An exclusive interview with Harrison Enofe Obamwonyi, a Senior Data Scientist known for turning complex datasets into practical, high-impact solutions.
His work spans industries from finance to healthcare, and he is particularly passionate about bridging the gap between model accuracy and interpretability.
Harrison, thank you for joining us. Let’s start at the beginning. How did your data science journey begin?
Harrison: Interestingly, I didn’t set out to become a data scientist. I initially saw myself working in pure analytics, but overtime, I kept finding myself drawn to the problem-solving aspect of machine learning. My first real taste came when I worked on a fraud detection project for a financial services company. Watching the model flag suspicious activity in real time was both thrilling and eye-opening. I realized then that data science was where I wanted to be.
You’ve worked on a variety of projects. What would you say was a turning point in your career?
Harrison: One project that changed my perspective was a healthcare initiative to predict patient readmission risk. We had highly accurate deep learning models, but the hospital needed to understand the predictions to trust them. That was when I truly grasped the importance of interpretability. Accuracy on paper is meaningless if stakeholders can’t act on your model’s output. That experience taught me to consider explainability as a first-class citizen in model design.
That ties into our main theme today: balancing model accuracy with interpretability. How do you approach this trade-off?
Harrison: It starts with the problem context. In some areas, like recommendation engines for e-commerce, you can lean toward complex, high-accuracy models because the cost of a wrong prediction is relatively low. But in regulated sectors like finance, healthcare, and insurance, interpretability is non-negotiable. I often use a hybrid approach: complex models for raw predictive power, paired with interpretable “shadow models” or explainability tools like SHAP or LIME to translate the decisions into human terms. The goal isn’t always to sacrifice accuracy for interpretability, but to design a workflow where both coexist.
When you have to explain model results to non-technical stakeholders, what’s your process?
Harrison: Storytelling. I strip away the technical jargon and focus on the cause-and-effect narrative. For example, instead of saying, “The model assigned a probability of 0.82,” I might say, “The model found that customers who delayed payment twice in the past three months are much more likely to churn.” I also use visuals to bridge the gap.
You’ve managed and mentored junior data scientists. What’s your leadership style?
Harrison: I’m collaborative but principles-driven. I encourage experimentation, but I also push for clarity in thought. I want my team to understand why they’re building a model before writing a single line of code. I also create space for them to challenge me. Sometimes the best improvements come from the newest voices in the room.
Can you share a time when a high-accuracy model still failed in production?
Harrison: Yes, and it was a humbling lesson. We had built a credit risk model with outstanding validation accuracy. But in production, its performance dropped sharply. The reason? The training data didn’t account for a sudden shift in the economic climate. That experience reinforced my belief that a model’s real test is in production, not in a Jupyter notebook. It also taught me to build for adaptability, not just performance.
For aspiring data scientists, what advice would you give about balancing technical skills with business impact?
Harrison: Don’t get lost in the code. Remember that your model exists to solve a problem, not to be an academic exercise. Learn to speak the language of your stakeholders. Measure your success not by the complexity of your algorithm, but by the clarity and impact of your solution. And most importantly, stay curious, because technology changes, but curiosity is timeless.
What’s the biggest misconception people have about data science?
Harrison: That it’s all about machine learning. In reality, a huge chunk of the work is data cleaning, framing the problem correctly, and aligning with business goals. You can have the fanciest model in the world, but if the problem is poorly defined, it’s wasted effort.
Finally, what keeps you going in this fast-changing field?
Harrison: The fact that there’s always something new to learn. Data science sits at the edge of technology, mathematics, and human behaviour. Every project feels like solving a new puzzle, and I find that endlessly exciting.