How does a company as massive as Mastercard decide where to deploy AI? Instead of centralizing every AI initiative, Mastercard operates under a “hub and spoke” model — balancing centralized AI leadership with decentralized innovation across business units.
In this episode, a16z Partner Marc Andrusko chats with Mastercard’s Chief AI and Data Officer Greg Ulrich about Mastercard’s long history of using AI, the opportunities (and potential risks) associated with integrating generative AI into fraud detection, determining what tech to employ based on use cases, and the best advice he’s ever gotten.
“In the Vault” is a podcast series by the a16z Fintech team, where we sit down with the most influential figures in financial services to explore key trends impacting the industry and the innovations that will shape our future.
Full Transcript
Marc: Hi. I’m Marc Andrusko, partner at a16z. In this episode of “In the Vault,” I talk to Greg Ulrich, Chief AI and Data Officer at Mastercard. We dive into Greg’s fascinating journey from working in the nonprofit sector to shaping AI strategy at one of the world’s largest financial services companies. Our conversation explores how Mastercard has been leveraging AI for decades, the transformative role of generative AI in payments and commerce, and Greg’s insights into building trusted AI ecosystems. Let’s dive in.
Marc: Greg, thank you so much for joining us on “In the Vault.” It’s great to see you.
Greg: Thank you, Marc.
Marc: Yes. Obviously, you and I have had the pleasure of knowing each other for a bit now, but maybe for our audience, you know, we’d love for you to share a bit about your background and your journey.
Greg: Yeah, sure. Thanks. So, Greg Ulrich. I’m the chief AI and data officer here at Mastercard, and it’s obviously a tremendous time to be in AI. We have so much excitement, so much energy, so much opportunity. At the same time, like, the future is still to be written on this. We’re very early days, like, what the future is going to hold, and more importantly, the path we take to get there is still unwritten. So it’s an awesome time to have this job.
My journey in this space, well, I guess it started in university, but I really got into it in working in the nonprofit sector a couple of decades ago. I was trying to help organizations understand the efficacy of different interventions. If you want to address clean water in certain communities, what’s the best way to do that? Or if you want to address malaria, what’s the organization that has the best approach? How do you determine what intervention, what strategy, what approach, what organization is having the most impact on whatever cause you care about? And I realized the data, the analytics, the ability to analyze that in the sector was limited, and ended up working at a company after that called APT or Applied Predictive Technologies. And that was really around how do you differentiate between causality and correlation and how do you identify the real impact of any intervention, any test, any experiment, anything you’re doing differentially in a market. And it has a lot of power for the nonprofit sector and what we’re looking at.
And I wanted two things there. One, how able you are if you have the right data and analytics to understand what’s working and measure that impact, and also how easy it is to misuse data and analytics if you’re not careful about it. And long story short, APT got bought by Mastercard about a decade ago. After that, I did a variety of roles here in our services division. And then, starting in 2000, I led strategy, M&A, and CorpDev for about four and a half years. And then, ironically, I was at an a16z event earlier this year, and our CEO called and asked if I would do this AI and data role. And I’ve been doing this since Q2 of this year and loving every minute of it.
Marc: So it’s fair to say, we’re solely responsible for you getting the job.
Greg: Entirely responsible.
Marc: Yeah, that’s fantastic.
Greg: I think that’s the takeaway.
Marc: Yeah, exactly. Amazing. Well, we’re super excited you’re in the role. And I think we’re in total alignment that, obviously, there’s never been a better time to be thinking about some of this stuff. I’m curious, a company as large and important to the financial and money movement ecosystem as Mastercard surely has been using artificial intelligence for a long time. But, you know, as of, call it, 2022, with the onset of ChatGPT and things of that nature, we’ve now kind of moved into this wave of, specifically, generative AI. So I’m curious, how do you think about the sort of distinction between those two sets of technologies and the various sets of opportunities that exist for Mastercard today?
Greg: Sure. And you’re right, we’ve been using AI for decades. It’s inherent in everything we do in fraud detection, right? I mean, we’ve been monitoring transactions to try to provide insights to merchants and issuers about if a transaction is fraudulent or not, to help make better decisions and make sure that the ecosystem moves in a secure and efficient way. And artificial intelligence is behind a lot of that, as well as what we do in personalization, forecasting, intelligence products we’re giving to our customers. So this has been inherent to what Mastercard has been doing for quite some time. And obviously, two years ago or, you know, two years and maybe a month ago, with the advent of GenAI really to the forefront, it’s created new opportunities. But for us, like, what technology to use, it’s really dependent on the use case.
