• yesterday
Pegah Ebrahimi, Co-founder and Managing Partner, FPV Ventures, Rajat Taneja, President, Technology, Visa Moderator: Terri Burns, Founder and General Partner, Type Capital
Transcript
00:00Okay. Pega, Rajat, really excited to have you both here today. Rajat, maybe we'll
00:05start with you. Visa, of course, is at the forefront of leveraging AI for risk
00:11prediction. Can you tell us a little bit more about this? Tell us how Visa
00:15approaches the challenges of scaling AI solutions to detect and mitigate AI risk
00:21globally. Absolutely, Terry. You know, Visa is the OG of SaaS companies and it was
00:27one of the original SaaS platforms and leaned in to AI back in 1991. So core to
00:37Visa, managing about a billion transactions a day, connecting 4.6
00:42billion credentials to over 150 million merchants, is the ability to do it in a
00:49trustworthy manner, in a secure manner, managing the payment security and
00:53cybersecurity. So our fraud models have been at the core of what we do with AI
00:59for over 30 years and we went from good old-fashioned AI to deep learning to
01:04neural networks to all the predictive AI pieces to now leaning in to generative
01:10AI, helping us with models with the problem of cold start for something like
01:15a real-time payment network to figure out if there is any problem going on
01:20with account that are being enumerated to see if bad guys are trying to see if
01:25cards are legitimate before they use it for nefarious activities. Real-time
01:31management of this and it's all about investing in the data platform. Spent
01:36about three and a half billion dollars in the last decade to create the data
01:40platform for the future and it's all coming together now. I want to talk more
01:44about how you and your team have really embraced this culture, but first,
01:47Pega, you have had a fascinating journey from CIO to investor. Can you tell us a
01:52little bit more about trends that you're seeing in AI and specifically what do
01:57you think could have the biggest impact on enterprise transformation or even CIO
02:02decision-making? Yeah, look, I was a CIO during the first mobile, when
02:07mobile wave was coming and it's like it's super interesting because you see
02:10you know everyone's like AI is so different than before and yes there's a
02:13lot of things that are different in the pace that it's growing but in a lot of
02:16ways history doesn't repeat itself but it rhymes and it reminds me so much of
02:21when mobile first came and everyone was there's all these mobile companies that
02:25existed right and for the most part nobody says a mobile company anymore.
02:29Every single company that came was on cloud is now born in the cloud and using
02:34mobile and it's part of being a company and I would say that's the same you see
02:38it happening with AI. At the beginning it was like AI company, AI company and I
02:42think now there's definitely going to be foundational infrastructure AI company
02:46but for a lot of the real great companies that are going to form it's
02:49going to be how are companies starting and taking advantage of the fact that
02:53these inherent advantages exist because of AI and to build something that they
02:59couldn't have built before, right? Similar to when Uber, nobody thinks of Uber as a
03:03mobile company but it wouldn't have existed without mobile. If you don't have
03:05your phone walking around you couldn't really do that business model so that's
03:09the part that's really exciting it's how do you take things that people just
03:13assumed would have to be services businesses, things that people assumed
03:16would have to you know depend on so many people to get it done or it couldn't
03:22have been done because of the scale it needed, those are the parts that are
03:25really exciting that will be the next wave of AI that I think are going to be
03:28exciting startups. Let's talk fintech. This is a question for both of you. Do
03:33you believe that fraud prevention presents the most significant
03:37application of AI and fintech today and if not what is something that is
03:41commonly overlooked that you each are really excited about? Look I think fraud
03:46is table stakes. Managing it, managing it well, staying ahead of the bad guys
03:51because they're very proficient. The same tools are available to them. If cyber
03:56crime was its own country in 22 it would be a seven trillion dollar economy
04:01growing at 32 percent. So it's very profitable, it's very big and so fraud
04:06and payment security and cybersecurity are existential. We have so many attacks
04:11on our network, millions that you have to stay ahead of and AI is a very
04:17important tool there. But I think there is the other side of AI. How do you use
04:21AI to reinvent and rethink commerce? If you think about it the last big major
04:26change in commerce was with e-commerce and then mobile and they continue to
04:31break barriers of shopping in the storefront or online. Now comes the
04:36opportunity to move from the hunt-and-peck era of shopping. We are
04:40still searching for things, you're analyzing it, you're trying to figure out
04:43what to buy, then going to the buying mechanisms to what we are calling the
04:48self-driving commerce era where AI takes away some of the mundane aspects of
