In this episode of Sloanies Talking with Sloanies, host Christopher Reichert, MOT ’04, interviews Yurui (Rui) Tong, MBAn ’21, a graduate of the MIT Sloan Master of Business Analytics program. Tong reflects on her educational and professional journey through a personal lens of self-discovery and curiosity. Though passionate about logic and reasoning, she shares her early struggles with math, eventually realizing that her strengths lie in storytelling and narrative-based problem-solving. This realization shaped her approach to data science and analytics, emphasizing clarity and meaning in complex concepts. At MIT, a key lesson came from Professor Dimitris Bertsimas, SM ’87, PhD ’88, who advised her to confront difficult problems head-on—a mindset that continues to guide her career. Tong’s path includes diverse roles, from WHO to Deloitte to Remitly, where she now serves as a senior product manager working on fraud detection powered by AI.
Tong explains that while AI is not new to data science professionals, its current wave of popularity stems from its broadened interaction with the public. At Remitly, she sees AI as a layered system—data, models, and decision-making—where the quality of data is as crucial as the model itself. She highlights the challenge of aligning business goals with probabilistic AI outputs, particularly in fraud detection, where trade-offs between customer experience and risk management are constant. Her podcast, Floating Questions, reflects her wide-ranging intellectual curiosity, covering topics from energy policy to personal identity. Tong’s time at Sloan fostered a deep appreciation for principled innovation and collaborative learning—values she continues to carry forward.
Sloanies Talking with Sloanies is a conversational podcast with alumni and faculty about the MIT Sloan experience and how it influences what they’re doing today. Subscribe and listen on Apple Podcasts, Spotify, or wherever you get your podcasts.
Episode Transcript
Rui Tong: One thing that I still remember was I was arguing with Professor Bertsimas about my grade. He said, the only thing that I want to make sure that you take away is that don't run away from the hard problems. You go straight into it and try to solve it. And I think it's that spirit of don't be afraid of hard problems that helps me after graduating Sloan.
Christopher Reichert: Welcome to Sloanies talking with Sloanies. A candid conversation with alumni and faculty about the MIT Sloan experience and how it influences what they're doing today. So what does it mean to be a Sloan? Over the course of this podcast, you'll hear from guests who are making a difference in their community, including our own very important one here at Sloan.
Christopher Reichert: Hi, I'm your host, Christopher Reichert, and welcome to Sloanies talking with Sloanies. My guest today is Yurui Tong, she goes by Rui, a 2021 Master of Business Analytics program graduate from MIT Sloan, which is an intense data science program where she was elected class president and served in the MIT Sloan Senate. So welcome,Rui.
Rui Tong: Thank you, Christopher. Glad to be here.
Christopher Reichert: Thanks for joining us. Let me give our listeners some quick background on you.Rui has always been interested in quantitative subjects and pursuits, statistics, and economics at the University of Illinois. She also has a varied background at the World Health Organization (WHO) in Geneva at Deloitte, as a researcher at the National Center for Supercomputing Applications. And while at Sloan, interned at Cloud Zero, a cloud cost optimization platform. And following Sloan, she's been with Remitly, an international payments company, first as a data scientist with increasing responsibilities, and now as a senior product manager, focusing on fraud using AI technology.
So more on that soon,Rui is also the host of Floating Questions, a podcast where curiosity guides the conversation and stories unfold. I guess you've always been quantitative by that and numerate, I guess, so maybe it came…
Rui Tong: Actually not, like I enjoy math as a subject, but I actually have a very complex journey with the whole STEM as a category. When I was a kid, sometimes I scored super high on math test, sometimes I scored really badly. And I couldn’t understand why.
I've been reflecting on what is my ability to score high on mathematic subjects fluctuating so much growing up. And why am I like in MIT, in one of those very quantitative subjects? And I just couldn't piece this together, like maybe MIT didn't do a good job selecting the candidate. As I was like thinking more and more and reflecting more and more, I realized that actually fundamentally I love reasoning. I love logic. But math expressed logic and reasoning away that it's not intuitive to me, meaning the symbol itself, it's not a language that I do the best. My reasoning and logic process is often expressing story. So, if I can somehow piece together some type of story in my head, that logic flow just like goes really smoothly and I can really quickly solve problem. But if it's in a very rigid form of this is exactly how we're going to express it, which comes with a precision for sure, I often have a lot of trouble with, but only when I penetrate the symbols of that mathematical things, then I was like, oh, I get it. Which is kind of interesting.
