1092. News: Finally, the robots take over: Starling gets smarter, Stripe buys OpenRouter, and Visa and Mastercard unite on agentic payments

22:53Laura Watkins:Yeah, that's such a great point. I think, David, you were nodding as Aditi was talking there, sort of back to the earlier point about sort of financial services companies becoming tech companies ultimately. And, you know, what's particularly interesting here is the premium that Stripe are prepared to pay for this company as well, which, you know, not that long ago was valued at sort of less than a quarter of the price. So, David, what's your take on sort of, to Adity's point about like the sort of overarching value, whether monetary or otherwise, that Stripe is seeing here to kind of bolster their business model?

23:31Yeah, I mean, to build on Adity's point, I think, you know, there's one thing in the valuation of kind of the actual open router operating model itself and the technology behind it, but as I said, the data that they have, right? It's, you know, data is the new oil in this world, I suppose. And so how do you put a price tag on what is arguably the hottest commodity at the moment, right? AI. So I think probably that is what's driving a lot of that reported price tag.

24:00Laura Watkins:For sure. And Stefano, kind of looking at it from the other side as well, sort of payments companies are investing in AI. AI companies still need the things that Stripe does, right? So payments, billing, financial infrastructure, you know, how kind of interconnected could these two ecosystems be? Seem interconnected and clearly Stripe now, to the point before, is on the control point where AI is consumed and measured. So this is a unique first repositioning, as I mentioned before, and can be definitely a set that become mainstream in the market now. I still think that being able to route between models is very different than controlling the asset.

24:48And if we go back to application in very sensitive, miscreepy application like payment, I think the star-washedness part is very important to take into account. But definitely the ability to route with such efficiency mounted on the financial infrastructure is definitely an insane business opportunity that they have for sure. I think on that billing point, there's another thing to think about, which is a few months ago, or last year, I don't exactly remember when, Stripe also acquired another company called Metronome, which does billing for AI. So now they are effectively doing both the invoicing and the billing and the collecting of the money for something that they are now routing to different companies.

25:36The other angle here is all of these model providers that Open Router will then be by default choosing between are big customers of Stripe for their own separate billing needs. So does that somewhere create friction or tension for Stripe where they are the people charging money for the billing for a company that they're telling we're not channeling tokens towards? So I think there are quite a few questions within this deal and I'm curious to see how it all, how the dust settles.

26:09Laura Watkins:Yeah, for sure. I think that's very interesting. On the one hand, Stripe are kind of like a magnet kind of pulling all of these big companies towards them. They're making huge acquisitions. I mean, you know, I should say at this particular moment, this one has not necessarily been confirmed. It has been, you know, rumored that they are making this one. Equally, they're sort of still pursuing that PayPal takeover, which obviously shows the scale of their ambitions, whether it comes to pass or otherwise. Like, you know, that is sort of showing where their heads are at the very least. So, yeah, it's kind of making a statement, if nothing else, about how big they're prepared to kind of go, I think.

26:50Laura Watkins:I think they are, yeah, just making an enormous sort of infrastructure play all over the shop at the moment, which is sort of, you know, going back to like that moving beyond fintech into just broader tech. Kind of what's your take on that, Anati? So I was just thinking as you were saying this, Laura, that until now, the payments were being made by people. Going forward, payments are going to be made by agents. When agents start making payments, effectively the currency that you're using, again, historically the currency we were using was fiat currency. Maybe now recently some stable coins, etc.

27:29In the future, once agents start making those payments, yes, they will be predicated on currency, but there will be so much more usage of AI and so much more exchange of AI tokens that will happen in the background and between this. perhaps Stripe have realized that the percentage margin that they get is much more from those tokens than from genuine currency transactions. And therefore, maybe it's about keeping the same business model, but changing the asset on which that business model operates. So I think it is very much an infrastructure play, but for the infrastructure of the future economy rather than the current economy.

28:09Laura Watkins:Yeah, that's such an interesting point and kind of backs up your earlier, you know, theory that they're sort of just future-proofing their business model as much as they possibly can as well. Stefano, what's your thoughts on that? I think, I mean, there's not much to add, but yes, I think the currency, I like what Aditi mentioned, the currency of C-token as a currency. And we see more and more in financial services as well, the tokenization of data as well that is becoming another important part. So I think from the financial infrastructure, we go to NAI common infrastructure, and all this thing will play an important role in the next couple of years.