If you have structured data, you’re doing forecasting models, a lot of the fraud management, traditional artificial intelligence/machine learning is going to be more efficient, more effective, and certainly more cost-effective as a way to do it. When you’re dealing with knowledge management, when you’re trying to synthesize or when you’re trying to create new content, when you’re dealing with unstructured data, then you’re looking at newer and different techniques. So for us, you know, it’s very use-case dependent. We’re trying to understand, what is the problem we’re trying to solve, and then what is the right analytic approach to get there?
And for some of it, we have fraud solutions that are based on, you know, machine learning that we’ve been using for years. Now, we’re able to bring in sort of new features using GenAI techniques to make an existing product better. GenAI, if you will, is a feature as opposed to a product. In other places, we’re using it for digital assistance and other things where it really is a GenAI-based solution. But again, it comes back to the problem. It comes back to the use case and what we’re trying to solve for.
Marc: A hundred percent. Yeah, no, that makes a ton of sense. And I imagine, if you were looking at the whiteboard of possibilities of all the places you could apply this technology, the list is sort of never-ending within the context of Mastercard. But as I was preparing for this conversation, I spent some time looking at some of the announcements you all have made, and it sounds like two of the early things you’ve shipped with respect specifically to generative AI have been, to the point you just made, one, a digital sort of onboarding assistant for some of your customers, and then two, are specific new fraud capabilities that maybe actually are leveraging the generative parts of AI and not just the traditional machine learning techniques you alluded to. Maybe you could just tell us a little bit about how you sort of landed on those two things as early use cases to launch in the organization and how they’re going thus far.
Greg: Sure. So I’ll get to those two specifically, but maybe I’ll give you the frame of more over how I think about where AI has opportunity for Mastercard, and I’ll bring it into four buckets as shorthand. They’re safer, smarter, more personal, and stronger. I’ll explain what I mean by those. But as I mentioned earlier, like, where we started in our AI journey, where a lot of what we’re doing today is in fraud management, fraud detection, fraud identification, and how we’re doing things to make the ecosystem secure. That’s on a transaction-by-transaction basis. And how do we provide intelligence to determine if a specific transaction is fraudulent or not? But also, how do we look at scams? How do we look at pain points across the entire ecosystem to see if there are hotspots in the network where we’re seeing more fraud and there might be malicious actors? And how do we shut that down, and how do we deal with it? So that’s how we make the eCommerce ecosystem safer.
The second one is how we make it smarter. That’s, how do we route transactions in the most efficient way, how do we provide insights to acquirers, issuers, merchants to help them optimize their portfolio and deal with their own customers in the most effective way, providing intelligence based on the data we see and the analytics we can deploy on top of that, when should you authorize a transaction, should you retry a transaction, etc.
The third one is how we make the ecosystem more personal. So we are a B2B2C business. We don’t have much in the direct-to-consumer realm, but we do help our partners, the banks, the merchants, personalize things so they can provide the right offer at the right time to the right customer. And we have software tools that do that, and a lot of those rely on artificial intelligence. And that’s the more personal element. And all of those are external. How do we create value for the ecosystem? How do we create value for our customers?
And the fourth bucket is, how do I make Mastercard stronger? How do we improve our own operations? A lot of that is on the productivity, the efficiency pieces. It’s around how do we put knowledge in the hands of our employees, how do we make their roles simpler, be it software engineers from coding, our sales team by providing better access to information. And those are the sort of areas in which we’re deploying AI and particularly GenAI overall. So I’ll put these things that we’ve shipped into those buckets.
The first one around fraud, the product we have is called Decision Intelligence. And that’s where, in the milliseconds we have between when a transaction passes through our network, from when you tap to buy something in a store, it passes through Mastercard to go to the issuer, we’re providing a score on that based on how likely that is a good or bad transaction. We use GenAI, recurrent neural networks, and other technologies to effectively add features to that by understanding not just your transaction history but the overall ecosystem of merchant behaviors and create, basically, a merchant vector database in there to help understand whether or not this is likely to be a good or bad transaction, even if you’ve never shopped at that merchant before, because we can see like behaviors from other consumers. And so it’s basically adding a feature to the model to give us a more accurate score, and we’ve been able to prove that over time. So that’s one of the things we’ve done in fraud.