Christopher Reichert: Interesting. It's an interesting insight. And when did you figure out that insight on how your brain works and where you find friction and where you find flow?
Rui Tong: Last year, I was just like, I was thinking of a conversation with my mom and reflecting on growing up, like, what kind of stuff that I had trouble with and why am I at MIT when like somehow it feels like I'm dumb at math, you know, these two things don't gel together. And then I realized that, you know what, actually I'm much better at telling stories. And like, for example, at work, there's a lot of very complex stuff. But if you really just form a story, it's actually really simple one. That's when I think see things very clearly that others might find a little bit more difficult to like put together the big picture.
But if you're asking me to really dive down in a specific way of like expressing that problem, my brain is just not enjoying it, it's just like pay a lot of attention to what I really enjoy and what I'm good at versus the places where I glitch and try to reconcile. Wait a minute. I think that I'm good at this, but why I'm not good at this way doing this thing in this specific way, then you sort of like starting to reconcile.
Christopher Reichert: What was your undergraduate subject?
Rui Tong: I majored in three things, statistics, economics and accounting, actually starting accounting/finance like business school because I wanted to learn how business works and accounting is supposed to be the language of the business, right? But then as I was like taking classes outside of the traditional track, like more probability theory and things like that, I was like, oh wow, I really like the philosophical aspect of it. So let me just pursue that as well. In economics, it's just something that I fell into.
I fell in love with economics with just one line. I still remember, I think this is in the “Textbook 101”. Economics is about allocation of scarce resources because at any given point of time you always have the problem of scarcity of resources, and it's all about how you allocate it. And I was like, oh, I love this. It’s beautiful as a problem statement.
Christopher Reichert: After your undergraduate, did you go to the World Health Organization? Was that during your undergraduate time?
Rui Tong: During the undergrad time. I managed to find a way again. I was trying to navigate some of the immigration issue because when you were on student visa, if you graduate from the school, you have to almost start working right away. You don't have much grace period, but I find a way to work with the system in a way that's just like, I find an internship at WHO because I was thinking, oh, maybe it will be great if I can feel that I'm building something that will be better for everyone in this world. And plus, that's a totally different career track, the international nonprofit track. But then when I got into it, I got a little bit disillusioned. So, I just ended up concluding that I like the capital markets a little bit more because you will get the feedback a lot quicker that way. And you are going to have to correct your course if you were not doing something right.
Christopher Reichert: And so then after the World Health Organization, you went to Columbia Business School as a research assistant and then Deloitte as an analytics consultant. And then I assume around that time you were thinking Sloan or maybe you applied to Sloan during that time and then worked at Cloud Zero as an intern while at Sloan. I was at maybe between your first and second years. It is a two-year program?
Rui Tong:No, it's actually one year, super intense one year program. It is part of the program that you are going to have to do an internship during the summer. So, the Columbia research internship helps me understand whether I want to go into PhD or not. I don't think I'm a good candidate for it because I think in order to do a PhD, you need two out of three things: tenacity, passion for that specific project, and intellectual power.
If you are really good at two of the three, you probably could make it through. So, I think my way of analyzing and reasoning things through are not that suited for potentially a PhD program.
Christopher Reichert: I mean, I think it's a good insight rather than spending five years and then dropping it, right? And finding yourself now really frustrated.
A question about Remitly and the, and I was wondering how it might relate to your experience of the World Health Organization, large, slow moving, you know, great mission, but perhaps, you know, even to accomplish part of that mission is ponderous and whatnot. How do you contrast that if you do with, with Remitly, which I assume is a much smaller organization and probably gives you some of that nimbleness that you, that was probably missing at World Health Organization?
Rui Tong: This is a great question. Actually, you touched upon something that I took a lot of time to reflect when I was searching for a job, when I was at MIT. Job searching is definitely challenging at that point in time. I've tried quite a few different positions at WHO, Deloitte Consulting, and then as a researcher at Columbia Business School, I just realized there are a couple of things that I would like to have in the next job.
