Flowcast’s Co-founder and CTO, Winnie Cheng was recently interviewed by Sandhya Krishanmurthy, COO of the FinTech School, for the latter’s FinTech Core 101 course. The FinTech School is dedicated to providing practical Fintech training developed by practicing entrepreneurs. It was founded in Mar 2016 by FinTech entrepreneurs and financial services professionals who have decades of financial service industry experience and have founded successful Fintech startups. FinTech Core 101 is an introductory course that provides the students a good overview of the FinTech industry sectors and future trends. Dr. Cheng’s interview was intended to give students a flavor of Artificial Intelligence and Machine Learning’s impact on the FinTech industry.
Sandhya from the FinTech School (Sandhya): It gives me great pleasure to introduce Winnie Cheng, a Fintech Entrepreneur and Data Scientist extraordinaire. She has worked for pure technology giants like IBM and has contributed to their Watson platform. She has also worked with Financial Services giants like JP Morgan and American Express. She holds a PhD in Computer Science and Artificial Intelligence from MIT and a MS in Computer Science from Stanford University. She is currently the Co-founder, CTO and Chief Data Scientist at Flowcast, a FinTech startup.
Winnie, it’s pleasure to be talking to you; thank you so much for agreeing to this interview.
Winnie from Flowcast (Winnie): It’s a pleasure to be here.
Sandhya: Can you please share more about your educational and professional background? And we’d love to hear what brought you to Data Sciences.
Winnie: I’m currently the co-founder & CTO of a data science company called Flowcast. My journey to data science is a bit unusual. I earned my MS in Electrical Engineering from Stanford and PhD in Computer Science from MIT. At Stanford, my main focus was a little different. I was interested in making software run really fast on hardware, whether it was with FPGAs or GPUs, so that we could crunch a lot of data very, very quickly using parallel architectures.
It wasn’t until I arrived at MIT that I discovered machine learning and AI. As a grad student in MIT’s Computer Science and AI Lab, I got to see a lot of cool projects across many sub-disciplines within computer science and met groups of very talented researchers. I learned about NLP, machine learning and control systems. A couple of us teamed up to build an underwater autonomous vehicle to compete in an international competition in San Diego. My hands-on experience came through many sleepless nights writing and debugging code poolside. It was a blast. When your robotic submarine starts swimming in a chaotic fashion, you learn very quickly that those math equations are for more than fancy research papers, and flipping a sign on a variable can have disastrous consequences.
While at MIT, I also learned a lot about finance, and in particular, quantitative strategies for trading and valuing portfolios. We had amazing professors like Prof Andrew Lo from the Sloan school that made the topics really interesting. When it comes to AI / ML, financial markets is a place with a lot of data, and the holy grail is being able to make sense of these data to discover insights that affect our capital market and economy.
After my PhD, I worked at, in my opinion, really cool companies–research labs and big banks. I got a sense of ‘fintech’ before the term existed. At IBM Research, I designed machine learning algorithms for the commercialization of IBM Watson for various business use cases. I still remember all of us gathering in the cafeteria in Yorktown, NY to watch Watson beat a human on Jeopardy. It felt a little strange cheering for a machine! Then, I joined American Express and later JP Morgan, at an interesting time, when they are trying to make sense of big data and machine learning. At American Express, we tried out some new approaches to building credit scoring models using massive amount of data. At JP Morgan, I lead innovation projects where my day-to-day was to learn about different functions within the investment bank, and explore and demonstrate how AI / ML can help our clients.
Finance is really fascinating to me. It impacts our society in so many different ways. Efficient use and allocation of capital is needed for a healthy economy. If you are a small business with a great idea, to make it a reality, you need money to fund your product development and costs to manufacture and ship your products to your customers. This is where Flowcast comes in. We use data science to come up with novel ways to assess risk in supply chain and trade financing.
Sandhya: Thanks for that background, Winnie. It’s fascinating! I’d love to talk more about Flowcast, but I’m still not able to shake the image of you making a robotic submarine run by San Diego poolside (laughs). As a STEM enthusiast, I have been in awe of IBM Watson’s capabilities. And I feel it’s an honor to be talking to someone who has contributed to its capabilities.
Now, let’s come back to Flowcast. Can you share more information and shed more light on how Flowcast came about?
Winnie: Flowcast came about with my friend and co-founder Ken So. We met back when I was at MIT and he was doing his MBA at Berkeley. We ended up doing a cross-school project together around how to launch a company–the thought process behind taking an idea from concept to market. Now with Flowcast, we are doing it for real. Both Ken and I have worked for many years in financial services. Ken comes from an investment banking and corporate M&A background. While he was acquiring companies, he saw how most companies still don’t have a very good understanding of their cash flow, and an even more limited way to anticipate their cash flow needs. So, this leads to poor working capital management and costly trade financing to bridge cash flow gaps. Flowcast is bringing data science to the rescue to service both the companies who want financing and the banks offering the financing. For these companies, we help them analyze their data to anticipate cash flow surpluses and deficits by understanding their clients’ payment behavior and their relationship with their vendors. For banks and funders, we help them understand the risks for underwriting trade financing using machine learning techniques.
Sandhya: So, while on one hand, your clients are Financial Service firms like Banks or Funders, on the other hand, your clients are small to mid size businesses and could be associated with any industry, right?
Winnie: Yes, that’s correct.
Sandhya: Thanks for that context, Winnie. Now let me shift gears and talk about something that is core to your value proposition. Flowcast’s value hinges on Data Science as well as Machine Learning. In that context, ‘Data Science’, ‘Machine Learning’ and ‘Artificial Intelligence’ are three different, but connect concepts. Can you please explain the difference between these three?