In personalization, we have something called Shopping Muse, which is basically enabling the in-store experience online. So you have a chatbot, effectively, where you can type in your own language and ask for recommendations. “I’m going to be on an a16z podcast with Marc. What should I wear?” And it’ll give recommendations. Or, “I’m going to a wedding,” so and so, “what should I do?” And it’s allowing you to interact like you would with an in-store agent using all of GenAI techniques to make, again, personalization, a better experience for the end consumer.
And then the digital assistant you mentioned, one of the things that we’re trying to do is making it easier for our customers, the banks, the merchants, to consume and integrate Mastercard products. And these have a lot of technical specs, a lot of steps sometimes to bring them on board within the bank ecosystem, within the merchant ecosystem. So what we’ve done is we’ve created a digital assistant that uses RAG and points to all of these technical databases, the Q&A’s we’ve had over time where we’ve tested and fine-tuned these models to enable customers to onboard our products easier by automating a lot of manual tasks and by creating an environment where they can ask questions and we can get answers back to them much more rapidly. We do put a human in the loop of this. So the chatbot is actually directed at our agents, but it still really reduces the time that it takes a customer to bring on a Mastercard product, which helps them drive value from those much more rapidly.
Marc: Sitting in my seat, I’ve had the privilege of speaking to a lot of folks at large enterprises who are trying to make sense of this GenAI wave and figure out what solutions do we want to buy, what solutions do we want to build, where do we want to partner, etc. And I think there is this notion that, you know, one of the most hazardous things a super early-stage company can do is go to a prospect and say, “Oh, well, we built this amazing AI thing, but in order for us to prove how valuable it is, you have to share a bunch of your enterprise’s data with us.” And it’s like, “Well, hold on just a second, you know, we have to safeguard that asset.” Whereas, again, you all occupy such a privileged position of being the trusted partner to millions of merchants and acquirers and banks. And I think it probably makes for a much easier ask to be able to leverage this data in ways that are mutually beneficial, understanding that, as you always have, you will continue to safeguard it. So I think that’s it.
Greg: Yeah. And you need to make sure they’re partners that have the same values, the same approach, the same commitment to working through these things as you do. Otherwise, you get into a difficult situation. But we put the safeguarding of data at the absolute pinnacle of what we need to do here because we recognize the downside and we recognize the importance of maintaining the position we have as a trusted brand and trusted ecosystem. I mean, it’s what enables our payment network to thrive the way it does. It’s when you’re in a place you’ve never been, shopping at a place that you’ve never shopped, that you know if you tap your card or dip your card or swipe your card, but the merchant knows they’re going to get paid, you know you’re going to get the product, you know if there’s a problem, you have recourse. The ecosystem works, and it’s because the network enables trust. And that goes the same with how we use data and how we treat data and how we treat the products and services that sit on top of that.
Marc: Yep. Well, I do have to ask, given I’m sure there are many founders and operators who are listening to this, many of whom don’t have the benefit of decades in business as a trusted and known entity but are very eager to work with the Mastercards of the world. I’m curious, as you think about which solutions you’re interested in sort of buying or partnering or plugging in a vendor for, and you’re approached by, you know, an earlier growth stage technology company, what are some of the criteria you and the team are using to help you decide, you know, where are you going to prioritize your time and resources and the types of players that you’re excited to do business with?
Greg: Yeah. So a couple of things there. Where we’re prioritizing, where we’re spending our time still comes back to where the areas that we think that we’re focused on AI, the safer, stronger, personal, smarter. We then look within those about, if it’s a new GenAI solution, where there are manual tasks or unstructured data, where GenAI is actually the right tool for the problem back to, you know, when you’re using GenAI versus traditional AI solutions. And then it comes down to sort of the feasibility, the viability of the different opportunities that we’re seeing in those spaces we’re looking to prioritize what we’re doing.