04:53discovery and of helping you with warranty management, price matches,
04:58registration of products and then brings the joys of commerce back to you in ways
05:02that you're discovering things that you would like to see whether it's products
05:06or travel or art and then helping you really enjoy and appreciate the
05:12pleasures of shopping. I think that's where a lot of the innovation is going
05:15to happen in addition to the basic foundation of core infrastructure, fraud
05:20and security. Yeah and I always loved it like as you know a CIO, it was
05:25always nobody, SaaS companies had a hard time selling to financials because at
05:30the time if you think about it all the things in financials were on-prem and it
05:33was just really difficult and I would say what's really exciting about this
05:36time around and how it took financial companies in a lot of ways so much
05:40longer to get advantage of the digital transformation trends because they had
05:44to really take their time, make sure their data was secure, do it in a much
05:47longer time horizon. Right now financial companies I'm seeing move way faster in
05:53AI than some of the other industries right because it almost was they had to
05:57get set up, they had to care about data, they had to care about some of these
06:00regulations and some of those parts make it easier in certain use cases for them
06:06to adopt faster and I was always when I went into tech I was at Cisco and it was
06:11always interesting how everyone thinks of financial companies as old stodgy
06:14companies. No offense to us right and in so many ways financial
06:19companies depended on such razor-thin margins that in so many areas they had
06:23to be super efficient. Yes they weren't the early adopters of technology but
06:27efficiency is so core to financial organizations and I think you see it now
06:32some of the financial companies, Visa, other they're moving so much faster on thinking
06:37where are the areas where we could actually take advantage of some
06:40of these technologies for margin expansion, for what services do we have
06:45on the fraud side, on customer service, on financial advisors. I think that's
06:48incredibly interesting you look at you know there's going to be 80 trillion of
06:52wealth transfer happening from baby boomers in the next few decades and so
06:57many of the you know it only 1% has access to financial advisors because
07:01it's just not a scalable by person business but so many people that are
07:05used to technology and all those things I think will get advantages of that that
07:08they couldn't have before and I also think there's just there's there's gonna
07:12be a myriad of other stuff that comes from building efficiencies in the orgs
07:16to all the to actually consumers getting the benefits of it. Yeah I think it's a
07:19really interesting point what in recent history might have been seen as kind of
07:23old-school is now really at the cutting edge and I think that both of you
07:26alluded to that in your responses. In a couple of minutes I'm going to turn it
07:31over to the audience so if anyone has questions start getting those ready but
07:35first I want to talk a little bit about AI transparency. So there's this ongoing
07:40debate with AI models particularly open source versus closed source. I know this
07:46is something I think about a lot as an early stage investor. What do you each
07:50think is the right approach for early stage companies? Well I mean I'll talk
07:55high level and then we're going to talk about what specifically they're doing.
07:58I think there's no one formula. I think it's so early that everyone's
08:02experimenting with a lot of different ones but I do some of the things that
08:06are saying parallel. Look I remember when the cloud migration was going and people
08:10were like oh do I choose AWS or Azure or G I don't want to get stuck on this
08:14cloud. Now it's even fundamentally even I would say more existential because
08:19people are forget cloud lock and needing a few years and a bunch of
08:23engineering resources to migrate. If you have like model lock you got seven years
08:28in and the prices change or the dynamics change and now you have all this
08:33investment that you did on really bringing your model up. So forget model
08:37lock I think that cloud lock model lock is even becoming more key and so
08:42because of that I think CIOs have saw that game played out so they're just
08:46going at it more eyes wide open this time and thinking I don't want to be
08:50stuck if something changes. If open gets closed which is happening to some
08:55companies right and if things like that happen and I think that's that's the
08:59learnings from that cycle have made CIOs just want a lot of different
09:03opportunities and knowing how to have to be more in control and not stuck on
09:07something and you see it at Visa you guys are doing it. I think that so well
09:10said Pega you know I think there are pros and cons to open and closed and you
09:15described it really well with the model lock. I think the core is for there to be
09:20radical transparency. A key pillar of a model is the data. You have the
09:27algorithms and you have the GPUs but it's really about the data and these are
09:31general purpose models that are built and trained on massive amounts of data.