28:53We see a lot of needs of digital assets, the tokenization of data that are ramping up in financial services. That is a double down and even reinforcing what Aditi was saying before. Fantastic. So we want to see other example, I think, on the sales side in particular of things like this.

29:15Laura Watkins:Absolutely. Well, as we said, this deal has not yet gone through, so we will keep an eye on it. But, you know, it would appear likely. So we'll keep an eye on it and see where it develops. But on that note, we're just going to take a quick pause here back very shortly.

29:33Hey, Chicago. Class it up with Crocs. You know back to school is coming in fast. So why wait to find your new fave footwear? Step into a local croc store and step into your new look. Try it. Style it. Make it yours. Because the right pair doesn't just show up. It shows off. First day fits, handled. Walk out ready for whatever's next. Visit your nearest croc store today.

30:01Laura Watkins:This episode is brought to you by Facebook. So you were scrolling on Marketplace, and there it was, the bike you'd been searching for. You sent a message, and it turned out the seller was super chatty, kind of funny, and an avid cyclist. The next thing you know, you're in a cycling crew. Well, a community cycling group. The thing about Facebook, you might find more than what you're looking for. From a browse to a bike ride, this summer, find more on Facebook.

30:35Laura Watkins:Before we get back to the news, we wanted to tell you about our next After Dark event, which is now just over four weeks away. FinTech Insider hosts David Brewer and Kate Moody will be joined live on stage by special guests to record this very podcast and break down the biggest news stories of the week. It's taking place in London on Thursday, the 17th of September, and tickets are available now at 11fs.com forward slash After Dark, all one word. There's also a link in the show notes of this episode below. If you want to be part of the fun, don't miss out. But now back to the news. And our final main story this week is Visa and MasterCard back new Agentec Payments Alliance.

31:16Laura Watkins:The story in Finnextra. Visa, MasterCard, Fiserv, Circle, Solana and Remitly are among more than 25 companies joining the Agentec Payments Alliance, a new industry coalition focused on developing standards and frameworks for AI-powered commerce. The alliance has been created by its stablecoin payment platform, RAINN, and will operate as a working coalition governed collectively by its founding members rather than being controlled by a single company. Its early work will focus on shared research and frameworks, testing standards for identity and authorization, and regulatory questions around AI agents making transactions on behalf of consumers and businesses.

31:57Laura Watkins:The initiative comes as companies prepare for AI agents to move beyond recommending products and services towards completing transactions themselves. Okay, so this feels like it's sort of backing up a lot of what we've been talking about up till now anyway around sort of companies preparing for the future, AI agents being able to shop and transact for us and then what kind of is needed in the background to safeguard it for us. Aditi, what's your initial take on this? Why is a coalition like this so important? Or indeed, do you think that it is? It's maybe the first place to start. Yeah, absolutely.

32:36And the answer to that, Laura, is 100 % yes. Like a coalition like this is absolutely important because I think there's two things here which are important to focus on. The first is a common, agreed and shared framework for any of these transactions to start happening. So for any kind of agentic commerce, any kind of agentic payments, you need to have a common framework on which all the participants in the market agree. And this coalition is a step in that direction. What is very encouraging and interesting is the fact that the instigator of that coalition or the capitalist rather to have a more positive spin on it is a company which is much, much smaller than the other behemoths that it has then brought to the table.

33:25So Reyn is a relatively small company. The last funding round they raised was just 25 million. But then they've then gotten Visa and MasterCard and all of these people to come and join that. So that is an extremely encouraging sign. The second part of it, which I didn't want to talk about, is in order for large changes of mechanisms also or operating methodologies to happen across the system, you've always had coalitions that have to come in to do the work. And interestingly, those coalitions are almost invariably competitors partnering and collaborating to create that baseline from which everyone works.

34:06So if we think of the KYC framework set up by banks in Scandinavia, Invidem and BankID, it's exactly the same principle. If we think of any of the Internet's protocols like HTML, like HTTP, etc., it was essentially competitors coming together to create those. It's the same thing with any data and governance frameworks. It's the same thing with open source software. unless you create that baseline for the entire market to operate off, you are not really going to get the same level of scale and adoption that you need to kick off an entirely new market space effectively. And I think that is exactly what has started happening here.