The first one is that the company would actually have a mission that I can get behind because I am motivated by the purpose of a mission, more than anything else. That's something that I figured out about myself throughout years of trying different things. The second thing that I was looking for was that a very dynamic team, because I love working in a fast-paced environment, maybe a little bit chaotic, but that's where also energy is. The last part is about building at the intersection of business strategy and also data and analytics in AI, because that's inherently the space that I'm really interested in. So, that's the three things I'm looking for in the next job. I end up actually writing an email to my network to say, hey, if you know any job that meet these criteria, please let me know.
Remitly fit the bill. It's a great group of people. I love its mission, and in general, we serve global immigrants, and that's something that I can totally relate to. It's a lot of immigrants coming to the United States were migrating to other countries, doing really difficult jobs to earn that money and send home to support their family back home, maybe their parents are sick, or maybe their children are still in school. That's why they use Remitly. So, serving this type of customer, and also on top of like it's in the fraud detection domain where the company and a customer interest are truly aligned. That makes me feel like, okay, I really can get behind this problem statement, and I want to help solve the problem in the space really well.
Christopher Reichert: You know, everyone's interested, scared, excited about AI, and it seems to have hit the zeitgeist really quickly, and I think I was probably about almost three years ago, less than three years ago, where all of a sudden Open AI was just on everyone's tongue, and there was this notion of ChatGPT, and then all of the cascading opportunities, the people who were positive for it were talking about the opportunities, and all the people who were scared about their job security and job experience and job skills are up in arms about it. You're incorporating AI for fraud detection, and so what does AI mean to you?
So as you came out of MIT with this degree, and presumably, you know, more about the underpinnings of this, than most people, say, how are you approaching what AI means to a business, to your job, and to maybe your career, for that matter?
Rui Tong: First of all, and if you talk to a lot of professionals that have been analytics for decades now, they wouldn't say that AI is something new, right?
Christopher Reichert: Of course, yeah.
Rui Tong: You know, people have built a lot of different types of algorithm in the past, even the simplest regression model that often are used in academics to understand, for example, the statistical significance of certain things in building certain predictive models or causal studies. Technically, those are already models, and you can call that within the realm of more classic and traditional simple form of AI.
So, I think for a lot of professionals in this space, AI has been here for a long time. It just reached another new sort of like interaction mode with the mass, and also the way that the models are built. But a lot of challenges have been there since the classical model errors, starting from the regression model that's been used in, let's say, the finance world, and loan prediction, right? Like predicting somebody's risks of defaulting or predicting whether a machine is going to break down. So, all of those use cases have been there. It's just now being also get carried into this hype of AI.
So, one, a lot of companies, I think more or less, have been using AI to some extent. Two, I think AI is a very broad term. The way that I think about AI is as long as an algorithm that you have to tell the algorithm what you want it to do. So, this algorithm is termed as supervised, or unsupervised machine learning models, or deep learning models with tons of tons of like neural layers, or it is an optimization algorithm to say, okay, we need to optimize for, let's say, the quantity x, y, z with the constraint of A, B, C. To me, that's also AI, right? Anything that has some element of like complexity in the algorithm to achieve certain business outcome or customer goal, that is AI to me.
Christopher Reichert: I mean, the way I've kind of observed it is the volume of historical data input into this, Witches Brew, which we call an LLM, right, is backward looking. In the sense that you throw in all of Shakespeare to gain, for example, all of the mathematical theory to date, all of maybe language rules, and a variety of languages that are out there, the rules for grammar, and the rules for conjugation, and then you throw in the accumulated knowledge of, I guess, that's called the last 25 years of, whatever has been crawled on the internet, right? Okay, so there's this large body of data, and there's that tension between believing whether or not you believe what's been output by these AI models as accurate, and then realizing that maybe it's doing a better job of predicting than humans can, and so does that, where do humans fit in? Does that make sense?
Riu Tong: I guess I just want to quickly piggy back on the common you say, like, there is the element of trying to discern in whether the model is actually accurate or not, that's definitely a huge component of it, but I think in the context of leveraging model as a product, in a lot of other use cases other than just, you know, a chat interface, the critical piece of that application is about taking the model output and decide what to do with it. Accuracy is one thing, but also ultimately, a lot of times those treat like classical LLM models, they don't produce what to do. You still have to make a decision for what to do. This is where what humans potentially, well, at least I think it will take a lot of long time for model to really catch up all the ambiguity and context that it takes for a decision.