Mastercard has a long history of partnering with early-stage companies. We do it in a variety of ways. We’ve embraced the fintech community, and we have relationships with so many early-stage fintechs in our ecosystem. We have a program called Start Path that sort of enables early-stage companies to plug into the Mastercard network, get the value of what we bring, as well as some support alongside that. We’ve had connections from my old job in corporate development about where we’ve invested and partnered with a lot of early-stage companies. So we embrace the early-stage community, and we’re looking to partner with them whenever we can. We’re strategy-first. It has to fit within where we’re trying to deploy solutions based on what needs we’re solving for a customer and then sort of the prioritization areas that we have within that. But then, if there’s a company that has sort of, we think, the best approach to that and we think that they fit within our approach to governance and security, then we’re always happy to partner if we think that that’s the most efficient way forward.
Marc: Yeah. And in that process, I’m curious, so you now occupy this really interesting sort of top-down horizontal leader charged with making sense of this AI wave and making sure that the organization at large is taking advantage of it. But I also imagine that there are P&L owners, GMs, folks in a more vertical setting who are making similar decisions for their specific business unit, and they can help answer those questions of how, you know, specific GenAI capabilities are going to make them smarter, stronger, and sort of all of those criteria. For founders who are looking to navigate the organization, both Mastercard but also generally speaking, where there is a senior horizontal decision maker charged with innovation or AI and then there are vertical business leaders, how do you think about…could you articulate sort of how decisions get made in that capacity? And I think you’ve described to me how that spoke in the past. Like, maybe just walk folks through how that works at a big company like Mastercard.
Greg: Sure. So what we have done in this structure that our CEO put in place earlier this year was create a new group for AI and data. Now, that’s not centralizing every activity in a company the size of ours. Again, artificial intelligence, like electricity, it’s just a part of so much of the organization it would be foolish, dangerous, and probably impossible to try to strip out all of those elements and centralize them into one place. At the same time, you can have people trying to solve similar problems in different places in an uncoordinated way, which is problematic. So the way the hub and spoke model works is I and my team from an enterprise-wide standpoint are looking at where we think the technology has the greatest opportunity to drive value for our customers and to drive value ultimately for our shareholders. Where do we think that there are opportunities? And I bring that to the business partners, in HR, in finance, in fraud securities, in open banking, in our core payments business, to talk about these opportunities.
At the same time, the leaders of all those businesses are always trying to innovate, always trying to find new solutions, and they’re coming up with ideas, and they’re bringing them to our team. And you have that two-way collaboration and discussion. And what we now enable is, one, if they’re trying to go about something, we have a lot of learnings across the organization about what’s working and what doesn’t. A lot of this requires the right access to the data, the right tools. We want to reuse things and patterns that we’ve seen work and avoid those that we haven’t. And so there’s a lot of that learning that my team can bring to that part of the organization to make sure we’re innovating in the best way and the most efficient way. And at the same time, we’re not trying to own the ideation and the product development and the business development that those groups are doing. That’s their job. That’s what they’re doing by running their businesses. And they’re coming up with those ideas and bringing them to us.
So it’s that two-way street communication where we get the best ideas and then also make sure that we’re not being duplicative, or if we found an approach or a vendor or model that’s efficient, that we build on top of that, because that’s going to be a faster time to market. It’s going to be a more effective solution. We know it’s going to work within our governance and our privacy rules and rubric.
Marc: Yep. And then, what about for, like, ongoing, okay, you’ve implemented a vendor or a model or you’ve leaned into a solution in some capacity, and it’s been in market for a year, and you’re trying to assess, did they deliver upon the ROI, or did we collectively deliver upon the ROI that we thought we were going to deliver? Are you thinking about that process separate and distinct from what I’m sure is a very robust vendor management motion that already exists at the firm? Or at that point, is it, “Okay, the current machinery we have in place already does a good job of figuring out which vendors are working, which vendors are duplicative?” Like, does it get kind of put into the machine, or is it still kind of carved out as, like, “This is new and exploratory in AI land and we need to figure out a different way to measure ROI?”
Greg: Yeah. Like, my team owns that. We’ve sort of spun up an approach where every new initiative has a key set of KPIs. We establish targets for those, and we have a plan to measure those over time. I think, particularly at this stage of these new GenAI use cases to find out what’s working, how we’re driving value for our customers, how we’re driving value for our employees, how we’re driving value for the ecosystem, it’s important to look at those specifically. So if I look at what we’re doing for our software engineers and developers around coding assistance, yeah, I want to understand what the efficiency is for them. I want to understand we have different interventions for training and pointing into different elements of the life cycle if that efficiency gets higher. But just as importantly, or more importantly, what’s the feedback? What is the customer satisfaction with these? Like, if we’re taking away parts of the job that people don’t find as exciting, then they can redeploy that time to answering harder questions and doing more thinking to improve the throughput of the ecosystem, and that increases satisfaction, that’s really what we’re after.
