamics and it can be the
• 34:42
  interpretation of the European regulation, it can be the talent
• 34:44
  that is available, it can be the companies that are operating
• 34:47
  into into the same country. It's very difficult to say a
• 34:50
  blank flavour. As much as we want to say this
• 34:52
  is A1 big European blog, there's still a lot of flavours in, in
• 34:58
  the, in the countries. But we the way we operate is
• 35:03
  through our ecosystem. And the only way to be able to
• 35:06
  make an impact is to empower the people that work in payments or
• 35:11
  support payments organization. And this was we have been doing
• 35:14
  quite consistently over the last, I don't know, I mean this
• 35:17
  has always been invidious strategy, leverage the people
• 35:20
  that understand the specific locally to be able to add value.
• 35:25
  And how's the progress with the FSA, how you kind of find them
• 35:28
  conversation? Very, very interesting.
• 35:31
  They had a demo day. I don't know if you had a chance
• 35:33
  to see. They had actually 2 demo days a
• 35:35
  few weeks ago. There's some very interesting
• 35:39
  examples coming up. They were showcasing not coming
• 35:42
  up. There's interesting different
• 35:46
  use cases, different companies in terms of scale, in terms of
• 35:50
  parts of the value chain that are participating.
• 35:52
  So I definitely see a lot of positivities and we can only
• 35:57
  welcome more and more of this this to happen.
• 36:00
  So if you could then give one piece of advice to organisations
• 36:03
  trying to balance innovation with audit readiness, what would
• 36:07
  it be? Audit readiness, yeah.
• 36:11
  I mean, for me it's, it's always the data, the insights and
• 36:17
  applications, right? So if you don't follow the the
• 36:20
  stack, then it's very difficult to to be able to control the
• 36:24
  flow of data and watch what you because most of the the
• 36:27
  financial services companies are afraid of not exposing
• 36:31
  confidential data to customers. So giving different given an
• 36:36
  advice that is not into that, if you have a understanding of how
• 36:40
  your data is transferred into the insight, what these insights
• 36:43
  then leads into in terms of actions, then for you to audit
• 36:48
  it becomes much more easier rather than have again hundred
• 36:51
  different off the shelf proposition that you have
• 36:55
  plugged to your data or different models that you don't
• 36:59
  fully understand, very difficult to track.
• 37:01
  And AI probably helps as well in terms of actually saying where
• 37:03
  that data came from, right? It can collate the information.
• 37:06
  A lot of a lot of areas that you can increase the explainability
• 37:10
  of I am. When we talk kind of about kind
• 37:13
  of the next evolution of of payments and intelligence, we
• 37:15
  know it doesn't stop. What do you think good looks
• 37:18
  like in in five years and kind of what's the challenge of
• 37:21
  holding that up? Yeah.
• 37:23
  I mean goods, we already have good examples of what goods
• 37:27
  looks into. There is, I said there are six
• 37:30
  companies that are using this payment foundation models.
• 37:34
  And hopefully next time we talk, well, I can share a lot more
• 37:37
  insights and we're thinking about how to spread some of the
• 37:40
  goodness that will come in from the GTC after the event.
• 37:44
  You see a real tangible use cases coming into that.
• 37:48
  I, I continue repeating it. I, I probably get, I start
• 37:51
  signing a little bit boring, but for me, this is the next stage.
• 37:54
  The next stage is data into insights.
• 37:57
  Today we don't have insights, right.
• 38:00
  We, we have some kind of patterns that we can track, but
• 38:04
  we don't necessarily get the most of the data.
• 38:07
  So if you can use insights, if this insights are then used to
• 38:11
  add either new services or add new benefits to your customers,
• 38:15
  it's a pretty good outcome. And if it actually helps to the
• 38:17
  regulators to be able to feel more comfortable with the, the
• 38:22
  things data is used in financial services, even better.
• 38:25
  And what I really hope, really hope at some point of time we'll
• 38:28
  get our act together and we can share.
• 38:31
  I don't even want to share, share data between institution
• 38:34
  to combat fraud. Everyone is talking about how
• 38:37
  big fraud is a a problem and how we how much increase in fraud
• 38:41
  has been seen in the last couple last years.
• 38:45
  But I hope there will be a framework for us, the financial
• 38:48
  services institution and beyond to share data and actually not
• 38:51
  share data, but share insights. So I don't know if you have
• 38:56
  heard about Federated learning where you don't necessarily need
• 38:59
  to share the data, but you have models that work independently
• 39:04
  in each work in each individual financial services organization
• 39:08
  and they share insights to a central models that can provide
• 39:13
  further intelligence. So I hope this is something that
• 39:16
  we will see coming. And what role do open models
• 39:20
  play in the future of financial AI systems?
• 39:23
  Link to that. For sure.
• 39:24
  I think open open models are quite important in terms of cost
• 39:29
  for sure they can make because think about any reasoning model
• 39:33
  that you're using today, they are becoming quite even single,
• 39:38
  single transaction. And if there's models there
• 39:40
  which will require a number of different tokens to be
• 39:42
  generated. So this will be quite can become
• 39:45
  quite expensive when you deploy and scale.
• 39:47
  So I definitely see from that side open models are very good
• 39:51
  for companies to be able to explore and to kind of build
• 39:54
  their unique experiences and retain some of the IP, right.