I think going back to my previous point, it is an algorithm that we tell it what to do. If we don't know what we want it to do, you wouldn't really have a good model as a product. So, the human layer, I think, at least at the output layer, I think a key role that we're playing is to figure out how to efficiently leverage the model output to maximize certain business objectives. And that involves a lot of complexity and figuring out how to really make that a scalable decisioning process.
Christopher Reichert: Tell me about a day in your job.
Rui Tong: Maybe let me start with, I think we have to think about AI product in three layers and then in an ecosystem supply chain fashion. And each layer is built on the top of the previous one. The first layer is raw data, all the numbers that you collect in your system, whether it's available publicly online or it's within your internal company system. The second layer is the model. The model sits on top of the data, doing all the crunching and then processing and then trying to produce certain output that potentially you can leverage for business use case. So, the third layer is the decision-making part. I think a lot of times people just say, oh, let's just build a model and then, you know, fraud detection. Oh yeah, let's just, you know, produce the probability of, let's say, whether a transaction could end up being a fraudulent one or not. But it's not so simple. Because each layer actually requires certain product governance in order for you to achieve the final layer of like, are you actually making a good, some business decision, right?
So, I think more and more professionals in the industry start to really pound on the importance of treating data as a product in itself. But this is a very tricky problem because, well, first of all, it's tricky in terms of technical challenges because, you know, there could be changes in the data. There could be online offline data drift, data reliability. It's a problem. But I think fundamentally what makes it tricky is it because it's a people and incentive problem. It's a classic tragedy of the commons where everyone would benefit from high quality well-documented data. But no one individual or team is naturally incentivized to ensure that happens.
Christopher Reichert: So how do you, how do you maintain quality control then?
Rui Tong: Right. So, this is actually a huge industry problem. I don't think companies have really nailed it and have actually solved the problem yet. But more and more people are realizing, actually, without treating data, raw data as a product in itself, in this new AI age, it will cause a lot of havoc downstream. And your model could break down and impact your users.
Christopher Reichert: So for Remitly, and I guess maybe we can say this is a broad question, but say, speaking for Remitly, there's a service, sending money from point A to point B.
There's a process for doing that efficiently, which is, I'm sure there's all sorts of regulatory rules and hurdles that you go through to communicate with banks in Rwanda or Nigeria or India or wherever. Then there's the accuracy, the speed that it gets done, the accuracy that it gets done. And that seems to rely on what you, what I heard you say, maybe I heard incorrectly, is kind of like the crown jewels of the business are the quality of the code that you are relying on to make sure that you make the right call on this is fraud or this is not fraud.
Rui Tong: You're right. So, remittance industry is very competitive, right? And in order to get that competitive advantage, nailing down fraud detection is critical, because that's how you really lower the cost. And Remitly is the one pay for the loss in a lot of cases. Now the tricky part is that all of these risk vectors have some similarities between like shared across, but also they're technically distinct as well. And how you detect them on what type of level requires a little bit of customization or adaptation. Or you can say that, you know what, I'm going to build a very generic one that will just capture them all, but then are you going to lose some type of model prediction accuracy because now you're trying to build a little bit more generic solution. Again, this is all about framing how your business is expected to evolve over time and then you kind of have to adapt at your strategy alongside it.
Christopher Reichert: And there's trade-offs in speed and accuracy, right? And that's a trade-off from a financial perspective of letting fraud get past and cost you or preventing it and slowing down the customer experience and going over time.
Rui Tong: Right. And that's the part that you just mentioned, then it comes the third layer, which is a decision layer, right? Only if the model is perfect, it just tells you that, oh, this transaction is definitely fraudulent. So, the probability will just be one and then that transaction is definitely from a good customer. So, the probability will be zero. In that perfect, idealistic world, everybody is happy, but the truth is the model will never be perfect because the world is deeply imperfect, honestly. So, then how do you think in a very probabilistic way that make very educated systematic trade-off between a customer friction and fraud loss become critical as well? Because that will definitely impact both through a top line bottom line. Most people think of risk as this back-end system that's mostly impacting the bottom line, your financial loss. But if you just raise it one step further, you will immediately say that it will impact the top line, right? Because the more control you put into place, the more customers you're going to slow down, the more likely that you're going to switch.