34:45Laura Watkins:Amazing. And yeah, as you say, like it's sort of, it's definitely headline worthy when you hear Visa and Mastercard doing something together. So I think that's kind of part of it and shows how seriously they're sort of taking this coalition. Stefano, what's your take on this in terms of the kind of interoperability, the sort of bringing together of all of these big names to try and put some sort of industry standards in? I totally see the point, as mentioned before, creating a common infrastructure where we can build innovation. And I can tell you that there is a lot of similar initiatives happen in the financial industry, non-necessary payment that are ramping up under discussion because if we don't define a common protocol in terms of platform, steps, governance, authentication, protocol, how we embed the current legal and compliance protocol inside the agent TKI, this thing will never fly in a compliant way to regulation.

35:49And so, there's absolutely need in every segment of the financial industry. We're going to see more and more this consortium. We saw this in the past with other providers. We saw this in the past with a lot of trading platform. And we're going to see happening now more and more because of the needs of the adoption of AI in this business protocol. For sure. There's just an example that comes to mind, which I wanted to chat about. You know, for any kind of AI or AI agents to work, the principle always is the data that they work across has to be good data, comprehensive data, reliable data. However, if you look at the move to the cloud that has happened in financial services, especially kind of during and after the COVID era, you still have the hyperscalers, so the Googles and the Microsofts and the Amazons of the world that are relatively closed in their own universes and their own landscapes.

36:49They each have their own separate offerings for everything across the stack from data storage to governance to AI tools and so on. The biggest challenge that organizations face at the moment is that they don't talk to each other so well. And therefore, as a large multinational bank with a multi-cloud strategy, my struggle is how do I get all of my data together? Now, in something like this, the analogy I think of is, I don't know, 30 years ago when you wanted to book a flight or when you wanted to travel somewhere, you'd go to a travel agent, I think. The travel agent would then pick up the phone to one airline, ask them what flights they had and at what time, and give you a printed ticket for that one airline.

37:34You would have no idea what other airlines are doing, what prices they are, where else they're flying, what other destinations can you optimize by clubbing together two different airlines. We can do all of that today when you go to a kayak or some sort of travel supermarket kind of website. And I think that is the change that is starting to happen now in the payment space with this kind of interoperability where even though they are different payment rails, even though they are different kind of providers of perhaps similar infrastructure, the moment you let agents lose across that landscape, those agents are going to try to optimize everything as much as they can.

38:12They're going to optimize the payment hops. They're going to optimize interchange fees. They're going to optimize the settlement and FX rates that they get. And you can only achieve that true value of optimization if everything is interconnected. So I think that's kind of how I see this change happening.

38:30Laura Watkins:Yeah, for sure. I mean, David, you were sort of nodding along there with the sort of travel agent analogy there. What's your kind of take on this and in terms of, you know, particularly future-proofing towards agents sort of operating on our behalves, essentially? Yeah, I think, you know, definitely double down on all the points that you've made. I think also from a purely technical perspective, if you don't have that kind of consistent protocol and set of interfaces, then AI has come a long way and there's incredibly sophisticated models out there now. But even those, if you don't give them a consistent landscape of protocols to work with, they fundamentally won't work.

39:13And I think this kind of alliance is really evidence that if this market is to grow and agentic commerce is really to take off, then actually all these payment providers, they have to come together and come together on a unified protocol. It's the only way this is going to work. So I think that's the first thing. But yeah, then in terms of trust, I think as consumers, dispatching an AI to buy something or spend your money on your behalf is a huge thing. We talked about building trust earlier on. everything we've talked about so far in this podcast, we've not even come close to the idea that AI would spend your money on your behalf.

39:46So the work that needs to go into building the trust and making sure that it's transparent, I think if every provider had a different flow for that, it's going to be a very disjointed experience for consumers. And so that's also not going to help with that kind of picture of building out trust.

39:59Laura Watkins:Yeah, that's so true. And Stefano, what's your take on that kind of trust element and the kind of consistency in the experience leading towards that? Maybe as you noticed from my comment, this podcast for me is all about trust. In Dominic, we did a lot of research. We published 20 something papers in a year, about 90 % of them were about trust. I still think that in having a GNTGI penetrating this mission-critical, regulated processes, we still need to remember that we are dealing with the probabilistic technology. It's a probabilistic approach and we can be probabilistic, we have to be deterministic in dealing with transaction, portfolio construction, loan evaluation, etc.

