Spargle meets Longhow Lam
Data Scientist at FedEx
Book: ‘The Elements of Statistical Learning’, by Jerome H. Friedman, Robert Tibshirani, and Trevor Hastie
Documentary: Star Talk with Neil Degrasse Tyson
“What I find fascinating about data is its universal nature. Comparable to mathematics, once you translate a real-world problem into variables and equations, it doesn’t matter what it’s about: you can still work with it.”
What is your role, and have you done this before?
I’m a Data Scientist and have been working as a freelancer for the past eight years. Currently, I’m working on a project at FedEx, where I focus on analyzing data to help solve business problems. As a freelancer, I typically work on short-term contracts, which has allowed me to gain experience across many industries. Before FedEx, I worked at Bidfood, and prior to that, at several other companies in different sectors (Adidas, Heineken and ING for example). My core responsibilities involve using machine learning, predictive modeling, and AI to understand and predict customer behavior. In short, I translate business problems into data science solutions. If a business wants to know how they can improve their sales, I can look at the data and see where the pain point is in order to find a solution.
What sparks your interest in working with data?
What I find fascinating about data is its universal nature. I’ve worked in shipping, banking, retail, and fashion. Whilst the industries differ, the way we model problems with data often looks similar. Comparable to mathematics, once you translate a real-world problem into variables and equations, it doesn’t matter if it’s about customers, products, or shipments: you can still work with the equations. That abstraction is what really thrills me. I also like uncovering hidden patterns or unexpected relationships in data that others may not have noticed. It feels a bit like detective work, which keeps it exciting.
How do you determine what to look for in the data?
It starts with a business problem, a pain point, this can be a drop in sales or a high customer churn.The goal is to understand why the problem is happening by collecting and analyzing relevant data. From there, I might use visualizations, algorithms, or statistical techniques to uncover insights. There are many tools in the data science toolbox, but which ones I use depends on the question at hand. Sometimes, I also work on hobby projects with publicly available datasets, which is a good way to experiment and learn new techniques. But in a professional setting, the end goal is always to solve a real business issue, which helps driving the business forward.
“One of the biggest challenges is simply having access to the right data. A company may have a problem they want to solve, but if the necessary data doesn’t exist (or isn’t stored properly) it becomes difficult.”
What are some challenges you face when solving these problems?
One of the biggest challenges is simply having access to the right data. A company may have a problem they want to solve, but if the necessary data doesn’t exist (or isn’t stored properly) it becomes difficult. Sometimes the data is scattered across systems, poorly documented, or has quality issues like missing values. I’ve even had situations where data was emailed to me manually, which is far from ideal. There’s also another big challenge: “unclear goals”. Some companies know they want to use AI or data science but haven’t clearly defined what problem they want to solve. As a data scientist, I can help guide that conversation, but I’m not a mind reader. I still need a clear business objective to get started and make an effective analysis.
How do you manage or overcome those challenges?
Sometimes you can’t fully solve these issues, but you learn to work around them. If the data is incomplete or of low quality, you look for alternative analyses or adjust your approach. For example, if I can’t access data on customers’ age, maybe I can use behavioral data as a proxy, or skip that part of the analysis entirely. Gathering new data takes time and resources. You often have to be pragmatic. On the modeling side, it’s also worth noting that predictive models can still work reasonably well even with imperfect data. So it’s about being flexible and realistic with what’s possible in each project.
Can behavior analytics help estimate missing data, like age?
Yes, sometimes we can make educated guesses based on behavior. For instance, when I worked at WE Fashion, we used purchase history to conclude age groups: people buying children’s clothes were likely younger parents. That’s a simple example, but in general, behavioral data can sometimes help us approximate missing information. Of course, these are still assumptions, and they won’t be perfect. It’s important to treat these estimates with caution, especially if they’re used in decisions. But when exact data isn’t available, these approximations can still be quite useful.
“Over the years, the field evolved, and I evolved with it. While the terminology has changed, the core of what I do has stayed the same for over 25 years.”
Did you always know you wanted to work in this field?
In a way, yes. In high school, I already knew I wanted to study mathematics, thanks to a great math teacher who encouraged and inspired me. When I started my university studies, the term “data science” didn’t exist yet. But during my final university years, I gravitated toward statistics and data analysis, which led to my first job as an applied statistician. That was back in 1997, when it was called data mining or applied statistics. Over the years, the field evolved, and I evolved with it. So while the terminology has changed, the core of what I do – turning data into insights – has stayed the same for over 25 years.
What trends do you foresee in the coming year?
AI and generative AI are still major trends, and companies are experimenting more and more with them. However, just like with earlier data science projects, many of these AI initiatives will struggle to produce results. One reason is that the data infrastructure often isn’t in place, so the data may be scattered or poorly maintained. I think we’ll see more companies realize this and invest in getting their data organized first. There’s also a distinction between personal productivity tools, like ChatGPT or Copilot, and large-scale business applications. The former are being widely adopted, but company-wide use cases remain difficult to implement effectively. So in 2025, I expect more focus on the data foundations, rather than just following the AI hype.
Who inspires you, and why?
A long time ago, my high school math teacher was a big inspiration. He made math exciting and gave me extra challenges that sparked my interest in the field. Currently, I draw inspiration from science communicators like Neil deGrasse Tyson and the team behind the Cool Worlds YouTube channel. They’re able to explain complex astrophysics concepts in a very clear and engaging way. I try to draw inspiration from that skill, and try to apply the same clarity when I explain technical topics to non-experts. It reminds me that how you communicate an idea can be just as important as the idea itself.
“The first question I ask is: “Let’s suppose I have this machine learning model for you right now. Now that it’s on your desk, what are you going to do with it?”
Do you find a lot of difficulty translating the data language to a company? I can imagine there could be a language barrier ?
That can also be the case with data science and depends a little bit on which company you go, but sometimes the use case is quite easy to understand: ‘We have customers that stop buying our products or services, i.e. customer churn. Can we predict which customers are likely to churn?’ Then I can look at data, of course. And then sometimes you also find that they haven’t really thought about that. And that’s always the first question I ask. Okay, let’s suppose I have this machine learning model for you right now. It is on your desk, what are you going to do with it? Can you test the model with a campaign?
What would that machine learning model look like, for example? What kind of value does a model like that add to a company?
A machine learning model (for example in a customer churn context) is in essence just a mathematical formula that calculates the probability that a customer will churn. As a marketing department knowing this probability for each of your customers can be valuable, you may start campaigns to target only specific groups of likely churners and prevent them from churning.
What is your favourite documentary podcast and book?
Star Talk with Neil Degrasse Tyson is one of my favourites. It’s a YouTube channel on astronomy. For books I can recommend The Elements of Statistical Learning. That book is already 25 years old, and it gives a nice overview of the different techniques in machine learning, which are still used today. That book was written 25 years ago. I read it when it was released then. And even now I still sometimes read it when I have nothing to do. I just browse through this book. It’s like my Bible.
What’s your life motto or your favourite quote?
My favourite quote is from Johan Cruyff, the famous Dutch soccer player. He says: ‘soccer is simple, but the most difficult thing to do is playing simple soccer’, which translates a little bit to my motto in data science. I like to keep things as simple as possible. So there was a period when people, for a simple problem, would throw in the most advanced and complex machine learning models. And that’s often an overkill. So then take one step back. How can you create a simple model that still describes what you’re trying to predict or still works? Sometimes that’s not even that simple. So then I have to think back on this quote. Some things might be simple, but it’s very difficult to do simple things.