Christopher Reichert:So let me change subjects here. I want to talk about Floating Questions, your podcast. You've had interesting topics ranging from hiking, creative writing, and navigating data science and major league baseball and climate policy. That was one podcast, by the way, folks. Exploration of ethics, innovation, and dualities that we all navigate. Part technical, part philosophical, and holy human with Brian Hsiu. And I guess the theme I'll detect is the intersection of data science, social enterprise, personal history. How do you any choose your topics and find your guests?
Riu Tong: In general, I'm curious about a wide range of subjects. To be honest, when I started this podcast, what I have in mind is that I'm going to eventually talk to a wide range of guests from all walks of life. Because I'm inherently interested in so many different subjects. And I would like to peek into those domain a little bit, just via the guest. And that's why you see that even within the same episode, we would go from solar panel energy to defense technology, to their personal story, and also childhood growing up experience, and how that really eventually shaped who they are and what they do.
Christopher Reichert: So, let's talk about your Sloan experience. How did you choose the master in business analytics at Sloan and Sloan in particular?
Rui Tong: Because I did an internship at Columbia Business School, working for a finance research finance professor and then doing some modeling. And then I realized, you know what, I'm not caught out to do the PhD and I don't really want to do it. However, I do love learning. So, I know at some point I'm going to go back to school, but I want to learn at the intersection of data, AI, and business. For me, because my interest is also data. So, I really want to go back to a program where that is present. And then I start researching what kind of programs actually offered this type of the intersection of business and analytics. MIT naturally came up. I chose MIT because when I was visiting to just get a taste of what's the vibe of the program, the people, I've come across a lot of super friendly and like collaborative folks at MIT. And that really helped me to imagine, okay, if I joined this program, this is how my life would be, this is a type of person I'm going to interact with. Am I going to enjoy it or not?
And I actually have to say the director who started the program, Professor Bertsimas, he approaches things in a very principled way. You can see his description and his thought process in, you know, MIT affiliate websites or his personal ones. And I was really impressed by that.
MIT aims to cultivate people who wants to do things in a principled way with integrity and infuse with the innovation. I think that's what resonate with me.
Christopher Reichert: Let's say if you think back on your time at Sloan, is there anything that you return to that changed you or as a framework that resonated with you.
Rui Tong: I can talk about the framework or the specific incident, but I just first want to say that I really love all the people that I came across, especially in my cohort. I think there are super smart humble and kind. And that's what I think what will like come with me for years.
The one thing that I still remember was, I was actually arguing with Professor Bertsimas about my grade. He said, you know what, in a few years you wouldn't even remember what exactly this course was about. But the only thing I want to make sure that you take away with you is that don't run away from the hard problem. You go straight into it and try to solve it. That's the type of principle that I want to make sure that I'm still in all of you. And I think it's that spirit of don't be afraid of hard problem that helps me after graduating Sloan as well.
Christopher Reichert: And if you had to go back to Sloan, does anything you think back and I wish I had more time to do?
Rui Tong: Oh, yeah, 100%. I mean, that program was one year and it was super intense. What I wished I could take is just like everything.
Christopher Reichert: I want to thank Rui Tong for joining us on this episode of Sloanies Talking with Sloanies. If you want to follow Rui's work, search for Floating Questions wherever you get your podcasts. And thanks for joining us today.
Rui Tong: Thank you, Christopher. This was a wonderful chat. Happy to share more. And you know, if any people have questions or just want to discuss a little more in the challenges they encounter in building model as a product or just in general AI as a product, I will love to chat with them.
Christopher Reichert: Sloanies Talking with Sloanies is produced by the Office of External Relations at MIT Sloan School of Management. You can subscribe to this podcast by visiting our website, mitsloan.mit.edu/alumni, or wherever you find your favorite podcasts. Support for this podcast comes in part from the Sloan Annual Fund, which provides essential flexible funding to ensure that our community can pursue excellence. Make your gift today by visiting giving.mit.edu/sloan.