40:50So I think the need of keep on investing in audit trail, traceability, live evaluation, this is not a problem that we solve with putting a stamp on the model and an agent that does a good job once, but there has to be a live monitoring of the audit behavior of the agent, in particular from issue of the navigation. So the typical example, we published a paper last year about positional bias for the enterprise model where you ask, given all the analytics and modeling that you can have available, you ask an agent which is the best investment between Tesla and Microsoft, he tells you Tesla in a very compelling way.

41:35Same prompt, you invert Maxwell and Tesla, he tells you Microsoft. So basically it's bias to choose the first asset security dimension in the prompt. So these are many, one of the thousand examples you can do where if you don't build a very strong protocol to control the behavior of this model, you build infrastructure. We are getting there. Now we are building the business cases, but in the middle we need to build very strong auditability protocols. Otherwise, this thing will never fly and we create operational risks.

42:07Laura Watkins:Absolutely. And then finally, how does that sort of then translate towards the consumer? So you've got everything sort of set up in the back end so that you know it's secure. But how do you convince me as the customer that it is in order that I trust it to transact for me? That's a very good question. I personally think that trust will come over time. And I think David at the beginning mentioned that the sophistication and how aggressive and penetrating is the business case has to come over time. First, you start for a simple one, then you gain trust on a simple one, then you keep on evolving the complexity and the sensitivity of the business case.

42:52I think that's the only way. But without the technology and the tools, without the MRI machine, the scan, the agent behavior, I don't think we're going to solve the problem.

43:01Laura Watkins:For sure. David, you're nodding. Anything to add to that? Yeah, you know, it's something like we're acutely aware of and with our assistant, we started with the ability that you could move money around within the account and now we're about to introduce the ability that you could transfer money between other styling accounts you own and then that'd be a stepping stone to external payments, right? But I think you've got to take that one step at a time. If you were to come out all guns blazing with you can instruct agents to make payments on your behalf, people aren't going to go for that. So I think it's all about incremental steps, kind of get people used to using the technology, what the experiences are like and to Stephano's point as well, wherever possible, you know, give that insight on what's going on under the hood, make sure it's as transparent as possible.

43:43There's a lot of work for us all to do there. Do you find, David, that the ability to build those apps and tools themselves is a step in the direction to getting them more comfortable with the AI? If I've built something, I am likely to trust it a lot more than if somebody else has built something. Are you seeing that change in usage, in behavior, in loyalty from your customers at all? Is that something that you're monitoring? like from the pre-build days and the post-build days? Yeah, absolutely. 100%. We as a team that built it, we'll go in and use the Assistant to do all manner of things. But again, yeah, we have that advantage of, well, we built the outside of that MRI machine and we know how it all works inside.

44:27So it is super interesting to see, actually. We've been kind of pleasantly surprised. We honestly didn't know how it would be received when we launched the Assistant and it had the ability to even move money within the account, even the idea of moving money to Spaces. We didn't know how people would go for that, but actually we were really surprised to see that within the first few months, 95 % that were shown that request for confirmation of you're about to move money, do you want to approve this? 95 % of people said yes, right? So that was really interesting to actually see that when people got to that part of the agendic flow, actually they were quite comfortable to instruct the model to do it.

45:02Whether that translates to payments between accounts or external accounts is yet to be seen, right? But for sure, we thought there was more work we would have to do to even get buy-in to moving money around the account. It's like I'm a lot more likely to eat food that I've cooked myself, even if it doesn't taste good at all, than from an unknown restaurant, maybe. So, yeah, fair enough. And I think that's actually, I think you kind of hit the nail on the head. That's a really important part about building AI, trusting AI is, I think, you know, a lot of people, if they're using these tools, if the AI is making decisions on their behalf and asking permission, you're a lot more likely to say no because you don't feel in control.

45:40Whereas actually the psychology of all this is if you have the idea, you're a lot more likely to kind of go along with the recommendations of the assistant if it's kind of reaffirming the original idea you had. We saw this, for example, with one of the first customer-facing AI features we launched was something called scam intelligence. We could upload a picture of something you were going to buy online, say eBay or social media marketplace. And you can upload the image and it would just tell you the red flags of any potential scams. And we saw that people that use that tool were three times more likely to not proceed with the payment versus if you showed the usual kind of payment screens, which are like, are you sure you want to do this?