And so you have different sort of measurement and KPIs for different initiatives, but right now, I am looking at everything that we’re rolling out there, trying to establish, like, what are we trying to measure, measuring that over time, and making sure those learnings come back so we figure out where we’re going to deploy effectively, you know, our capital and scarce resources to have the biggest bang for the buck. While we do that for everything we do, I am taking a slightly bigger focus and putting a shinier light or a brighter light on some of the GenAI solutions that we’re putting out to market.
Marc: Yep. One other question I had was your role is really interesting in the sense that I think Mastercard has almost 33,000 people, it’s a very large company, and in order to figure out where AI could have the highest potential impact, you have to have a very deep understanding of the operation of the organization, which, for a company that large, you could spend all of your time kind of mapping out and figuring out. Simultaneously, to make sure you’re picking things that are best in class and the breakneck speed at which this software is developing, you kind of also have to have an ear to the ground of what’s happening outside the four walls of the organization and kind of stay up to date with all of the rapid progress that’s being made. Like, are there any either sources or processes or rituals you have to make sure that you’re kind of staying up to date on on the latest?
Greg: Yeah. I say it’s a combination of a few things. One, I guess some of the good news is in my last job, and the strategy role gave me pretty good training of trying to keep an eye over a very broad ecosystem and what’s happening in that. But the amount of time I spend listening to podcasts, reading external news, and also just networking and talking to others both about how they’re deploying AI, how they’re organizing their organization, like, “This hub and spoke model, there are a lot of ways to do it. What do you federate? What do you centralize? How do you coordinate most effectively? How do you balance innovation and speed with the right governance?” like, there’s a lot that you can learn from others. So I spend a lot…like, just today, if I think about it, I’ve listened to a handful of podcasts. I’ve read a bunch of articles. I’ve done two networking calls. We’re doing this podcast. It’s a lot more time external. We had an event last night with a lot of external parties at a dinner, which was just sort of a sharing ecosystem and talking about what we’re doing and trying to learn from others.
When we presented to our board of directors on AI in September, one way we structured is I had seven external speakers come in, right, to give different perspectives from an investor perspective, the CEO of one of the leading LLMs. We partner with Databricks quite a bit. We had someone from their C-suite come in. We had academics come in providing these perspectives. So our senior leadership, myself, my team are hearing a variety of different perspectives because it’s changing rapidly. And unless you keep an ear to the ground, like, you’re not going to stay up on the latest and greatest.
Marc: Yep. And while you and I sit here and, I think, can clearly articulate why we’re so excited about this technology and where we see potential for ROI, I’m curious, you know, Mastercard, again, sits in this really interesting position at the nexus of issuing banks and acquiring banks and processors and merchants, all sorts of distinct constituents in the same sort of payments and commerce ecosystem. As you’ve been out in the market, thinking about the future of this technology, are you finding that everyone is ubiquitously excited and eager to adopt it, or are there any constituents in that map that I just drew out who are really hesitant or, you know, are worried? Like, would you say it’s been unbridled excitement, or are you seeing kind of both sides of the coin?
Greg: I don’t think it’s unbridled excitement. I think there’s a lot of concern out there, particularly in a regulated industry, like some of the places in which we touch and in which we work. There’s enthusiasm, but it’s definitely not unbridled, as people are very worried about the accuracy, the hallucination, the efficacy of these. And particularly as you start pointing these to customer-facing solutions, there’s a very high bar on what that has to achieve, and I don’t think a lot of these solutions are at that point, even with fine-tuning, even with pointing at the right data. And so a lot of what we’re seeing is a human in the loop or putting things behind a human to enable your own employees to get the insights and information but sort of holding off on the direct application to the consumer.
Now, not every organization is the same, and some people are leaning into this more, but I think there’s a lot of people that want to see this proven out and want to see the continued improvements in the underlying models, the improvements in the accuracy. And we are early days. Like, we’re seeing improvements every time there’s a new iteration in accuracy, in speed, in latency, in cost performance, you know, and these things evolve over time. And so I think everybody’s a little bit patient on this as well to make sure that we take advantage of this while also managing the “do no harm” element of some of it as well, particularly in financial services.
