Some of the most useful lessons in data science have little to do with the model. For Catalina Ramos (’23SPS), they came from a Capstone project in Columbia’s M.S. in Applied Analytics program, where her team worked with one of the largest commercial banks in the U.S. to predict employee attrition. The data was limited, the client contacts weren’t technical and the goal was a recommendation the bank could act on, not just a correct answer.
Ramos, an economist who spent several years in investment banking before moving into data science, is now head of AI and data science at Blooms, where she builds production machine learning systems, including credit risk models and multi-agent pipelines. In a new interview with SPS, she shares how the Capstone gave her a repeatable framework for every project since, what working within constraints taught her about creative feature engineering and why communication is the skill that turns a technical solution into real business value.
Can you tell us about your background and what led you to pursue the M.S. in Applied Analytics at Columbia?
I’m an economist and spent several years in investment banking before deciding to move into data science. In banking, I saw how much decision-making depended on data, and I wanted to be the person building the models rather than just consuming the outputs. When I looked at programs, I wanted one that balanced business context with real technical depth, and that came with a strong support network of career advisors, faculty and alumni who could guide me through a career change. Columbia’s program offered all of that. Once I was in the program, I leaned heavily into the technical courses and looked for every opportunity to get hands-on experience, including a research position at Columbia’s Data Science Institute, which became my entry point into the field.
Walk us through your Capstone project. What problem did your team set out to solve, and what models, algorithms and tools did you use to develop your solution?
Our team partnered with one of the largest commercial banks in the U.S. to understand and predict employee attrition. Attrition is expensive for any large organization, and the bank wanted to identify which employees were at higher risk of leaving and what factors were driving it. We framed it as a binary classification problem and worked through a range of models, from logistic regression as an interpretable baseline to tree-based methods such as decision trees and random forests. We built the pipeline in both Python and R, covering data cleaning, exploratory analysis, feature engineering, model training and evaluation.
How have you been able to apply the skills you gained from your Capstone project in your career?
The Capstone was my first experience running a data science project end to end in a real-world setting: scoping the problem with a client, exploring the data, engineering features, building and evaluating models and presenting recommendations. That structure has carried directly into my work. Today, as head of AI and data, I build production ML systems such as credit risk models and multi-agent pipelines, and I still follow the same or similar process. The project also strengthened how I communicate with both technical teams and business stakeholders, which has become one of the most important parts of leading a data function.
What role did your Capstone teammates, lecturer and course associate play in shaping your approach to solving complex data problems?
My teammates were essential because we each brought different strengths. Some had more experience wrangling raw data, others excelled at modeling, and others at presenting insights. We learned a lot from each other’s approaches. Our lecturer and course associate pushed us to treat the project like a real engagement rather than an academic exercise: how to structure deliverables, manage client communication and present findings that lead to decisions. That shift in mindset, from “getting the right answer” to “delivering something the client can use,” shaped how I approach every data problem.
What was the biggest challenge you faced, and how did you overcome it?
Our biggest challenge was limited data, which is very common in real-world machine learning projects. We asked the client for more data, but compliance restrictions made that impossible. Instead of letting that stall us, we focused on getting the most out of what we had. We engineered new variables from existing fields, giving our models more signal to work with. It taught me early on that constraints are part of the job and that creative feature engineering can often matter more than having more data.
What lessons from your Capstone Project have stayed with you as you transitioned from investment banking to AI and data science? What was the biggest takeaway from the experience?
The biggest lesson I learned was the value of a clear project structure. The Capstone gave me a repeatable framework, which I’ve used on every data science project since, from my first roles after Columbia to today. Coming from banking, it also showed me that my business background was an asset because understanding what the client actually needs is half the work.
My biggest takeaway is that communication is everything. Our client contacts weren’t technical, and much of the feedback we received from our lecturer and course associate focused on how to present our findings so that someone without a data science background could understand and act on them. That situation comes up in every organization. You’ll always be working with people from different backgrounds, and a project’s success depends as much on how well you explain your work as on the model itself. Clear communication earns buy-in and turns a technical solution into real business value.
About the Program
Columbia University’s Master of Science in Applied Analytics prepares students with the practical data and leadership skills to succeed. The program combines in-depth knowledge of data analytics with the leadership, management, and communication principles and tactics necessary to impact decision-making across industries and organizational functions.
Learn more about the program here. The program is available full-time and part-time, online and on-campus.