46:20So people usually just click through those. And once you're in the position of no, I'm going to make this payment as a bank, no matter how many screens you show warning, are you sure people are just going to more often than not proceed through it right but if you actually you're given that information about the red flags and you make a conscious decision of oh i'm not sure about this the psychology is a lot stronger a lot more likely to go on the recommendations of the ai so i think there's a lot of that as well to actually building

46:48Laura Watkins:uh trust and engagement in these tools for sure i think we've uh come full circle it's so much like human behavior um that needs to be factored into uh into how the technology is used as well but on that note we're going to take another quick pause and we'll be back very shortly

47:08hey Chicago class it up with Crocs you know back to school is coming in fast so why wait to find your new fave footwear step into a local croc store and step into your new look try it style it make it yours because the right pair doesn't just show up it shows off first day fits and

47:40Laura Watkins:Okay, now for a quick look at stories we don't have time to cover in full. And this week we're looking at Seoul's financial watchdog targets AI hallucinations. The story in FinTech Global. South Korea's Financial Security Institute has developed the country's first dedicated evaluation framework for assessing the reliability and safety of AI used by financial institutions. The framework is designed to address risks including AI hallucinations, model errors, bias, security breaches, and unreliable outputs, as financial institutions increasingly deploy AI across core banking functions. The framework draws on existing South Korean regulation and guidance, as well as international standards, including ISO IEC 42001 and tools developed by the UK AI Safety Institute.

48:28Laura Watkins:The Financial Security Institute plans to begin pilot testing in the second half of 2026, with full evaluations expected to start in 2027, initially amongst its member financial institutions. So, the Korean market is not one that we cover very often on this show, even though it's very easy to imagine them being a very tech-first market. So, it's good to see them leading from the front here, but also drawing on local data and regulation, as well as looking further afield for inspiration. Obviously, sitting here in the UK as I am, it's interesting to see them choose the UK AI Safety Institute's tool in particular.

49:05Laura Watkins:And what they're looking to do is assess AI systems across 10 criterias in two areas, reliability and safety. Reliability covers areas such as model performance, data quality, fairness and bias, and the ongoing monitoring of hallucinations and performance degradation. and the safety assessment covers areas including AI-specific cyber threats, protection of AI assets, third-party models, supply chain security and leakage. So obviously when you're talking about AI, the best things you can do is make it reliable and secure and they are launching this in early 2027, which probably can't come soon enough at the rate that AI is being used as discussed in basically all of the stories in today's show.

49:47Laura Watkins:So it'll be very interesting to see if other markets follow suit and or borrow the same frameworks that South Korea are laying down. So we'll keep an eye on this one and see how it evolves. And now I'm going to move us to our and finally section. Our and finally this week is£173 ,000 of government spent on influencers, a good value for money exercise. So this story on the BBC, the Isle of Man government has spent£173 ,000 on influencer marketing over the past five years, according to figures released following a Freedom of Information request. Visit Isle of Man has used social media creators to promote the island as a tourism destination, arguing that influencers can help it reach audiences that might not be reached through traditional advertising.

50:31Laura Watkins:Around£35 ,000 of the total was spent hosting influencers on the island, with that money going towards local tourism businesses involved in producing the content. Australian travel creator Daniel Hertzberg, who has more than 68 ,000 Instagram followers, visited in 2024 and said content from the trip became some of his most viral that year. He also said that people have contacted him saying they have subsequently planned trips to the island. The spending has prompted debate over how governments should measure the return from influencer marketing and whether authorities should instead work with local creators who already promote their destination.

51:08Laura Watkins:Okay, so what do we think on this? I think, you know, influencer marketing is very increasingly popular. You know, we've seen so-called finfluences and celebrity crypto endorsements. You know, what do we think of this? Is this a good use of government spending? Does anyone have any strong thoughts on this? No, it's not my set of expertise, so I can use the common sense here. And I think in my small experience in processing with AI, Social media data, I think is very easy to understand the traffic about some influencer activity. It's very difficult to do the attribution and so link to the real value.