Winnie: That’s a great question. I see these terms used almost interchangeably these days, but there are actually subtle differences. Data Science is best understood if you think of the two words, ‘data’ and ‘science’, separately. It’s applying the scientific process on data. In science, whether it be biology, chemistry, or physics, the scientific process teaches us to start by asking a question, then forming the hypothesis, and executing an experiment to validate the hypothesis. In Data Science, we are doing the same thing but using data to answer the question and validate our hypothesis. For a data scientist to be successful at carrying out this process, he/she needs to have a bunch of tools in the tool box, and that’s where business domain knowledge, statistics, machine learning, programming and visualization skills come in. Data science is really an interdisciplinary practice that encompasses machine learning.
Machine Learning is a subset of AI. AI refers to any computer programs that appear to do something smart. It’s possible for a program to do something smart without using data and without being adaptive to changing conditions. AI programs can be as simple as a set of rules that are enforced or automated by a computer. Machine Learning enters the picture to make a really smart AI program. Machine Learning is a set of techniques that teaches the computer to learn from data. It may be classification algorithms that learn the different patterns between email that are spam vs those that are not. Or it can be used to identify unusual patterns or anomalies from all the examples it has seen, for example, in the case of fraud detection. When you train a machine to learn, it becomes a much smarter AI program.
One can think of IBM Watson playing jeopardy. AI includes Watson’s ability to listen to the question, figure out what is being asked as well as coming up with the answer. It includes all the sensors and engineering that go into making a complete system. Data science helps Watson to answer, for example, “how do we train Watson to figure out the book that a given movie is based on?” The data science approach then looks at what data sources make sense and what machine learning algorithm can best extract the relevant info from these data sources.
Sandhya: That example of IBM Watson was helpful in bringing the concepts together. Now, let’s back to Financial Services and FinTech. You have been involved with FinTech long before the word ‘FinTech’ was coined. Given that perspective, how do you think Financial Services is using Machine Learning and Artificial Intelligence now-a-days? And how do you think Machine Learning and Artificial Intelligence will be used in FinTech in the near future – say 2 years or 5 years from now?
Winnie: The past 5 years has been about Big Data. Financial services invested heavily in Big Data infrastructure. This is a great thing–it means that a lot of the data are being stored and maintained, and ready to be put to good use by data scientists. In the next 2 years, I expect that predictive models will be as common as databases in developing fintech applications. We will see more user interfaces that can sort of ‘read your mind’ and suggest or highlight actions that are relevant to you. The entire banking user experience will be transformed, with design thinking and data analytics being a core part of designing any business process. The way we apply for mortgages, for business loans will be much less painful than it is now.
Also, we have seen an explosion in data aggregators and APIs. There is a lot more sharing of data between financial institutions, and I expect this trend to continue.
While the exact roles of financial institutions are changing with these technological advances, one thing that won’t change, as the well-known fintech VC, Pascal Bouvier, puts it, it is that, “a bank is in a TRUST business.” In the next 5 years, I see that banks will move to use this data to strengthen trust with their clients. With AI and ML, banks have the capability to help businesses decide who to trust, who to lend to and from whom to borrow. We will move towards the paradigm where it is “In Data We Trust”, and financial institutions will play a crucial role.
Sandhya: I love the phrase “In Data we Trust” (laughs). I too read that article from Pascal where he challenges Banks to re-examine what business they are in and he builds a case systematically to convince Banks that their current opportunity lies in providing ‘Trust’ services around their customer’s data. Thanks for bringing that up. We will provide our students a link to that article which was published on LinkedIn Pulse and Medium; they can read if they are interested.
Coming back to students, we have students from various backgrounds taking this course. Some of them have Technology background, some of them have Financial Services background and some come from a different industry altogether. And they are taking this course because they are interested in FinTech. I can say that most do not have PhD in Computer Sciences like you do. So given this info, if they want to get theoretical knowledge or hands-on experience on Machine Learning, what is your advice to them?
Winnie: Ultimately, data science is about problem solving. I think a good way to get hands-on experience is to expose yourself to many real-world use cases. For each one, think about what are the problems you are really trying to solve. Without looking at the data or thinking about specific techniques, ask yourself, how you would intuitively solve this problem. Then, decide whether machine learning is really a good solution to this problem or if the same problem can be solved by simple design or process change. If machine learning is indeed needed, look at the data and debate about the pros and cons of different techniques on the given problem. You can learn a lot in this evaluation process.
For someone who wants to get into AI / ML, there are ways to be part of the innovation journey without coding. You can learn about the capabilities and limitations of AI / ML, and think about processes and experiences around you, how they might change if you are armed with this very powerful technology and be the evangelist for innovation within organizations. What are new questions you can now answer? From a technology perspective, there are many startups, Flowcast being one of them, that are doing cutting-edge work that you can partner with without re-inventing the wheel.
Sandhya: I really appreciate your bringing the focus back to the problem and problem solving. Sometimes we forget, especially in the face of shiny new technology, that our investment in cutting technology – whether in our effort or funds – is so much more impactful and we get ROI (return on investment) when diverted to solving real world problems. That way, I admire what you are doing at Flowcast; you are applying Machine Learning towards solving a dire need in the market place today. Because I know that small businesses in the US, and I would say across the world, struggle with cash flow problems. And planning and solving for this cash flow problem many times decides between survival and closing shop for a business. We will provide our students a link to Flowcast’s website so that they can check out for themselves the wonderful work you are doing.
Once again, thanks for agreeing to this interview. It’s been an absolute pleasure!
Winnie: Yeah, thank you very much for this opportunity.
Recorded on July 27, 2016