Marc: Yep. Maybe just to close out, you know, the world is your oyster in some senses in this seat. Like, if I had to put you on the spot and you had to pick one or two things that you’re just brimming with excitement over, like, what are you thinking about most for maybe the next year or two? Like, what has you most excited?
Greg: So I think the things I have my eye on for the next year are, one is how we interact with models and AI is changing, I think, in a couple of ways that we’re seeing more and more, one around sort of multimodality and being able to look at not just text but the integration of text and images and voice and video, and for a lot of instances that we think about. Like, you could think of what the insurance companies are now able to do, where they can piece together photos of an accident with the data from the policy, from the information from, you know, their manuals and what they have internally to come up with a much faster and more effective estimate. When we think about things in financial services, like invoice reconciliation, bill pay, you’ll have these things come in as PDFs, you’ll have them come in as images, you’ll have them come in as text, in a variety of ways, as we start looking at things holistically in different modalities and still come up with a single source of truth that’s very exciting.
And then I think the reasoning models and the evolution we’re seeing on those is very exciting as well. It opens up new use cases. And for me, what’s exciting is some of the experimentation I’ve seen on those where it’s not just the accuracy that is improving but the percentage of time, the significant increase in percentage of time that those models will basically say, “I don’t know the answer to a question,” which, you know, as we go back to risk and we go back to, you know, how models are improving and what we need to see, knowing the limits, just take AI out of it, if you’re a human being, knowing the limits of your knowledge is incredibly important.
I remember when I took my last job and I was presenting to the board every couple of months, the best advice I ever got was, “The best thing to say when you don’t know the answer is I don’t know the answer. The worst thing is to try to fake your way through it.” And that’s true in life, but it’s certainly the case with AI. And AI, if it doesn’t know the answer, it still sounds very lucid and erudite its wrong answer. So it’s not just wrong, but it’s confidently wrong. I go back to where we started this conversation, Marc, around APT and the ability to misuse analytics and data, and the challenge and the danger that has is real. And so as these reasoning models are understanding their limits better and knowing, “Hey, I don’t know the answer to that question,” much more effectively than some of the foundational models, I think that’s an exciting evolution in this space. So the first one is around how we’re dealing with AI is changing.
The second one is I do think that this move to trust, this moved to responsibility, how that’s incorporated, how we bring transparency in, still has its day. And we’re leaning in hard on that dimension. I think that’s going to be a big trend, not just in 25 but for the next few years. But I continue to think that how we do this in a trusted way is really critical. And then I think the third thing is we’re seeing…and who knows where the scaling laws will go and how much better the models will get as we add more data, as we add more compute, as we add more time to train them. But I do think that the use of data becomes a more critical differentiator, inference becomes a more critical differentiator, and I think that’s the other thing that we’re looking at. And that matters quite a bit for us, given the data that we have and the data we have inside our organization that we can use and leverage to help bring better insights to our employees and help make better solutions for our customers.
Marc: Yep. Well, I got to say, talking to you reminds me that we’re both very fortunate to sit at the front row of all this stuff that’s unfolding. I’m leaving this conversation fired up about the many opportunities that lie ahead for Mastercard with both generative AI but also generally. And so I really appreciate you spending time with us today. Thanks for coming on.
Greg: Thank you so much, Marc. I really appreciate the time.
Timestamps
00:00 – Intro
00:28 – Greg Ulrich Background
02:49 – Distinction between AI and generative AI
05:22 – AI opportunities at Mastercard
06:24 – Handling fraud, making the commerce ecosystem smarter and more personal
10:02 – Digital Assistant to allow customer integration
10:59 – Safe guarding customer data
12:50 – Criteria for resource allocation and early stage partnerships
15:11 – How AI decisions are made
18:37 – Measuring ROI and KPIs
21:03 – Staying up-to-date on AI technology
23:35 – AI adoption and concerns
25:51 – AI forecast in the coming year
Resources
Find “In The Vault” on a16z.com: https://a16z.com/podcasts/in-the-vault/
Find more about a16z’s vision for the future of fintech: https://a16z.com/banking-on-ai/
Find Marc Andrusko on X: https://x.com/mandrusko1
Find Greg Ulrich on LinkedIn: https://www.linkedin.com/in/gregu/
Find a16z on X: https://x.com/a16z
Find a16z on LinkedIn: https://www.linkedin.com/company/a16z