51:57That's my only comment I can make. It will not be an expert on the field, but from the pure data science perspective, I noticed that it's very difficult the physical attribution or the return. Yeah, for sure.

52:10Laura Watkins:That's 100 % the biggest problem with it. I think you can spend such a lot of money and you don't know if it worked. You know, the fact that they're highlighting kind of qualitative data that somebody told the guy, they've gone ever since seeing his post, but other than that, they don't know. It's a difficult one to kind of show return of investment for sure. I actually laured the principle behind it. The fact that the Isle of Man government thought about doing something else and they're open to non-traditional, non-government-like ways of promoting tourism. I actually think that's a great thing.

52:52If you think about that number also, across five years, it is not really that big a number. It's maybe the salary of like one person who would then spend time marketing and spend time kind of doing the traditional ways of promoting tourism over there. So from that perspective, as long as the return on investment holds up, I think it's something that should be both admired and perhaps built up a bit more the right way rather than dismissed out of hand because it's unusual.

53:29Laura Watkins:That's an interesting perspective. And yeah, like kind of, yeah, 173 ,000 over five years is not an enormous marketing budget. Let's be clear on that. David, any strong thoughts on this? I guess on the theme of the podcast, if it was mine, maybe I'd spend it on tokens. But again, I'm not sure. $173 ,000 over five years is a terribly great token budget these days either. So I'm not sure how much return you'd get on that. But I now need to go and find videos from this Daniel Haysburg guy to see what he actually did. He came all the way from Australia to the Isle of Man. It ought to have offered him something good, I think.

54:10Laura Watkins:I wonder if he was disappointed when he got there. Maybe not. There's that other angle also, where like in financial services also, we know, for example, that a majority of Gen Z are getting their financial services knowledge from TikTok, for example. I wonder if there is something to the fact that more people are likely to believe him precisely because he is not a local and therefore he might not be biased. He has seen Australia. He has seen multiple other places. And then when he says that this place is also good, it gives it a bit more credibility than somebody who's obviously from there loving that place talking about it.

54:45So that's another interesting angle.

54:49Laura Watkins:That's true. We're kind of back to our trust and human behavior debate all over again. And that maybe is a full circle moment to end the show on. So thank you so much to you guys for joining me. where can people find out a little bit more about you and your companies starting with you David yeah LinkedIn please do feel free to reach out to me and the Starling Bank website we have a blog on there we post we have a newsroom we post whenever we're launching new features so lots of information there as well which apparently is now weekly we hear weekly for the smart tools yes we'll be busy thank you and Stefano our website provides a lot of information LinkedIn we're very active in promoting themes about trustworthy AI, AI you can own and not rent.

55:42So we are very active on LinkedIn and our website and publication. We publish a lot of technical paper. Brilliant. Thank you so much. And Aditi, please find me on LinkedIn, message me, talk to me and find out more about Snowflake at www.snowflake.com. Fantastic.

56:00Laura Watkins:Thank you. As for me, you can find me, Laura Watkins on LinkedIn on 11FS.com, 11FS.com forward slash after dark, if you want to get your tickets and come and join us, or otherwise just on this podcast. And that wraps up today's episode. Thank you so much for listening. If you like what you've heard, please make sure you follow us on your favorite podcast platform of choice and feel free to share the podcast with a colleague or friend. And if you want to join the conversation, find us on social media, just search for 11FS or Fintech Insider or email podcasts at 11FS.com. Thanks again and goodbye.

56:39You matched on Hinge. You're vibing. Then her energy completely changes. What do you do? I'm Raven Smith and I wrote and read a real love story about this exact sitch. Listen to the free audiobook now. The Viore Core Short moves with you. With everyday versatility and classic athletic fit. it's the one short for everything your day brings invest in your happiness and get 20 % off your first purchase at viore.com slash 420 that's v-u-o-r-i dot com slash k-o-r-e-2-0 exclusions apply visit the website for full terms and conditions hello look what tj max dragged in the devil wears prada 2 is now streaming on disney plus We are digital.

57:32We are downloadable. We are streamable. The fashion event of the year is certified fresh. Pull yourself together. We have work to do. Critics say it's smart and witty and the perfect sequel. That's all. Get runway ready for The Devil Wears Prada 2 on Disney Plus and Hulu. Rated PG-13.