• 39:59
  Interesting some of the. The board discussions which I'm
• 40:02
  having, it's a simple question is who owns your models, right?
• 40:08
  And it's a very difficult question to answer for any
• 40:10
  payments organization because it's such fundamental for the
• 40:14
  wake of the company operate. But the answer is not
• 40:16
  straightforward. So the more you can control for
• 40:19
  your models, the more of the IP you keep for yourself rather
• 40:23
  than transferring it to a third party do.
• 40:25
  You think we're heading towards a fully autonomous financial
• 40:28
  decision making system or do you think it still needs that that
• 40:32
  that money? To bench how many of the what
• 40:34
  percentage of the current transactions are automatically
• 40:37
  approved? 80%, Ninety percent?
• 40:40
  95%? You can argue that we already
• 40:44
  not far from this, right? Well, the other 5%, there's
• 40:48
  reasons why they're there. You can optimise it, but again,
• 40:52
  we are optimising only on a a sliver of the entire margins.
• 40:57
  Yeah. So then you when you think, when
• 40:58
  you connect them all together, this can become quite
• 41:01
  interesting. I mean, I don't some of the
• 41:04
  areas. So we did some work on
• 41:07
  suspicious, suspicious activity report creation, even automating
• 41:13
  the the work for the agents to review these activities.
• 41:17
  And these reports can generate quite significant benefits in
• 41:21
  optimising the time to come to a decision.
• 41:25
  Will there be all still processes that will require
• 41:28
  human for sure. What percentage of that?
• 41:31
  Very hard to say, but I think payment is already quite well
• 41:34
  automated. It's well automated, but I think
• 41:37
  the insight and data bit that you talk about definitely needs.
• 41:40
  Yeah. It's ironic to me how many
• 41:41
  payments companies have tools called insights.
• 41:43
  And everything I'm saying is that like, actually those
• 41:47
  insights need a hell of a lot of of work and effort.
• 41:49
  I mean, it's been a fascinating chat today.
• 41:51
  Before we we wrap up, there is a final part of the show, The
• 41:54
  Shelf of Shame, where all of our guests nominate something in
• 41:57
  payments business, fintech, whatever it might be.
• 42:00
  What are you bringing to the shelf today?
• 42:02
  Add on applications, I'm always there's a lot of discussion
• 42:06
  about BIBU and buy, build or or build or buy.
• 42:12
  Yeah, yeah. So a lot of the things I it's
• 42:15
  quite interesting also with some of the boards there's always
• 42:17
  someone in the boards meeting that has a very passionate about
• 42:20
  particular use case or particular company.
• 42:22
  So getting this company and being plucked into everything
• 42:26
  that operates today is fascinating, but it doesn't hack
• 42:30
  a long term value. So if I can remove this notion
• 42:33
  immediately to be looking for short term impact with a simple
• 42:39
  plug on to existing operation, I think this will do a lot of good
• 42:42
  to a lot of companies. So.
• 42:44
  You're in the Bicamp, then a little bit of.
• 42:47
  Both a build, build I, I think people and this is where kind of
• 42:51
  a lot of the, the build is. When you build yourself, you,
• 42:57
  you keep a lot of the IP on your side.
• 42:59
  You need to be very, very specific.
• 43:01
  What I'm saying is more about urgent or kind of impulsive
• 43:07
  add-ons, let's call it. That's why impulsive add-ons,
• 43:10
  right? I used to in one of the other
• 43:12
  businesses I worked. So Davos was the event I was
• 43:14
  dreading the most because most of our execs were going to
• 43:18
  Davos. And then I knew that after
• 43:20
  Davos, I'll have tons of different.
• 43:23
  That's a quite exciting. Businesses that I need to figure
• 43:25
  out what to do to do about it. But yes, it's it's sometimes
• 43:32
  this impulsive decisions can can be quite distracted.
• 43:35
  People that are like can just acquire a business and plug a
• 43:37
  gap within their their ecosystem and it just doesn't work like
• 43:41
  that, right? And it takes years and years and
• 43:43
  years and many mistakes. And actually, you actually could
• 43:45
  have built it in a shorter amount of time.
• 43:48
  Correct. I mean it's yes strategy, long
• 43:51
  term vision, try to cut off the noise.
• 43:53
  Seems to to be a difficult skill to master.
• 43:56
  Which is why you now work at NVIDIA, right?
• 43:58
  Which is actually doing that for.
• 43:59
  Businesses. So it's leading brilliant
• 44:01
  conversation. Thank you so much.
• 44:04
  It's all about AI, agentic commerce and architectures.
• 44:08
  Thank you so much for joining us, Georges.
• 44:09
  Where can people find you? LinkedIn I I like LinkedIn.
• 44:13
  We're always open to conversations.
• 44:16
  We'll be doing a lot. As mentioned, our developer
• 44:18
  conference is coming the. What's the date of that?
• 44:22
  The 16th to 19th of March and you can follow most of the
• 44:27
  session online, which will be quite good.
• 44:30
  But then definitely, well, I'll make sure that a lot of the
• 44:33
  payments inside throughout foundation models about the
• 44:35
  future of agentic commerce and agentic payments will be broadly
• 44:38
  shared there. We're pretty sad.
• 44:40
  We wanted to be there but we did whatever the clash next.
• 44:42
  Time we'll get you next time thank you so much for joining us
• 44:45
  you're just. Collection.
• 44:46
  Thank you.