09:34We know the data out there has got noise in it there is data that is error
09:40prone there is data that is just blatantly wrong and so as more whether
09:45it's an open model or a closed model as they're building it to explain with
09:49transparency what an explainability how they use the model how the algorithm
09:54created fairness consent privacy copyright infringement issues is super
10:00important and for companies. Rajat can I interrupt you though can you have true
10:04radical transparency with a closed model? It depends well it's difficult right
10:09because it depends on what they are saying and the community is not able to
10:13go in and therefore you need to think about the regulations Terry that come in
10:17and help create a framework for understanding of what happened in the
10:21model so that the practitioners can use correctly. If models are going to be
10:25transformated in medical and health care and diagnostic tools or finding new
10:29drugs you need to know that right and therefore I think the role the
10:33regulators will play in creating that framework use based interoperable
10:40because models don't know boundaries of countries so you have to have global
10:45regulations that allow us to have guardrails and think through what went
10:49into it and how it can be tuned and how it can involve so they are helpful and
10:54they are able to do what they are supposed to do. Yeah I think that's fair
10:58I think my personal bias is towards open source but I think you're exactly right
11:01there are very clear use cases where for data protection or sensitive situations
11:06closed source but with still some transparency makes sense. So right now we
11:09have an ensemble we have a number of different models at Visa and we have
11:13abstraction layers on top of it much for the ways that Pega was saying but not
11:17getting a model lock-in. Yeah for sure. All right let me turn it over to the
11:21audience let's see if there are any questions. Okay it looks like there's a
11:25hand raised here we'll give it one second for our mic runner please do say
11:30your name. Hi Maria from Freestyle Capital I'm curious how you guys think
11:34about agentic payments evolving. I know that Visa has been thinking about I'm
11:39sure everyone has been but do you think that it will evolve on existing
11:42infrastructure or do you think that a new infrastructure will need to be
11:45created for it to work at scale? Look I think payments as a part of commerce is
11:50incredible right so you need high security you need it to be seamless you
11:54need it to adhere to all the regulations and compliance of every country and as
11:59you move into an agentic model where a agent is conducting commerce on behalf
12:04of a consumer it has to have all those attributes and explainability and
12:10management of compliance but it will require a new infrastructure because
12:15you're running models you're in the models are doing inferences the models
12:19are conducting transactions so it will require additional investment in
12:23infrastructure for sure. You sort of mentioned earlier Rajat you also
12:29mentioned this Pega regulation and so I want to kind of double click on on this
12:33idea how can companies balance regulatory compliance with the rapid
12:40rate of AI development how do you both think about that? So I mean I think one
12:46of the look and we're talking about financials which is interesting because
12:50it's been regulated for so long which is why it's been so hard to sell to it and
12:53you need to have a lot of data management and all these aspects I think
12:57what's gonna be interesting is that for for regulation is usually a little bit
13:03delayed I mean from how fast companies want to go and I do think that on for
13:09most folks that are trying to enable AI it's hard for them if you're a large
13:13company that has a high brand risk it's hard for you to be at the forefront and
13:17then get pulled back so what's happening is I'm seeing a lot of companies go in
13:21the areas when they know they're not going to be necessarily hitting some of
13:24those pieces like can you pick low-hanging fruit that you can drive a
13:28ton of efficiency versus trying to go for the moon and trying to go for
13:32something that has a really high risk and I think that's what you're seeing
13:34some of the pullback in companies where at the beginning there's this crazy
13:37excitement about I want to have POC's of every single use case on using AI and
13:43taking it back and saying okay what are the areas where we can do it and we can
13:46have some risk exposure it's not going to be you know existential for the
13:50company and where are those areas that we want to go and I think that's what
13:53you're gonna see the next you know few years until the regulation catches up
13:58because I think people kind of are like I don't know if it's worth it to take
14:01that leap and and have some existential risk but there's so I think you're in
14:05not all use cases are the same like if you're touching personal data versus if
14:08you're having an FAQ that's way more you know way more robust that's that's
14:15totally different areas to it I think as a risk you look I think you have to
14:19think about the spirit of any regulation right if a safe secure trustworthy
14:25models use of data explainability fairness so we have a set of guiding
14:30principles that are called data use policies in our company and we adhere to
14:34them very closely so as regulations evolve whether it is the EU Act of 24
14:39whether it is the executive order in the u.s. the year before that or it's new
14:44toolkits coming from Singapore or from the African Union I think you just have
14:49to stay with the spirit of what you're doing and collaborate with regulators
14:53and academia to create the right framework but if you are open and fair
14:57and transparent and explain things I think you will be always safer with your
15:03models in our last minute and a half here you both manage teams you have lots
15:09of stakeholders and clients and startups that you work with how do you all spot
15:13true value and opportunity from hype as we know there's a lot of hype here what
15:19are both of your really quickly frameworks for how you think about that
15:22I think I think the number one thing is I don't you you can always use a hype
15:27but to me just the best companies don't talk about AI they don't say oh we're an
15:32AI company they say here's the big problem we're solving and here's the
15:35unique insight on how we're actually solving it and I think that's at the end
15:39of the day that's where the huge companies will form we're solving real
15:43problems you can obviously use the advantages of AI you you should because
15:46you're a startup and you cannot a lot but I think those are the best ones and
15:50I think there's a lot of opportunities in the services or look at large
15:53companies they have massive services that they have already figured out how
15:57to extend you it outwards and I think there's gonna be a huge number of
16:00companies form there I think you know exactly what you said move away from
16:05PowerPoint and go to code and let the code talk to you and utilize it and take
16:11it for a test drive and if you're helping it write code software
16:16development use it to write software development and see how it compares or
16:19contrast that visa we gave access to every single employee access to the
16:25models back in 23 we had been using it years before that and had patents and
16:31papers and so we knew that there was great potential and but the real rubber
16:36meets the road when you use it absolutely
16:38Pega Rajat thank you so much for your time

Recommended